Research On Object Detection Of Dairy Goat Based On Improved YOLOv3

Intelligent management is the basis and development direction of the large-scale farming.In order to realize the intelligence and large-scale of dairy goat farm and improve the management level of the farm,we took the dairy goat images of the Animal Husbandry and Experiment Base of Northwest A & F University as the research object,and applied the YOLOv3 algorithm based on the channel attention module SeNet and GIOU loss function to realize the object detection and counting of the dairy goats.The main contents and conclusions of this paper are as follows:(1)Construction and preprocessing of dairy goat object detection datasets.In order to solve the problem of lack of public object detection datasets of dairy goat,firstly,a remote video monitoring device was set up in the dairy goat farm to obtain a large number of dairy goat videos;secondly,FFMpeg was used to extract image key frames every 5s;after that,the dairy goat image was enhanced by flipping,changing the contrast and saturation to expand the number of dairy goat datasets;finally,the image of dairy goat was labeled by the annotation wizard,and the coordinate data and category data of dairy goat in the image were converted into yolo format,which was used by the YOLOv3 model.(2)YOLOv3 algorithm based on channel attention module SeNet.In view of the low detection accuracy of the YOLOv3 algorithm on the dairy goat datasets,three SeNet modules were added to the neural network of the YOLOv3 algorithm to increase the network model’s attention to the dairy goat,reduce background interference,and improve the detection accuracy of the model.Experimental results show that using the channel attention module SeNet to improve the YOLOv3 model,the detection accuracy is increased by 1.75% while keeping the detection speed unchanged.(3)YOLOv3 algorithm based on GIOU loss function.In view of the poor convergence effect of the YOLOv3 algorithm when training on the dairy goat datasets,the GIOU loss function is used to replace the location loss function in the original loss function,and the focus of the location loss is transferred from the distance between the prediction box and the real box to the IOU between the prediction box and the real box,the convergence effect of model training is improved.Experimental results show that using the YOLOv3 algorithm based on the GIOU loss function,the model convergence effect is better,the Loss curve is smoother,and the detection accuracy is improved by0.93%.In summary,the YOLOv3 algorithm based on channel attention module SeNet and GIOU loss function realized the object detection and counting of dairy goats,which has important theoretical and practical significance for promoting the large-scale breeding of dairy goats and improving the management efficiency of dairy goat farms.