- Email: [email protected]
Image analysis method to evaluate beak and head motion of broiler chickens during feeding
Computers and Electronics in Agriculture 114 (2015) 88–95
Contents lists available at ScienceDirect
Computers and Electronics in Agriculture journal homepage: www.elsevier.com/locate/compag
Image analysis method to evaluate beak and head motion of broiler chickens during feeding S. Abdanan Mehdizadeh a, D.P. Neves b, M. Tscharke c, I.A. Nääs b,⇑, T.M. Banhazi c a
Department of Biosystem Mechanics, College of Agricultural Engineering, Ramin Khuzestan, University of Agriculture and Natural Resources, Mollasani, Ahvaz, Khuzestan, Iran Agricultural Engineering College, State University of Campinas, Campinas, SP, Brazil c NCEA, University of Southern Queensland, Toowoomba Campus, QLD, Australia b
a r t i c l e
i n f o
Article history: Received 18 June 2014 Received in revised form 23 March 2015 Accepted 26 March 2015
Keywords: Biomechanics Eating behaviour High-speed camera Image analysis Jaw apparatus
a b s t r a c t While feeding broiler chickens may exhibit different biomechanical movements in relation to the physical properties of feed such as size, shape and hardness. Furthermore, the chicken’s anatomical features at various ages, genders and breeds in conjunction with feed type and feeder design parameters may also have an influence on biomechanical movement. To determine the significance of these parameters during feeding, kinematic measurements related to the biomechanical motions are required. However, determining this information manually from video by a human operator is tedious and prone to errors. The aim of this study was to develop a machine vision technique which visually identifies the relevant biomechanical variables attributed to broiler feeding behaviour from high-speed video footages. A total of 369 frames from three broiler chicks of 5 days old were manually measured and compared to the automatic measurement. For each bird six mandibulations (i.e. a cycle of opening and closing the beak) were manually selected, which were two different sequences of three consecutive mandibulations starting right after a feed grasping. The kinematics variables considered were: (i) head displacement (eye centre position; x- and y-axis); (ii) beak opening speed (given in mm ms1); (iii) beak closing speed (measured in mm ms1); (iv) beak opening acceleration (given in mm ms2); and (v) beak closing acceleration (given in mm ms2). Results indicated that the highest error for eye position detection was 1.05 mm for x-axis and 0.67 for the y-axis. The difference between manual and automatic (algorithm output) measurements for the beak gape was 0.22 ± 0.009 mm, in which the maximum difference was 0.76 mm. Regression analysis indicated that both measures are highly correlated (R2 = 99.2%). Statistical tests suggested that the primary probably causes of error are the speed and acceleration of the beak motion (i.e. blurred image), as well as the presence of feed particles in the first and second mandibulations right after the feed grasping (i.e. occluded beak tips by feed particles). The presented method calculated automatically the position of the eye centre (x- and y-axis) and both upper and lower beak tips distance in a high level of accuracy, but the model can be improved by using a camera with higher resolution, a higher acquisition rate, and infrared-reflective markers. The method has the potential to facilitate efficient and repeatable acquisition of biomechanical data of broiler chickens during feeding, and be used to benchmark the feed physical properties and its processing methods, likewise evolving knowledge for futures studies in feeders’ design. Ó 2015 Elsevier B.V. All rights reserved.
1. Introduction The poultry industry is considered as one of the most active meat producing industries requiring frequent increases in production to satisfy the worldwide demand for poultry meat. The largest broiler chicken producers by country are the United States, China and Brazil with the United States and Brazil contributing to ⇑ Corresponding author at: Agricultural Engineering College, State University of Campinas, Av. Marechal Rondon, 80 Campinas, SP, Brazil. Tel.: +55 19 35211039. E-mail address: [email protected] (I.A. Nääs). http://dx.doi.org/10.1016/j.compag.2015.03.017 0168-1699/Ó 2015 Elsevier B.V. All rights reserved.
two-thirds of poultry meat exports globally (FAO, 2012; USDA, 2012). Feed costs are the main drivers of profitability on commercial poultry farms, so minimizing feed wastage is desirable. Advances in poultry nutrition are mainly responsible for the exceptional growth rate responses of current domesticated species. In addition to nutritional value, the feed properties should also ensure it is palatable and easy to consume and digest by the birds. Past research have investigated the impact of both chemical and physical characteristics of the feed on animal responses, and the economic feasibility regarding feed processing methods (Thomas et al., 1998; Perez and Oliva-Teles, 2002), feed particle size (Nir et al.,
89
S. Abdanan Mehdizadeh et al. / Computers and Electronics in Agriculture 114 (2015) 88–95
1994a; Amerah et al., 2007), feed material form (Greenwood et al., 2004; Nir et al., 1994b; Skinner-Noble et al., 2005; Zang et al., 2009), and more recently the influence of feeders’ features on birds’ preferences (Neves et al., 2010; Roll et al., 2010). To date, the scientific research on broilers’ feeding performance and behaviour are based on productivity indexes, physiological responses and the impact of environmental conditions. However, little is known about the biomechanical responses of birds during feed consumption. Biomechanics can be described as a physical (mechanical) movement displayed or produced by living systems (Mclester and Pierre, 2008). Studies in biomechanics are of interest to various professional fields, such as zoology, medical, biomedical engineering, and kinesiology (study of human movement) (Hall, 1999). The detection of jaw movement has been previously examined by identifying features that characterize prey ingestion in captive animal species of mammals (Ropert-Coudert et al., 2004), birds (Van Der Heuvel and Berkhoudt, 1998), fishes (Wainwright, 1991), carnivorous and herbivorous feeding habits and either marine or terrestrial environments were investigated (Pennycuick, 1992). Furthermore, Stefen et al. (2011) used the digital bi-planar highspeed X-ray system to investigate jaw movements during incisor action and mastication in beaver. The mechanical process exhibited by domestic chicken while feeding is similar to that of pigeons and can be divided into different phases starting with the identification of a potential feed particle and ending with ingestion. Terminology for the various stages exhibited includes fixation, approach or pecking, grasping, withdrawal, stationing, transporting, collecting and swallowing (Zweers, 1982; Bermejo et al., 1989; Van Der Heuvel and Berkhoudt, 1998). These phases are described according to the position of the feed within the beak, and the motion and position of the particular parts of the chicken’s body during the eating process (head height, upper and lower beak displacement, sliding movement of the tongue and the eye blink). Furthermore, the feeding behaviour of these birds was described as stereotyped movement patterns, considering both duration and temporal organization of the variables involved the process (Zweers, 1982; Van Der Heuvel and Berkhoudt, 1998). The bird can adapt certain movement patterns depending on the type of feed, but such behaviours are subordinate to limitations of morphological structure and mechanical construction (Zweers, 1982). Most research on broiler feeding behaviour addresses the productivity indices and birds’ physiological responses. This paper aims to present a methodology to evaluate the biomechanical motion of broiler chickens while feeding through computational image analysis, considering the movement characteristics of the birds’ beak and head.
2.2. Experimental procedure The broiler chicks were placed individually in a rectangular glass box containing a feed tray. For birds 1 and 2, it was offered mash-type ration and for bird 3 grinded corn. A high-speed camera Ò (Mikrotron EoSens , Mikrotron GmbH, Unterschleißheim, Bavaria, Germany) with Nikon lens 50 mm/F 1.4 was set up with an acquisition rate of 300 fps (frames per second) and used to record the birds’ feeding behaviour. This acquisition rate resulted in 3 ms (ms) time delay between frames. With the aid of a tripod, the camera was positioned to fit the bird’s head and the feed tray in a perpendicular-lateral direction, in the field of view of the camera. A white paper sheet was placed in the background, to provide better contrast between the bird and its surrounds (Estrella and Masero, 2007). The high-speed camera was connected to a personal computer to store and manage the data. The trials occurred outdoor during a short period, so no artificial light source was used, since the natural daylight with small variations was capable of illuminating the scene.
2.3. Frames sequences classification and kinematic variables The analysed frames sequences were first selected and classified manually. For each one of the three birds, it was considered six mandibulations, which were two different sequences of three consecutive mandibulations starting right after the feed grasping (Fig. 1; Table 1). The mandibulation consisted of a cycle of opening and closing the beak, in which the beak in the first opening frame and the last closing frame was not necessary completely closed. The chosen mandibulations sequences were those which the head remained in a perpendicular lateral orientation to the camera view. The kinematic variables analysed were: (i) head displacement (eye centre position; x- and y-axis); (ii) beak opening speed (given in mm ms1); (iii) beak closing speed (given in mm ms1); (iv) beak opening acceleration (given in mm ms2); and (v) beak closing acceleration (given in mm ms2). Displacement is defined as the change in position (expressed in mm). Speed is the time derivative of displacement, which is the rate of change of displacement regarding time (given in mm ms1). Acceleration is the time derivative of speed, which is the change of velocity with respect to the time (expressed in mm ms2) (Robertson and Caldwell, 2004). The feed tray diameter (47 mm) was used for calibration.
