一种改进的PCAC-YOLO目标算法在无人机参与城市作战目标检测中的应用  

Application of an improved PCAC-YOLO target detection algorithm in urban combat target detection with the participation of unmanned aerial vehicles

作　　者：彭富伦[1] 裴昊晨 刘超 李双全[1] 赵妍 PENG Fulun;PEI Haochen;LIU Chao;LI Shuangquan;ZHAO Yan(Xi’an Institute of Applied Optics,Xi’an 710065,China;School of Weapon Science and Technology,Xi’an Technological University,Xi’an 710021,China;Xi’an Military Representative Bureau,Xi’an 710021,China;School of Electronic Information Engineering,Xi’an Technological University,Xi’an 710021,China)

机构地区：[1]西安应用光学研究所,西安710065 [2]西安工业大学兵器科学与技术学院,西安710021 [3]西安军代局,西安710021 [4]西安工业大学电子信息工程学院,西安710021

出　　处：《国外电子测量技术》2025年第1期103-109,共7页Foreign Electronic Measurement Technology

摘　　要：无人机在城市作战的复杂环境中,目标对象不仅种类繁多且密集分布,极易遭遇相互遮挡的情况;此外,由于无人机视角下的目标尺寸显著缩小,因此,检测过程中会出现漏检和误检的问题出现。针对以上问题提出了一种无人机目标检测算法PCAC-YOLO。为了增强遮挡目标边缘信息的表征能力,通过裁剪CBS层和添加基于相似度的激活模块(Similarity-Aware Activation Module,SimAM)注意力机制,设计了新的空间池化连接自注意力机制(Spatial Pooling Connect Self-Attention Mechanisms,SPCSM)模块。同时引入卷积特征提取模块Conv2Former,提高了模型对小目标特征的关注能力。实验结果表明,在AU-AIR数据集中,相较于原始的YOLOv7算法,Precision值增加至52.8%,提升了14.1%;mAP@0.5值增加至41.4%,提升了6.1%;mAP@0.5:0.95|small值为16.2%,提升了1.7%。该目标算法有效提升了在城市作战环境无人机视角下的目标检测准确率,证明了设计算法的有效性。In the complex environment of unmanned aerial vehicles(UAV)in urban combat,the target objects are not only diverse and densely distributed,which are very likely to encounter mutual occlusion,in addition,the detection process will trigger the situation of omission and misdetection problems due to the significant reduction of the target size under the UAV viewpoint.A UAV target detection algorithm,PCAC-YOLO,was proposed to address the above problems,and in order to enhance the characterization ability of the edge information of the occluded target,a new SPCSM module was designed by cropping the CBS layer and adding the SimAM attention mechanism.The ability of the model to recognize small-size target features was enhanced by integrating the convolutional feature extraction unit Conv2Former.The experimental data show that on the AU-AIR dataset,compared with the original YOLOv7 algorithm,the Precision value is 52.8%,which is improved by 14.1%;mAP@0.5 value is 41.4%,which is improved by 6.1%;and mAP@0.5:0.95|small,which is improved by 16.2%,which is improved by 1.7%.It effectively improves the accuracy of target detection under UAV view in urban combat environment and proves the effectiveness of the designed algorithm.

关 键 词：城市作战 目标检测 YOLOv7 PCAC-YOLO 无人机视角 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] V18[自动化与计算机技术—控制科学与工程]