- Email: [email protected]
Prediction of chick hatching time using visible transmission spectroscopy combined with partial least squares regression
Engineering in Agriculture, Environment and Food xxx (2014) 1e6
Contents lists available at ScienceDirect
Engineering in Agriculture, Environment and Food journal homepage: http://www.sciencedirect.com/eaef
Research paper
Prediction of chick hatching time using visible transmission spectroscopy combined with partial least squares regression Md. Hamidul Islam a, *, Naoshi Kondo a, Yuichi Ogawa a, Tateshi Fujiura a, Tetsuhito Suzuki a, Shusaku Nakajima a, Shinichi Fujitani b a b
Graduate School of Agriculture, Kyoto University, Kitashirakawa-Oiwakecho, Sakyo-ku, Kyoto 606-8502, Japan NABEL Co. Ltd., 86 Morimotocho, Nishikujo, Minami-ku, Kyoto 606-8444, Japan
a r t i c l e i n f o
a b s t r a c t
Article history: Received 7 March 2014 Received in revised form 10 September 2014 Accepted 5 October 2014 Available online xxx
The feasibility of using visible transmission spectroscopy for the prediction of chick hatching time was investigated. An experiment was conducted with 100 chicken eggs in which transmission spectra were measured between incubation day 0 (non-incubated) and days 8 and subsequent hatching time was recorded. Spectral transmittance in the range of 500e750 nm was used in analysis. Spectral data were linked to hatching time using a partial least squares (PLS) regression method. Different pre-processing procedures were compared. The calibration model using incubation day 4 spectra with multiplicative scatter correction (MSC) resulted in the lowest root mean square error of prediction (RMSEP) ¼ 3.41 h. The result indicates that the use of visible transmission spectroscopy combined with multivariate analysis has potential to predict the chick hatching time. © 2014, Asian Agricultural and Biological Engineering Association. Published by Elsevier B.V. All rights reserved.
Keywords: Hatching time Chicken egg Hatch window Visible spectroscopy Transmission PLS regression Calibration model
1. Introduction Day-old chicks are an important starting input for broiler farms, and the end product of the poultry hatchery. Overall success in broiler production depends on the quality of these chicks. The quality of day-old chicks is determined from their post-hatched survivability and growth potential, i.e. growth rate, breast meat yield and feed conversion ratio (Decuypere and Bruggeman, 2007). For broiler farmers, the aim is to obtain a batch of homogeneous, high quality chicks. However a batch of day-old chicks is frequently not homogeneous in quality due to the spread of the hatch window (Tona et al., 2003). Ideally, all chicks in a batch are desired to hatch at the same time, but in reality, chicks hatch at different moments within a time period called the “hatch window”. As a result of this broad hatch window, a production manager is dealing with chicks of different biological ages at take-off. Though the chronological ages of the chicks in a batch are defined by convention as being the same, the earlier hatched chicks are in reality older than one day. These chicks often show signs of dehydration at takeoff and are a poor quality stock for later
* Corresponding author. E-mail address: [email protected] (Md.H. Islam).
production. In addition, the delay in the collection of the chick batch also delays further hatchery procedures, such as sexing, vaccinations, packaging, and transportation, which also ultimately delays the first feed and access to water. Previous researchers have reported that the delay in the feed intake is associated with higher early mortality in chicks and impaired performance throughout the growth period (Chou et al., 2004; Gonzales et al., 2003). Therefore, a narrow hatch window is desired by poultry hatchery managers in order to produce the best quality chicks. However, it is very difficult for the hatchery managers to correctly estimate hatching time and consequently the spread of the hatch window. This is because the hatching time of individual chicks within a batch is influenced by multiple factors, such as the age of parent flock, egg handling, egg storage duration, and incubation conditions (Decuypere et al., 2001). Since a narrow hatch window results in higher quality chicks due to the smaller variation in the biological ages of chicks, hatchery managers undertake several techniques in order to obtain a narrow hatch window. The most common method used is to collect chicks that hatch in the first 24 h window. The main drawback of this method is, if the incubation is ended prematurely, eggs with viable chicks inside them are thrown away, decreasing the hatchability of the cohort and resulting in economic losses. Other techniques to optimize hatchability include exchanging the
http://dx.doi.org/10.1016/j.eaef.2014.10.001 1881-8366/© 2014, Asian Agricultural and Biological Engineering Association. Published by Elsevier B.V. All rights reserved.
Please cite this article in press as: Islam MdH, et al., Prediction of chick hatching time using visible transmission spectroscopy combined with partial least squares regression, Engineering in Agriculture, Environment and Food (2014), http://dx.doi.org/10.1016/j.eaef.2014.10.001
2
Md.H. Islam et al. / Engineering in Agriculture, Environment and Food xxx (2014) 1e6
position of the egg trays from cold to hot areas inside of the incubator and vice versa, increasing either temperature or carbondioxide (CO2) levels in the hatcher, as reviewed by Hill (2011). These practices can improve the hatchability of chicks to some extent, but cannot produce the best quality chick cohort, because individual chicks have their own natural hatch window and a short hatch window created by overheating the embryos or increasing CO2 level in the hatcher is not a natural hatch window. Therefore, development of a new hatch window control technique is demanded by the poultry hatchery in order to obtain the best quality chicks with high hatchability. Hatching time of chicks is closely related to embryonic development. The incubation length required by an egg is predetermined by this embryonic development. In fertile hatching eggs blood formation takes place from day 2 of incubation, as reviewed by Bamelis et al. (2002), and subsequent physiological changes occur in live embryos during the incubation period. Thus information about the stage of embryonic development could be used to estimate the incubation length needed by the egg to hatch. Nowadays, spectral measurements are commonly used to extract internal information about eggs, such as blood and meat spots, presence of embryo, embryonic growth, and egg freshness. Such spectral measurements have the advantage of being fast, nondestructive, and noncontact. Moreover, this technique can be implemented at a reasonable price. Preliminary investigations have revealed that significant variations in the spectra were found during incubation. Kemps et al. (2010) used visible transmission spectroscopy coupled with a multivariate analysis technique to assess embryonic weight in chicken eggs. They linked spectral information to embryonic weight using a partial least square (PLS) regression method and reported correlation coefficients of 0.97 based on the multiplicative scatter correction (MSC) spectra for the prediction of embryonic weight within the spectral range 570e750 nm. To our knowledge, however, no attempts have been made to estimate the hatching time of chicks using spectral information. The goal of this study is to investigate the potential to predict hatching time of chicks using visible transmission spectroscopy combined with an appropriate multivariate analysis method. A partial least square (PLS) regression algorithm was employed to build a calibration model for the prediction of hatching time of chicks. Spectral information in the range 500e750 nm was used to develop a calibration model. 2. Material and methods 2.1. Material A total of 100 light brown-shell chicken eggs laid by a commercial broiler breed (Ross 308 strain) were used in this study. All eggs were collected from a commercial poultry hatchery (Yamamoto Co. Ltd., Kameoka, Kyoto, Japan). At the moment of egg collection, the laying hens were 36 weeks of age. Since large variation in the size of eggs effects on hatching time, all samples were selected within a range of 42.5e44.5 mm (diameter), 54.4e59.4 mm (height) and 55.5e65.5 g (weight) to minimize their effects. In addition, to minimize the effect of egg shell pigmentation on transmission spectra, eggs with nearly similar shell color (selected by color image analysis method) were selected. Prior to incubation, all eggs were stored for 3 days following standard poultry hatchery practice (at 15.0 (±0.5) C and 80 (±5) % of relative humidity (RH)). 2.2. Spectral acquisition system The experimental setup used for the measurement of the transmission spectra of an intact egg is shown in Fig. 1. The egg is