2. Material and methods 2.1. Birds and facilities The experiment was carried out at the Construction and Environmental Laboratory in the Agricultural Engineering College (FEAGRI), the State University of Campinas (UNICAMP), Campinas-SP, Brazil, in July of 2011. For the present study, three male broiler chicks (CobbÒ strain) of 5 days old were randomly chosen from a climate chamber of another experiment, which standard broiler housing was adopted (Cobb-Vantress, 2009). At this stage of the present study, the use of three specimens seems to be enough to present the proposed methodology, in accordance with previous studies approaching kinematic analysis of fishes (Tran et al., 2010; Wassenbergh and Rechter, 2011), turtles (Natchev et al., 2009), lizards (Schaerlaeken et al., 2007), and birds (Gussekloo and Bout, 2005; Dawson et al., 2011).
Fig. 1. Schematic representation of frames sequences classification for one bird.
Table 1 Number of analysed frames by birds, mandibulation order, and motion type. Bird
Frames
Mandibulation
Frames
Motion type
Frames
1 2 3 Total
130 141 98 369
1st 2nd 3rd –
68 143 158 369
Opening Maximum beak gape Closing –
224 18 127 369
90
S. Abdanan Mehdizadeh et al. / Computers and Electronics in Agriculture 114 (2015) 88–95
particle removal to prevent a mistake during the beak tip detection when a particle occludes the beak tip. Unlike the methods utilized by Horster et al. (2002), in which involved physic markers placed on the birds’ body, the study involved calculation of reliable reference point on the chicken so that body features relative to this point could be identified. The methodology in this study required no bird training or unusual management, and the machine vision algorithm was able to determine the eyeball and tip of the upper and lower beak automatically. Step 1: Eye detection
Fig. 2. Schematic representation of the manual measurements (a) and an example of the beak tips occluded by feed particles (b).
A thresholding process was applied to the video frame based on the colour difference between the eyeball and the body of the chicken, for calculating the eye position. All artefacts except the eyeball were then removed from the resulting image (Fig. 4a). After segmentation, the coordinate of the centre area of the eye was measured from the x- and y-axis of the image (Fig. 4b; red1 point refers to the centre of eyeball area). Step 2: Removal of redundant segments After finding the eye position, a region of interest (corresponding to a rectangle of 610 740 pixels) was defined in the centre of the eye for all video frames. This area was then extracted to remove other parts of the body in order to avoid redundant information and to enhancing the beak tip detection Step 3: Beak tip detection In order to find the beak tips, Otsu’s threshold was applied to the image in order to convert it into binary format (Otsu, 1979). The algorithm then started a search for the beak tips at the bottom left of the binary image (Fig. 5; arrow directed from the left to the right). If the beak was opened, the first non-zero pixel was classified as part the lower beak tip. If the beak was closed, the first non-zero pixel was identified as part of the upper beak. Step 4: Removing feed particle
Fig. 3. The four steps used for image analysis to measure eye position, beak gape, beak speed and beak acceleration.
2.4. Manual measurements The manual measurements were carried out by a human operator through ImageJ software (National Institutes of Health, Bethesda, Maryland, USA). In this procedure (Fig. 2a), the ‘Polygon selections’ tool were used to define the border of the birds’ eye and the coordinates of its centroid were the point of interest (x- and y-axis). The beak gape was measured from the Euclidian distance (‘Straight line’ tool) between the upper and lower beak tips. At some frames, the feed particles occluded the beak tips (Fig. 2b), so it was necessary to advance and return the frames to intuitively estimate the beak tip position. 2.5. Automatic measurements The automatic image analysis procedure comprised of four steps (Fig. 3) using MatlabÒ software (MathWorks, Inc., Natick, Massachusetts, USA). These four steps were eye detection as a reference point to determine the position of bird’s head; head extraction to remove redundant background information during analysis, beak tips detection to analyse the biomechanical behaviour (maximum beak-gape, speed and acceleration) and feed
The feed particles in some frames occluded the beak tip hindering its precise detection (illustrated in Fig. 6a and c). In order to identify and remove the feed particle from the image the following algorithm was applied:
Inewx;y
0 r x;y j160 < r x;y 255 ¼ 1 r x;y j0 r x;y 160
ð1Þ
Here r is the red channel of the unsigned 8 bit image, and x and y denotes the Cartesian co-ordinates of the old image r and the new image with the feed particle removed Inewx;y (Fig. 6d). After this process, a region of interest was defined around the beak (250 210 pixels) so that the maximum beak gape could be defined. First, the boundary of the beak was found within the area so that the two beak tips corresponding to the endpoints could be identified (Fig. 7a). Then, the Euclidian distance between the two beak tips (blue line) was measured. When the beak was closed, only one tip was detected, and the Euclidian distance was zero (Fig. 7b). The Euclidian distance was also used to measure the beak tips distance to calculate both motion speed and acceleration (Fig. 8). 1 For interpretation of colour in Figs. 4 and 7, the reader is referred to the web version of this article.
S. Abdanan Mehdizadeh et al. / Computers and Electronics in Agriculture 114 (2015) 88–95
91
Fig. 4. The eyeball segmentation (a), and the detection of its centre in x and y coordinates (b).
2.6. Statistical analysis
Upper beak tip Lower beak tip
Fig. 5. Binarized image of the chicken’s head and starting point of search algorithm to find the beak tips location.
Manual and automatic measurements were compared by calculating the difference between them. The unit considered in this study was the number of frames analysed (N = 369). General descriptive analysis was used. Regression analysis was applied to the difference between manual and automatic beak gape calculations. Then, ANOVA (Tukey test at 95% of confidence level) was applied to verify the differences between different birds (1–3), mandibulation order (1st, 2nd, and 3rd), and motion type (opening, maximum beak gape, and closing). The manual measurements were considered the ‘gold standards’, so it was used to calculate the speed and acceleration of the results. Additionally, frames regarded to the maximum beak gape were considered as the last Ò frame of the beak opening motion. Minitab 17 software (Minitab
Fig. 6. Schematic representation of the (a) original frame showing the feed particle occluding the lower beak tip; (b) binarized image with the feed represented in blue; (c) detail of the feed particle occluding the beak; and extracted feed particle from the image (d). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Fig. 7. Upside down picture showing the calculation of beak tips when it is opened (a) and closed (b).
92
S. Abdanan Mehdizadeh et al. / Computers and Electronics in Agriculture 114 (2015) 88–95
Opening sequence
1
2
3
4
5
2
3
4
5
6
Closing sequence
1
Fig. 8. Representation of beak opening and closing sequences.
Inc., Pennsylvania, USA) was used to carry out the statistical analysis.
Table 2 Differences between manual and automatic measurements (mm) of beak gape and head position (x- and y-axis).
3. Results
Variable
Method
Mean ± SE
P-value
Error (mean ± SE)
Maximum error
3.1. General differences between manual and automatic methods
Beak gape
Manual Automatic Manual Automatic Manual Automatic
2.43 ± 0.11 2.22 ± 0.11 31.95 ± 0.19 31.64 ± 0.19 24.18 ± 0.25 24.00 ± 0.24
0.162
0.22 ± 0.01
0.76
0.247
0.38 ± 0.01
1.05
0.605
0.24 ± 0.01
0.67
3.2. Beak gape, speed, and acceleration To calculate beak speed and acceleration, data from manual method (i.e. ‘gold standards’) were considered, besides the frames regarding to the maximum beak gape were considered as the last frame of opening motion for these calculations. In order to identify possible causes of the highest errors, ANOVA (Tukey test at 95% of confidence level) was applied to verify the influence of birds, mandibulation order, and motion type, likewise beak speed and acceleration. Results (Table 3) indicated that the speed of the beak was significantly (P-value < 0.005) higher for the first, second, and third mandibulations, respectively. The acceleration was considerably greater in the first mandibulation than the third one. Moreover, the speed of the beak was significantly higher during the closing motion (0.27 mm ms1) than opening (0.14 mm ms1). About the error between both methods, results indicated significantly (P-value < 0.005) differences between birds and mandibulation order, but no differences in motion type. The higher errors were found in the first and second mandibulations for speed, and first mandibulation for acceleration, in which the maximum error were 0.36 mm ms1 for speed and 0.08 mm ms2 for acceleration. 4. Discussion The feeding behaviour of animals can be divided into appetitive phase, corresponding to the demand for feed and consummatory act, i.e. the real feed intake. The appetitive phase of birds is characterized by the foraging behaviour, which is the time to explore the environment searching for food. Moreover, it was
x-axis y-axis
Tukey test at 95% of confidence level; SE = standard error.
10
Automatic Beak gape
Overall, the error between manual and automatic (algorithm output) measurements for the birds’ beak gape was 0.22 ± 0.01 mm (mean ± standard error), in which the maximum difference was 0.76 mm. For head position, the maximum error for x- and y-axis was 1.05 mm and 0.67 mm, respectively. ANOVA (Table 2) and regression analysis (Fig. 9) indicated no significant differences (P-value > 0.005) between both methods. Thus, they were highly correlated.
