In view of the limitations of the existing object detection networks in feature extraction and multi-scale fusion,the group convolution and Gaussian pyramid are introduced into the deep neural network,and an object detection method based on group attention mechanism and Gaussian multi-scale fusion is designed.This method uses image graying and histogram equalization to enrich the input information and reduce the influence of illumination;Secondly,a two-stage feature extraction structure is constructed,the object features are initially extracted by deep separable convolution,and then the weight of key information is enhanced by group attention structure to further refine the object features;At the same time,in order to fully capture the multi-scale information of the object,a Gaussian multi-scale structure is designed.The Gauss pyramid feature is constructed by using adaptive fusion of different scale features,and then the object detection is realized jointly with the corresponding scale features.The experimental results on ImageNet,MS COCO and PASCAL VOC datasets show that the proposed method has a great improvement in object information enrichment,effective feature extraction and network scale invariance,and also has high robustness and generalization ability in complex scenarios.
维普
万方数据