RCSA-YOLO: Improved SAR Ship Instance Segmentation of YOLOv8

doi:10.3778/j.issn.1002-8331.2401-0445

Abstract

Abstract: Addressing the challenges of low segmentation accuracy in synthetic aperture radar (SAR) images due to complex backgrounds, small targets, and large scale variations, the RCSA-YOLO algorithm is proposed. Firstly, the structural reparameterization technique is used to design the RepBlock feature extraction module, which replaces the original C2f module in the network. This improves the ability of network to extract and represent features, effectively filtering out interference from complex background noise. Secondly, the content-aware reassembly of features (CARAFE) module replaces the nearest neighbor upsampling method, reducing the loss of information in small targets and improving segmentation refinement. Finally, the switchable atrous convolution (SAC) is used for downsampling operations, allowing for the dynamic adjustment of receptive field size. This gives the model better adaptability to multiple scales and ensures segmentation accuracy for different sizes of vessels. Experimental evaluation on the HRSID dataset demonstrates that the proposed algorithm increases the AP50 value of the YOLOv8 model from 87.7% to 90.7%, surpassing the original algorithm by 3 percentage points. Comparative analysis against mainstream instance segmentation algorithms further reveals significant enhancement in SAR ship instance segmentation accuracy, validating the effectiveness of RCSA-YOLO.

Key words: synthetic aperture radar, structural reparameterization, upsampling, switchable atrous convolution

摘要： 针对合成孔径雷达（synthetic aperture radar，SAR）图像中背景复杂、目标小和尺度变化大等导致分割精度低的问题，提出了一种基于改进YOLOv8的SAR图像舰船实例分割算法RCSA-YOLO。利用结构重参数技术设计特征提取模块RepBlock，用以替换原网络中的C2f模块，增强网络的特征提取和特征表达能力，有效过滤了复杂背景噪声的干扰。使用基于内容感知的特征重组模块（content-aware reassembly of features，CARAFE）替换最近邻上采样方法，有效缓解了小目标信息丢失现象，提升了分割精细化程度。使用可切换空洞卷积（switchable atrous convolution，SAC）进行下采样操作，动态调整感受野大小，使模型具备更强的多尺度适应能力，确保了在不同尺寸舰船目标上的分割精度。在HRSID数据集上的实验结果表明，提出的算法可以将YOLOv8模型的AP50值从87.7%提高到90.7%，较原算法提高了3个百分点。与主流的实例分割算法对比，SAR舰船实例分割精度也明显提升，证明了RCSA-YOLO的有效性。

关键词: 合成孔径雷达, 结构重参数化, 上采样, 可切换空洞卷积

WANG Lei, ZHANG Bin, WU Qihong. RCSA-YOLO: Improved SAR Ship Instance Segmentation of YOLOv8[J]. Computer Engineering and Applications, 2024, 60(18): 103-113.

王磊, 张斌, 吴奇鸿. RCSA-YOLO：改进YOLOv8的SAR舰船实例分割[J]. 计算机工程与应用, 2024, 60(18): 103-113.

