摘 要：针对装甲目标图像背景复杂、目标尺度小等问题，提出一种基于 YOLOv5s 的装甲目标检测算法。首先在 FPN 结构中增加一个浅层分支，增强对小目标特征的提取能力；其次通过 Focal Loss 损失函数来平衡正负样本；再次将 CIoU_loss 用作边框回归损失函数，用以提升识别精度；最后将 ECA 注意力模块引入算法中，加强重要特征的表达。实验结果表明，改进算法在自制数据集上 AP 达到 92.9%，相较于原始算法提高了 4.2%，能够很好地满足装甲目标检测任务的精度与速度需求。
关键词：装甲目标；YOLOv5s；特征金字塔；ECA 注意力模块；Focal_loss
DOI:10.19850/j.cnki.2096-4706.2023.05.017
中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2023）05-0073-05
An Improved Armored Target Detection Algorithm Based on YOLOv5s
YI Tuming1, WANG Xianquan2, YUAN Wei 1, KONG Qingyong1
(1.Southwest Computer Co., Ltd., Chongqing 400060, China; 2.Chongqing University of Technology, Chongqing 400054, China)
Abstract: Aiming at the problems of complex background and small target scale of armored target image, an armored target detection algorithm based on YOLOv5s is proposed. First, a shallow branch is added to the FPN structure to enhance the ability of extracting small target features; Secondly, the Focal Loss loss function is used to balance the positive and negative samples; CIoU_ Loss is used as the loss function of frame regression to improve the recognition accuracy; Finally, ECA attention module is introduced into the algorithm to enhance the expression of important features. The experimental results show that AP of the improved algorithm on the self-made data set achieves 92.9%, which is 4.2% higher than that of the original algorithm, and can well meet the accuracy and speed requirements of the armored target detection task.
Keywords: armored target; YOLOv5s; characteristic pyramid; ECA attention module; Focal_loss
参考文献：
[1] 邓磊，李海芳 . 基于多尺度整体嵌套池化语义的装甲目标检测 [J]. 激光与红外，2022，52（2）：295-304.
[2] 孙皓泽，常天庆，张雷，等 . 基于轻量级网络的装甲目标快速检测 [J]. 计算机辅助设计与图形学学报，2019，31（7）：1110-1121.
[3] 王全东，常天庆，张雷，等 . 基于深度学习算法的坦克装甲目标自动检测与跟踪系统 [J].系统工程与电子技术，2018，40（9）：2143-2156.
[4] 王曙光，吕攀飞 . 改进 YOLOv2 的装甲车辆目标识别 [J].计算机与现化，2018（9）：68-71+79.
[5] HU J，SHEN L，SAMUEL A，et al. Squeeze-and-excitation networks [C]//Proceedings of IEEE conference on computer vision and pattern recognition. Salt Lake City：IEEE，2018：7132-7141.
[6] WANG Q，WU B，ZHU P，et al. ECA-Net: efficient channel attention for deep convolutional neural networks [C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle： IEEE，2020：11531-11539.
[7] GIRSHICK R，DONAHUE J，DARRELL T，et al. Rich feature hierarchies for accurate object detection and semantic segmentation [C]//2014 IEEE Conference on Computer Vision and Pattern Recognition. Columbus：IEEE，2014：580-587.
[8] REN S Q，HE K M，ROSS G，et al. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks [J].IEEE transactions on pattern analysis and machine intelligence，2017，39（6）：1137-1149.
[9] REDMON J，DIVVALA S K，GIRSHICK R B，et al. You Only Look Once: Unified, Real-Time Object Detection [EB/OL].[2022- 09-25].https://blog.csdn.net/weixin_54546190/article/details/123028952.[10] LIU W，ANGUELOV D，ERHAN D，et al. Ssd: Single
shot multibox detector [C]//European conference on computer vision. Amsterdam：Cham，2016：21-37.
[11] HOWARD，ANDREW G. Mobilenets: Efficient convolutional neural networks for mobile vision applications [EB/OL]. [2022-01-2].https://arxiv.org/pdf/1704.04861v1.pdf.
[12] ZHANG X Y，ZHOU X Y，LIN M X，et al. ShuffleNet: An Extr-emely Efficient Convolutional Neural Network for Mobile Devices [C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Salt Lake City：IEEE，2018：6848-6856.
[13] 丛眸，张平，王宁 . 基于改进 YOLOv3 的装甲车辆检测方法 [J]. 兵器装备工程学报，2021，42（4）：258-262.
[14] SANDLER M，HOWARD A G，ZHU M L，et al. Mobilenetv2: Inverted residuals and linear bottlenecks [C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City：IEEE，2018：4510-4520.
[15] HOWARD A，SANDLER M，CHU G，et al. Searching for mobilenetv3 [C]//Proceedings of the IEEE International Conference on Computer Vision. Seoul：IEEE Computer Society，2019：1314-1324.
[16] MA N N，ZHANG X Y，ZHENG H T，et al. Shufflenetv2:Practical guidelines for efficient CNN architecture design [C]// Proceedings of the European Conference on Computer Vision. Munich： Springer，2018：122-138.
[17] 王燕妮，余丽仙 . 注意力与多尺度有效融合的 SSD 目标检测算法 [J]. 计算机科学与探索，2022，16（2）：438-447.
作者简介：易图明（1969.10—），男，汉族，四川南充人，正高级工程师，国务院政府特殊津贴专家，本科，主要研究方向：通信技术；通讯作者：王先全（1968.09—），男，汉族，四川华蓥人，教授，硕士研究生，主要研究方向：计算机软件技术和智能仪器。