8 6 4 2 0 0
2
4
6
8
10
Manual Beak gape Fig. 9. Regression analysis between manual and automatic measurements for the whole dataset of the birds beak gape (R2 = 99.2%).
Table 3 The speed and acceleration of the birds’ beak motion by different birds, mandibulation order, and motion type. Variable
Factor
Speed (mm ms1)
Acceleration (mm ms2)
Bird
1 2 3
0.17 ± 0.02 a 0.21 ± 0.02 a 0.15 ± 0.01 a
0.03 ± 0.00 a 0.04 ± 0.00 a 0.03 ± 0.00 a
Mandibulation
1st 2nd 3rd
0.36 ± 0.04 a 0.19 ± 0.02 b 0.10 ± 0.01 c
0.07 ± 0.00 a 0.03 ± 0.00 b 0.02 ± 0.00 b
Motion type
Opening Closing
0.14 ± 0.01 a 0.27 ± 0.02 b
0.03 ± 0.00 a 0.04 ± 0.00 a
Tukey test; means that do not share a letter are significantly different (P-value < 0.005).
93
S. Abdanan Mehdizadeh et al. / Computers and Electronics in Agriculture 114 (2015) 88–95
reported that two-thirds of young bird pecks do not result in the catchment of a feed particle (Yo et al., 1997). Therefore, biomechanical assessments can be an effective method to study the Table 4 The error (mean ± standard error) between manual and automatic measurements for the beak gape (mm), beak speed (mm ms1) and beak acceleration (mm ms2) by bird, mandibulation order, and motion type. Variable
Factor
Beak gape
Speed
Acceleration
Bird
1 2 3
0.25 ± 0.01 a 0.24 ± 0.01 a 0.16 ± 0.01 b
0.03 ± 0.00 a 0.04 ± 0.00 a 0.03 ± 0.00 a
0.020 ± 0.002 a 0.017 ± 0.001 a 0.019 ± 0.002 a
Mandibulation order
1st
0.24 ± 0.01 ab
0.07 ± 0.01 a
0.023 ± 0.002 a
2nd 3rd
0.25 ± 0.01 a 0.19 ± 0.01 b
0.03 ± 0.00 b 0.02 ± 0.00 b
0.019 ± 0.002 ab 0.017 ± 0.001 b
Opening Maximum beak gape Closing
0.23 ± 0.01 a 0.19 ± 0.03 a
0.03 ± 0.00 a –
0.019 ± 0.02 a –
0.21 ± 0.01 a
0.04 ± 0.00 a
0.019 ± 0.001 a
Motion type
Tukey test; means that do not share a letter are significantly different (P-value < 0.005).
feeding behaviour of broiler chickens in order to improve the feeding efficiency while respecting their natural behaviour. The feeding assessments of animals may be related to bite events, and/or visits to feeders (Slater, 1974; Berdoy, 1993; Nielsen, 1999), in which these could be considered a unit of the study (Nielsen et al., 1995). The aim of the present study was to validate the automatic method (algorithm output) considering the calculations of kinematic variables of the birds in each mandibulation. Thus, the unit found here was the number of frames. The errors between manual and automatic methods of head motion (i.e. tracking the centre of the eye area) can be considered low, which the maximum errors were 1.05 mm and 0.67 mm for x- and y-axis, respectively. Thus, for this particular feeding phase (mandibulation), the algorithm presented a high accuracy. However, it is possible that improvements are needed to apply in other feeding phases, which the eye blink will potentially lead to misconstrued calculations. About the beak gape measurements, the errors between manual and automatic methods were significantly different for distinct individuals. However, little can be concluded from this observation (inter-individual differences), since a few sampling of individuals were observed. On the other hand, the highest errors of automatic
Motion type
Mandibulation order 0,28
Beak gape error (mm)
0,28
a
0,26 0,24
0,24
0,22
0,22
0,20
0,20
0,18
0,18
0,16
0,16
0,14
0,14
0,12
0,12 1st
2nd
3rd
Beak speed (mm ms-1)
0,42
Closing
c
0,38
0,34 0,30
0,26
0,26
0,22
0,22
0,18
0,18
0,14
0,14
0,10
0,10 3rd
d
Closing
Opening
0,09
0,09
Beak acceleration (mm ms-2)
Opening
0,38
0,30
2nd
Maximum beak gape
0,42
0,34
1st
b
0,26
e
0,08
f
0,08
0,07
0,07
0,06
0,06
0,05
0,05
0,04
0,04
0,03
0,03
0,02
0,02 0,01
0,01 1st
2nd
3rd
Closing
Opening
Fig. 10. Interval plots of the beak gape error (a and b), beak speed (c and d), and beak acceleration (e and f) by mandibulation order and motion type; refers to Tables 3 and 4.
94
S. Abdanan Mehdizadeh et al. / Computers and Electronics in Agriculture 114 (2015) 88–95
measurements were found in the first and second mandibulations right after the feed grasping. This could be explained by the amount of feed particles presented in the scene that occludes the upper and/or lower beak tips. The methodology presented in this paper overcomes this problem. Another explanation, however, might be the beak speed and acceleration. The higher speed of the beak creates a blurred image on the frame, which might lead the algorithm to errors due to difficult to detect the beak tip precisely. The higher beak speeds were observed at the first, second, and third mandibulations, respectively (Fig. 10c), likewise a larger beak acceleration in the first mandibulation (Fig. 10e). Interestingly, no significant difference was found in the opening and closing beak motions, even though they presented different speeds (Fig. 10d). This could be explained by the higher speeds detected in the 1st mandibulation compared with the speeds of the closing motion (Fig. 10c and d), which might explain the higher errors in the higher beak speeds. In this sense, it can be highlighted the importance of an adequate camera setup (resolution and acquisition rate) in order to avoid blurred images of the object of interest, and balance it with data amount to be storage and analysed. During broilers growth, it is important that feeders and drinkers be properly arranged and well managed. Furthermore, several studies indicate that some design features, such as size, location, geometry, spacing and angle of feeders can affect the behaviour of animals (Buskirk et al., 2003; Wolter et al., 2009). Developing a precise and non-invasive method for assessing the motion in relation to the feeders’ usage remains a challenge (Lu and Chang, 2012). In this sense, some advantages are seen when using high speed cameras and computational image analysis for motion assessments, especially because its relatively small cost, versatility in analysis, commercial availability of the hardware, and possibility to upgrade the system as needed (Sakatani and Isa, 2004). In this study, it was observed that sometimes the chicks rotate their heads while feeding. This is an instinct behaviour in order to kill the prey, and previous study has described this kind of behaviour in pigeons (Horster et al., 2002). Thus, in a 2-diomensional analysis a calibration line on the birds’ head is required to correct the measurements. This method was not applied in the present study, because it was not considered mandibulation sequences when the birds showed this kind of behaviour. The algorithm for automatic eye detection works based on colour difference, so the background should not be set with a similar colour. To compensate different colours of feed particles, colour threshold was modified based on the colour difference of beak and feed. This modification can be acquired manually or automatically to get best discrimination results between beak and feed. Furthermore, a camera with higher acquisition rate (number of frames per second) and higher resolution combined with improvements in image capturing techniques has the potential to enhance the accuracy of the model. A possible solution to improve computational image analysis is the use of infrared-reflective motion capture markers instead just physical markers, which might enhance its identification (Heysea et al., 2014; Paxton et al., 2013). Birds select different sizes of feed particles on the first week of life. The format and structure of the beak determine the size and type of feed to be ingested, and thus the granulometry of the particle is of high importance for the regulation of the consumption (Nir et al., 1990; Nir et al., 1994a; Addo et al., 2012). The contact perception contributes to the identification of the feed and broiler chickens can discriminate different types of diets by associating the feed physical features with nutritional content (Emmans and Kyriazakis, 2001). Thus, the study of kinematics during birds’ feeding behaviour combined with physiological responses should aid the decision regarding the optimum feed type and granulometry for different rearing environments. This should be considered in relation to the growth phase, strain, the diet composition, and