positioned vertically in a plastic holder with the blunt end pointed upward between the illuminating fiber and the collecting optical fiber. The illuminating fiber was used for cool illumination of the eggs and the optical fiber was used to collect and transport the transmitted light to a Hamamatsu C 7473-36 model spectrometer (Hamamatsu Photonics K. K, Japan). The distance between the illuminating fiber and the optical fiber was kept at 100 mm to obtain a consistent transmission signal. A halogen light source (FHL-10, Asahi Spectra Co. Ltd., Japan) was used for illuminating the samples. This light source consisted of a dichroic reflector type halogen lamp (capable of cutting off infrared energy) in order to prevent warming of the egg surface where the light beam is shone onto the eggshell. The light was focused on the egg surface by an illuminating fiber of 5 mm in diameter and only the light that transmitted through the egg was received and transported by the optical fiber (1 mm effective light receiving diameter) to the spectrometer. The software package PMA-11 Spectral Analyzer for Windows (PMA Software U6039-01, Hamamatsu Photonics K.K., Japan) was used to control the spectral acquisition process. The transmission spectra of the eggs were measured over the spectral range of 200e950 nm at 1 nm intervals. The integration time for one scan was 100 ms and the spectrum of each egg contains an average of 10 scans. Since the characteristics of a halogen lamp changes over time, the spectrometer was calibrated before each measurement using a Teflon block (PTFE push rod Ø45 mm, Chukoh Chemical Industries, Ltd., Japan) of 30 mm thickness (Kemps et al., 2010). In addition, the reference spectra were measured after each 10 samples to evaluate any changes in the reference spectra with time. Furthermore, a correction was made for the electrical noise by taking the spectra with no light exposure to the spectrometer. All measurements were done inside a black box to shield any stray light from entering. The transmission values of light passing through an egg are expressed as a ratio of the amount of light passing through the eggs to the amount of light passing through the reference at the same wavelength (Kemps et al., 2006). The relative transmission (T) was calculated using Eq. (1).
TðlÞ ¼
SðlÞ RðlÞ
(1)
Where: T (l) is relative transmission at wavelength l nm S (l) is intensity of sample at wavelength l nm R (l) is intensity of Teflon reference at wavelength l nm It should be noted that the term “transmission spectra” used throughout this text refers to the relative transmission spectra. 2.3. Experimental design Prior to setting the eggs into the incubator, eggs were preheated for 16 h (first 6 h at 28 C and the remaining 10 h at 30 C) to bring the embryos to a uniform temperature when they are placed into the incubator. Just prior to incubation (referred to as “Day 0”), the transmission spectra of all eggs were measured. Upon completion of measurements, the eggs were immediately placed in a commercial incubator (SSH-02 all in one type, Showa Furanki Co. Ltd., Saitama, Japan) to incubate at 37.8 C and 55% of relative humidity according to Lourens et al. (2005). During incubation eggs were turned automatically every hour through an angle of 90 until incubation day 18. Between incubation day 1 to day 8, eggs were taken out from the incubator every 24 h to measure the spectral transmission of each egg. To minimize exposure time of the egg
Please cite this article in press as: Islam MdH, et al., Prediction of chick hatching time using visible transmission spectroscopy combined with partial least squares regression, Engineering in Agriculture, Environment and Food (2014), http://dx.doi.org/10.1016/j.eaef.2014.10.001
Md.H. Islam et al. / Engineering in Agriculture, Environment and Food xxx (2014) 1e6
3
Fig. 1. Experimental setup used for the measurement of transmission spectra.
outside the incubator, eggs were immediately placed back into the incubator after the measurements. At incubation day 19, eggs with living embryos were placed into hatcher baskets at 37.8 C and 60% of RH as mentioned by Walstra et al. (2010) and allowed to hatch. From incubation day 19, hatching chicks were monitored by a web camera to record the time of hatching for each egg until the hatching process ended. During the monitoring period, images were captured at 5 min intervals and saved automatically on the hard disk of a personal computer. The time of capture was also recorded simultaneously. Later, incubation time required by each egg to hatch was calculated as the duration between the time of emergence of the chick and the time when incubation started. 2.4. Data analysis The transmission spectra of eggs in the visible region from 500 to 750 nm were used for statistical analysis, because most of the transmitted light below 500 nm absorbed by the calciferous shell of the egg (Brant et al., 1953). Among 100 eggs, 1 egg was infertile and 2 eggs were failed to hatch. At first, relative transmission spectra of successfully hatched eggs were normalized by a peak normalization technique taking transmission at 500 nm as 1 to minimize the effect of variation in egg shell thickness on transmission spectra. Principal component analysis (PCA) was performed on the complete data set to detect outliers. After removal of outliers, samples were then divided randomly into two groups i.e. the calibration set and prediction set. The calibration sample set was used to develop a calibration model and prediction set was used for the validation of the calibration model. Hatching time of two sample sets are shown in Table 1. Several pre-processing procedures including moving average smoothing, Savitzky-Golay smoothing, multiplicative scatter correction (MSC), baseline offset correction, Savitzky-Golay first derivative were compared during this work. The averaging technique is used to reduce the number of wavelength or to smooth the spectra of the egg. It is also used to optimize the signal to noise ratio (Cen and He, 2007). The MSC transformation method is applied to compensate additive and multiplicative scatter effects included in the spectra. While baseline offset correction is done to correct the baseline shift of the spectra. The derivation of spectra is needed to remove baseline shift and superposed peaks (Swierenga
