遥感图像目标检测与识别主要通过自动分析遥感图像内容，快速确定关键目标的位置、并判断其类别，在制导、瞄准、侦察和防御等军事应用中，目标检测技术是关键的技术支撑。本研究探讨了特征金字塔结构的优化，通过融合高低两级特征，对比了基于多尺度特征融合的特征金字塔Faster R-CNN算法在检测不同尺寸舰船目标时的精度，实现了一种可以针对目标旋转进行自适应的回归网络用于目标检测框的回归，对于简单背景下的舰船检测识别精度较好。本研究在目标尺度变化范围大、方向多变、形状多变的目标具有较好的检测效果。
Object detection and recognition in remote sensing images is mainly based on automatic analysis of the content of remote sensing images to quickly determine the location and category of key objects. In military applications such as guidance, aiming, reconnaissance and defense, target detection technology is the key technical support. In this study, the optimization of the feature pyramid structure is discussed. By fusing the lower-level and higher-level features, the accuracy of the feature pyramid Faster R-CNN algorithm based on multi-scale feature fusion in detecting ship targets of different sizes is compared, and a regression network that can adapt to the rotation of the target is realized for the regression of the target detection frame. The accuracy of ship detection and recognition under simple background is better. This study has a good detection effect on targets with large scale variation range, variable direction and variable shape.
2025,47(20): 170-174 收稿日期：2025-2-18
DOI：10.3404/j.issn.1672-7649.2025.20.026
分类号：U675.79
作者简介：黄猛(1977-),男,硕士,高级工程师,研究方向为软件工程
参考文献：
[1] 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.
[2] HE K, GKIOXARI G, DOLLÁR P, et al. Mask r-cnn[C]//Proceedings of the IEEE international conference on computer vision, 2017.
[3] QIAN X, ZHU X, WANG Y, et al. Rotated Text Detection with Transformer-based Localization Refinement[C]//AAAI. 2021.
[4] LI X, WANG S, ZHANG C, et al. Arbitrary-oriented remote sensing text detection via adaptive rotation feature aggregation[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 1-15.
[5] 王冰. 基于深度学习的舰船目标检测研究[D]. 厦门: 厦门大学2017.
[6] 王腾飞. 高分辨率遥感图像深度学习舰船检测技术研究[D]. 哈尔滨: 哈尔滨工业大学, 2017.
[7] 刘昱龙. 高分辨率光学遥感图像舰船检测研究[D]. 厦门: 厦门大学, 2017.
[8] CHEN S, WANG J, PANG J, et al. Faster R-CNN Revisited: An end-to-end training framework[C]//NeurIPS, 2021.
[9] WANG Y, JIANG Y, CHEN Y, et al. Efficient convolutional networks for multi-task visual recognition[J]. IEEE Transactions on Image Processing, 2020, 29: 7893-7906.
[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] HE K, ZHANG X, REN S, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE transactions on pattern analysis and machine intelligence, 2015, 37(9): 1904-1916.
[12] ZHANG Y, LI H, WANG G, et al. Enhanced faster R-CNN for aerial object detection[J]. Journal of Applied Remote Sensing, 2021, 15(2): 026018.
[13] 殷文斌. 卷积神经网络在遥感目标识别中的应用研究[D]. 北京: 中国科学院大学 (中国科学院遥感与数字地球研究所), 2017.
[14] 刘敏. 高分遥感图像目标在线检测技术研究[D]. 武汉: 武汉大学, 2020.
[15] LI X, WANG L, ZHANG T, et al. Multi-scale feature fusion for remote sensing object detection[J]. Remote Sensing, 2023, 15(5): 1324.