the available technology in the feed plant. In this study, the algorithm presented accurate results for chicks’ mandibulation assessments (1.05 mm was the highest error), but improvements in recording techniques should help its adaptation for further kinematics analysis; e.g. other feeding phases and others species. 5. Conclusions The presented method calculates automatically the position of the eye centre (x- and y-axis) in order to track the head displacement of broiler chicks, and also both upper and lower beak tips distance. The algorithm presented an error less than 1.05 mm (the highest error) and can be improved by using a camera with higher resolution, a higher acquisition rate, and infrared-reflective markers. The highest errors were found in the first and second mandibulations right after the feed grasping. This is explained by the presence of feed particles that occluded the beak tips and the higher beak speed and acceleration that created a blurred image. A better understanding of the mechanical process of the birds’ jaw apparatus during feeding might be an efficient method for determining the relationship between different types of feeds and the biomechanical patterns exhibited by the birds. Moreover, it also should consider the anatomical variations between different strains, gender, growth phase, and also the influence of feeder design. The high-speed camera combined with techniques of computational image analysis is a useful technology to aid such assessments. Acknowledgement The authors thank CNPq – National Council for Scientific and Technological Development for supporting this research. References Addo, A., Bart-Plange, A., Akowuah, J.O., 2012. Particle size evaluation of feed ingredient produced in the Kumasi Metropolis, Ghana. ARPN J. Agric. Biol. Sci. 7, 177–181. Amerah, A.M., Ravindran, V., Lentle, R.G., Thomas, D.G., 2007. Feed particle size: implications on the digestion and performance of poultry. World Poult. Sci J 63, 439–451. Berdoy, M., 1993. Defining bouts of behaviour: a three process model. Anim. Behav. 46, 387–396. Bermejo, R., Allan, R.W., Houben, D., Deich, J.D., Zeigler, H.P., 1989. Prehension in the pigeon I: descriptive analysis. Exp. Brain Res. 75, 569–576. Buskirk, D.D., Zanella, A.J., Harrigan, T.M., Van Lente, J.L., Gnagey, L.M., Kaercher, M.J., 2003. Large round bale design affects hay utilization and beef cow behavior. J. Anim. Sci. 81, 109–115. Cobb-Vantress. 2009. Manual de manejo de frangos de corte (Broiler Management Guide). L-1020-02 PT. Dawson, M.M., Metzger, K.A., Baier, D.B., Brainerd, E.L., 2011. Kinematics of the quadrate bone during feeding in mallard ducks. J. Exp. Biol. 214, 2036–2046. Emmans, G., Kyriazakis, I., 2001. Consequences of genetic change in farm animals on food intake and feeding behavior. Proc. Nutr. Soc. 60, 115–125. Estrella, S.M., Masero, J.A., 2007. The use of distal rhyncokinesis by birds feeding in water. J. Exp. Biol. 210, 3757–3762. FAO. 2012. Food Outlook, Rome: Trade and Market Division of FAO. Greenwood, M.W., Cramer, K.R., Clark, P.M., Behnke, K.C., Beyer, R.S., 2004. Influence of feed form on dietary lysine and energy intake and utilization of broilers from 14 to 30 days of age. Int. J. Poult. Sci. 3, 189–194. Gussekloo, S.W.S., Bout, R.G., 2005. Cranial kinesis in palaeognathous birds. J. Exp. Biol. 208, 3409–3419. Hall, S.J., 1999. Basic Biomechanics, third ed. Mcgraw-Hill, Singapore. Heysea, T.J., El-Zayata, B.F., Corteb, R. De, Scheys, L., Chevalierd, Y., FuchsWinkelmanna, S., Labeye, L., 2014. Biomechanics of medial unicondylar in combination with patellofemoral knee Arthroplasty. Knee 21 (S1), 53–59. Horster, W., Krumm, E., Mohr, C., Delius, D.J., 2002. Conditioning the pecking motions of pigeons. Behav. Process 58, 27–43. Lu, T., Chang, C., 2012. Biomechanics of human movement and its clinical applications. Kaohsiung J. Med. Sci. 28, S13–S25. Mclester, J., Pierre, P.S., 2008. Applied Biomechanics: Concepts and Connections. Thomson Wadsworth, Belmont. Natchev, N., Heiss, E., Lemell, P., Stratev, D., Weisgram, J., 2009. Analysis of prey capture and food transport kinematics in two Asian box turtles, Cuora
S. Abdanan Mehdizadeh et al. / Computers and Electronics in Agriculture 114 (2015) 88–95 amboinensis and Cuora flavomarginata (Chelonia, Geoemydidae), with emphasis on terrestrial feeding patterns. Zoology 112, 113–127. Neves, D.P., Nääs, I.A., Vercellino, R.A., Moura, D.J., 2010. Do broilers prefer to eat from a certain type of feeder? Braz. J. Poult. Sci. 12, 179–187. Nielsen, B.L., 1999. On the interpretation of feeding behaviour measures and the use of feeding rate as an indicator of social constraint. Appl. Anim. Behav. Sci. 63, 79–91. Nielsen, B.L., Lawrence, A.B., Whittemore, C.T., 1995. Effects of single-space feeder design of feeding behaviour and performance of growing pigs. Anim. Sci. 61, 575–579. Nir, I., Melcion, J.P., Picard, M., 1990. Effect of particle size of sorghum grains on feed intake and performance of young broilers. Poult. Sci. 69, 2177–2184. Nir, I., Hillel, R., Shefet, G., Nitsan, Z., 1994a. Effect of grain particle size on performance. 2. Grain texture interactions. Poult. Sci. 73, 781–791. Nir, I., Shefet, G., Aaroni, Y., 1994b. Effect of particle size on performance. 1. Corn. Poult. Sci. 73, 45–49. Otsu, N., 1979. A threshold selection method from grey level histograms. IEEE Trans. Syst. Man Cyb. 9, 62–66. Paxton, H., Daley, M.A., Corr, S.A., Hutchinson, J.R., 2013. The gait dynamics of the modern broiler chicken: a cautionary tale of selective breeding. J. Exp. Biol. 216, 3237–3248. Pennycuick, C.J., 1992. Newton Rules Biology: A Physical Approach to Biological Problems. Oxford University Press, New York. Perez, H., Oliva-Teles, A., 2002. Utilization of raw and gelatinized starch by European sea bass (Dicentrarchuslabrax) juveniles. Aquaculture 205, 287– 299. Robertson, D.G.E., Caldwell, G.E., 2004. Planar kinematics. In: Robertson, D.G.E., Caldwell, G.E., Hamill, J., Kamen, G., Whittlesey, S.N. (Eds.), Research Methods in Biomechanics. Human Kinetics, Champaign, IL, USA, pp. 9–52. Roll, V.F.B., Dai Prá, M.A., Roll, A.A.P., Xavier, E.G., Rossi, P., Anciuti, M.A., Rutz, F., 2010. Effect of tubular feeder height on ingestive behavior of broiler. Arch. Zootec. 59, 115–122. Ropert-Coudert, Y., Kato, A., Liebsch, N., Wilson, R.P., Muller, G., Baubet, E., 2004. Monitoring jaw movements: a cue to feeding activity. Game Wildlife Sci. 20, 1–19.
95
Sakatani, T., Isa, T., 2004. PC-based high-speed video-oculography for measuring rapid eye movements in mice. Neurosci. Res. 49, 123–131. Schaerlaeken, V., Meyers, J.J., Herrel, A., 2007. Modulation of prey capture kinematics and the role of lingual sensory feedback in the lizard Pogona vitticeps. Zoology 110, 127–138. Skinner-Noble, D.O., Mckinney, L.J., Teeter, R.G., 2005. Predicting effective caloric value of non-nutritive factors: III. Feed form affects broiler performance by modifying behavior patterns. Poult. Sci. 84, 403–411. Slater, P.J.B., 1974. The temporal pattern of feeding in the Zebra finch. Anim. Behav. 22, 506–515. Stefen, C., Ibeb, P., Fischer, M.S., 2011. Bi-planar X-ray motion analysis of the lower jaw movement during incisor interaction and mastication in the beaver (Castor fiber L. 1758). Mamm. Biol. 76, 534–539. Thomas, M., Van Vliet, T., van der Poel, A.F.B., 1998. Physical quality of pelleted animal feed 3. Contribution of feedstuff components. Anim. Feed Sci. Technol. 70, 59–78. Tran, H.Q., Mehta, R.S., Wainwright, P.C., 2010. Effects of ram speed on prey capture kinematics of juvenile Indo-Pacific tarpon, Megalops cyprinoides. Zoology 113, 75–84. USDA. 2012. Livestock and Poultry: World Markets and Trade, s.l.: USDA. Van Der Heuvel, W.F., Berkhoudt, H., 1998. Pecking in the chicken (Gallus gallus domesticus) motion analysis and stereotypy. Neth. J. Zool. 48, 273–303. Wainwright, P.C., 1991. Ecomorphology: experimental functional anatomy for ecological problems. Am. Zool. 31, 680–693. Wassenbergh, S.V., Rechter, D.D., 2011. Piscivorous cyprinid fish modulates suction feeding kinematics to capture elusive prey. Zoology 114, 46–52. Wolter, B.F., Ellis, M., Curtis, S.E., Parr, E.N., Webel, D.M., 2009. Feeder location did not affect performance of welling pigs in large groups. J. Anim. Sci. 78, 2784– 2789. Yo, T., Vilariño, M., Faure, J.M., Picard, M., 1997. Feed pecking in young chickens: new techniques of evaluation. Physiol. Behav. 61, 803–810. Zang, J.J., Piao, X.S., Huang, D.S., Wang, J.J., Ma, X., Ma, Y.X., 2009. Effects of feed particle size and feed form on growth performance, nutrient metabolizability and intestinal morphology in broiler chickens. Asian Austral J. Anim. Sci. 22, 107–112. Zweers, G.A., 1982. Pecking of the pigeon (Columba livia L.). Behaviour 81, 173–230.
