Research On Detection Of Diarrhea In Weaned Piglets Based On Machine Vision

The weaning period is the most critical moment in the pig breeding industry.During weaning,a piglet’s antibodies decrease,resulting in low disease resistance.Together with other factors such as environmental conditions.Piglets became vulnerable to diseases and stunt growth,leading to high piglet mortality.Diarrhea in weaning piglets is a clear indication of a disease occurrence and is always challenging to prevent and control in large-scale production.Diarrhea in piglets affects not only the farm economy but also an indicator of poor welfare conditions.The primary method to detect diarrhea is by observing the excretion behavior and normiality of piglet droppings.Currently,the conventional way for diarrhea detection in weaning piglets is by visual observation,which is subjective,time-consuming,laborious,difficult to detect a specific disease,and challenging to incorporate in the big data platforms.In this study,machine vision technology and the Internet of Things technology were used to achieve the information perception of diarrhea in piglets.A deep convolution neural network combined with a spatiotemporal information fusion method was used to identify diarrhea in weaned piglets automatically.The system of diarrhea information in the excretion area of the nursery was monitored by video was developed.This system provided a reliable data source for pig breeding information comprehensive perception big data platform.The main contents of this paper were as follows:(1)Model of diarrhea signal object detection in weaned piglets based on deep convolution neural network:Frame image from video data was obtained to prepare the PASCAL VOC dataset,deep convolution neural network was designed,and model parameters were trained.FPN algorithm was used to fuse high-level and low-level features to accurately and quickly detect the diarrhea posture,abnormal manure,and back number of piglets.The diarrhea signal detection of weaned piglets in the excretion area of the nursery was realized.The impact of different iteration times and different picture sizes on model performance was analyzed,and excellent performance was selected.(2)Behavior recognition algorithm based on posture segment:The concept of posture segment was proposed in this study to efficiently perform the excretion behavior recognition and diarrhea pig detection in the process of video analysis.The calculation and simplification method of the posture segment was developed,which made the model transit from static excretion posture recognition to dynamic excretion behavior recognition and improved the efficiency of video analysis.This method could be used to identify other behaviors related to attitude maintenance time,such as eating,drinking,climbing,etc.(3)Diarrhea detection algorithm of weaned piglets based on spatiotemporal information fusion method:Spatiotemporal information fusion method was proposed in this research study,to associate abnormal droppings and piglet excretion behavior to realize the diagnosis of diarrhea disease.Excretion behavior with abnormal droppings from the temporal domain and spatial domain was associated with detecting weaned piglet diarrhea in the surveillance video.The disease was considered to belong to the pig when the excretion behavior appears in the front,and abnormal dropping appears in the back.The distance between the excretion behavior and the center of abnormal manure does not exceed the given threshold.A video dataset was built to test the algorithm performance.(4)The software system of diarrhea detection platform for weaned piglets:The system completed information management and storage in the diarrhea detection process by deploying middleware and database on a cloud server,and Android APP with MVP mode and multi-threaded processing mode was realized,which improved the robustness and operation efficiency of the system.The UI interface was optimized,and a reference for pig farm managers to take targeted diagnosis and treatment of diarrhea pigs was provided.