Design And Implementation Of Weaned Piglet Behavior Recognition System Based On Deep Learning

As one of the important domestic animals in China,the demand of pig is increasing with the vigorous development of animal husbandry industry.The resistance of young piglets is low,especially the autoimmune system of weaned piglets has not been fully established,and the piglets have changed from relying on mother’s milk for nutrition to feeding completely independently.The resistance of piglets without the nutritional support of maternal antibodies will be weakened,and the stress will also occur when the piglets are away from the living area where their parents live together.As a result,the piglet stage has become the most vulnerable and critical part of the pig feeding process,and the health status of weaned piglets has been paid much attention by the industry.The behavior status of weaned piglets is an important basis to judge whether the piglets are healthy or not.Timely treatment of abnormal behavior of piglets can minimize the loss of pig farm.In the actual production of large-scale pig farms,a certain number of diseased and dead piglets appear every day.Finding and dealing with these diseased and dead piglets requires a lot of manpower,which is inefficient and easy to cause stress of piglets.In view of the current problems of pig in automation in the process of breeding and technical problems to be solved,this thesis designs a set of weaned piglets behavior recognition system based on depth of learning,the system can real-time gesture recognition piglet behavior status,and synchronous behavior attitude information to alert administrators,timely treatment of abnormal piglets by and disposal measures such as isolation,It can shorten the retention time of abnormal piglets with infection risk in piggery,reduce the risk of pathogen infection,improve the survival rate and breeding welfare of piglets,and ensure the safe and orderly production of piglets in piggery.In jiangsu province town of yangzhou city jiangdu district Shao Bo piglet weaning period between changbai piglets,as the research object,the image data acquisition system for real-time acquisition piglets image automatically,and will collect images of piglets wireless transmission to the cloud,the cloud received via a gesture recognition system based on deep learning network of piglets behavior after processing,It can realize real-time identification of piglets’ behavior and posture status,and display it on the Android App terminal to help administrators find abnormal behavior and posture of piglets.Main contents of the study:(1)The behavior characteristics of weaned piglets and the acquisition and processing of image data were analyzed.Two kinds of behavior of weaned piglets during feeding and drinking and two main postures of piglets during non-feeding and drinking were summarized,and the characteristics of these four kinds of piglets were analyzed.The collected data will be screened,annotated and enhanced successively,and finally made into data sets in two formats: PASCAL VOC and COCO.(2)Behavior and attitude recognition model of weaned piglets based on YOLOV4,CenterNet and FCOS algorithm.The behavior and attitude recognition model of weaned piglets based on YOLOV4,Centernet and FCOS algorithm was designed,and the detection performance was compared to establish the behavior and attitude recognition model of weaned piglets based on YOLOV4 deep convolutional network algorithm.The results showed that the recall rate and accuracy rate of the model on the data set of weaned piglets were more than 98%,and the detection speed reached 29.2FPS.The model could accurately identify the feeding and drinking behavior of weaned piglets and the standing and lying posture of weaned piglets during non-feeding and non-drinking.(3)The behavior and posture recognition model of piglets during weaning period based on YOLOV4 algorithm was used to detect and analyze the effects of three different feeding materials on feeding and drinking behavior of piglets and standing and lying posture of piglets during non-feeding and non-drinking.The conclusions are as follows:1)The feed intake of piglets fed the antibiotic aureomycin diet and those fed the Lactulose-Bacillus coagulans compound diet were higher than those fed the normal control group;2)The piglets fed the antibiotic aureomycin diet drank less water than the control piglets fed the normal diet;3)There was no effect on standing time of weaned piglets fed diets supplemented with aureomycin and supplemented with lactulose-Bacillus coagulans synbiotin;4)The sleeping time of weaned piglets supplemented with Lactulose-Bacillus coagulans synbiotin was shorter than that of normal piglets.(4)Using YOLOV4 algorithm based piglet behavior recognition model during weaning period to detect and analyze the differences of piglets’ standing and lying posture under acute stress test conditions and those under normal conditions.The conclusions are as follows:1)The acute stress test significantly reduced the drinking time of weaning piglets fed with the three feedstuffs;2)The acute stress test significantly reduced the standing time of weaning piglets fed with the three feedstocks;3)The acute stress test significantly increased the lying time of weaning piglets fed with the three feedstuffs.(5)Implementation of behavior and posture information management platform for weaned piglets.In this thesis,an Android-based behavioral posture information management platform for weaned piglets is designed.The platform mainly includes three design parts: middleware,database and Android client.The platform can realize the collection,storage,processing and analysis of pigsty video data,and finally present the processing results on the Android interface in real time,realizing the automatic detection of weaned piglets’ behavior and helping pig farm managers find abnormal behavior and posture piglets.