Recognition Method Of Identification And Drinking Behavior For Individual Pigs Based On Machine Vision

With continuous improvement of the pig industry in the scale and intelligent,intelligent video surveillance technology is being widely used and studied.Traditional pig industry requires breeders to obtain status information of pigs in real time to detect abnormalities in time.This is not only a time-consuming and laborious work,but also can interfere with the normal growth of pigs.To solve these problems,a recognition method of identification and drinking behavior for individual pigs based on machine vision is presented in this paper.This method can improve the production efficiency of pig farms and reduce the workload of breeders.Firstly,due to the non-rigid body of the pig,the characteristic regions with stability and uniqueness were extracted from overlooking surveillance video.Then color information entropy,shape parameters and Tamura texture features were extracted successively to form multidimensional eigenvectors which was used to characterize pig identity.Combining with the vectors similarity calculation method,the similarity between of the identifying pigs and training sample pigs was calculated to achieve identification of individual pigs.Secondly,aimed at the characteristic of relatively fixed posture when pigs were drinking,the improved Douglas-Peukcer polygon approximation method was used to fit the pig contours in the drinking area.Then the angle and distance characteristics were extracted.This two-dimensional feature which had scale invariance and rotation invariance was used to characterize the drinking state of the pigs.The optimal match between the contour fragments could be obtained by Hungarian algorithm.Then the matching cost with the optimal matching was calculated to complete the matching of the contours and achieve identification of drinking behavior of pigs.Finally,for the identification recognition algorithm,the optimal edge length of back characteristic region was selected by testing the recognition rate and the average time of recognizing the single pig with different parameters.Meanwhile,for the drinking behavior recognition algorithm,the optimal similarity threshold was selected by testing the recognition rate with different thresholds.Then MATLAB GUI was used to design the image processing interface.Parameter setting,identification recognition,drinking behavior recognition and other functions are completed to achieve visualization operation of recognizing identification and drinking behavior.The experimental results showed that the identification rate of individual pigs was 86.7%,and the average time of recognizing the single pig was 1.9154 s.Compared with other typical methods,the relatively high recognition rate was ensured with good performance of time.Meanwhile,the recognition rate of drinking behavior was 94.05%.This method distinguished the drinking behavior and non-drinking behavior well,and reached the expected effect.In this paper,machine vision technology was used to realize the intelligent monitoring and identification of pigs' identity and drinking behavior.This laid a foundation for the future research on the behavior of feeding,defecation and so on.This study also provide a new way to explore the identification and drinking behavior of livestock.