一种SSD-RTF输电线路小目标检测算法研究OA北大核心CSTPCD

To address the problems of the traditional transmission line UAV inspection image detection method,such as weak detection ability,high error detection and missing rate,and insufficient shallow network semantic information,this paper proposes a small target detection algorithm model of SSD-RTF transmission line.In the shallow network layer of VGG-16 in the original backbone of SSD algorithm,a visual mechanism is added to enlarge the receptive field and a three-pronged trunk feature fusion module is introduced to extract multi-scale features of the feature map to increase the robustness of the feature map.FusionNet shallow feature modules are integrated to increase the extraction capability of small targets.Attention mechanism is employed to improve the learning efficiency of key information and thus further enhance the efficiency of target detection.Improved non-maximum suppression improves the representation capability of the network.Our experimental results of the improved SSD-RTF algorithm on an independently-built power line data set show the accuracy and real-time detection of small targets improve to a certain extent,the overall mAP is up by 7.5％,and the wrong and missing detection decreases.