一种SSD-RTF输电线路小目标检测算法研究  

Research on a small target detection algorithm for SSD-RTF transmission lines

作　　者：唐心亮 李少杰 王建超 王震洲[1] TANG Xinliang;LI Shaojie;WANG Jianchao;WANG Zhenzhou(School of Information Science and Engineering,Hebei University of Science and Technology,Shijiazhuang 050018,China)

机构地区：[1]河北科技大学信息科学与工程学院,石家庄050018

出　　处：《重庆理工大学学报（自然科学）》2024年第9期150-157,共8页Journal of Chongqing University of Technology：Natural Science

基　　金：河北省高等学校科学技术研究项目(ZD2020318);河北省教育厅青年基金项目(QN2023185)。

摘　　要：针对传统输电线路无人机巡检图像检测方法对输电线路小目标检测能力弱,并且存在错检和漏检率高、浅层网络语义信息不足等问题,提出了一种SSD-RTF输电线路小目标检测算法模型。在原始的SSD算法主干网络VGG-16浅层网络层添加视觉机制扩大感受野,并且引入三叉乾特征融合模块提取特征图的多尺度特征,增加特征图的鲁棒性;融合FusionNet浅层特征模块,从而增加小目标的提取能力;使用注意力机制提升对关键信息的学习效果,从而提高目标检测效率;改进非极大值抑制提高网络的表示能力。改进后的SSD-RTF算法在自行构建的输电线数据集上的实验结果显示,在检测小目标时的准确性和实时性都有了一定的提高,整体mAP上升了7.5%,同时减少了错检和漏检。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.

关 键 词：目标检测 浅层融合 三叉乾特征融合 视觉机制 非极大值抑制 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]