Worldwide trends in animal welfare have resulted in an increased interest in individual management of sows housed in groups within hog barns. Estrus detection has been shown to be one of the greatest determinants of sow productivity. Purpose: We conducted this study to develop a method that can auto...
Worldwide trends in animal welfare have resulted in an increased interest in individual management of sows housed in groups within hog barns. Estrus detection has been shown to be one of the greatest determinants of sow productivity. Purpose: We conducted this study to develop a method that can automatically detect the estrus state of a sow by selecting optimal texture parameters from images of a sow's pudendum and by optimizing the number of neurons in the hidden layer of an artificial neural network. Methods: Texture parameters were analyzed according to changes in a sow's pudendum in estrus such as mucus secretion and expansion. Of the texture parameters, eight gray level co-occurrence matrix (GLCM) parameters were used for image analysis. The image states were classified into ten grades for each GLCM parameter, and an artificial neural network was formed using the values for each grade as inputs to discriminate the estrus state of sows. The number of hidden layer neurons in the artificial neural network is an important parameter in neural network design. Therefore, we determined the optimal number of hidden layer units using a trial and error method while increasing the number of neurons. Results: Fifteen hidden layers were determined to be optimal for use in the artificial neural network designed in this study. Thirty images of 10 sows were used for learning, and then 30 different images of 10 sows were used for verification. Conclusions: For learning, the back propagation neural network (BPN) algorithm was used to successful estimate six texture parameters (homogeneity, angular second moment, energy, maximum probability, entropy, and GLCM correlation). Based on the verification results, homogeneity was determined to be the most important texture parameter, and resulted in an estrus detection rate of 70%.
Worldwide trends in animal welfare have resulted in an increased interest in individual management of sows housed in groups within hog barns. Estrus detection has been shown to be one of the greatest determinants of sow productivity. Purpose: We conducted this study to develop a method that can automatically detect the estrus state of a sow by selecting optimal texture parameters from images of a sow's pudendum and by optimizing the number of neurons in the hidden layer of an artificial neural network. Methods: Texture parameters were analyzed according to changes in a sow's pudendum in estrus such as mucus secretion and expansion. Of the texture parameters, eight gray level co-occurrence matrix (GLCM) parameters were used for image analysis. The image states were classified into ten grades for each GLCM parameter, and an artificial neural network was formed using the values for each grade as inputs to discriminate the estrus state of sows. The number of hidden layer neurons in the artificial neural network is an important parameter in neural network design. Therefore, we determined the optimal number of hidden layer units using a trial and error method while increasing the number of neurons. Results: Fifteen hidden layers were determined to be optimal for use in the artificial neural network designed in this study. Thirty images of 10 sows were used for learning, and then 30 different images of 10 sows were used for verification. Conclusions: For learning, the back propagation neural network (BPN) algorithm was used to successful estimate six texture parameters (homogeneity, angular second moment, energy, maximum probability, entropy, and GLCM correlation). Based on the verification results, homogeneity was determined to be the most important texture parameter, and resulted in an estrus detection rate of 70%.
* AI 자동 식별 결과로 적합하지 않은 문장이 있을 수 있으니, 이용에 유의하시기 바랍니다.
문제 정의
This study is being developed on group breeding control and management system, performed to judge their estrus condition in breeding facilitate as one part of research. The estrus detecting system was designed to detect the estrus condition by a sow image from the camera at the first step, and in the physical method at the second step (Seo et al.
제안 방법
Changes in images due to events associated with estrus, including pudendum hyperemia and enlargement, and increased mucus secretion, were analyzed to discriminate estrus using an artificial neural network and to determine the most important texture parameter. A total of 60 images were obtained from 20 sows.
Learning was successful with six texture parameters: homogeneity, angular second moment, energy, maximum probability, entropy, and GLCM correlation, but not with contrast and dissimilarity. Contrast and dissimilarity were the parameters used to analyze the differences in brightness between pixels. Errors occurred when the images were acquired in different environments.