Contents lists available at ScienceDirect
Computers and Electronics in Agriculture journal homepage: www.elsevier.com/locate/compag
Image analysis method to evaluate beak and head motion of broiler chickens during feeding S. Abdanan Mehdizadeh a, D.P. Neves b, M. Tscharke c, I.A. Nääs b,⇑, T.M. Banhazi c a
Department of Biosystem Mechanics, College of Agricultural Engineering, Ramin Khuzestan, University of Agriculture and Natural Resources, Mollasani, Ahvaz, Khuzestan, Iran Agricultural Engineering College, State University of Campinas, Campinas, SP, Brazil c NCEA, University of Southern Queensland, Toowoomba Campus, QLD, Australia b
a r t i c l e
i n f o
Article history: Received 18 June 2014 Received in revised form 23 March 2015 Accepted 26 March 2015
Keywords: Biomechanics Eating behaviour High-speed camera Image analysis Jaw apparatus
a b s t r a c t While feeding broiler chickens may exhibit different biomechanical movements in relation to the physical properties of feed such as size, shape and hardness. Furthermore, the chicken’s anatomical features at various ages, genders and breeds in conjunction with feed type and feeder design parameters may also have an influence on biomechanical movement. To determine the significance of these parameters during feeding, kinematic measurements related to the biomechanical motions are required. However, determining this information manually from video by a human operator is tedious and prone to errors. The aim of this study was to develop a machine vision technique which visually identifies the relevant biomechanical variables attributed to broiler feeding behaviour from high-speed video footages. A total of 369 frames from three broiler chicks of 5 days old were manually measured and compared to the automatic measurement. For each bird six mandibulations (i.e. a cycle of opening and closing the beak) were manually selected, which were two different sequences of three consecutive mandibulations starting right after a feed grasping. The kinematics variables considered were: (i) head displacement (eye centre position; x- and y-axis); (ii) beak opening speed (given in mm ms1); (iii) beak closing speed (measured in mm ms1); (iv) beak opening acceleration (given in mm ms2); and (v) beak closing acceleration (given in mm ms2). Results indicated that the highest error for eye position detection was 1.05 mm for x-axis and 0.67 for the y-axis. The difference between manual and automatic (algorithm output) measurements for the beak gape was 0.22 ± 0.009 mm, in which the maximum difference was 0.76 mm. Regression analysis indicated that both measures are highly correlated (R2 = 99.2%). Statistical tests suggested that the primary probably causes of error are the speed and acceleration of the beak motion (i.e. blurred image), as well as the presence of feed particles in the first and second mandibulations right after the feed grasping (i.e. occluded beak tips by feed particles). The presented method calculated automatically the position of the eye centre (x- and y-axis) and both upper and lower beak tips distance in a high level of accuracy, but the model can be improved by using a camera with higher resolution, a higher acquisition rate, and infrared-reflective markers. The method has the potential to facilitate efficient and repeatable acquisition of biomechanical data of broiler chickens during feeding, and be used to benchmark the feed physical properties and its processing methods, likewise evolving knowledge for futures studies in feeders’ design. Ó 2015 Elsevier B.V. All rights reserved.
1. Introduction The poultry industry is considered as one of the most active meat producing industries requiring frequent increases in production to satisfy the worldwide demand for poultry meat. The largest broiler chicken producers by country are the United States, China and Brazil with the United States and Brazil contributing to ⇑ Corresponding author at: Agricultural Engineering College, State University of Campinas, Av. Marechal Rondon, 80 Campinas, SP, Brazil. Tel.: +55 19 35211039. E-mail address: [email protected] (I.A. Nääs). http://dx.doi.org/10.1016/j.compag.2015.03.017 0168-1699/Ó 2015 Elsevier B.V. All rights reserved.
two-thirds of poultry meat exports globally (FAO, 2012; USDA, 2012). Feed costs are the main drivers of profitability on commercial poultry farms, so minimizing feed wastage is desirable. Advances in poultry nutrition are mainly responsible for the exceptional growth rate responses of current domesticated species. In addition to nutritional value, the feed properties should also ensure it is palatable and easy to consume and digest by the birds. Past research have investigated the impact of both chemical and physical characteristics of the feed on animal responses, and the economic feasibility regarding feed processing methods (Thomas et al., 1998; Perez and Oliva-Teles, 2002), feed particle size (Nir et al.,
89
S. Abdanan Mehdizadeh et al. / Computers and Electronics in Agriculture 114 (2015) 88–95
1994a; Amerah et al., 2007), feed material form (Greenwood et al., 2004; Nir et al., 1994b; Skinner-Noble et al., 2005; Zang et al., 2009), and more recently the influence of feeders’ features on birds’ preferences (Neves et al., 2010; Roll et al., 2010). To date, the scientific research on broilers’ feeding performance and behaviour are based on productivity indexes, physiological responses and the impact of environmental conditions. However, little is known about the biomechanical responses of birds during feed consumption. Biomechanics can be described as a physical (mechanical) movement displayed or produced by living systems (Mclester and Pierre, 2008). Studies in biomechanics are of interest to various professional fields, such as zoology, medical, biomedical engineering, and kinesiology (study of human movement) (Hall, 1999). The detection of jaw movement has been previously examined by identifying features that characterize prey ingestion in captive animal species of mammals (Ropert-Coudert et al., 2004), birds (Van Der Heuvel and Berkhoudt, 1998), fishes (Wainwright, 1991), carnivorous and herbivorous feeding habits and either marine or terrestrial environments were investigated (Pennycuick, 1992). Furthermore, Stefen et al. (2011) used the digital bi-planar highspeed X-ray system to investigate jaw movements during incisor action and mastication in beaver. The mechanical process exhibited by domestic chicken while feeding is similar to that of pigeons and can be divided into different phases starting with the identification of a potential feed particle and ending with ingestion. Terminology for the various stages exhibited includes fixation, approach or pecking, grasping, withdrawal, stationing, transporting, collecting and swallowing (Zweers, 1982; Bermejo et al., 1989; Van Der Heuvel and Berkhoudt, 1998). These phases are described according to the position of the feed within the beak, and the motion and position of the particular parts of the chicken’s body during the eating process (head height, upper and lower beak displacement, sliding movement of the tongue and the eye blink). Furthermore, the feeding behaviour of these birds was described as stereotyped movement patterns, considering both duration and temporal organization of the variables involved the process (Zweers, 1982; Van Der Heuvel and Berkhoudt, 1998). The bird can adapt certain movement patterns depending on the type of feed, but such behaviours are subordinate to limitations of morphological structure and mechanical construction (Zweers, 1982). Most research on broiler feeding behaviour addresses the productivity indices and birds’ physiological responses. This paper aims to present a methodology to evaluate the biomechanical motion of broiler chickens while feeding through computational image analysis, considering the movement characteristics of the birds’ beak and head.
2.2. Experimental procedure The broiler chicks were placed individually in a rectangular glass box containing a feed tray. For birds 1 and 2, it was offered mash-type ration and for bird 3 grinded corn. A high-speed camera Ò (Mikrotron EoSens , Mikrotron GmbH, Unterschleißheim, Bavaria, Germany) with Nikon lens 50 mm/F 1.4 was set up with an acquisition rate of 300 fps (frames per second) and used to record the birds’ feeding behaviour. This acquisition rate resulted in 3 ms (ms) time delay between frames. With the aid of a tripod, the camera was positioned to fit the bird’s head and the feed tray in a perpendicular-lateral direction, in the field of view of the camera. A white paper sheet was placed in the background, to provide better contrast between the bird and its surrounds (Estrella and Masero, 2007). The high-speed camera was connected to a personal computer to store and manage the data. The trials occurred outdoor during a short period, so no artificial light source was used, since the natural daylight with small variations was capable of illuminating the scene.
2.3. Frames sequences classification and kinematic variables The analysed frames sequences were first selected and classified manually. For each one of the three birds, it was considered six mandibulations, which were two different sequences of three consecutive mandibulations starting right after the feed grasping (Fig. 1; Table 1). The mandibulation consisted of a cycle of opening and closing the beak, in which the beak in the first opening frame and the last closing frame was not necessary completely closed. The chosen mandibulations sequences were those which the head remained in a perpendicular lateral orientation to the camera view. The kinematic variables analysed were: (i) head displacement (eye centre position; x- and y-axis); (ii) beak opening speed (given in mm ms1); (iii) beak closing speed (given in mm ms1); (iv) beak opening acceleration (given in mm ms2); and (v) beak closing acceleration (given in mm ms2). Displacement is defined as the change in position (expressed in mm). Speed is the time derivative of displacement, which is the rate of change of displacement regarding time (given in mm ms1). Acceleration is the time derivative of speed, which is the change of velocity with respect to the time (expressed in mm ms2) (Robertson and Caldwell, 2004). The feed tray diameter (47 mm) was used for calibration.
