强化学习驱动的复杂遥感场景目标检测方法-航天遥感图像智能处理与分析

  • 刘文林 ,
  • 胡锡坤 ,
  • 钟平
展开
  • 国防科技大学电子科学学院

收稿日期: 2025-10-09

  修回日期: 2025-11-27

基金资助

国家自然科学基金

Reinforcement learning-driven object detection method for complex remote sensing scenes

  • LIU Wen-Lin ,
  • HU Xi-Kun ,
  • ZHONG Ping
Expand

Received date: 2025-10-09

  Revised date: 2025-11-27

Supported by

National Natural Science Foundation of China

摘要

卫星遥感图像目标检测技术是当前对地观测与智能解译的重要手段。然而,现有研究多集中于理想成像条件下,对于复杂天气、大气扰动及噪声干扰等复杂退化条件下的目标检测能力仍显不足。针对这一问题,本文提出了一种强化学习驱动的退化遥感图像目标检测方法,通过动态编排图像预处理算子以实现复杂场景下的鲁棒检测。该方法的核心思想是以目标检测性能为优化目标,利用强化学习的决策优势,自适应地迭代选择并组合图像去噪、去模糊、对比度增强等预处理操作,从而提升遥感影像的质量与检测精度。在使用YOLO11-OBB检测器基于DIOR和DOTA卫星遥感数据集构建的退化场景上进行的实验表明,所提方法均取得优异表现:在DIOR数据集上,相较于Raw-Syn(原始数据训练、退化场景数据验证)和Syn-Syn(退化场景数据训练与验证)方案,mAP50分别提升11.1%和2.5%,最终达80.8%;在DOTA数据集上,mAP50较Raw-Syn和Syn-Syn分别提升7.2%和2.8%,最终达76.6%。同时,处理后遥感影像的图像质量明显提升(PSNR>25dB),充分验证了所提方法在复杂环境下的有效性与适用性。

本文引用格式

刘文林 , 胡锡坤 , 钟平 . 强化学习驱动的复杂遥感场景目标检测方法-航天遥感图像智能处理与分析[J]. 航空学报, 0 : 0 -0 . DOI: 10.7527/S1000-6893.2025.32861

Abstract

Satellite remote sensing image object detection constitutes a pivotal technique for earth observation and intelligent interpretation. However, most existing research has concentrated on ideal imaging conditions, and the capability for object detection under complex degradations, such as adverse weather, atmospheric turbulence, and noise interference, remains notably inadequate. To address this limitation, this paper proposes a reinforcement learning-based adaptive object detection methodology that dynamically orchestrates image preprocessing operators to achieve robust detection in complex scenarios. The core principle is to optimize object detection performance by leveraging reinforcement learning's decision-making capability to adaptively and iteratively select and compose preprocessing operations, including denoising, deblurring, and contrast enhancement, thereby enhancing both remote sensing imagery quality and detection precision. Experiments on the degraded scenarios constructed from the DIOR and DOTA satellite remote-sensing datasets with the YOLO11-OBB detector demonstrate that the proposed method achieves superior performance in all cases: on DIOR, the proposed method achieves mAP50 improvements of 11.1 and 2.5 points over Raw-Syn (trained on pristine data, validated on degraded data) and Syn-Syn (trained and validated on degraded data) baselines, respectively, attaining a final mAP50 of 80.8; on DOTA, it improves mAP50 by 7.2 and 2.8 points over the same baselines, reaching 76.6. Furthermore, the quality of processed remote sensing imagery is significantly enhanced (PSNR > 25 dB), substantiating the efficacy and applicability of the proposed approach in challenging environments.