et al., 1999). These pre-processing operations were performed for both calibration and prediction sample sets. All types of spectral pre-processing were performed using The Unscrambler® software: version 9.8 (CAMO, Oslo, Norway). 2.5. Calibration model for the prediction of chick hatching time Partial least square (PLS) regression multivariate analysis was used to develop a calibration model for the estimation of chick hatching time. The PLS regression approach is one of the most popular chemometric algorithms for calibration model development due to its simplicity and small volume of calculations. The spectral data were linked to hatching time of chicks using a PLSR multivariate analysis to develop a calibration model. A low number of factors were desirable in order to avoid inclusion of any signal noise in the developed model (Xiaobo et al., 2007). The PLS regression analysis of the samples was performed using The Unscrambler® software. Initial PLS regression analysis was done on normalized spectra of each incubation day from day 3 to day 8 to develop the calibration models to select most effective incubation day. Further, PLS regression was done on selected incubation day normalized spectra combined with different pre-processing operations. During development of the calibration models, 15 principal components (PCs) were set up as a maximum. 2.6. Evaluation of the calibration model The prediction capacity of the developed calibration model was evaluated using several statistical parameters, including correlation coefficient (r) of prediction, root mean square error of calibration (RMSEC), root mean square error of prediction (RMSEP) and Bias. Bias is the mean difference between predicted and observed value of hatching time. A good calibration model should have a high correlation coefficient (r) and a low RMSEP. In addition, the difference between RMSEC and RMSEP should be small for a good calibration model (Camps and Christen, 2009). The quality of the final model was evaluated according to lowest difference between correlation coefficient of calibration and prediction set, lowest difference between RMSEC and RMSEP and lowest bias value. 3. Results and discussion
Table 1 Hatching time of calibration and prediction sample sets used for data analysis. Items
Calibration set
Prediction set
Number of samples Range (hour) Mean (hour) Standard deviation
60 483.1e497.75 483.1 5.97
35 483.3e497.5 483.3 6.25
3.1. Characteristics of visible transmission spectra of an incubating egg Typical transmission spectra of an incubating egg at different incubation times in the spectral range 500e750 nm are shown in Fig. 2. The overall shape of the incubated egg spectra was
Please cite this article in press as: Islam MdH, et al., Prediction of chick hatching time using visible transmission spectroscopy combined with partial least squares regression, Engineering in Agriculture, Environment and Food (2014), http://dx.doi.org/10.1016/j.eaef.2014.10.001
4
Md.H. Islam et al. / Engineering in Agriculture, Environment and Food xxx (2014) 1e6
qualitatively examined in order to illustrate the influence of embryonic development on the spectral profile. From Fig. 2, it is apparent that the transmission spectra shifted up and down depending on the incubation length. The most significant spectral difference appeared around 570e590 nm, 620e630 nm, and 640e650 nm when considering the different incubation lengths. The large decreases in the transmission spectra at 589 and 643 nm prior to incubation day 4 are due to light absorbed by the shell color pigment “protoporphyrin IX” (Gielen et al., 1979). However, the changes in the shape of the transmission spectra around 570e580 nm from incubation day 4 onward are because of the light absorbed by hemoglobin, which is the detected pigment in the blood (Bamelis et al., 2002). The shifting down of the transmission spectra from incubation day 3 to onward are related to the formation of sub embryonic fluid (SEF). With the formation of SEF, the albumen of the egg gradually lost water and became thicker reported by Baggott (2001). This thick albumen reduced the transmission of light through it and results in lower signal received at the detector. Therefore, the spectral differences displayed in Fig. 2 are mainly dependent on differences in the volumetric fractions of water in the albumin of eggs at the different incubation times.
Table 2 PLS regression results for the prediction of chick hatching time for different preprocessing techniques at incubation day 4. Pre-processing method
PCs
rcal
RMSEC
rpred
RMSEP
Bias
Normalized spectra Moving Ave. smoothinga SG smoothingb MSC Baseline offset SG first derivativeb
7 7 7 4 7 6
0.872 0.878 0.878 0.870 0.871 0.914
2.81 2.74 2.75 2.94 2.83 2.34
0.888 0.885 0.882 0.873 0.884 0.848
3.49 3.53 3.63 3.41 3.71 3.96
0.753 0.756 0.761 0. 651 0.865 0.992
rcal and rpred are the correlation coefficient for calibration and prediction set, respectively. SG: Savitzky-Golay method. a Number of segments: 5. b Number of segments: 5, Polynomial order: 2.
spectra based model was the less complex because the number of principal components was minimal. This result indicates that visible transmission spectroscopy has the potential for the estimation of chick hatching time at early stage of incubation. The scatter plot between actual and predicted hatching time of the best calibration model is presented in Fig. 3.
3.2. Calibration model for the prediction of hatching time 3.3. Validation of the calibration model A PLS regression method was used to develop the calibration model for the estimation of chick hatching time. Spectral information in the range of 500e750 nm was used for developing the calibration model and for validating the model. Initial PLS regression results (correlation coefficient obtained from incubation days 3 to days 8 are 0.536, 0.872, 0.145, 0.168, 0.157 and 0.128 respectively) reveal that spectral information at incubation day 4 is more effective to predict chick hatching time. Therefore, in rest of the manuscript we discussed only the results that are obtained from incubation day 4 transmission spectra. The PLS regression results for the prediction of chick hatching time using different preprocessing procedures at incubation day 4 is shown in Table 2. At first raw spectra was normalized and then other pre-processing operations were done on the normalized spectra. From Table 2, it is apparent that all calibration models had a good correlation coefficient i.e. r ¼ 0.870e0.914. The RMSEC values varied from 2.34 to 2.94 h. The RMSEC value less than 3 h is acceptable for the hatchery manager because, in practice the hatch window spans from 24 to 48 h (Careghi et al., 2005). Among all calibration models, the MSC
The capability of the developed calibration models for the estimation of chick hatching time of the independent sample was confirmed. For this purpose, transmission spectra of 35 chicken eggs (with different hatching time) were used to predict hatching time. The RMSEP obtained from different calibration models were varied from 3.41 to 3.96 h. As seen in Table 2, the calibration model using MSC spectra resulted in the lowest RMSEP ¼ 3.41 h and a high r ¼ 0.873. In addition, this calibration model had the lowest difference between RMSEC and RMSEP and lowest bias value. Also the difference between correlation coefficient of calibration and prediction is minimal for the MSC spectra based model. Hence, this calibration model was selected as the best calibration model for the prediction of chick hatching time. The scatter plot between actual and predicted hatching time using the MSC spectra based calibration model is presented in Fig. 4. From the figure it is observed that some samples have prediction error above 5 h, this is because of individual egg weight difference (Hassan and Nordskog, 1971). This result shows that a calibration model for prediction of chick
Fig. 2. Typical transmission spectra of an incubating egg at different incubation time in visible region.
Fig. 3. Scatter plot between actual and predicted value of hatching time for the calibration model developed using incubation day 4 spectra with MSC.
