改进YOLOv7的SAR舰船检测算法

doi:10.3778/j.issn.1002-8331.2304-0109

摘要/Abstract

摘要： 为了解决合成孔径雷达（synthetic aperture radar，SAR）图像中小目标舰船和复杂背景下舰船检测精度低的问题，并使模型更加轻量化，提出了一种改进YOLOv7的SAR舰船检测算法。在YOLOv7主干网络构建REP-PConv-ELAN模块替换原ELAN，减少网络的计算量和内存占用，加快推理速度，同时增强网络对目标特征的提取能力；在特征融合部分融入ConvNeXt Block加速网络提取和融合复杂目标的特征信息；再把全局注意力机制（GAM）加入至下采样阶段，构建一种用来捕捉全局特征的采样模块（MP-GAM），在通道维度和空间维度上进行特征捕捉和特征融合，实现多维信息的交互，提高网络对复杂背景下舰船的关键特征捕捉能力；在检测头的回归损失函数处引入新度量NWD替换IoU，增强对小目标的检测能力。在HRSID数据集上进行了实验对比，改进后的方法相比于YOLOv7，模型的参数量和计算量显著减少；AP值提升了10.04个百分点，准确率提升了3.61个百分点，召回率提升了15.15个百分点。与目前的主流算法对比，精度明显提高。实验结果表明，改进算法能有效提升SAR舰船检测精度，显著改善复杂舰船的误检和漏检情况。

关键词: SAR图像, 舰船检测, YOLOv7, 部分卷积, 全局注意力机制, NWD度量, ConvNeXt Block

Abstract: In order to solve the problem of low accuracy of ship detection for small target ships and complex backgrounds in synthetic aperture radar（SAR） images, while making the model more lightweight, a SAR ship detection algorithm with improved YOLOv7 is proposed. The REP-PConv-ELAN module is built in the YOLOv7 backbone network to replace the original ELAN, which reduces the computation and memory consumption of the network, speeds up the inference speed, and enhances the extraction capability of the network for target features; the ConvNeXt Block is incorporated in the feature fusion part to accelerate the network to extract and fuse the feature information of complex targets. Then the global attention mechanism（GAM） is added to the down-sampling stage to build a sampling module for capturing global features（MP-GAM）, which performs feature capture and feature fusion in channel dimension and spatial dimension to realize the interaction of multidimensional information and improve the ability of network to capture key features of ships in complex backgrounds. A new metric NWD is introduced at the regression loss function of the detection head to replace IoU to enhance the detection capability of small targets. Experimental comparisons are conducted on the HRSID dataset, and the improved method improves the AP value by 10.04?percentage points, the accuracy by 3.61?percentage points, and the recall by 15.15?percentage points compared to YOLOv7. The accuracy is significantly improved compared with the current mainstream algorithm. The experimental results show that the improved algorithm can effectively improve the false and missed detections of ships.

Key words: SAR images, ship detection, YOLOv7, partial convolution, global attention mechanism, NWD metric, ConvNeXt Block

肖振久, 林渤翰, 曲海成. 改进YOLOv7的SAR舰船检测算法[J]. 计算机工程与应用, 2023, 59(15): 243-252.

XIAO Zhenjiu, LIN Bohan, QU Haicheng. Improved SAR Ship Detection Algorithm for YOLOv7[J]. Computer Engineering and Applications, 2023, 59(15): 243-252.

