Heuristic Research on Mathematical Models of Air Refueling Support

Available online at www.sciencedirect.com

Procedia Environmental Sciences 10 (2011) 748 – 752

2011 3rd International Conference on Environmental www.elsevier.com/locate/procedia Science and Information Application Technology (ESIAT 2011)

Analysis of Image Fusion of TM and CBERS Based on Pixel Level Wu-xue Cheng*1.a, Jie-ming Zhou1.b, Cun-jian Yang1.c, Wan-cun Zhou2.d 1

Key Laboratory of Land Resources Evaluation and Monitoring in Southwest, Ministry of Education,Chengdu, China 2 Institute Of Mountain Hazards And Environment, CAS Cheng du ˈChina a [email protected] ,[email protected] ,[email protected] d [email protected]

Abstract At presentˈthe research in the field of remote sensing image fusion is frequently. But the case of remote image fusion research by TM and CBERS image is less. In this paper, firstly, author apply six image fusion methods fused the TM and CBERS image in Anju district, Sichuan province. Secondly, author using the programming which writing by matlab software for image fusion quality evaluation .The results show that the qulity fused by WAV is the best one in six methods. The results of research can supply a good basic data for recognition of remote sensing. Keywords:Image fusion; matlab; Quality evaluation; Fusion methods

1. Preface Image fusion of remote sensing is the process of generating new images, which according to certain algorithm based on the same coordinate system. It not only improve the spatial resolution of multispectral images ,but also save the spectrum characteristics of images. So, the results of image fusion can provide good data for image pattern recognition. At present, image fusion mainly divided into three categories, namely pixel level, feature level and decision level. This research mainly applied in the method of pixel level to fusion. Then compile program to evaluate the quality of fusion results. 2. Data Resource In this study, the study area is Anju district ,SuiNingShi ,in sichuan province. We mainly applied the image come from China-Brazil Earth Resource Satellite(short for CBERS) and Landsat satellite of study area in 2000 years as the study data. The resolution of CBERS data is 20 meters , and its scanning width is113km; The cycle is repeated observations is 26 days because of CCD camera extracellularly function, observation with same area of the shortest period can be for 3 days. TM image data from the U.S. satellite, the Landsat each piece ground cover area of 185km x 185km, the spatial resolution is 30m, and have seven band, the band of six have loss, so only six bands of TM data was used for study.

1878-0296 © 2011 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of Conference ESIAT2011 Organization Committee. doi:10.1016/j.proenv.2011.09.121

Wu-xue Cheng et al. / Procedia Environmental Sciences 10 (2011) 748 – 752

3. The Technological Process and Fusion Methods 3.1. Technical processes The technical process as shown in fig.2, data pretreatment mainly including data geometric correction, and image enhanced imaging and so on. Data fusion methods The main data fusion methods of this paper including Wavelet transform method(short for WT), Multiplication transform method(short for MT), High-pass filter method(short for HPF) ,Brovey transform method(short for BT), intensity, hue and saturation method(short for IHS). Low special resolution image

High special resolution image

Data preprocessing

Data preprocessing

Fusion model

Comprehensive evaluation

Fusion results Figuer1. Data fusion technology process 4. The Fusion Results and Quality Evaluation 4.1. Fusion results All kinds of image fusion is done with software of ERDAS, specific results as shown in figure 2.

Figuer2 .Fusion results

749

750

Wu-xue Cheng et al. / Procedia Environmental Sciences 10 (2011) 748 – 752

4.2. The quality evaluation The six indexes are applied to evaluate the quality of fusion results. There are the mean, standard deviation(SD), information entropy(IE), spectrum fidelity(SF), correlation coefficient(CC) and square root error(RMSE). Generally speaking, as the mean increase the fusion image contains the image information is more higher; As the standard deviation change more bigger, the grayscale of the fusion image became more disperse, the contrast became more big, and the information change more abundant.Information entropy is bigger show the contents of image is more abundant. The fidelity of spectrum is more bigger indicate the quality of image is better. The correlation coefficient is bigger indicate the high-frequency information of image is more abundant. At last, if the value of RMSE is less indicate the high-frequency information of high-resolution image contain more in new fusion image. In this study mainly applied the program of matlab software to evaluated the quality of fusion image. The detailed code shown as below: [fliename1, pathname1]= uigetfile({'*.tif';'*.jpg';'*.bmp'},' Select the source images '); str1 = [pathname1 fliename1];im1 = imread(str1);im1 = double(im1); [fliename2, pathname2]= uigetfile({'*.tif';'*.jpg';'*.bmp'},' Choose fusion image '); str2 = [pathname2 fliename2];im2 = imread(str2);im2 = double(im2); %im2= imresize(im2,[size(im1,1),size(im1,2)],'bilinear'); [m1,n1,p1] = size(im1);[m2,n2,p2] = size(im2);sum1=zeros(m1,n1); sum2=zeros(m2,n2); for i=p1 sum1 = im1(:,:,i) + sum1;end im1_a = sum1/p1; for i=p2 sum2 = im2(:,:,i) + sum2;end im2_a = sum2/p2;im1_aver = mean(im1_a(:));im2_aver = mean(im2_a(:)); im1_std = std(im1_a(:));im2_std = std(im2_a(:));im1_Entropy = H(im1_a); im2_Entropy = H(im2_a);d = (im2_a - im1_a)./im1_a;[m,n]=size(d); D = sum(d(:))/(m*n); % A = (im2_a - im2_aver).*(im1_a - im1_aver); % B = ((im2_a - im2_aver).^2).*((im1_a - im1_aver).^2); % P = sum(A(:))./(sqrt(sum(B(:)))); P = corrcoef(im1_a,im2_a);p = P(1,2); RMSE = sqrt(sum((im2_a(:)-im1_a(:)).^2)/((m2-1)*(n2-1))); result1 = [im1_aver,im1_std,im1_Entropy]; result2 = [im2_aver,im2_std,im2_Entropy]; result3 = [D,p,RMSE]; fprintf('\n mean standard deviation entropy\n'); fprintf('\n source image˖ %.4f %.4f %.4f ',result1); fprintf('\n fusion image˖ %.4f %.4f %.4f \n\n',result2); fprintf('\n fidelity correlation coefficient RMSE\n'); fprintf('\n two pieces of image˖ %.4f %.4f %.4f \n\n\n',result3); [m,n]=size(h);h=uint8(h);X=zeros(1,256);result=0; for i=1:m for j=1:n X(h(i,j)+1)=X(h(i,j)+1)+1; End end

