YOLOv3-DSN Object Detection Algorithm Based on Depth Wise Separable Convolution

In order to realize the real-time detection of dairy goat objects in the sheep farm, a neural network detection algorithm based on depth wise separable convolution YOLOv3-DSN is proposed. Firstly, the video frames are used to screen out the key frames containing the dairy goats based on the surveillance video of the sheep farm, and construct the dairy goat sample set. Then the K-means clustering method is used to determine the number and dimensions of the object candidate box on the data set, and the GIOU box regression loss function is used to improve the positioning accuracy of the dairy goat regression box. At the same time, the model is optimized through multi-scale training, and the depth wise separable convolutionYOLOv3-DSN network is used to return the object category and position,which realizes end-to-end object detection.Under the circumstance of taking into account accuracy and speed, realize the object detection of sheep farm surveillance video.The experimental results show that compared with SSD and YOLOv3, it can obtain better object detection results in terms of efficiency and accuracy.Provide basic technology for the development of intelligent video surveillance systems for sheep farms and reduce the workload of experimenters.