Optimization and Comparative Analysis of YOLOV3 Target Detection Method Based on Lightweight Network Structure

As an open source target detection network, YOLOV3 has clear superiority in terms of accuracy and speed. However, the hardware configuration requirements are relatively high in actual applications, and the detection effect and real-time performance of Tiny-Yolov3 for embedded platforms are difficult to achieve expectations. In order to solve these problems, an improved YOLOV3 model optimization method based on a lightweight network structure is proposed, using lightweight network structure as the backbone network to replace the original convolution structure and residual module in YOLOV3. The size of the model network is reduced. The problem of high computational complexity and slow inference speed appearing on the embedded end is solved based on this targeted optimization. In case of slight loss of detection performance, this method can significantly reduce the model size and improving the detection speed compared with traditional methods on multiple public data sets.