Please cite this article in press as: Islam MdH, et al., Prediction of chick hatching time using visible transmission spectroscopy combined with partial least squares regression, Engineering in Agriculture, Environment and Food (2014), http://dx.doi.org/10.1016/j.eaef.2014.10.001
Md.H. Islam et al. / Engineering in Agriculture, Environment and Food xxx (2014) 1e6
5
there is no concrete evidence, further investigation is needed to clarify this finding. 4. Summary and conclusions
Fig. 4. Scatter plot between actual and predicted value of hatching time in validation step. Predicted value of hatching time was calculated using MSC spectra based calibration model.
hatching time using visible transmission spectroscopy was successfully developed and was well validated. 3.4. Evaluating the structure of the calibration model The PLS regression was also a useful method for investigating the location of information related to the hatching time. In order to clarify the behavior of the calibration model, the regression coefficient of the best calibration model for the prediction of hatching time of chick was plotted against wavelength as demonstrated in Fig. 5. From this, it is apparent that around 577 nm a very slight peak is observed which indicates that the formation of the pigment hemoglobin has a minor contribution to the prediction of the hatching time of chick. From Fig. 5, it can be seen that the spectral range 610e750 nm makes an important contribution to the prediction of hatching time. The spectral range 610e750 nm is associated with the shifting of albumen water content, as mentioned in an earlier section (see explanation for Fig. 2). The hatching time of chicks influenced by the embryonic growth rate, which depends on the formation of SEF, as reported by Baggott (2001). From this result, the prediction of hatching time of chicks in the visible region is mainly driven by the SEF because formation of SEF makes albumen of egg thicker by reducing water content. However, since
Fig. 5. Regression coefficients of the calibration model used for the prediction of chick hatching time.
In this study, the potential of using visible transmission spectroscopy for the prediction of hatching time of chicks has been demonstrated. Spectral information combined with PLSR multivariate analysis was performed to develop a calibration model and for validation of the model. The transmission spectra (500e750 nm) through intact eggs were applied to develop the prediction model. In the subsequent analysis, relative spectra were normalized with a selected waveband to exclude the egg shell thickness effect and then several pre-processing procedures were applied on the spectra. The calibration model using incubation day 4 spectra with MSC achieved a satisfactory result i.e. RMSEP ¼ 3.41 h and r ¼ 0.873. These results indicate that visible transmission spectroscopy combined with multivariate analysis has significant potential to predict chick hatching time. Furthermore, early prediction of hatching time for individual eggs will enable hatchery managers to more finely control the spread of the hatch window, and subsequently the profitability of broiler production. Although this result is very promising, further study is needed for an example with heterogeneous egg sample (size, color, different storage condition). Acknowledgments The authors express their gratitude to Professor Garry Piller, Graduate School of Agriculture, Kyoto University for editing and proof reading of this manuscript. References Baggott GK. Development of extra-embryonic membranes and fluid compartments. In: Deeming DC, editor. Perspectives in fertilization and embryonic development in poultry. Lincolnshire, UK: Ratite Conference Books; 2001. p. 23e9. ISBN 095275844. Bamelis FR, Tona K, De Baerdemaeker JG, Decuypere EM. Detection of early embryonic development in chicken eggs using visible light transmission. Br Poult Sci 2002;43(2):204e12. Brant AW, Norris KH, Chin G. A spectrophotometric method for detecting blood in white shelled eggs. Poult Sci 1953;32:357e63. Camps C, Christen D. On-tree follow-up of apricot fruit development using a handheld NIR instrument. J Food Agric Environ 2009;7(2):394e400. Careghi C, Tona K, Onagbesan O, Buyse J, Decuypere E, Bruggeman V. The effects of the spread of hatch and interaction with delayed feed access after hatch on broiler performance until seven days of age. Poult Sci 2005;84(8):1314e20. Cen H, He Y. Theory and application of near infrared reflectance spectroscopy in determination of food quality. Trends Food Sci Technol 2007;18(2):72e83. Chou CC, Jiang DD, Hung YP. Risk factors for cumulative mortality in broiler chicken flocks in the first week of life in Taiwan. Br Poult Sci 2004;45(5):573e7. Decuypere E, Tona K, Bruggeman V, Bamelis F. The day-old chick: a crucial hinge between broilers and breeders. World's Poult Sci J 2001;57(2):127e38. Decuypere E, Bruggeman V. The endocrine interface of environmental and egg factors affecting chick quality. Poult Sci 2007;86(5):1037e42. Gielen RMAM, de Jong LP, Kerkvliet HMM. Electro-optical blood-spot detection in intact eggs. IEEE Trans Instrum Meas 1979;28(3):177e83. Gonzales E, Kondo N, Saldanha ES, Loddy MM, Careghi C, Decuypere E. Performance and physiological parameters of broiler chickens subjected to fasting on the neonatal period. Poult Sci 2003;82(8):1250e6. Hassan GM, Nordskog AW. Effects of egg size and heterozygosis on embryonic growth and hatching speed. Genetics 1971;67:279e85. Hill D. The hatch window. In: Paper presented on the XXII Latin American Poultry Congress, Buenos Aires, Argentina, 6e9 September, 2011; 2011. Kemps BJ, Bamelis FR, De Ketelaere B, Mertens K, Tona K, Decuypere EM, et al. Visible transmission spectroscopy for the assessment of egg freshness. J Sci Food Agric 2006;86(9):1399e406. Kemps BJ, Bamelis FR, Mertens K, Decuypere EM, De Baerdemaeker JG, De Ketelaere B. Assessment of embryonic growth in chicken eggs by means of visible transmission spectroscopy. Biotechnol Prog 2010;26(2):512e6. Lourens A, van den Brand H, Meijerhof R, Kemp B. Effect of eggshell temperature during incubation on embryo development, hatchability, and posthatch development. Poult Sci 2005;84(6):914e20.
Please cite this article in press as: Islam MdH, et al., Prediction of chick hatching time using visible transmission spectroscopy combined with partial least squares regression, Engineering in Agriculture, Environment and Food (2014), http://dx.doi.org/10.1016/j.eaef.2014.10.001
6
Md.H. Islam et al. / Engineering in Agriculture, Environment and Food xxx (2014) 1e6
Swierenga H, de Weijer AP, van Wijk RJ, Buydens LMC. Strategy for constructing robust multivariate calibration models. Chemom Intell Lab Syst 1999;49(1): 1e17. Tona K, Bamelis F, De Ketelaere B, Bruggeman V, Moraes VM, Buyse J, et al. Effects of egg storage time on spread of hatch, chick quality, and chick juvenile growth. Poult Sci 2003;82(5):736e41.
Walstra I, Ten Napel J, Kemp B, van den Brand H. Temperature manipulation during layer chick embryogenesis. Poult Sci 2010;89(7):1502e8. Xiaobo Z, Jiewen Z, Xingyi H, Yanxiao L. Use of FT-NIR spectrometry in non-invasive measurements of soluble solid contents (SSC) of ‘Fuji’ apple based on different PLS models. Chemom Intell Lab Syst 2007;87(1):43e51.
