Pig Delivery Identification And Piglet Tracking System Based On Deep Learning

In pig farms,the use of intelligent monitoring technology instead of manual monitoring,accurate identification of pig delivery and intelligent tracking of piglets that have just been delivered in situations with complex backgrounds and severe occlusion is a valuable research topic,and this subject is one of the main key technologies in a new product research and development of enterprise cooperation project.In this paper,in order to improve the time performance and accuracy performance of the object detection algorithm,we improved the Faster R-CNN object detection algorithm based on deep learning.In our pig dataset,the improved algorithm greatly improved the detection speed while improving the object positioning accuracy of pigs,and then based on this algorithm,we proposed a pig delivery identification algorithm and a piglet intelligent tracking algorithm.The specific work is as follows:? Improve the Faster R-CNN algorithm.Deep learning through the study and imitation of the ability to learn the brain,can effectively extract and analyze the original data,is the mainstream algorithm of the current object detection task.Based on the Faster R-CNN algorithm,this paper improves the precision and recall rate of the algorithm on the pig dataset by redesigning the Anchor.By deleting the low-confidence candidate object frame generated by the network forward propagation step by step,The unnecessary non-maximum suppression calculations are eliminated,and the detection speed of the algorithm is greatly improved without affecting the accuracy.? Pig identification and piglet tracking.This paper first proposes the piglet birth area demarcation algorithm,so that for each picture to be detected under the video,if there is a large pig,the precise piglet birth area is delineated;then the current piglet number determination algorithm is proposed for the video.Each frame,the algorithm will calculate the current number of piglets;Finally,this paper proposes a distance-based object box matching algorithm,when the piglet is born,according to the video frame before and after the relevant information,making the frame before and after the two frames of the piglet numbered one by one Correspond to achieve the purpose of tracking.? System engineering and project deployment.Based on the above algorithm proposed in this paper,we have developed a prototype of pig identification and piglet tracking system.The system is written in C++ and deployed on the Windows 7 operating system.In the actual delivery scenario provided by the cooperating enterprise,parallel monitoring of multiple HIKVISION cameras is performed.After a large number of test experiments,the results show that in the complex scenario,the system is more than 97%accurate in determining the moment of birth of piglets,and it can also track piglets well in case of severe obstruction,which has good commercial value.