Research Of Pig Face Recognition System Prototype Based On Deep Learning

The rapid development of computer vision technology and deep learning has brought new opportunities for large-scale,unmanned farming farms.The identification of individual pigs using computer vision technology to improve the management efficiency of pig farms has become a research hotspot in recent years.The facial features of pigs are more obvious,and the individual identity of pigs can be determined by identifying pig faces,but the dirty environment in actual farming brings great difficulty to pig face recognition.In this thesis,a pig face recognition algorithm based on multi-scale classification network is proposed.The algorithm improves the feature extraction ability of the network on the pig face image by deepening the network layer number and expanding the network width.The neural network is optimized by symmetrically splitting the asymmetric convolution kernel and the facial features extracted by convolution kernels of different sizes are merged to obtain the abstract features of the pig face image.Aiming at the problem that the algorithm of pig face recognition has low recognition rate of pig face image in harsh environment,this thesis proposes an improved hierarchical network.The model of pig face detection and the improved residual network are combined to form a hierarchical network,which reduces the extraction of redundant information and the influence of individual feature weights on the recognition results.The improved residual network utilizes normalization algorithm,random deactivation,jump connection and other methods,which not only reduces the phenomenon of training offset,but also increases the sparsity of the network to solve the degradation problem of deep network.Based on the algorithm of pig face recognition,the overall architecture and functional modules of the system prototype for pig face recognition are designed.Firstly,the system of pig face recognition is divided into hardware subsystem and host computer software according to different functions.The hardware subsystem is designed with the FPGA chip as the core,combined with the image acquisition equipment,to realize the function of collecting the pig face image in the breeding environment and transmitting it to the upper computer.The upper computer software uses the deep learning platform to realize the reception of the pig face image and calls the pig face recognition algorithm model for face recognition to obtain the pig individual number.The system of pig face recognition was used to collect multi-angle pig face images,and the algorithm of pig face recognition was tested.The recognition rate reached 97.67%,which was better than other algorithms in complex environments.The system prototype of the pig face recognition studied in this thesis has a high recognition rate,and can collect and identify the images of the pigs in the limit column in real time.This research is an innovative application of the deep learning algorithm in the field of intelligent livestock breeding.The prototype of the non-contact face recognition system combining deep learning algorithm with embedded system is of great significance to promote the development of intelligent animal husbandry.