摘 要：遥感图像普遍目标尺度变化较大，背景较为复杂，这些问题导致当前目标检测算法出现漏检或检测效果不佳等现象。文章选择 YOLOv3 算法作为基础网络进行改进，对训练数据做 Mosaic 数据增强，增加困难样本数量。结合 ECA 通道注意力机制，丰富特征的表达能力。使用 CIOU 损失作为定位损失，增强了目标回归框的检测精度和收敛速度。在遥感图像数据集 RSOD 上进行训练和测试，实验结果表明，改进的 YOLOv3 检测效果更佳，鲁棒性更强，具有很好的实用价值。
关键词：深度学习；目标检测；数据增强；注意力机制；损失函数
DOI:10.19850/j.cnki.2096-4706.2022.10.021
中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2022）10-0087-04
Remote Sensing Image Object Detection Based on Improved YOLO Algorithm in Complex Scenes
BU Huili
(School of Computer Science and Engineering, Anhui University of Science and Technology, Huainan 232001, China)
Abstract: The scale of common objects in remote sensing images varies greatly, and the background is more complex. These problems lead to the phenomenon of missing detection or poor detection effect in current object detection algorithms. In this paper, YOLOv3 algorithm is selected as the basic network for improvement and Mosaic data is used to enhance the training data to increase the number of difficult samples. Combined with the attention mechanism of ECA channel, it enriches the expression ability of features. Using CIOU loss as location loss, the detection accuracy and convergence speed of the target regression frame are enhanced. Training and testing are carried out on remote sensing image data set RSOD. Experimental results show that the improved YOLOv3 has better detection effect and stronger robustness, and it has good practical value.
Keywords: deep learning; object detection; data enhancement; attention mechanism; loss function
参考文献：
[1] KRIZHEVSKY A，SUTSKEVER I，Hin ton G E. Imagenet classification with deep convolutional neural networks [J].Communications of the ACM，2017，60（6）：84-90.
[2] GIRSHICK R. Fast r-cnn [C]//Proceedings of the IEEE international conference on computer vision. IEEE， 2015：1440-1448.
[3] REN S，HE K，GIRSHICK R，et al. Faster R-CNN： Towards Real-Time Object Detection with Region Proposal Networks [J].IEEE Transactions on Pattern Analysis & Machine Intelligence，2017，39（6）：1137-1149.
[4] HE K，GKIOXARI G，DOLLÁR P，et al. Mask r-cnn [C]//Proceedings of the IEEE international conference on computer vision.IEEE，2017：2961-2969.
[5] REDMON J，FARHADI A. Yolov3：An incremental improvement [J].arXiv：1804.02767 [cs.CV].[2022-03-08]. https：//arxiv.org/abs/1804.02767?context=cs.CV.
[6] LIU W，ANGUELOV D，ERHAN D，et al. Ssd： Single shot multibox detector [C]//European conference on computer vision.Springer，2016：21-37.
[7] 李婕，周顺，朱鑫潮，等 . 结合多通道注意力的遥感图像飞机目标检测 [J]. 计算机工程与应用，2022，58（1）：209-217.
[8] 周雪珂，刘畅，周滨 . 多尺度特征融合与特征通道关系校准的 SAR 图像船舶检测 [J]. 雷达学报，2021，10（4）：531-543.
[9] YE K，FANG Z，HUANG X，et al. Research on small target detection algorithm based on improved yolov3 [C]//2020 5th International Conference on Mechanical， Control and Computer Engineering （ICMCCE）.Harbin： IEEE，2020：1467-1470.
[10] 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 （CVPR）.Seattle：IEEE， 2020：11531-11539.
[11] ZHENG Z，WANG P，LIU W，et al. Distance-IoU loss：Faster and better learning for bounding box regression [C]//Proceedings of the AAAI Conference on Artificial Intelligence，2020：12993-13000.
[12] 王艺皓，丁洪伟，李波，等 . 复杂场景下基于改进YOLOv3 的口罩佩戴检测算法 [J]. 计算机工程，2020，46（11）：12-22.
作者简介：卜荟力（1996—），男，汉族，江苏徐州人，硕士研究生在读，研究方向：深度学习目标检测。