Algorithms to determine the optimal number of neurons in the hidden layer have been studied by various authors. In our study, we determined the number of neurons that yielded the highest discrimination rate using repeated learning and validation tests. We varied the number of neurons in the hidden layer from 2 to 20, and selected the number of units that resulted in the highest discrimination.
In this study, the image histograms that were normalized to have values between 0 and 1 using scaling maximum and minimum values in the texture parameter images were classified using increments of 0.1. Accordingly, a single texture map contained a total of 144 data points, and 10 data points which means grades by texture histogram at 0.
Because camera images of estrus state are used by leaning input-data, Thirty images from 10 sows make sense. Learning was successful with six texture parameters: homogeneity, angular second moment, energy, maximum probability, entropy, and GLCM correlation, but not with contrast and dissimilarity. Contrast and dissimilarity were the parameters used to analyze the differences in brightness between pixels.
Texture parameters in the images were analyzed to examine the relationship between neighboring pixels and the repeated pixel patterns to detect signs of estrus. A gray level co-occurrence matrix (GLCM) algorithm was used to analyze the texture parameters.
After the learning of 30 texture histogram data, the artificial neural network was verified using 30 images. The learning was successful with six texture parameters such as homogeneity, angular second moment, energy, maximum probability, entropy, and GLCM correlation, but not with contrast and dissimilarity, which were based on brightness data. The number of neurons in the hidden layer ranged from 2 to 20, and the number of units with the highest discrimination rate was selected at 15.
The neural network used in this study was a multi-layer neural network that consisted of input, hidden, and output layers. The number of hidden layer units has been shown to be an important parameter for data learning and accurate identification.
The pudendum images were converted into 16-level gray scale images using red-values by considering that sows in estrus have an enlarged pudendum, pudendum hyperemia, and increased mucus secretion. Figure 2 shows the grade that was used to acquire and process estrus site images for texture analysis.
, 2010). Therefore, the research verified the previous study and also improved the method to increase the success rate of the estrus detection of sows. Sows which were come into estrus after parturition for one month were surveyed and measured their estrus at a farm in Korea.
However, it doesn’t use camera images but activity and estrus days of cow as input-data. We therefore performed this current study to determine better parameters for detecting estrus by analyzing texture parameters in anestrus and estrus images of the same sow with the ultimate goal of detecting estrus using an artificial neural network with an optimized number of neurons in the hidden layer.
대상 데이터
A 640 × 480 color CCD camera that was situated 50 cm from the sows was used to acquire the images.
Design of the estrus discrimination network. The network consists of 10 input layer units, 15 hidden layer units, and 1 output layer unit.
Estrus images were obtained from the same sows after 5 days. The subject sows were housed in the stalls used for artificial insemination at a pig farm in Hapchen, Gyeongsangnamdo, Korea. The skilled farmer determined whether the sow was in estrus or anestrus manually.
To discriminate the estrus state, images of the same sow in estrus and anestrus were acquired, and the analyzed textures were processed via the neural network system. Thirty images from 10 sows were used for learning. Research about estrus of cows which used neural networks (Firk et al.
데이터처리
The estrus detection sites were divided into 5 × 5 pixel local sites. Computations were conducted for eight GLCM descriptors: contrast and dissimilarity represented differences in brightness between pixels; homogeneity represented the uniformity of pixels; angular second moment and energy represented the uniformity of brightness; maximum probability represented the directivity of texture according to the frequency of brightness; entropy represented the existence of feature points according to the randomness of brightness distribution; and correlation represented the changes between pixels. For a matrix with a local area of i × j (Pij), the eight descriptors of the GLCM algorithm were calculated using equations 1-8 (Parker, 1997).
이론/모형
Texture parameters in the images were analyzed to examine the relationship between neighboring pixels and the repeated pixel patterns to detect signs of estrus. A gray level co-occurrence matrix (GLCM) algorithm was used to analyze the texture parameters. The GLCM algorithm is a statistical analysis method that depicts the pixel patterns numerically.
성능/효과
The number of neurons in the hidden layer ranged from 2 to 20, and the number of units with the highest discrimination rate was selected at 15. In the verification results, the estrus discrimination rate using the texture parameter homogeneity was highest at 70%.