751

Wu-xue Cheng et al. / Procedia Environmental Sciences 10 (2011) 748 – 752

for k=1:256 P(k)=X(k)/(m*n); if (P(k)~=0) result=P(k)*log2(P(k))+result; end end result=-result; f=result; Table1 The quality evaluation results of fusion image IMA ME S I S C R RESU GE AN D E F C MSE LTS 62.8 6 4 SI 5 .03 .52 12.6 2 3 0 0 50. HPF 11.49 6 .87 .50 .8 .63 08 28.0 1 2 0 0 34. IHS 11.30 5 .96 .70 .56 .76 88 46.2 5 4 0 0 16. BT 12.17 4 .13 .12 .02 .97 57 17.5 3 3 0 0 45. PC 11.73 0 .12 .57 .05 .79 41 WA 64.5 6 3 0 0 8.8 13.95 V 7 .36 .77 .38 .16 6 MU 16.3 5 4 0 0 46. 12.09 L 3 .00 .32 .27 .33 55 5. Conclusion and Discussion The fusion methods of this study includes the most algorithm of the pixel level. The results of evaluate by the weighted average method showed the methods of highest score is wavelet transform, which is the best fusion algorithm for TM and CBERS. The fusion effect from high to low in turn is BT algorithm, MUL algorithm, PC algorithm, HPF algorithm and HIS algorithm. All the fusion results show in figure3. The same weight value to every data in this paper. So the fusion quality effect intensity of evaluation indexes is insufficient responses. We should be in-depth in this field in later research. This study is based on the TM images of 20 meters resolution and CBERS images of 30 meters resolution for image fusion with pixel fusion methods, not involves character class and decision-level fusion methods. In the later study, we should try these methods.

0 H D Q

       







 





+3)

,+6

%7

3&

,PDJH )XVLRQ0HWKRGV

A. Mean contrast diagram







RXUFH







:$9

08/



RXUFH ,PDJH

+3)

,+6

%7

3&

)XVLRQ0HWKRGV

B. Standard deviation contrast diagram

:$9

08/

752

Wu-xue Cheng et al. / Procedia Environmental Sciences 10 (2011) 748 – 752

Acknowledge Thanks all the teachers to help me in writing this paper! *[email protected]; phone:13880988540 Sichuan normal university’s University-level research projects ˖Analysis of landuse change based on the multi-source remote sensing images and the BP neural network model in Chengdu city˄124197˅. Sichuan normal university’s key research projects ˖ Analysis of Remote Sensing Quantitative Estimation of Forest Volume Based on Sublot˄037265˅ Fund Project: (NSFC) Study On The Multi-source Remote Sensing Renewal Of Forest Resources Database Based On Small Class˄40771144˅ References [1]Dang Anrong, Wang Xiaodong, Chen Xiaofeng. Remote sensing image processing method with Erdas Image [M]. Beijing: Tsinghua University Press, 2003. [2]Jia Yonghong. Digital Image Processin[M]. Wuhan: Wuhan University Press, 2003. [3]Wangyi, Hushen, Sun Jiabing. The multi-source remote sensing image quality evaluation method research [J]. Bulletin of Surveying and Mapping, 2002,˄5˅: 15-18. [4]Pohl CˈVan Garderen J L.Multisensor Image Fusion in Remote Sensing:Concepts,Methods,Application[J].Int J Remote Sensing,1998,19(5):362~367. [5]Wang Haihun, Peng Jiaxiong, Wu Wei.A Study of Evaluation Methods on Performance of the MultiSource Remote Sensing Image Fusion [J]. Computer Engineering and Applications, 2003, 39(25) :33~37.









 



 



  

  



 



   





 





 +3)

,+6

%7

3&

:$9

+3)

08/

)XVLRQ 0HWKRGV

C. Entropy value contrast diagram

,+6





%7

3&

:$9

08/

)XVLRQ 0HWKRGV

D. Fidelity of contrast diagram

   

  



     

  







  

 

  

   

   



   

 





 









 ⑤೒‫ڣ‬

 



































 







 



+3)

,+6

%7

3&

)XVLRQ0HWKRGV

:$9

08/



+3)

,+6

%7

3&

:$9

)XVLRQ0HWKRGV

E. Correlation coefficient contrast diagram F. RMS error comparison chart Fig.3 Results contrast of Image fusion quality evaluation

08/