Application Research Of Individual Identification Technology Based On Deep Learning In Dairy Farming

With the rapid development of domestic dairy industry towards refinement and intelligence,individual recognition technology has gradually become an important part of dairy farming.Traditional physical recognition methods not only cause physical damage to dairy cows,but also easily lose labels.In this paper,deep learning is applied to the scene of cow individual recognition.The main work is as follows:In order to realize the non-contact,high-precision and effective identification of individual dairy cows in the farm environment,the SSD algorithm was improved in view of its own defects and the low average accuracy in the identification of dairy cows.In the third chapter,an improved SSD(shallow feature module SSD)algorithm based on shallow feature module is proposed.Firstly,in order to reduce the network computation and improve the detection speed of the algorithm,the backbone network of traditional SSD algorithm is replaced by mobile netv2 network instead of vgg16 network,Secondly,according to SSD algorithm is a multi-scale target detection algorithm,and SSD algorithm network structure of the shallow feature map through convolution layer less,feature extraction ability is poor,in order to improve SSD algorithm shallow feature map on the cow feature extraction ability,design shallow feature module,expand SSD algorithm Finally,K-means clustering algorithm is used to re set the region candidate box in SSD algorithm to improve the detection accuracy of SSD algorithm.The experimental results show that the average accuracy of sfm-ssd algorithm is 3.13% higher than the original SSD algorithm.At the same time,the real-time performance of the algorithm is also improved.In order to realize the accurate recognition of dairy cattle in complex farm environment,the SSD algorithm is improved to solve the problem that the original algorithm is not good at identifying overlapping dairy individuals.Firstly,according to the characteristics of different feature graphs in SSD algorithm,different feature graphs in SSD algorithm can not be complementary.The different feature graphs are fused to make the feature information of different feature maps complementary to improve the detection effect of overlapping objects.then,according to the characteristics of friesiancattle2017 data set,there are no small object features,and conv4 in the network framework of SSD algorithm is removed＿At the same time,the number of candidate boxes of other feature graphs in SSD algorithm is increased,which not only ensures the real-time performance of the algorithm,but also improves the detection accuracy.Finally,the transfer learning method is introduced.The experimental results show that the average accuracy AP(average precision)is increased by 4.32% when the real-time detection is satisfied,and the AP of the improved SSD algorithm is improved by3.85% after migration.