Please cite this article in press as: Islam MdH, et al., Prediction of chick hatching time using visible transmission spectroscopy combined with partial least squares regression, Engineering in Agriculture, Environment and Food (2014), http://dx.doi.org/10.1016/j.eaef.2014.10.001
Contents lists available at ScienceDirect
Engineering in Agriculture, Environment and Food journal homepage: http://www.sciencedirect.com/eaef
Research paper
Prediction of chick hatching time using visible transmission spectroscopy combined with partial least squares regression Md. Hamidul Islam a, *, Naoshi Kondo a, Yuichi Ogawa a, Tateshi Fujiura a, Tetsuhito Suzuki a, Shusaku Nakajima a, Shinichi Fujitani b a b
Graduate School of Agriculture, Kyoto University, Kitashirakawa-Oiwakecho, Sakyo-ku, Kyoto 606-8502, Japan NABEL Co. Ltd., 86 Morimotocho, Nishikujo, Minami-ku, Kyoto 606-8444, Japan
a r t i c l e i n f o
a b s t r a c t
Article history: Received 7 March 2014 Received in revised form 10 September 2014 Accepted 5 October 2014 Available online xxx
The feasibility of using visible transmission spectroscopy for the prediction of chick hatching time was investigated. An experiment was conducted with 100 chicken eggs in which transmission spectra were measured between incubation day 0 (non-incubated) and days 8 and subsequent hatching time was recorded. Spectral transmittance in the range of 500e750 nm was used in analysis. Spectral data were linked to hatching time using a partial least squares (PLS) regression method. Different pre-processing procedures were compared. The calibration model using incubation day 4 spectra with multiplicative scatter correction (MSC) resulted in the lowest root mean square error of prediction (RMSEP) ¼ 3.41 h. The result indicates that the use of visible transmission spectroscopy combined with multivariate analysis has potential to predict the chick hatching time. © 2014, Asian Agricultural and Biological Engineering Association. Published by Elsevier B.V. All rights reserved.
Keywords: Hatching time Chicken egg Hatch window Visible spectroscopy Transmission PLS regression Calibration model
1. Introduction Day-old chicks are an important starting input for broiler farms, and the end product of the poultry hatchery. Overall success in broiler production depends on the quality of these chicks. The quality of day-old chicks is determined from their post-hatched survivability and growth potential, i.e. growth rate, breast meat yield and feed conversion ratio (Decuypere and Bruggeman, 2007). For broiler farmers, the aim is to obtain a batch of homogeneous, high quality chicks. However a batch of day-old chicks is frequently not homogeneous in quality due to the spread of the hatch window (Tona et al., 2003). Ideally, all chicks in a batch are desired to hatch at the same time, but in reality, chicks hatch at different moments within a time period called the “hatch window”. As a result of this broad hatch window, a production manager is dealing with chicks of different biological ages at take-off. Though the chronological ages of the chicks in a batch are defined by convention as being the same, the earlier hatched chicks are in reality older than one day. These chicks often show signs of dehydration at takeoff and are a poor quality stock for later
* Corresponding author. E-mail address: [email protected] (Md.H. Islam).
production. In addition, the delay in the collection of the chick batch also delays further hatchery procedures, such as sexing, vaccinations, packaging, and transportation, which also ultimately delays the first feed and access to water. Previous researchers have reported that the delay in the feed intake is associated with higher early mortality in chicks and impaired performance throughout the growth period (Chou et al., 2004; Gonzales et al., 2003). Therefore, a narrow hatch window is desired by poultry hatchery managers in order to produce the best quality chicks. However, it is very difficult for the hatchery managers to correctly estimate hatching time and consequently the spread of the hatch window. This is because the hatching time of individual chicks within a batch is influenced by multiple factors, such as the age of parent flock, egg handling, egg storage duration, and incubation conditions (Decuypere et al., 2001). Since a narrow hatch window results in higher quality chicks due to the smaller variation in the biological ages of chicks, hatchery managers undertake several techniques in order to obtain a narrow hatch window. The most common method used is to collect chicks that hatch in the first 24 h window. The main drawback of this method is, if the incubation is ended prematurely, eggs with viable chicks inside them are thrown away, decreasing the hatchability of the cohort and resulting in economic losses. Other techniques to optimize hatchability include exchanging the
http://dx.doi.org/10.1016/j.eaef.2014.10.001 1881-8366/© 2014, Asian Agricultural and Biological Engineering Association. Published by Elsevier B.V. All rights reserved.
Please cite this article in press as: Islam MdH, et al., Prediction of chick hatching time using visible transmission spectroscopy combined with partial least squares regression, Engineering in Agriculture, Environment and Food (2014), http://dx.doi.org/10.1016/j.eaef.2014.10.001
2
Md.H. Islam et al. / Engineering in Agriculture, Environment and Food xxx (2014) 1e6
position of the egg trays from cold to hot areas inside of the incubator and vice versa, increasing either temperature or carbondioxide (CO2) levels in the hatcher, as reviewed by Hill (2011). These practices can improve the hatchability of chicks to some extent, but cannot produce the best quality chick cohort, because individual chicks have their own natural hatch window and a short hatch window created by overheating the embryos or increasing CO2 level in the hatcher is not a natural hatch window. Therefore, development of a new hatch window control technique is demanded by the poultry hatchery in order to obtain the best quality chicks with high hatchability. Hatching time of chicks is closely related to embryonic development. The incubation length required by an egg is predetermined by this embryonic development. In fertile hatching eggs blood formation takes place from day 2 of incubation, as reviewed by Bamelis et al. (2002), and subsequent physiological changes occur in live embryos during the incubation period. Thus information about the stage of embryonic development could be used to estimate the incubation length needed by the egg to hatch. Nowadays, spectral measurements are commonly used to extract internal information about eggs, such as blood and meat spots, presence of embryo, embryonic growth, and egg freshness. Such spectral measurements have the advantage of being fast, nondestructive, and noncontact. Moreover, this technique can be implemented at a reasonable price. Preliminary investigations have revealed that significant variations in the spectra were found during incubation. Kemps et al. (2010) used visible transmission spectroscopy coupled with a multivariate analysis technique to assess embryonic weight in chicken eggs. They linked spectral information to embryonic weight using a partial least square (PLS) regression method and reported correlation coefficients of 0.97 based on the multiplicative scatter correction (MSC) spectra for the prediction of embryonic weight within the spectral range 570e750 nm. To our knowledge, however, no attempts have been made to estimate the hatching time of chicks using spectral information. The goal of this study is to investigate the potential to predict hatching time of chicks using visible transmission spectroscopy combined with an appropriate multivariate analysis method. A partial least square (PLS) regression algorithm was employed to build a calibration model for the prediction of hatching time of chicks. Spectral information in the range 500e750 nm was used to develop a calibration model. 