Thus, this texture parameter is not suitable for estrus state discrimination. The discrimination results using homogeneity and angular second moment reflected true positives (a sow in estrus was determined as a anestrus), but homogeneity was the optimal texture parameter because it had the highest total discrimination success rate. The discrimination rate of our neural network as a whole was lower than that reported in other studies, but our neural network has good reliability because the estrus states for 10 sows were correctly discriminated.
참고문헌 (17)
Artmann, R. 1999. Electronic identification systems: state of the art and their further development. Computers and Electronics in Agriculture 24 (1-2):5-26.
Backus, G. B. C., H. M. Vermeer, P. F. M. M. Roelofs, P. C. Vesseur, J. H. A. N. Adams, G. P. Binnendijk, J. J. J. Smeets, C. M. C. PeetSchwering and F. J. F. J. Wilt. 1997. Comparative study of four housing systems for nonlactating sows. Proceedings of the Fifth International Symposium. Livestock environment Volume 5(2) 1997-279.
Bates, R. O., D. B. Edwards and R. L. Korthals. 2003. Sow performance when housed either in groups with electronic sow feeders or stalls 1. Livestock Production Science 79 (1):29-35.
Kim, D. J., S. C. Yeon and H. H. Chang. 2007. Development of a device for estimating the optimal artificial insemination time of individually stalled sows using image processing. Journal of Animal Science and Techonology (Kor.) 49 (5):677-688.
Kim, D. W., B. R. Min, K. W. Seo, M. Y. Lee, J. T. Hong, Y. I. Hwa, D. H. Kam, J. K. Kim and D. W. Lee. 2010. The Image Texture Analysis for Estrus Detecting of Sows. Proceedings of the Korean Society for Agricultural Machinery Conference 15 (1):45-49.
Eradus, W. J. and M. B. Fansen. 1999. Animal identification and monitoring. Computers and Electronics in Agriculture 24 (1-2):91-98.
Firk, R., E. Stamer, W. Junge and J. Krieter. 2002. Automation of oestrus detection in dairy cows: a review. Livestock Production Science 75 (3):219-232.
Freson, L., S. Godrie, N. Bos, J. Jourquin and R. Geers. 1998. Validation of an infra-red sensor for oestrus detection of individually housed sows. Computers and Electronics in Agriculture 20 (1):21-29.
Geers, R. 1994. Electronic monitoring of farm animals: a review of research and development requirements and expected benefits, Computers and Electronics in Agriculture 10 (1):1-9.
Jeon, J. H., S. C. Yeon and H. H. Chang. 2005. Comparative analysis for general and estrus-related vocalizations in sows. Journal of Animal Science and Techonology 47 (1):133-140.
Seo, K., B. Min and D. Lee 2006. The detection of esophagitis by using back propagation network algorithm. Journal of Mechanical Science and Technology 20(11):1873-1880.
Seo, K., B. Min, H. Kim, S. Lee and D. Lee 2008. Classification of esophagitis grade by neural network and texture analysis. Journal of Mechanical Science and Technology. 22(12):2475-2480.
Seo, K. W., B. R. Min, D. W. Kim, M. Y. Lee, D. H. Kam, J. K. Kim and D. W. Lee. 2010. The estrus auto-Detecting system of sows for welfare rearing managements. Proceedings of the Korean Society for Agricultural Machinery Conference 15 (1):355-359.
Teoh, E. J., C. Xiang and C. T. Kay. 2006. Estimating the number of hidden neurons in a feedforward network using the singular value decomposition. Lecture Notes in Computer Science 3971(1):858-865.
Parker, J. R. 1997. Algorithms for Image Processing and Computer Vision. Wiley. 150-175.
Shuxiang, X. and C. Ling. 2008. A novel approach for determining the optimal number of hidden layer neurons for FNN's and its application in data mining. 5th International Conference on Information Technology and Applications 683-686.
Stuyft, E., C. P. Schofield, J. M. Randall, P. Wambacq and V. Goedseels, 1991. Development and application of computer vision systems for use in livestock production. Computers and Electronics in Agriculture 6 (3):243-265.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.