Recognition Of Drinking Behavior Of Group Pigs Based On Deep Learning And Area Occupancy

With the rapid development of the intelligent pig industry,intelligent video surveillance technology is being widely studied for monitoring the welfare and health of pigs.The drinking water status of pigs is an important indicator to measure the health status of pigs.Monitoring the drinking behavior of pigs and extracting the rules of drinking behavior can effectively warn pigs of abnormal health and improve their growth welfare.Pigs have difficult characteristics and poor universality.This paper proposes a pig drinking water behavior recognition method based on deep learning and area occupancy.The main research contents and conclusions are:First of all,because the identification of swine drinking water behavior needs to study pig individuals and their pig head regions,a segmentation algorithm for pig body and pig head based on improved Mask R-CNN is proposed.This algorithm can be used on group pig images at the same time.Segment each pig individual and its pig head area.Since the segmentation task in this paper is the segmentation of multi-objects in the same category,it is proposed to introduce non-local feature vectors into the region proposal network of the Mask R-CNN model,and add an edge detection filter to the output mask branch.After the improved model,the segmentation accuracy of pig body and pig head reached 90.3% and 88.1%,respectively increased by 4.2% and 3.3%.Secondly,since the position of the drinking fountain in the pig house is fixed and the drinking behavior is a process in which the pig head and the drinking fountain continuously and continuously contact for a period of time,a drinking behavior judgment method based on area occupancy is proposed.First,divide the designated drinking area and drinking fountain area in the group pig image,extract the pig individuals who are completely contained in the drinking area,and then pass the target detection frame of the pig head and the drinking fountain in a single frame image.Predict the behavior,and then finely segment the pig head area,further judge the drinking behavior by calculating the area occupancy,and finally calculate the movement status of the pig in consecutive frames,and finally complete the recognition of drinking behavior.The accuracy of the algorithm on drinking and non-drinking behaviors on the test set reached 94.76%.Finally,an algorithm for identifying drinking pigs using a convolutional neural network model is proposed.First,in order to enable the convolutional neural network to better learn the local features of each part of the pig,the extracted drinking pigs are rotated through the minimum The circumscribed rectangle method is horizontally aligned to obtain the input image of the neural network.Then manually build the structure of the network,and optimize the parameters through experiments.The recognition rate under the optimal parameters has reached 91.55%.Finally,through comparison experiments with other pigs' identity recognition,the superiority of this algorithm is proved.This article combines deep learning technology and computer vision technology to identify the drinking behavior of group pigs.Compared with other algorithms,this algorithm is more universal and has a higher recognition rate,which has achieved the expected research results.The pig's other behavior recognition research laid the foundation.