2. Material and methods 2.1. Material A total of 100 light brown-shell chicken eggs laid by a commercial broiler breed (Ross 308 strain) were used in this study. All eggs were collected from a commercial poultry hatchery (Yamamoto Co. Ltd., Kameoka, Kyoto, Japan). At the moment of egg collection, the laying hens were 36 weeks of age. Since large variation in the size of eggs effects on hatching time, all samples were selected within a range of 42.5e44.5 mm (diameter), 54.4e59.4 mm (height) and 55.5e65.5 g (weight) to minimize their effects. In addition, to minimize the effect of egg shell pigmentation on transmission spectra, eggs with nearly similar shell color (selected by color image analysis method) were selected. Prior to incubation, all eggs were stored for 3 days following standard poultry hatchery practice (at 15.0 (±0.5) C and 80 (±5) % of relative humidity (RH)). 2.2. Spectral acquisition system The experimental setup used for the measurement of the transmission spectra of an intact egg is shown in Fig. 1. The egg is
positioned vertically in a plastic holder with the blunt end pointed upward between the illuminating fiber and the collecting optical fiber. The illuminating fiber was used for cool illumination of the eggs and the optical fiber was used to collect and transport the transmitted light to a Hamamatsu C 7473-36 model spectrometer (Hamamatsu Photonics K. K, Japan). The distance between the illuminating fiber and the optical fiber was kept at 100 mm to obtain a consistent transmission signal. A halogen light source (FHL-10, Asahi Spectra Co. Ltd., Japan) was used for illuminating the samples. This light source consisted of a dichroic reflector type halogen lamp (capable of cutting off infrared energy) in order to prevent warming of the egg surface where the light beam is shone onto the eggshell. The light was focused on the egg surface by an illuminating fiber of 5 mm in diameter and only the light that transmitted through the egg was received and transported by the optical fiber (1 mm effective light receiving diameter) to the spectrometer. The software package PMA-11 Spectral Analyzer for Windows (PMA Software U6039-01, Hamamatsu Photonics K.K., Japan) was used to control the spectral acquisition process. The transmission spectra of the eggs were measured over the spectral range of 200e950 nm at 1 nm intervals. The integration time for one scan was 100 ms and the spectrum of each egg contains an average of 10 scans. Since the characteristics of a halogen lamp changes over time, the spectrometer was calibrated before each measurement using a Teflon block (PTFE push rod Ø45 mm, Chukoh Chemical Industries, Ltd., Japan) of 30 mm thickness (Kemps et al., 2010). In addition, the reference spectra were measured after each 10 samples to evaluate any changes in the reference spectra with time. Furthermore, a correction was made for the electrical noise by taking the spectra with no light exposure to the spectrometer. All measurements were done inside a black box to shield any stray light from entering. The transmission values of light passing through an egg are expressed as a ratio of the amount of light passing through the eggs to the amount of light passing through the reference at the same wavelength (Kemps et al., 2006). The relative transmission (T) was calculated using Eq. (1).
TðlÞ ¼
SðlÞ RðlÞ
(1)
Where: T (l) is relative transmission at wavelength l nm S (l) is intensity of sample at wavelength l nm R (l) is intensity of Teflon reference at wavelength l nm It should be noted that the term “transmission spectra” used throughout this text refers to the relative transmission spectra. 2.3. Experimental design Prior to setting the eggs into the incubator, eggs were preheated for 16 h (first 6 h at 28 C and the remaining 10 h at 30 C) to bring the embryos to a uniform temperature when they are placed into the incubator. Just prior to incubation (referred to as “Day 0”), the transmission spectra of all eggs were measured. Upon completion of measurements, the eggs were immediately placed in a commercial incubator (SSH-02 all in one type, Showa Furanki Co. Ltd., Saitama, Japan) to incubate at 37.8 C and 55% of relative humidity according to Lourens et al. (2005). During incubation eggs were turned automatically every hour through an angle of 90 until incubation day 18. Between incubation day 1 to day 8, eggs were taken out from the incubator every 24 h to measure the spectral transmission of each egg. To minimize exposure time of the egg
Please cite this article in press as: Islam MdH, et al., Prediction of chick hatching time using visible transmission spectroscopy combined with partial least squares regression, Engineering in Agriculture, Environment and Food (2014), http://dx.doi.org/10.1016/j.eaef.2014.10.001
Md.H. Islam et al. / Engineering in Agriculture, Environment and Food xxx (2014) 1e6
3
Fig. 1. Experimental setup used for the measurement of transmission spectra.
outside the incubator, eggs were immediately placed back into the incubator after the measurements. At incubation day 19, eggs with living embryos were placed into hatcher baskets at 37.8 C and 60% of RH as mentioned by Walstra et al. (2010) and allowed to hatch. From incubation day 19, hatching chicks were monitored by a web camera to record the time of hatching for each egg until the hatching process ended. During the monitoring period, images were captured at 5 min intervals and saved automatically on the hard disk of a personal computer. The time of capture was also recorded simultaneously. Later, incubation time required by each egg to hatch was calculated as the duration between the time of emergence of the chick and the time when incubation started. 2.4. Data analysis The transmission spectra of eggs in the visible region from 500 to 750 nm were used for statistical analysis, because most of the transmitted light below 500 nm absorbed by the calciferous shell of the egg (Brant et al., 1953). Among 100 eggs, 1 egg was infertile and 2 eggs were failed to hatch. At first, relative transmission spectra of successfully hatched eggs were normalized by a peak normalization technique taking transmission at 500 nm as 1 to minimize the effect of variation in egg shell thickness on transmission spectra. Principal component analysis (PCA) was performed on the complete data set to detect outliers. After removal of outliers, samples were then divided randomly into two groups i.e. the calibration set and prediction set. The calibration sample set was used to develop a calibration model and prediction set was used for the validation of the calibration model. Hatching time of two sample sets are shown in Table 1. Several pre-processing procedures including moving average smoothing, Savitzky-Golay smoothing, multiplicative scatter correction (MSC), baseline offset correction, Savitzky-Golay first derivative were compared during this work. The averaging technique is used to reduce the number of wavelength or to smooth the spectra of the egg. It is also used to optimize the signal to noise ratio (Cen and He, 2007). The MSC transformation method is applied to compensate additive and multiplicative scatter effects included in the spectra. While baseline offset correction is done to correct the baseline shift of the spectra. The derivation of spectra is needed to remove baseline shift and superposed peaks (Swierenga
