针对目前基于深度学习的合成孔径雷达（SAR）图像舰船目标斜框检测方法难以满足实时化检测需求的问题，提出了一种轻量化SAR舰船目标斜框检测方法。该方法基于无锚框框架，设计了轻量化的网络结构，对模型参数量和运行速度进行了优化，可直接从头训练。同时，为解决基于角度回归的斜框参数表示方法存在的角度敏感性问题，提出了基于旋转向量的斜框参数表示方法。在公开的SAR图像舰船目标检测数据集上的实验结果表明，所提方法在无需预训练的情况下取得了与迁移学习方法相近的检测精度，模型参数量和运行速度取得了最优结果，充分验证了所提方法的有效性。
Current oriented ship detection methods in Synthetic Aperture Radar （SAR） images based on deep learning cannot meet the requirements of real-time detection. This paper proposes a lightweight oriented ship detection method. A lightweight network structure is designed based on the anchor-free framework. The number of model parameters and running speed are optimized， so that the model can be directly trained from scratch. To solve the problem of angle sensitivity existing in the method based on angle regression， an oriented bounding box representation method is proposed based on the rotated vector. Experiments are conducted on public SAR ship detection dataset. The results show that the proposed method can reduce model parameters and improve the detection speed while maintaining the detection accuracy， which fully verifies the effectiveness of the method.
[1] HE F S, HE Y, LIU Z G, et al. Research and development on applications of convolutional neural networks of radar automatic target recognition[J]. Journal of Electronics & Information Technology, 2020, 42(1): 119-131 (in Chinese). 贺丰收, 何友, 刘准钆, 等. 卷积神经网络在雷达自动目标识别中的研究进展[J]. 电子与信息学报, 2020, 42(1): 119-131.
[2]
[3] SUN Z Z, DAI M C, LENG X G, et al. An anchor-free detection method for ship targets in high-resolution SAR images[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2021, 14: 7799-7816.
[4] ZHANG T W, ZHANG X L, KE X. Quad-FPN: A novel quad feature pyramid network for SAR ship detection[J]. Remote Sensing, 2021, 13(14): 2771.
[5] DENG Z P, SUN H, ZHOU S L, et al. Learning deep ship detector in SAR images from scratch[J]. IEEE Transactions on Geoscience and Remote Sensing, 2019, 57(6): 4021-4039.
[6] ZHANG P, TANG J S, ZHONG H P, et al. Self-trained target detection of radar and sonar images using automatic deep learning[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 1-14.
[7] ZHANG X H, YAO L B, Lü Y F, et al. Center based model for arbitrary-oriented ship detection in remote sensing images[J]. Acta Photonica Sinica, 2020, 49(4): 0410005 (in Chinese). 张筱晗, 姚力波, 吕亚飞, 等. 基于中心点的遥感图像多方向舰船目标检测[J]. 光子学报, 2020, 49(4): 0410005.
[8] HE Y S, GAO F, WANG J, et al. Learning polar encodings for arbitrary-oriented ship detection in SAR images[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2021, 14: 3846-3859.
[9] LI J W, XU C A, SU H, et al. Ship detection in SAR images based on recurrent feature pyramid network and rotatable bounding box[J]. Journal of Applied Remote Sensing, 2021, 15: 044502.
[10] ZHOU X Y, WANG D Q, KR?HENBüHL P. Objects as points[DB/OL]. arXiv preprint: 1904.07850, 2019.
[11]
[12]
[13]
[14] LAW H, DENG J. CornerNet: Detecting objects as paired keypoints[J]. International Journal of Computer Vision, 2020, 128(3): 642-656.
[15]
[16]
[17]
[18]
[19]
CJA