YOLO Object Detection Algorithm with Hybrid Atrous Convolutional Pyramid

Aiming at the problems of insufficient detection accuracy of small targets and insufficient receptive field of feature points in the field of target detection, a YOLO target detection algorithm with hybrid atrous convolution pyramid is proposed. The algorithm firstly introduces atrous convolutions with different dilation rates into the feature pyramid network, and builds a hybrid receptive field module (HRFM), which enhances the ability to obtain global information by increasing the receptive field, and solves the problem of target occlusion; Secondly, design a pyramid network of atrous path aggregation, fusion of shallow feature information and high-level semantic information, improve the global detail information and representation ability of feature maps, and enhance the multiscale adaptability of the model. Three progressive schemes are designed for testing on the VOC dataset. The experimental results show that the algorithm can effectively solve the problem of target occlusion and improve the detection accuracy of small targets.