et al., 1999). These pre-processing operations were performed for both calibration and prediction sample sets. All types of spectral pre-processing were performed using The Unscrambler® software: version 9.8 (CAMO, Oslo, Norway). 2.5. Calibration model for the prediction of chick hatching time Partial least square (PLS) regression multivariate analysis was used to develop a calibration model for the estimation of chick hatching time. The PLS regression approach is one of the most popular chemometric algorithms for calibration model development due to its simplicity and small volume of calculations. The spectral data were linked to hatching time of chicks using a PLSR multivariate analysis to develop a calibration model. A low number of factors were desirable in order to avoid inclusion of any signal noise in the developed model (Xiaobo et al., 2007). The PLS regression analysis of the samples was performed using The Unscrambler® software. Initial PLS regression analysis was done on normalized spectra of each incubation day from day 3 to day 8 to develop the calibration models to select most effective incubation day. Further, PLS regression was done on selected incubation day normalized spectra combined with different pre-processing operations. During development of the calibration models, 15 principal components (PCs) were set up as a maximum. 2.6. Evaluation of the calibration model The prediction capacity of the developed calibration model was evaluated using several statistical parameters, including correlation coefficient (r) of prediction, root mean square error of calibration (RMSEC), root mean square error of prediction (RMSEP) and Bias. Bias is the mean difference between predicted and observed value of hatching time. A good calibration model should have a high correlation coefficient (r) and a low RMSEP. In addition, the difference between RMSEC and RMSEP should be small for a good calibration model (Camps and Christen, 2009). The quality of the final model was evaluated according to lowest difference between correlation coefficient of calibration and prediction set, lowest difference between RMSEC and RMSEP and lowest bias value. 3. Results and discussion
Table 1 Hatching time of calibration and prediction sample sets used for data analysis. Items
Calibration set
Prediction set
Number of samples Range (hour) Mean (hour) Standard deviation
60 483.1e497.75 483.1 5.97
35 483.3e497.5 483.3 6.25
3.1. Characteristics of visible transmission spectra of an incubating egg Typical transmission spectra of an incubating egg at different incubation times in the spectral range 500e750 nm are shown in Fig. 2. The overall shape of the incubated egg spectra was
Please cite this article in press as: Islam MdH, et al., Prediction of chick hatching time using visible transmission spectroscopy combined with partial least squares regression, Engineering in Agriculture, Environment and Food (2014), http://dx.doi.org/10.1016/j.eaef.2014.10.001
4
Md.H. Islam et al. / Engineering in Agriculture, Environment and Food xxx (2014) 1e6
qualitatively examined in order to illustrate the influence of embryonic development on the spectral profile. From Fig. 2, it is apparent that the transmission spectra shifted up and down depending on the incubation length. The most significant spectral difference appeared around 570e590 nm, 620e630 nm, and 640e650 nm when considering the different incubation lengths. The large decreases in the transmission spectra at 589 and 643 nm prior to incubation day 4 are due to light absorbed by the shell color pigment “protoporphyrin IX” (Gielen et al., 1979). However, the changes in the shape of the transmission spectra around 570e580 nm from incubation day 4 onward are because of the light absorbed by hemoglobin, which is the detected pigment in the blood (Bamelis et al., 2002). The shifting down of the transmission spectra from incubation day 3 to onward are related to the formation of sub embryonic fluid (SEF). With the formation of SEF, the albumen of the egg gradually lost water and became thicker reported by Baggott (2001). This thick albumen reduced the transmission of light through it and results in lower signal received at the detector. Therefore, the spectral differences displayed in Fig. 2 are mainly dependent on differences in the volumetric fractions of water in the albumin of eggs at the different incubation times.
Table 2 PLS regression results for the prediction of chick hatching time for different preprocessing techniques at incubation day 4. Pre-processing method
PCs
rcal
RMSEC
rpred
RMSEP
Bias
Normalized spectra Moving Ave. smoothinga SG smoothingb MSC Baseline offset SG first derivativeb
7 7 7 4 7 6
0.872 0.878 0.878 0.870 0.871 0.914
2.81 2.74 2.75 2.94 2.83 2.34
0.888 0.885 0.882 0.873 0.884 0.848
3.49 3.53 3.63 3.41 3.71 3.96
0.753 0.756 0.761 0. 651 0.865 0.992
rcal and rpred are the correlation coefficient for calibration and prediction set, respectively. SG: Savitzky-Golay method. a Number of segments: 5. b Number of segments: 5, Polynomial order: 2.
spectra based model was the less complex because the number of principal components was minimal. This result indicates that visible transmission spectroscopy has the potential for the estimation of chick hatching time at early stage of incubation. The scatter plot between actual and predicted hatching time of the best calibration model is presented in Fig. 3.
3.2. Calibration model for the prediction of hatching time 3.3. Validation of the calibration model A PLS regression method was used to develop the calibration model for the estimation of chick hatching time. Spectral information in the range of 500e750 nm was used for developing the calibration model and for validating the model. Initial PLS regression results (correlation coefficient obtained from incubation days 3 to days 8 are 0.536, 0.872, 0.145, 0.168, 0.157 and 0.128 respectively) reveal that spectral information at incubation day 4 is more effective to predict chick hatching time. Therefore, in rest of the manuscript we discussed only the results that are obtained from incubation day 4 transmission spectra. The PLS regression results for the prediction of chick hatching time using different preprocessing procedures at incubation day 4 is shown in Table 2. At first raw spectra was normalized and then other pre-processing operations were done on the normalized spectra. From Table 2, it is apparent that all calibration models had a good correlation coefficient i.e. r ¼ 0.870e0.914. The RMSEC values varied from 2.34 to 2.94 h. The RMSEC value less than 3 h is acceptable for the hatchery manager because, in practice the hatch window spans from 24 to 48 h (Careghi et al., 2005). Among all calibration models, the MSC
The capability of the developed calibration models for the estimation of chick hatching time of the independent sample was confirmed. For this purpose, transmission spectra of 35 chicken eggs (with different hatching time) were used to predict hatching time. The RMSEP obtained from different calibration models were varied from 3.41 to 3.96 h. As seen in Table 2, the calibration model using MSC spectra resulted in the lowest RMSEP ¼ 3.41 h and a high r ¼ 0.873. In addition, this calibration model had the lowest difference between RMSEC and RMSEP and lowest bias value. Also the difference between correlation coefficient of calibration and prediction is minimal for the MSC spectra based model. Hence, this calibration model was selected as the best calibration model for the prediction of chick hatching time. The scatter plot between actual and predicted hatching time using the MSC spectra based calibration model is presented in Fig. 4. From the figure it is observed that some samples have prediction error above 5 h, this is because of individual egg weight difference (Hassan and Nordskog, 1971). This result shows that a calibration model for prediction of chick
Fig. 2. Typical transmission spectra of an incubating egg at different incubation time in visible region.
Fig. 3. Scatter plot between actual and predicted value of hatching time for the calibration model developed using incubation day 4 spectra with MSC.
