Research On Pig Face Recognition Based On Deep Learning

In recent years,with the rise of intelligent breeding industry,the breeding methods of pigs are constantly changing,gradually changing from traditional breeding methods to high-tech intelligent breeding.Building intelligent pig farms has become the development trend of the whole industry.However,with the continuous innovation of breeding methods,it also brings greater challenges to the whole breeding industry.How to effectively identify and manage individual pigs has become a very key link.Based on the pig face recognition in intelligent breeding,the main research contents are as follows:Firstly,by processing the data collected from the pig farm,the pig face data set used in the cost subject is constructed.For pig face recognition,there is no public and available data set,which needs to be built by ourselves.The video data collected from the pig farm is processed in frames.In order to prevent over fitting of the model,the similarity of two adjacent frames is calculated through the structural similarity algorithm to remove the duplicate of the image.The de duplicated image is sent to YOLOv5 target detection algorithm for pig face position detection to reduce the extraction of redundant information.According to the detection results,the pig face is segmented and constructed into a new data set.At the same time,in order to prevent distortion caused by image size scaling during pig face recognition model training,the segmented pig face image is edge filled without changing the image proportion.The final pig face data set contains 50 categories and 16466 image data.Secondly,the improved weak supervised data enhancement network is used to recognize the pig face data set.According to the characteristics of small difference between classes and large difference within classes of pig face dataset,and considering the deployment of the model at the mobile end,this topic selects the lightweight MobileNetV3-Small network as the backbone network of weak supervised data enhancement network algorithm,and improves the MobileNetV3-Small backbone network from the activation function and network structure.The improved network structure is named MobileNetV3-SS.Through several groups of comparative experiments,the effectiveness of the improved algorithm is verified.Finally,the local trained model is deployed to the mobile device Raspberry Pi 4B to realize pig face recognition on the mobile terminal.In order to meet the needs of pig face recognition in the field pig farm,this topic selects Raspberry Pi 4B as the embedded platform of the mobile terminal.Using the lightweight mobile terminal reasoning framework mobile neural network,the pig face recognition model trained at the local terminal is transformed and deployed to the mobile terminal equipment,and finally the pig face recognition on the mobile terminal equipment is completed.