Research On Individual Identification Of Goat Based On Deep Learning

Individual identification has important research significance in the fields of intelligent security,biometric identification,criminal tracking and so on,and also lays the foundation for livestock identification analysis such as goat individual identification based on images and videos.In the research process of goat identification with high similarity,due to the difficulty of image acquisition caused by goat clustering and the high similarity of goat,this paper proposes a goat identification algorithm based on deep learning,which integrates a variety of optimization algorithm modules to achieve accurate identification of highly similar goat,in order to provide effective guidance for the identity management of goat.The main contents of this paper are as follows:(1)Research on the method of goat data enhancement based on adversarial generative network.In this study,the time span of goat image data acquisition is large,including different angles such as early,middle,late,front and side.Because some goats are aggressive,the data set is unbalanced.Traditional image enhancement methods,such as translation and rotation,are difficult to improve the image quality.In order to solve the problems of data imbalance between classes and insufficient utilization of image data,this study uses the generative antagonistic network model to enhance data,and proposes a nonlinear image data enhancement method based on the improved Cycle-Gan network model,which can stably and effectively fuse the characteristics of goat images and balance data sets.(2)High similarity goat identification method based on joint loss optimizationDue to the high similarity of goat individuals,this paper constructs a joint optimization loss function model for high similarity of goat identification,and uses the triple loss function and cross entropy loss function to jointly optimize the model to enhance the network characterization ability,thereby improving the recognition accuracy.The accuracy rate of dairy goat and cashmere goat can reach 93.077% and 89.677%,respectively.(3)Research on the construction of network model based on goat images.To solve the problem of low individual recognition rate of cashmere goats,transfer learning is introduced in this section to pre-learn the goat data in this study by training the pretraining model with only the cross-entropy loss function,and then the best model is used as the basic pre-training model for the joint loss function training in the next stage.In this study,the recognition rate of cashmere goats reaches 93.75%.Then,in order to further optimize the model proposed in this paper,a network model based on cascade deep learning is proposed in this section to effectively improve the accuracy of goat identification;The experimental results show that the accuracy of the model for dairy goats can reach 93.846%,and the accuracy of cashmere goats can reach 95%,which shows that the network model based on goat images in this section can effectively improve the accuracy of goat identification.(4)Integration of high similarity goat identification system based on deep learningIn order to promote the application scenario of goat identification,the goat identity model is transplanted to the server,and a high similarity intelligent identification monitoring platform for goat is designed and developed.Based on the video uploaded by users or the camera monitoring video obtained by IP address,the platform can identify the identity of goat online,realize two interactive modules of video recognition and real-time camera monitoring,and identify two breeds of cashmere goats and dairy goats.