Please cite this article in press as: Islam MdH, et al., Prediction of chick hatching time using visible transmission spectroscopy combined with partial least squares regression, Engineering in Agriculture, Environment and Food (2014), http://dx.doi.org/10.1016/j.eaef.2014.10.001
Md.H. Islam et al. / Engineering in Agriculture, Environment and Food xxx (2014) 1e6
5
there is no concrete evidence, further investigation is needed to clarify this finding. 4. Summary and conclusions
Fig. 4. Scatter plot between actual and predicted value of hatching time in validation step. Predicted value of hatching time was calculated using MSC spectra based calibration model.
hatching time using visible transmission spectroscopy was successfully developed and was well validated. 3.4. Evaluating the structure of the calibration model The PLS regression was also a useful method for investigating the location of information related to the hatching time. In order to clarify the behavior of the calibration model, the regression coefficient of the best calibration model for the prediction of hatching time of chick was plotted against wavelength as demonstrated in Fig. 5. From this, it is apparent that around 577 nm a very slight peak is observed which indicates that the formation of the pigment hemoglobin has a minor contribution to the prediction of the hatching time of chick. From Fig. 5, it can be seen that the spectral range 610e750 nm makes an important contribution to the prediction of hatching time. The spectral range 610e750 nm is associated with the shifting of albumen water content, as mentioned in an earlier section (see explanation for Fig. 2). The hatching time of chicks influenced by the embryonic growth rate, which depends on the formation of SEF, as reported by Baggott (2001). From this result, the prediction of hatching time of chicks in the visible region is mainly driven by the SEF because formation of SEF makes albumen of egg thicker by reducing water content. However, since
Fig. 5. Regression coefficients of the calibration model used for the prediction of chick hatching time.
In this study, the potential of using visible transmission spectroscopy for the prediction of hatching time of chicks has been demonstrated. Spectral information combined with PLSR multivariate analysis was performed to develop a calibration model and for validation of the model. The transmission spectra (500e750 nm) through intact eggs were applied to develop the prediction model. In the subsequent analysis, relative spectra were normalized with a selected waveband to exclude the egg shell thickness effect and then several pre-processing procedures were applied on the spectra. The calibration model using incubation day 4 spectra with MSC achieved a satisfactory result i.e. RMSEP ¼ 3.41 h and r ¼ 0.873. These results indicate that visible transmission spectroscopy combined with multivariate analysis has significant potential to predict chick hatching time. Furthermore, early prediction of hatching time for individual eggs will enable hatchery managers to more finely control the spread of the hatch window, and subsequently the profitability of broiler production. Although this result is very promising, further study is needed for an example with heterogeneous egg sample (size, color, different storage condition). Acknowledgments The authors express their gratitude to Professor Garry Piller, Graduate School of Agriculture, Kyoto University for editing and proof reading of this manuscript. References Baggott GK. Development of extra-embryonic membranes and fluid compartments. In: Deeming DC, editor. Perspectives in fertilization and embryonic development in poultry. Lincolnshire, UK: Ratite Conference Books; 2001. p. 23e9. ISBN 095275844. Bamelis FR, Tona K, De Baerdemaeker JG, Decuypere EM. Detection of early embryonic development in chicken eggs using visible light transmission. Br Poult Sci 2002;43(2):204e12. Brant AW, Norris KH, Chin G. A spectrophotometric method for detecting blood in white shelled eggs. Poult Sci 1953;32:357e63. Camps C, Christen D. On-tree follow-up of apricot fruit development using a handheld NIR instrument. J Food Agric Environ 2009;7(2):394e400. Careghi C, Tona K, Onagbesan O, Buyse J, Decuypere E, Bruggeman V. The effects of the spread of hatch and interaction with delayed feed access after hatch on broiler performance until seven days of age. Poult Sci 2005;84(8):1314e20. Cen H, He Y. Theory and application of near infrared reflectance spectroscopy in determination of food quality. Trends Food Sci Technol 2007;18(2):72e83. Chou CC, Jiang DD, Hung YP. Risk factors for cumulative mortality in broiler chicken flocks in the first week of life in Taiwan. Br Poult Sci 2004;45(5):573e7. Decuypere E, Tona K, Bruggeman V, Bamelis F. The day-old chick: a crucial hinge between broilers and breeders. World's Poult Sci J 2001;57(2):127e38. Decuypere E, Bruggeman V. The endocrine interface of environmental and egg factors affecting chick quality. Poult Sci 2007;86(5):1037e42. Gielen RMAM, de Jong LP, Kerkvliet HMM. Electro-optical blood-spot detection in intact eggs. IEEE Trans Instrum Meas 1979;28(3):177e83. Gonzales E, Kondo N, Saldanha ES, Loddy MM, Careghi C, Decuypere E. Performance and physiological parameters of broiler chickens subjected to fasting on the neonatal period. Poult Sci 2003;82(8):1250e6. Hassan GM, Nordskog AW. Effects of egg size and heterozygosis on embryonic growth and hatching speed. Genetics 1971;67:279e85. Hill D. The hatch window. In: Paper presented on the XXII Latin American Poultry Congress, Buenos Aires, Argentina, 6e9 September, 2011; 2011. Kemps BJ, Bamelis FR, De Ketelaere B, Mertens K, Tona K, Decuypere EM, et al. Visible transmission spectroscopy for the assessment of egg freshness. J Sci Food Agric 2006;86(9):1399e406. Kemps BJ, Bamelis FR, Mertens K, Decuypere EM, De Baerdemaeker JG, De Ketelaere B. Assessment of embryonic growth in chicken eggs by means of visible transmission spectroscopy. Biotechnol Prog 2010;26(2):512e6. Lourens A, van den Brand H, Meijerhof R, Kemp B. Effect of eggshell temperature during incubation on embryo development, hatchability, and posthatch development. Poult Sci 2005;84(6):914e20.
Please cite this article in press as: Islam MdH, et al., Prediction of chick hatching time using visible transmission spectroscopy combined with partial least squares regression, Engineering in Agriculture, Environment and Food (2014), http://dx.doi.org/10.1016/j.eaef.2014.10.001
6
Md.H. Islam et al. / Engineering in Agriculture, Environment and Food xxx (2014) 1e6
Swierenga H, de Weijer AP, van Wijk RJ, Buydens LMC. Strategy for constructing robust multivariate calibration models. Chemom Intell Lab Syst 1999;49(1): 1e17. Tona K, Bamelis F, De Ketelaere B, Bruggeman V, Moraes VM, Buyse J, et al. Effects of egg storage time on spread of hatch, chick quality, and chick juvenile growth. Poult Sci 2003;82(5):736e41.
Walstra I, Ten Napel J, Kemp B, van den Brand H. Temperature manipulation during layer chick embryogenesis. Poult Sci 2010;89(7):1502e8. Xiaobo Z, Jiewen Z, Xingyi H, Yanxiao L. Use of FT-NIR spectrometry in non-invasive measurements of soluble solid contents (SSC) of ‘Fuji’ apple based on different PLS models. Chemom Intell Lab Syst 2007;87(1):43e51.
Please cite this article in press as: Islam MdH, et al., Prediction of chick hatching time using visible transmission spectroscopy combined with partial least squares regression, Engineering in Agriculture, Environment and Food (2014), http://dx.doi.org/10.1016/j.eaef.2014.10.001