2. Material and methods 2.1. Birds and facilities The experiment was carried out at the Construction and Environmental Laboratory in the Agricultural Engineering College (FEAGRI), the State University of Campinas (UNICAMP), Campinas-SP, Brazil, in July of 2011. For the present study, three male broiler chicks (CobbÒ strain) of 5 days old were randomly chosen from a climate chamber of another experiment, which standard broiler housing was adopted (Cobb-Vantress, 2009). At this stage of the present study, the use of three specimens seems to be enough to present the proposed methodology, in accordance with previous studies approaching kinematic analysis of fishes (Tran et al., 2010; Wassenbergh and Rechter, 2011), turtles (Natchev et al., 2009), lizards (Schaerlaeken et al., 2007), and birds (Gussekloo and Bout, 2005; Dawson et al., 2011).
Fig. 1. Schematic representation of frames sequences classification for one bird.
Table 1 Number of analysed frames by birds, mandibulation order, and motion type. Bird
Frames
Mandibulation
Frames
Motion type
Frames
1 2 3 Total
130 141 98 369
1st 2nd 3rd –
68 143 158 369
Opening Maximum beak gape Closing –
224 18 127 369
90
S. Abdanan Mehdizadeh et al. / Computers and Electronics in Agriculture 114 (2015) 88–95
particle removal to prevent a mistake during the beak tip detection when a particle occludes the beak tip. Unlike the methods utilized by Horster et al. (2002), in which involved physic markers placed on the birds’ body, the study involved calculation of reliable reference point on the chicken so that body features relative to this point could be identified. The methodology in this study required no bird training or unusual management, and the machine vision algorithm was able to determine the eyeball and tip of the upper and lower beak automatically. Step 1: Eye detection
Fig. 2. Schematic representation of the manual measurements (a) and an example of the beak tips occluded by feed particles (b).
A thresholding process was applied to the video frame based on the colour difference between the eyeball and the body of the chicken, for calculating the eye position. All artefacts except the eyeball were then removed from the resulting image (Fig. 4a). After segmentation, the coordinate of the centre area of the eye was measured from the x- and y-axis of the image (Fig. 4b; red1 point refers to the centre of eyeball area). Step 2: Removal of redundant segments After finding the eye position, a region of interest (corresponding to a rectangle of 610 740 pixels) was defined in the centre of the eye for all video frames. This area was then extracted to remove other parts of the body in order to avoid redundant information and to enhancing the beak tip detection Step 3: Beak tip detection In order to find the beak tips, Otsu’s threshold was applied to the image in order to convert it into binary format (Otsu, 1979). The algorithm then started a search for the beak tips at the bottom left of the binary image (Fig. 5; arrow directed from the left to the right). If the beak was opened, the first non-zero pixel was classified as part the lower beak tip. If the beak was closed, the first non-zero pixel was identified as part of the upper beak. Step 4: Removing feed particle
Fig. 3. The four steps used for image analysis to measure eye position, beak gape, beak speed and beak acceleration.
2.4. Manual measurements The manual measurements were carried out by a human operator through ImageJ software (National Institutes of Health, Bethesda, Maryland, USA). In this procedure (Fig. 2a), the ‘Polygon selections’ tool were used to define the border of the birds’ eye and the coordinates of its centroid were the point of interest (x- and y-axis). The beak gape was measured from the Euclidian distance (‘Straight line’ tool) between the upper and lower beak tips. At some frames, the feed particles occluded the beak tips (Fig. 2b), so it was necessary to advance and return the frames to intuitively estimate the beak tip position. 2.5. Automatic measurements The automatic image analysis procedure comprised of four steps (Fig. 3) using MatlabÒ software (MathWorks, Inc., Natick, Massachusetts, USA). These four steps were eye detection as a reference point to determine the position of bird’s head; head extraction to remove redundant background information during analysis, beak tips detection to analyse the biomechanical behaviour (maximum beak-gape, speed and acceleration) and feed
The feed particles in some frames occluded the beak tip hindering its precise detection (illustrated in Fig. 6a and c). In order to identify and remove the feed particle from the image the following algorithm was applied:
Inewx;y
0 r x;y j160 < r x;y 255 ¼ 1 r x;y j0 r x;y 160
ð1Þ
Here r is the red channel of the unsigned 8 bit image, and x and y denotes the Cartesian co-ordinates of the old image r and the new image with the feed particle removed Inewx;y (Fig. 6d). After this process, a region of interest was defined around the beak (250 210 pixels) so that the maximum beak gape could be defined. First, the boundary of the beak was found within the area so that the two beak tips corresponding to the endpoints could be identified (Fig. 7a). Then, the Euclidian distance between the two beak tips (blue line) was measured. When the beak was closed, only one tip was detected, and the Euclidian distance was zero (Fig. 7b). The Euclidian distance was also used to measure the beak tips distance to calculate both motion speed and acceleration (Fig. 8). 1 For interpretation of colour in Figs. 4 and 7, the reader is referred to the web version of this article.
S. Abdanan Mehdizadeh et al. / Computers and Electronics in Agriculture 114 (2015) 88–95
91
Fig. 4. The eyeball segmentation (a), and the detection of its centre in x and y coordinates (b).
2.6. Statistical analysis
Upper beak tip Lower beak tip
Fig. 5. Binarized image of the chicken’s head and starting point of search algorithm to find the beak tips location.
Manual and automatic measurements were compared by calculating the difference between them. The unit considered in this study was the number of frames analysed (N = 369). General descriptive analysis was used. Regression analysis was applied to the difference between manual and automatic beak gape calculations. Then, ANOVA (Tukey test at 95% of confidence level) was applied to verify the differences between different birds (1–3), mandibulation order (1st, 2nd, and 3rd), and motion type (opening, maximum beak gape, and closing). The manual measurements were considered the ‘gold standards’, so it was used to calculate the speed and acceleration of the results. Additionally, frames regarded to the maximum beak gape were considered as the last Ò frame of the beak opening motion. Minitab 17 software (Minitab
Fig. 6. Schematic representation of the (a) original frame showing the feed particle occluding the lower beak tip; (b) binarized image with the feed represented in blue; (c) detail of the feed particle occluding the beak; and extracted feed particle from the image (d). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Fig. 7. Upside down picture showing the calculation of beak tips when it is opened (a) and closed (b).
92
S. Abdanan Mehdizadeh et al. / Computers and Electronics in Agriculture 114 (2015) 88–95
Opening sequence
1
2
3
4
5
2
3
4
5
6
Closing sequence
1
Fig. 8. Representation of beak opening and closing sequences.
Inc., Pennsylvania, USA) was used to carry out the statistical analysis.
Table 2 Differences between manual and automatic measurements (mm) of beak gape and head position (x- and y-axis).
3. Results
Variable
Method
Mean ± SE
P-value
Error (mean ± SE)
Maximum error
3.1. General differences between manual and automatic methods
Beak gape
Manual Automatic Manual Automatic Manual Automatic
2.43 ± 0.11 2.22 ± 0.11 31.95 ± 0.19 31.64 ± 0.19 24.18 ± 0.25 24.00 ± 0.24
0.162
0.22 ± 0.01
0.76
0.247
0.38 ± 0.01
1.05
0.605
0.24 ± 0.01
0.67
3.2. Beak gape, speed, and acceleration To calculate beak speed and acceleration, data from manual method (i.e. ‘gold standards’) were considered, besides the frames regarding to the maximum beak gape were considered as the last frame of opening motion for these calculations. In order to identify possible causes of the highest errors, ANOVA (Tukey test at 95% of confidence level) was applied to verify the influence of birds, mandibulation order, and motion type, likewise beak speed and acceleration. Results (Table 3) indicated that the speed of the beak was significantly (P-value < 0.005) higher for the first, second, and third mandibulations, respectively. The acceleration was considerably greater in the first mandibulation than the third one. Moreover, the speed of the beak was significantly higher during the closing motion (0.27 mm ms1) than opening (0.14 mm ms1). About the error between both methods, results indicated significantly (P-value < 0.005) differences between birds and mandibulation order, but no differences in motion type. The higher errors were found in the first and second mandibulations for speed, and first mandibulation for acceleration, in which the maximum error were 0.36 mm ms1 for speed and 0.08 mm ms2 for acceleration. 4. Discussion The feeding behaviour of animals can be divided into appetitive phase, corresponding to the demand for feed and consummatory act, i.e. the real feed intake. The appetitive phase of birds is characterized by the foraging behaviour, which is the time to explore the environment searching for food. Moreover, it was
x-axis y-axis
Tukey test at 95% of confidence level; SE = standard error.
10
Automatic Beak gape
Overall, the error between manual and automatic (algorithm output) measurements for the birds’ beak gape was 0.22 ± 0.01 mm (mean ± standard error), in which the maximum difference was 0.76 mm. For head position, the maximum error for x- and y-axis was 1.05 mm and 0.67 mm, respectively. ANOVA (Table 2) and regression analysis (Fig. 9) indicated no significant differences (P-value > 0.005) between both methods. Thus, they were highly correlated.
