Research On Sows’ Estrus Checking Method Based On Machine Vision

In the past ten years,the development of the pig industry has turned our country into the world’s largest pig raising country.At the same time it is also facing the problem of insufficient production levels.With the development of artificial intelligence technology,smart devices have begun to be applied to pig companies to increase productivity.Estrus detection plays an important role in raising the litter size of pigs.Based on the research of domestic and foreign literatures and the actual demand of pig enterprises,the abnormal behaviors of pigs in estrus were mainly static standing,vulva redness and swelling,pig-arched standing,restlessness,etc.Standing reaction is the most important external performance of pigs during estrus.This paper mainly studies the identification of pig static reaction,introduces artificial intelligence method into the estrus detection process,and proposes a sows’ estrus checking method based on machine vision.In this method,the standing determination is divided into two stages.The first stage of the model is sows contour recognition,and the second stage is pigs contour matching.This paper mainly completes the following aspects:1.Segmentation model for contour recognition,this paper proposes a network model(PCNN)based on the Mask RCNN model,which is more suitable for pig contour extraction.The model uses deep separable convolution as the basic operating unit of the model to reduce the computational burden of the model and reduce the generation of parameters.It uses GA-RPN to replace the original RPN network in Mask RCNN model to reduce the generation of useless anchor.Experiments show that the performance of the improved contour extraction algorithm is greatly improved.In the pig contour recognition,the recognition accuracy of the mask RCNN algorithm was improved by 1.01%,and the recognition speed was reduced from682 ms to 372 ms.2.Judgment model of standing behavior for pigs,The static characteristics of pigs and the image data collected continuously were analyzed.Based on Vibe algorithm,this paper proposes the contour matching algorithm for sows in static standing.In this algorithm,the template is matched with the background before and after,and the background space is initialized and the model is updated according to the characteristics of the data.The template matching similarity threshold and static determination time threshold in the model were set through data analysis.The error caused by the low accuracy of contour recognition and the influence of contour change caused by the shaking of pig’s ear or tail on the static determination are solved.The experimental results show that the contour matching algorithm based on Vi Be proposed in this paper is far superior to the traditional contour matching algorithm in the detection and judgment of pig estrus,with an accuracy of up to 92%This paper realizes a sows’ estrus checking method based on machine vision by recognizing and matching pig contours.In order to test the overall robustness and recognition accuracy of the model,this paper uses the real environment of the pig farm to verify the program.Experiments shows that the program can help(or replace)professionals in sows’ estrus detection.