参考文献

[1] 臧晶, 李成华, 田野. 卫星遥感农业监测系统中实例检索算法研究[J]. 宇航学报, 2019, 40(11): 1358-1366.</br> ZANG J, LI C H, TIAN Y. Research on instance retrieval algorithm in satellite remote sensing agricultural monitoring system[J]. Journal of Astronautics, 2019, 40(11): 1358-1366 (in Chinese).</br>[2]李志忠, 卫征, 付垒, 等. 我国遥感卫星技术与应用重要进展[J]. 卫星应用, 2025, (4): 16-19.</br>LI Z Z, WEI Z, FU L, et al. Important progress of China's remote sensing satellite technology and application[J]. Satellite Application, 2025, (4): 16-19 (in Chinese).</br>[3]王俊杰, 李清泉, 邬国锋. 红树林定量遥感研究进展[J]. 遥感学报, 2025, (6): 1769-1787.</br>WANG J J, LI Q Q, WU G F. Progress in quantitative remote sensing of mangroves[J]. Journal of Remote Sensing, 2025, (6): 1769-1787 (in Chinese).</br>[4]莫妮卡. 卫星遥感图像舰船目标检测系统[D]. 杭州:浙江大学,2022.</br>MO N K. Ship target detection system for satellite remote sensing images[D]. Hangzhou: Zhejiang University, 2022 (in Chinese).</br>[5]刘瑞锦, 何章鸣. 基于YOLOv8的卫星遥感图像快速目标检测方法[J]. 空间控制技术与应用, 2023, 49(5): 89-97.</br> LIU R J, HE Z M. Fast object detection method for satellite remote sensing images based on YOLOv8[J]. Aerospace Control and Application, 2023, 49(5): 89-97 (in Chinese).</br>[6]赵其昌, 吴一全, 苑玉彬. 光学遥感图像舰船目标检测与识别方法研究进展[J]. 航空学报, 2024, 45(8): 51-84.</br> ZHAO Q C, WU Y Q, YUAN Y B. Research progress of ship target detection and recognition methods for optical remote sensing images[J]. Acta Aeronautica et Astronautica Sinica, 2024, 45(8): 51-84 (in Chinese).</br>[7]XI Y, JIA W J, MIAO Q G, et al. CoDerainNet: Collaborative deraining network for drone-view object detection in rainy weather conditions[J]. Remote Sensing, 2023, 15(1487): 1487. </br>[8]Aswini N, Uma S V. Drone image de-noising and feature extraction[C]//2020 IEEE International Conference for Innovation in Technology (INOCON). Bengaluru, India: IEEE, 2020: 1-6.</br>[9]Jae-In K, Chang-Uk H, Hyangsun H, et al. Digital surface model generation for drifting Arctic sea ice with low-textured surfaces based on drone images[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2021, 172: 147-159. </br>[10]QIAN G, WANG Y, GU J, et al. Rethinking learning-based demosaicing, denoising, and super-resolution pipeline[C]//2022 IEEE International Conference on Computational Photography (ICCP). Cluj-Napoca, Romania: IEEE, 2022: 1-12.</br>[11]XING W Z, EGIAZARIAN K. End-to-end learning for joint image demosaicking, denoising and super-resolution[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2021: 3507-3516. </br>[12]SUGANUMA M, LIU X, OKATANI T. Attention-based adaptive selection of operations for image restoration in the presence of unknown combined distortions[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2019: 9039-9048.</br>[13]KIM C, KIM T H, BAIK S. LAN: Learning to adapt noise for image denoising[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024: 25193-25202.</br>[14]LIU Y, LI W, GUAN J, et al. Effective cloud removal for remote sensing images by an improved mean-reverting denoising model with elucidated design space[C]//Proceedings of the IEEE/CVF Computer Vision and Pattern Recognition Conference. 2025: 17851-17861.</br>[15]ZHANG J, ZHANG Q, ZHAO X, et al. Boosting denoisers with reinforcement learning for image restoration[J]. Soft Computing, 2022, 26(7): 3261-3272.</br>[16]RYUSUKE F, NAOTO I, TOSHIHIKO Y. PixelRL: Fully convolutional network with reinforcement learning for image processing[J]. IEEE Transactions on Multimedia, 2020, 22(7): 1704-1719.</br>[17]YU K, DONG C, LI L, et al. Crafting a toolchain for image restoration by deep reinforcement learning[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA: CVF, 2018: 2443-2452.</br>[18]UKCHEOL S, KYUNGHYUN L, IN S K. DRL-ISP: Multi-objective camera ISP with deep reinforcement learning[C]//2022 IEEE/RSJ International Conference on Intelligent Robots and Systems, Kyoto, Japan: IEEE, 2022: 7044-7051.</br>[19]YU K, WANG X T, DONG C, et al. Path-restore: Learning network path selection for image restoration[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 44(10): 7078-7092.</br>[20]WEI Z Y, CHEN H H, NAN L L, et al. PathNet: Path-Selective Point Cloud Denoising[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024, 46(6): 4426-4442, doi: 10.1109/TPAMI.2024.3355988.</br>[21]范天麒, 邹征夏, 史振威. 基于强化学习数据合成的典型遥感目标检测[J]. 航空学报, doi: 10.7527/S1000-6893.2025.31955. </br>FAN T Q, ZOU Z X, SHI Z W. Typical remote sensing object detection based on reinforcement learning data synthesis[J]. Acta Aeronautica et Astronautica Sinica, doi: 10.7527/s1000-6893.2025.31955 (in Chinese).</br>[22]HE K M, ZHANG X Y, REN S Q, et al. Deep Residual Learning for Image Recognition[C]//2016 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, NV, USA: CVF, 2016:770-778.</br>[23]HAARNOJA T, ZHOU A, ABBEEL P, et al. Soft actor-critic: Off-policy maximum entropy deep reinforcement learning with a stochastic actor[C]//Proceedings of the 36th International Conference on Machine Learning. New York: PMLR, 2019: 2879-2888.</br>[24]RONNEBERGER O, FISCHER P, BROX T. U-Net: Convolutional networks for biomedical image segmentation[C]//International Conference on Medical Image Computing and Computer-Assisted Intervention. Munich, Germany, 2015-10: 234-241. Cham: Springer International Publishing, 2015.</br>[25]Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A. A., Veness, J., Bellemare, M. G., ... & Petersen, S. (2015). Human-level control through deep reinforcement learning.?Nature, 518(7540), 529-533. DOI: 10.1038/nature14236.</br>[26]Schulman, J., Wolski, F., Dhariwal, P., Radford, A., & Klimov, O. (2017). Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347.</br>[27]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 and Machine Intelligence, 2016, 39(6): 1137-1149.</br>[28]ULTRALYTICS. YOLOv11: Multi-feature fusion real-time object detection framework (Version 1.0)[EB/OL]. Hangzhou: Ultralytics Inc., (2025-02-03) [2025-09-30]. https://github.com/ultralytics/ultralytics/tree/main/yolo1</br>[29]XIE X X, CHENG G, WANG J B, et al. Oriented R-CNN for object detection[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. IEEE, 2021: 3520-3529.</br>[30]LIU Z, LIN Y T, CAO Y, et al. Swin Transformer: Hierarchical vision transformer using shifted windows[C]//Proceedings of the 2021 IEEE/CVF International Conference on Computer Vision (ICCV). Montreal, Canada: IEEE/CVF, 2021: 1001-1010.</br>[31]LI K, WAMG G, CHENG G, et al. Object detection in optical remote sensing images: A survey and a new benchmark[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2020, 159: 296-307.</br>[32]XIA G S, BAI X, DING J, et al. DOTA: A large-scale dataset for object detection in aerial images[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. IEEE, 2018: 3974-3983.
Options
文章导航

/