8 6 4 2 0 0
2
4
6
8
10
Manual Beak gape Fig. 9. Regression analysis between manual and automatic measurements for the whole dataset of the birds beak gape (R2 = 99.2%).
Table 3 The speed and acceleration of the birds’ beak motion by different birds, mandibulation order, and motion type. Variable
Factor
Speed (mm ms1)
Acceleration (mm ms2)
Bird
1 2 3
0.17 ± 0.02 a 0.21 ± 0.02 a 0.15 ± 0.01 a
0.03 ± 0.00 a 0.04 ± 0.00 a 0.03 ± 0.00 a
Mandibulation
1st 2nd 3rd
0.36 ± 0.04 a 0.19 ± 0.02 b 0.10 ± 0.01 c
0.07 ± 0.00 a 0.03 ± 0.00 b 0.02 ± 0.00 b
Motion type
Opening Closing
0.14 ± 0.01 a 0.27 ± 0.02 b
0.03 ± 0.00 a 0.04 ± 0.00 a
Tukey test; means that do not share a letter are significantly different (P-value < 0.005).
93
S. Abdanan Mehdizadeh et al. / Computers and Electronics in Agriculture 114 (2015) 88–95
reported that two-thirds of young bird pecks do not result in the catchment of a feed particle (Yo et al., 1997). Therefore, biomechanical assessments can be an effective method to study the Table 4 The error (mean ± standard error) between manual and automatic measurements for the beak gape (mm), beak speed (mm ms1) and beak acceleration (mm ms2) by bird, mandibulation order, and motion type. Variable
Factor
Beak gape
Speed
Acceleration
Bird
1 2 3
0.25 ± 0.01 a 0.24 ± 0.01 a 0.16 ± 0.01 b
0.03 ± 0.00 a 0.04 ± 0.00 a 0.03 ± 0.00 a
0.020 ± 0.002 a 0.017 ± 0.001 a 0.019 ± 0.002 a
Mandibulation order
1st
0.24 ± 0.01 ab
0.07 ± 0.01 a
0.023 ± 0.002 a
2nd 3rd
0.25 ± 0.01 a 0.19 ± 0.01 b
0.03 ± 0.00 b 0.02 ± 0.00 b
0.019 ± 0.002 ab 0.017 ± 0.001 b
Opening Maximum beak gape Closing
0.23 ± 0.01 a 0.19 ± 0.03 a
0.03 ± 0.00 a –
0.019 ± 0.02 a –
0.21 ± 0.01 a
0.04 ± 0.00 a
0.019 ± 0.001 a
Motion type
Tukey test; means that do not share a letter are significantly different (P-value < 0.005).
feeding behaviour of broiler chickens in order to improve the feeding efficiency while respecting their natural behaviour. The feeding assessments of animals may be related to bite events, and/or visits to feeders (Slater, 1974; Berdoy, 1993; Nielsen, 1999), in which these could be considered a unit of the study (Nielsen et al., 1995). The aim of the present study was to validate the automatic method (algorithm output) considering the calculations of kinematic variables of the birds in each mandibulation. Thus, the unit found here was the number of frames. The errors between manual and automatic methods of head motion (i.e. tracking the centre of the eye area) can be considered low, which the maximum errors were 1.05 mm and 0.67 mm for x- and y-axis, respectively. Thus, for this particular feeding phase (mandibulation), the algorithm presented a high accuracy. However, it is possible that improvements are needed to apply in other feeding phases, which the eye blink will potentially lead to misconstrued calculations. About the beak gape measurements, the errors between manual and automatic methods were significantly different for distinct individuals. However, little can be concluded from this observation (inter-individual differences), since a few sampling of individuals were observed. On the other hand, the highest errors of automatic
Motion type
Mandibulation order 0,28
Beak gape error (mm)
0,28
a
0,26 0,24
0,24
0,22
0,22
0,20
0,20
0,18
0,18
0,16
0,16
0,14
0,14
0,12
0,12 1st
2nd
3rd
Beak speed (mm ms-1)
0,42
Closing
c
0,38
0,34 0,30
0,26
0,26
0,22
0,22
0,18
0,18
0,14
0,14
0,10
0,10 3rd
d
Closing
Opening
0,09
0,09
Beak acceleration (mm ms-2)
Opening
0,38
0,30
2nd
Maximum beak gape
0,42
0,34
1st
b
0,26
e
0,08
f
0,08
0,07
0,07
0,06
0,06
0,05
0,05
0,04
0,04
0,03
0,03
0,02
0,02 0,01
0,01 1st
2nd
3rd
Closing
Opening
Fig. 10. Interval plots of the beak gape error (a and b), beak speed (c and d), and beak acceleration (e and f) by mandibulation order and motion type; refers to Tables 3 and 4.
94
S. Abdanan Mehdizadeh et al. / Computers and Electronics in Agriculture 114 (2015) 88–95
measurements were found in the first and second mandibulations right after the feed grasping. This could be explained by the amount of feed particles presented in the scene that occludes the upper and/or lower beak tips. The methodology presented in this paper overcomes this problem. Another explanation, however, might be the beak speed and acceleration. The higher speed of the beak creates a blurred image on the frame, which might lead the algorithm to errors due to difficult to detect the beak tip precisely. The higher beak speeds were observed at the first, second, and third mandibulations, respectively (Fig. 10c), likewise a larger beak acceleration in the first mandibulation (Fig. 10e). Interestingly, no significant difference was found in the opening and closing beak motions, even though they presented different speeds (Fig. 10d). This could be explained by the higher speeds detected in the 1st mandibulation compared with the speeds of the closing motion (Fig. 10c and d), which might explain the higher errors in the higher beak speeds. In this sense, it can be highlighted the importance of an adequate camera setup (resolution and acquisition rate) in order to avoid blurred images of the object of interest, and balance it with data amount to be storage and analysed. During broilers growth, it is important that feeders and drinkers be properly arranged and well managed. Furthermore, several studies indicate that some design features, such as size, location, geometry, spacing and angle of feeders can affect the behaviour of animals (Buskirk et al., 2003; Wolter et al., 2009). Developing a precise and non-invasive method for assessing the motion in relation to the feeders’ usage remains a challenge (Lu and Chang, 2012). In this sense, some advantages are seen when using high speed cameras and computational image analysis for motion assessments, especially because its relatively small cost, versatility in analysis, commercial availability of the hardware, and possibility to upgrade the system as needed (Sakatani and Isa, 2004). In this study, it was observed that sometimes the chicks rotate their heads while feeding. This is an instinct behaviour in order to kill the prey, and previous study has described this kind of behaviour in pigeons (Horster et al., 2002). Thus, in a 2-diomensional analysis a calibration line on the birds’ head is required to correct the measurements. This method was not applied in the present study, because it was not considered mandibulation sequences when the birds showed this kind of behaviour. The algorithm for automatic eye detection works based on colour difference, so the background should not be set with a similar colour. To compensate different colours of feed particles, colour threshold was modified based on the colour difference of beak and feed. This modification can be acquired manually or automatically to get best discrimination results between beak and feed. Furthermore, a camera with higher acquisition rate (number of frames per second) and higher resolution combined with improvements in image capturing techniques has the potential to enhance the accuracy of the model. A possible solution to improve computational image analysis is the use of infrared-reflective motion capture markers instead just physical markers, which might enhance its identification (Heysea et al., 2014; Paxton et al., 2013). Birds select different sizes of feed particles on the first week of life. The format and structure of the beak determine the size and type of feed to be ingested, and thus the granulometry of the particle is of high importance for the regulation of the consumption (Nir et al., 1990; Nir et al., 1994a; Addo et al., 2012). The contact perception contributes to the identification of the feed and broiler chickens can discriminate different types of diets by associating the feed physical features with nutritional content (Emmans and Kyriazakis, 2001). Thus, the study of kinematics during birds’ feeding behaviour combined with physiological responses should aid the decision regarding the optimum feed type and granulometry for different rearing environments. This should be considered in relation to the growth phase, strain, the diet composition, and
