Improved YOLOv3 model based on ResNeXt for target detection

In order to solve the problem of deep learning models that are difficult to balance accuracy and speed, this paper selects the YOLOv3 model with higher detection speed but lower accuracy to improve it, and intends to improve the accuracy of YOLOv3 to construct a high-speed and high-precision target detection model. First, we used the ResNeXt residual block to construct a backbone network with few parameters and high accuracy. Then we used the K-means++ clustering algorithm to obtain anchor boxes that is closer to the real box to help the model perform regression detection. The experimental results on the VOC data set show that the accuracy of the improved model is significantly improved, and achieved high-precision and high-speed detection.