References

[1] 张上, 陈益方, 王申涛, 等. 基于YOLOv5的改进舰船目标检测算法[J]. 电光与控制, 2023, 30(12): 66-72.
ZHANG S, CHEN Y F, WANG S T, et al. An improved ship target detection algorithm based on YOLOv5[J]. Electronics Optics ＆ Control, 2023, 30(12): 66-72.
[2] LI W, MA P, WANG H, et al. SAR-TSCC: a novel approach for long time series SAR image change detection and pattern analysis[J]. IEEE Transactions on Geoscience and Remote Sensing, 2023, 61: 1-16.
[3] HAMIDI E, PETER B G, MU?OZ D F, et al. Fast flood extent monitoring with SAR change detection using Google earth engine[J]. IEEE Transactions on Geoscience and Remote Sensing, 2023, 61: 1-19.
[4] LI S, WANG Y, CAI H, et al. MF-SRCDNet: multi-feature fusion super-resolution building change detection framework for multi-sensor high-resolution remote sensing imagery[J]. International Journal of Applied Earth Observation and Geoinformation, 2023, 119: 103303.
[5] 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 Z J, LIN B H, QU H C. Improved SAR ship detec-tion algorithm for YOLOv7[J]. Computer Engineering and Applications, 2023, 59(15): 243-252.
[6] SHAO Z, ZHANG X, WEI S, et al. Scale in scale for SAR ship instance segmentation[J]. Remote Sensing, 2023, 15(3): 629.
[7] YASIR M, ZHAN L, LIU S, et al. Instance segmentation ship detection based on improved Yolov7 using complex background SAR images[J]. Frontiers in Marine Science, 2023, 10: 1113669.
[8] 严春满, 王铖. 基于选择性坐标注意力的SAR图像舰船目标检测[J]. 电子学报, 2023, 51(9): 2481-2491.
YAN C M, WANG C. Ship target detection in SAR image based on selective coordinate attention[J]. Acta Electronica Sinica, 2023, 51(9): 2481-2491.
[9] 张帆, 陆圣涛, 项德良, 等. 一种改进的高分辨率SAR图像超像素CFAR舰船检测算法[J]. 雷达学报, 2023, 12(1): 120-139.
ZHANG F, LU S T, XIANG D L, et al. An improved superpixel-based CFAR method for high-resolution SAR image ship target detection[J]. Journal of Radars, 2023, 12(1): 120-139.
[10] 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.
[11] KANG M, LENG X, LIN Z, et al. A modified faster R-CNN based on CFAR algorithm for SAR ship detection[C]//Proceedings of the 2017 International Workshop on Remote Sensing with Intelligent Processing (RSIP), 2017: 1-4.
[12] ZHANG T, ZHANG X. ShipDeNet-20: an only 20 convolu-tion layers and<1-MB lightweight SAR ship detector[J]. IEEE Geoscience and Remote Sensing Letters, 2020, 18(7): 1234-1238.
[13] AN Q, PAN Z, LIU L, et al. DRBox-v2: an improved detector with rotatable boxes for target detection in SAR images[J]. IEEE Transactions on Geoscience and Remote Sensing, 2019, 57(11): 8333-8349.
[14] DOSOVITSKIY A, BEYER L, KOLESNIKOV A, et al. An image is worth 16x16 words: transformers for image recog-nition at scale[J]. arXiv:2010.11929, 2020.
[15] XIA R, CHEN J, HUANG Z, et al. CRTransSar: a visual transformer based on contextual joint representation learning for SAR ship detection[J]. Remote Sensing, 2022, 14(6): 1488.
[16] WEI S, ZENG X, QU Q, et al. HRSID: a high-resolution SAR images dataset for ship detection and instance segmen-tation[J]. IEEE Access, 2020, 8: 120234-120254.
[17] GAO F, ZHONG F, SUN J, et al. BBox-Free SAR ship instance segmentation method based on Gaussian heatmap[J]. IEEE Transactions on Geoscience and Remote Sensing, 2024, 62: 5206218.
[18] YANG X, ZHANG Q, DONG Q, et al. Ship instance segmentation based on rotated bounding boxes for SAR images[J]. Remote Sensing, 2023, 15(5): 1324.
[19] SU H, WEI S, LIU S, et al. HQ-ISNet: high-quality instance segmentation for remote sensing imagery[J]. Remote Sensing, 2020, 12(6): 989.
[20] ZHAO D, ZHU C, QI J, et al. Synergistic attention for ship instance segmentation in SAR images[J]. Remote Sensing, 2021, 13(21): 4384.
[21] GAO F, HUO Y, WANG J, et al. Anchor-free SAR ship instance segmentation with centroid-distance based loss[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2021, 14: 11352-11371.
[22] 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.
[23] WANG J, CHEN K, XU R, et al. Carafe++: unified content-aware reassembly of features[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 44(9): 4674-4687.
[24] QIAO S, CHEN L C, YUILLE A. Detectors: detecting objects with recursive feature pyramid and switchable atrous convolution[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021: 10213-10224.
[25] WANG X, ZHANG R, KONG T, et al. Solov2: dynamic and fast instance segmentation[C]//Advances in Neural Information Processing Systems, 2020: 17721-17732.
[26] BOLYA D, ZHOU C, XIAO F, et al. Yolact: real-time instance segmentation[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2019: 9157-9166.
[27] HE K, GKIOXARI G, DOLLáR P, et al. Mask R-CNN[C]//Proceedings of the IEEE International Conference on Computer Vision, 2017: 2961-2969.
[28] HUANG Z, HUANG L, GONG Y, et al. Mask scoring R-CNN[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019: 6409-6418.
[29] CAI Z, VASCONCELOS N. Cascade R-CNN: high quality object detection and instance segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019, 43(5): 1483-1498.
[30] 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.
[31] LYU C, ZHANG W, HUANG H, et al. Rtmdet: an empirical study of designing real-time object detectors[J]. arXiv:2212.07784, 2022.