the available technology in the feed plant. In this study, the algorithm presented accurate results for chicks’ mandibulation assessments (1.05 mm was the highest error), but improvements in recording techniques should help its adaptation for further kinematics analysis; e.g. other feeding phases and others species. 5. Conclusions The presented method calculates automatically the position of the eye centre (x- and y-axis) in order to track the head displacement of broiler chicks, and also both upper and lower beak tips distance. The algorithm presented an error less than 1.05 mm (the highest error) and can be improved by using a camera with higher resolution, a higher acquisition rate, and infrared-reflective markers. The highest errors were found in the first and second mandibulations right after the feed grasping. This is explained by the presence of feed particles that occluded the beak tips and the higher beak speed and acceleration that created a blurred image. A better understanding of the mechanical process of the birds’ jaw apparatus during feeding might be an efficient method for determining the relationship between different types of feeds and the biomechanical patterns exhibited by the birds. Moreover, it also should consider the anatomical variations between different strains, gender, growth phase, and also the influence of feeder design. The high-speed camera combined with techniques of computational image analysis is a useful technology to aid such assessments. Acknowledgement The authors thank CNPq – National Council for Scientific and Technological Development for supporting this research. References Addo, A., Bart-Plange, A., Akowuah, J.O., 2012. Particle size evaluation of feed ingredient produced in the Kumasi Metropolis, Ghana. ARPN J. Agric. Biol. Sci. 7, 177–181. Amerah, A.M., Ravindran, V., Lentle, R.G., Thomas, D.G., 2007. Feed particle size: implications on the digestion and performance of poultry. World Poult. Sci J 63, 439–451. Berdoy, M., 1993. Defining bouts of behaviour: a three process model. Anim. Behav. 46, 387–396. Bermejo, R., Allan, R.W., Houben, D., Deich, J.D., Zeigler, H.P., 1989. Prehension in the pigeon I: descriptive analysis. Exp. Brain Res. 75, 569–576. Buskirk, D.D., Zanella, A.J., Harrigan, T.M., Van Lente, J.L., Gnagey, L.M., Kaercher, M.J., 2003. Large round bale design affects hay utilization and beef cow behavior. J. Anim. Sci. 81, 109–115. Cobb-Vantress. 2009. Manual de manejo de frangos de corte (Broiler Management Guide). L-1020-02 PT. Dawson, M.M., Metzger, K.A., Baier, D.B., Brainerd, E.L., 2011. Kinematics of the quadrate bone during feeding in mallard ducks. J. Exp. Biol. 214, 2036–2046. Emmans, G., Kyriazakis, I., 2001. Consequences of genetic change in farm animals on food intake and feeding behavior. Proc. Nutr. Soc. 60, 115–125. Estrella, S.M., Masero, J.A., 2007. The use of distal rhyncokinesis by birds feeding in water. J. Exp. Biol. 210, 3757–3762. FAO. 2012. Food Outlook, Rome: Trade and Market Division of FAO. Greenwood, M.W., Cramer, K.R., Clark, P.M., Behnke, K.C., Beyer, R.S., 2004. Influence of feed form on dietary lysine and energy intake and utilization of broilers from 14 to 30 days of age. Int. J. Poult. Sci. 3, 189–194. Gussekloo, S.W.S., Bout, R.G., 2005. Cranial kinesis in palaeognathous birds. J. Exp. Biol. 208, 3409–3419. Hall, S.J., 1999. Basic Biomechanics, third ed. Mcgraw-Hill, Singapore. Heysea, T.J., El-Zayata, B.F., Corteb, R. De, Scheys, L., Chevalierd, Y., FuchsWinkelmanna, S., Labeye, L., 2014. Biomechanics of medial unicondylar in combination with patellofemoral knee Arthroplasty. Knee 21 (S1), 53–59. Horster, W., Krumm, E., Mohr, C., Delius, D.J., 2002. Conditioning the pecking motions of pigeons. Behav. Process 58, 27–43. Lu, T., Chang, C., 2012. Biomechanics of human movement and its clinical applications. Kaohsiung J. Med. Sci. 28, S13–S25. Mclester, J., Pierre, P.S., 2008. Applied Biomechanics: Concepts and Connections. Thomson Wadsworth, Belmont. Natchev, N., Heiss, E., Lemell, P., Stratev, D., Weisgram, J., 2009. Analysis of prey capture and food transport kinematics in two Asian box turtles, Cuora
S. Abdanan Mehdizadeh et al. / Computers and Electronics in Agriculture 114 (2015) 88–95 amboinensis and Cuora flavomarginata (Chelonia, Geoemydidae), with emphasis on terrestrial feeding patterns. Zoology 112, 113–127. Neves, D.P., Nääs, I.A., Vercellino, R.A., Moura, D.J., 2010. Do broilers prefer to eat from a certain type of feeder? Braz. J. Poult. Sci. 12, 179–187. Nielsen, B.L., 1999. On the interpretation of feeding behaviour measures and the use of feeding rate as an indicator of social constraint. Appl. Anim. Behav. Sci. 63, 79–91. Nielsen, B.L., Lawrence, A.B., Whittemore, C.T., 1995. Effects of single-space feeder design of feeding behaviour and performance of growing pigs. Anim. Sci. 61, 575–579. Nir, I., Melcion, J.P., Picard, M., 1990. Effect of particle size of sorghum grains on feed intake and performance of young broilers. Poult. Sci. 69, 2177–2184. Nir, I., Hillel, R., Shefet, G., Nitsan, Z., 1994a. Effect of grain particle size on performance. 2. Grain texture interactions. Poult. Sci. 73, 781–791. Nir, I., Shefet, G., Aaroni, Y., 1994b. Effect of particle size on performance. 1. Corn. Poult. Sci. 73, 45–49. Otsu, N., 1979. A threshold selection method from grey level histograms. IEEE Trans. Syst. Man Cyb. 9, 62–66. Paxton, H., Daley, M.A., Corr, S.A., Hutchinson, J.R., 2013. The gait dynamics of the modern broiler chicken: a cautionary tale of selective breeding. J. Exp. Biol. 216, 3237–3248. Pennycuick, C.J., 1992. Newton Rules Biology: A Physical Approach to Biological Problems. Oxford University Press, New York. Perez, H., Oliva-Teles, A., 2002. Utilization of raw and gelatinized starch by European sea bass (Dicentrarchuslabrax) juveniles. Aquaculture 205, 287– 299. Robertson, D.G.E., Caldwell, G.E., 2004. Planar kinematics. In: Robertson, D.G.E., Caldwell, G.E., Hamill, J., Kamen, G., Whittlesey, S.N. (Eds.), Research Methods in Biomechanics. Human Kinetics, Champaign, IL, USA, pp. 9–52. Roll, V.F.B., Dai Prá, M.A., Roll, A.A.P., Xavier, E.G., Rossi, P., Anciuti, M.A., Rutz, F., 2010. Effect of tubular feeder height on ingestive behavior of broiler. Arch. Zootec. 59, 115–122. Ropert-Coudert, Y., Kato, A., Liebsch, N., Wilson, R.P., Muller, G., Baubet, E., 2004. Monitoring jaw movements: a cue to feeding activity. Game Wildlife Sci. 20, 1–19.
95
Sakatani, T., Isa, T., 2004. PC-based high-speed video-oculography for measuring rapid eye movements in mice. Neurosci. Res. 49, 123–131. Schaerlaeken, V., Meyers, J.J., Herrel, A., 2007. Modulation of prey capture kinematics and the role of lingual sensory feedback in the lizard Pogona vitticeps. Zoology 110, 127–138. Skinner-Noble, D.O., Mckinney, L.J., Teeter, R.G., 2005. Predicting effective caloric value of non-nutritive factors: III. Feed form affects broiler performance by modifying behavior patterns. Poult. Sci. 84, 403–411. Slater, P.J.B., 1974. The temporal pattern of feeding in the Zebra finch. Anim. Behav. 22, 506–515. Stefen, C., Ibeb, P., Fischer, M.S., 2011. Bi-planar X-ray motion analysis of the lower jaw movement during incisor interaction and mastication in the beaver (Castor fiber L. 1758). Mamm. Biol. 76, 534–539. Thomas, M., Van Vliet, T., van der Poel, A.F.B., 1998. Physical quality of pelleted animal feed 3. Contribution of feedstuff components. Anim. Feed Sci. Technol. 70, 59–78. Tran, H.Q., Mehta, R.S., Wainwright, P.C., 2010. Effects of ram speed on prey capture kinematics of juvenile Indo-Pacific tarpon, Megalops cyprinoides. Zoology 113, 75–84. USDA. 2012. Livestock and Poultry: World Markets and Trade, s.l.: USDA. Van Der Heuvel, W.F., Berkhoudt, H., 1998. Pecking in the chicken (Gallus gallus domesticus) motion analysis and stereotypy. Neth. J. Zool. 48, 273–303. Wainwright, P.C., 1991. Ecomorphology: experimental functional anatomy for ecological problems. Am. Zool. 31, 680–693. Wassenbergh, S.V., Rechter, D.D., 2011. Piscivorous cyprinid fish modulates suction feeding kinematics to capture elusive prey. Zoology 114, 46–52. Wolter, B.F., Ellis, M., Curtis, S.E., Parr, E.N., Webel, D.M., 2009. Feeder location did not affect performance of welling pigs in large groups. J. Anim. Sci. 78, 2784– 2789. Yo, T., Vilariño, M., Faure, J.M., Picard, M., 1997. Feed pecking in young chickens: new techniques of evaluation. Physiol. Behav. 61, 803–810. Zang, J.J., Piao, X.S., Huang, D.S., Wang, J.J., Ma, X., Ma, Y.X., 2009. Effects of feed particle size and feed form on growth performance, nutrient metabolizability and intestinal morphology in broiler chickens. Asian Austral J. Anim. Sci. 22, 107–112. Zweers, G.A., 1982. Pecking of the pigeon (Columba livia L.). Behaviour 81, 173–230.