参考文献

[1] KEEL B M.Constant false alarm rate detectors[J].IET Digital Library，2010.
[2] REN S，HE K，GIRSHICK R，et al.Faster R-CNN：towards real-time object detection with region proposal networks[C]//Advances in Neural Information Processing Systems，2015.
[3] HE K，GKIOXARI G，DOLLáR P，et al.Mask R-CNN[C]//Proceedings of the IEEE International Conference on Computer Vision，2017：2961-2969.
[4] CARION N，MASSA F，SYNNAEVE G，et al.End-to-end object detection with transformers[C]//16th European Conference on Computer Vision，Glasgow，UK，August 23-28，2020：213-229.
[5] LIU W，ANGUELOV D，ERHAN D，et al.SSD：single shot multibox detector[C]//14th European Conference on Computer Vision，Amsterdam，The Netherlands，October 11-14，2016：21-37.
[6] LIN T Y，GOYAL P，GIRSHICK R，et al.Focal loss for dense object detection[C]//Proceedings of the IEEE International Conference on Computer Vision，2017：2980-2988.
[7] DUAN K，BAI S，XIE L，et al.Centernet：keypoint triplets for object detection[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision，2019：6569-6578.
[8] TAN M，PANG R，LE Q V.Efficientdet：scalable and efficient object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition，2020：10781-10790.
[9] TIAN Z，SHEN C，CHEN H，et al.Fcos：fully convolutional one-stage object detection[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision，2019：9627-9636.
[10] REDMON J，DIVVALA S，GIRSHICK R，et al.You only look once：unified，real-time object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition，2016：779-788.
[11] REDMON J，FARHADI A.YOLO9000：better，faster，stronger[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition，2017：7263-7271.
[12] REDMON J，FARHADI A.Yolov3：an incremental improvement[J].arXiv：1804.02767，2018.
[13] BOCHKOVSKIY A，WANG C Y，LIAO H Y M.Yolov4：optimal speed and accuracy of object detection[J].arXiv：2004.10934，2020.
[14] GLENN J.YOLOv5 releases v6.1[EB/OL].（2022-02-07）[2022?03?10].https：//github.com/ultralytics/yolov5/releases/tag/v6.1.
[15] GE Z，LIU S，WANG F，et al.Yolox：exceeding yolo series in 2021[J].arXiv：2107.08430，2021.
[16] WANG C Y，BOCHKOVSKIY A，LIAO H Y M.YOLOv7：trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[J].arXiv：2207.02696，2022.
[17] 徐守坤，顾佳楠，庄丽华，等.基于两阶段计算Transformer的小目标检测[J/OL].计算机科学与探索：1-21[2023-04-14].http：//kns.cnki.net/kcms/detail/11.5602.tp.20230411.
1050.006.html.
XU S K，GU J N，ZHUANG L H，et al.Small object detection based on two-stage calculation transformer[J/OL].Journal of Frontiers of Computer Science and Technology：1-21[2023-04-14].http：//kns.cnki.net/kcms/detail/11.5602.
tp.20230411.1050.006.html
[18]   赵鹏飞，谢林柏，彭力.融合注意力机制的深层次小目标检测算法[J].计算机科学与探索，2022，16（4）：927-937.
ZHAO P F，XIE L B，PENG L I.Deep small object detection algorithm integrating attention mechanism[J].Journal of Frontiers of Computer Science and Technology，2022，16（4）：927-937.
[19] CHEN Z，LIU C，FILARETOV V F，et al.Multi-scale ship detection algorithm based on YOLOv7 for complex scene SAR images[J].Remote Sensing，2023，15（8）：2071.
[20] ZHOU K，ZHANG M，WANG H，et al.Ship detection in SAR images based on multi-scale feature extraction and adaptive feature fusion[J].Remote Sensing，2022，14（3）：755.
[21] ZHU H，XIE Y，HUANG H，et al.DB-YOLO：a duplicate bilateral YOLO network for multi-scale ship detection in SAR images[J].Sensors，2021，21（23）：8146.
[22] GUO H，BAI H，YUAN Y，et al.Fully deformable convolutional network for ship detection in remote sensing imagery[J].Remote Sensing，2022，14（8）：1850.
[23]   谭显东，彭辉.改进YOLOv5的SAR图像舰船目标检测[J].计算机工程与应用，2022，58（4）：247-254.
TAN X D，PENG H.Improved YOLOv5 ship target detection in SAR image[J].Computer Engineering and Applications，2022，58（4）：247-254.
[24]   张廓，陈章进，张岩.改进YOLOv4-Tiny的SAR图像目标快速检测方法[J].计算机工程与应用，2023，59（14）：209-216.
ZHANG K，CHEN Z J，ZHANG Y.Improved YOLOv4-tiny fast target detection method for SAR image[J].Computer Engineering and Applications，2023，59（14）：209-216.
[25] LIU Y，WANG X.SAR ship detection based on improved YOLOv7-tiny[C]//2022 IEEE 8th International Conference on Computer and Communications（ICCC），2022：2166-2170.
[26] CHEN J，KAO S，HE H，et al.Run，don’t walk：chasing higher FLOPS for faster neural networks[J].arXiv：2303.03667，2023.
[27] DING X，ZHANG X，MA N，et al.Repvgg：making vgg-style convnets great again[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition，2021：13733-13742.
[28] LIU Z，MAO H，WU C Y，et al.A convnet for the 2020s[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition，2022：11976-11986.
[29] LIU Y，SHAO Z，HOFFMANN N.Global attention mechanism：retain information to enhance channel-spatial interactions[J].arXiv：2112.05561，2021.
[30] WANG J，XU C，YANG W，et al.A normalized Gaussian Wasserstein distance for tiny object detection[J].arXiv：2110.13389，2021.
[31] WEI S，ZENG X，QU Q，et al.HRSID：a high-resolution SAR images dataset for ship detection and instance segmentation[J].IEEE Access，2020，8：120234-120254.