Research On Target Tracking Of Dairy Goats Based On Attention Model

The target tracking of animals is the basis of studying their health status and behavior analysis.In order to accurately and real-time obtain the individual information and action status of animals,this paper takes the video of dairy goat farm in animal husbandry teaching experimental base of Northwest A & F University as the research object,based on TargetAware Deep Tracking(TADT)algorithm,combined with attention mechanism and template updating method,is used to track the dairy goat object and evaluate the tracking accuracy,which verifies the effectiveness of the improved algorithm.The main research contents and conclusions are as follows:(1)Data set construction and preprocessing.In order to obtain a sufficient number of data sets,remote video monitoring equipment was built in the outdoor sports ground of dairy goat to capture and preprocess the video of dairy goat.The invalid fragments were deleted,and the data were enhanced by power-law transform to integrate into high-quality milk goat motion video.Then,according to the diversity of dairy goat activities,Labelhub software is used to label the dairy goat according to its head and body position,and the Ground Truth is generated for the target tracking of dairy goat in the sports field.The experimental results show that when tracking based on the head of dairy goat,the background interference in the video image is relatively less,and the features are more obvious,so its performance is better than that of tracking based on the body of dairy goat.(2)Tadt dairy goat target tracking model based on spatial sensing sampling.In the past,twin network only uses shallow classification network for pre training,which can not extract deep features rich in semantic information.In this study,deep convolution classification network vgg-16 and deep residual network RESNET are used for feature extraction,and spatial sensing sampling strategy is introduced to eliminate the strict translation invariance of Siamese network.Combined with the target sensing features,a dairy goat target tracking model based on tadt is constructed.The experimental results show that the tadt model based on RESNET has the best tracking effect,improves the discrimination ability of the original model,and the tracking accuracy on the dairy goat data set can reach 80.3%.(3)The construction of TADT dairy goat target tracking model based on attention mechanism.In order to further optimize the tracking model of dairy goat,attention mechanism is added in the regression network to enhance the capture of the partial dependence of attention,enlarge the similarities and differences between features,and extract more detailed features.Secondly,template updating is introduced in the tracking process,which fuses the dynamic template with the initial static template to make it closer to the current tracking target and enhance the accuracy of template matching.The experimental results show that the improved algorithm has better performance,and the success rate and accuracy are improved by 1.1%and 1.8% respectively on the dairy goat data set,and the real-time tracking is guaranteed.To sum up,this paper achieves the goal tracking of dairy goats based on TADT algorithm and attention mechanism.Based on the tracking results,individual information of dairy goats can be assessed for health,and then intellectualized breeding can be achieved.