Research on multi-scale object detection algorithm based on improved ResNet50 and FPN
For the ResNet50 and FPN structures,the shallow detail information and the deep semantic information cannot be fully integrated and utilized.An algorithm to improve the structure of ResNet50 and FPN is proposed,introducing improved channel and spatial attention modules in ResNet50 network structure to take advantage of detailed and semantic information from different feature layers.In addi-tion,in the FPN structure,in order to allow the shallow feature layer to better utilize the semantic in-formation of the deep feature layer,in the top-down path of FPN,a bypass is added between different feature layers to enhance feature reuse.The experimental results show that the method performs well in PASCAL after training on the MS COCO dataset.The mean average precision(mAP)of the VOC 2012 test reached 83.2%,with an increase of 2.7%,and the mAP on the MS COCO dataset increased by 1.5%.It has good detection performance.
attention mechanismfeature pyramidfeature reusefeature fusionfeature layer information
NETL
NSTL
万方数据
维普
