Infrared Ship Detection Based on Multi-scale Semantic Network
-
摘要: 为了增强舰船检测的抗干扰性能,本文提出了一种有效且稳定的单阶段舰船检测网络,该网络主要由3个模块组成:特征优化模块,特征金字塔融合模块和上下文增强模块,其中特征优化模块是提取多尺度上下文信息,并进一步细化和增强顶层特征输入特性,增强弱小目标检测性能;特征金字塔融合模块能够生成表征能力更强的语义信息;上下文增强模块则是整合局部和全局特征增强网络特征表达能力,以降低复杂背景对检测性影响,平衡前景和背景的不均衡差异,消除鱼鳞波的影响。为了验证本文所提方法的有效性和鲁棒性,本文对自建的舰船数据集进行了定性定量验证。实验结果表明,相比现有最新基准对比模型,本文所提网络在自建数据集上均达到了最优性能,在不增加复杂度的情况下极大提升了检测精度。Abstract: To enhance the anti-jamming performance of ship detection, an effective and stable single-stage ship detection network is proposed in this study. The network is composed of three modules: feature optimization, feature pyramid fusion, and context enhancement modules. The feature optimization module extracts multi-scale context information and further refines the high-level feature input characteristics, to enhance the performance of dim–small object detection. The feature pyramid fusion module can generate semantic information with stronger representation ability. The context enhancement module integrates local and global features to enhance the network feature expression ability, reduce the impact of a complex background on detectability, adjust the imbalance between the foreground and background, and eliminate the impact of scale-wave. To verify the effectiveness and robustness of the proposed method, qualitative and quantitative verifications are performed on a self-built dataset. Experimental results show that the proposed network achieves optimal performance compared with the latest benchmark comparison model and considerably improves the detection accuracy without increasing complexity.
-
Key words:
- object detection /
- infrared ship /
- single-stage network /
- pyramid pooling /
- context enhancement
-
表 1 不同模块的消融结果
Table 1. Ablation results of different modules
MCI SI Fusion P mAP R F1 71.1 76.3 82.7 86.5 √ 74.5 76.9 83.2 86.6 √ √ 78.2 78.2 83.5 87.2 √ √ √ 80.5 79.2 85.0 88.8 表 2 自建数据集上的检测结果对比
Table 2. Comparison of results on non-public data sets
Models P mAP R F1 YOLOv3 75.5 74.2 81.3 83.9 RetinaNet 77.3 80.6 78.9 77.4 RefineNet 78.4 83.1 79.3 81.1 CenterNet 77.1 78.6 84.5 88.7 FCOS 78.7 85.1 76.6 86.5 Ours 80.5 79.2 85.0 88.8 表 3 不同数据子集上的检测结果对比
Table 3. Comparison results for different sub-set
Models SOS CBC Others P mAP R F1 P mAP R F1 P mAP R F1 YOLOv3 67.3 67.4 70.6 68.9 72.1 80.6 83.0 88.3 76.4 85.1 81.5 76.0 RetinaNet 66.6 70.3 72.5 69.4 75.4 81.1 83.1 83.3 78.5 84.5 85.8 76.6 RefineNet 64.8 78.8 78.3 70.9 73.7 82.3 85.0 89.1 77.2 89.6 86.4 75.7 CenterNet 67.8 74.6 79.6 73.2 73.6 77.1 81.9 93.0 79.3 78.9 80.1 85.5 FCOS 64.8 80.8 78.3 70.9 72.5 76.3 82.4 86.2 78.7 77.7 78.5 85.9 Ours 68.0 83.3 83.6 74.9 73.9 85.2 84.9 90.1 83.5 85.4 87.6 86.0 -
[1] 邢莎, 吉林, 雍杨, 等. 复杂场景下的自动红外舰船目标检测[J]. 红外技术, 2014, 36(4): 320-325. http://hwjs.nvir.cn/article/id/hwjs201404013XING Sha, JI Lin, YONG Yang, et al. Automatic infrared warship target detection under complicated background[J]. Infrared Technology, 2014, 36(4): 320-325. http://hwjs.nvir.cn/article/id/hwjs201404013 [2] 赵文涛, 曹昕鸷, 田志勇. 基于自适应阈值区域生长的红外舰船目标分割方法[J]. 红外技术, 2018, 40(2): 158-163. http://hwjs.nvir.cn/article/id/hwjs201802010ZHAO Wentao, CAO Xinzhi, TIAN Zhiyong. An infrared ship target segmentation method based on adaptive threshold region growth[J]. Infrared Technology, 2018, 40(2): 158-163. http://hwjs.nvir.cn/article/id/hwjs201802010 [3] 丁荣莉, 韩传钊, 谢宝蓉, 等. 红外遥感图像舰船目标检测[J]. 红外技术, 2019, 41(2): 127-133. http://hwjs.nvir.cn/article/id/hwjs201902004DING Rongli, HAN Chuanzhao, XIE Baorong, et al. Ship target detection in infrared remote-sensing image[J]. Infrared Technology, 2019, 41(2): 127-133. http://hwjs.nvir.cn/article/id/hwjs201902004 [4] Kumar D, ZHANG X. Ship detection based on faster R-CNN in SAR imagery by anchor box optimization[C]//2019 International Conference on Control, Automation and Information Sciences (ICCAIS), 2019: 309-313. [5] Redmon J, Farhadi A. YOLOv3: an incremental improvement[J]. arXiv: 1804.02767, 2018. [6] LIU W, Anguelov D, Erhan D, et al. SSD: single shot multibox detector[C]//European Conference on Computer Vision, 2016: 21-37. [7] LIN T Y, Goyal P, Girshick R, et al. Focal loss for dense object detection[C]//International Conference on Computer Vision, 2017: 2999-3007. [8] LAW H, DENG J. CornerNet: detecting objects as paired keypoints[C]// European Conference on Computer Vision, 2018: 765-781. [9] DUAN K, BAI S, XIE L, et al. CenterNet: keypoint triplets for object detection[J/OL]. arXiv: 1904.08189, 2019. [10] TAN M, PANG R, LE Q V, et al. EfficientDet: scalable and efficient object detection[J/OL]. arXiv: 1911.09070, 2019. [11] 杨涛, 戴军, 吴钟建, 等. 基于深度学习的红外舰船目标识别[J]. 红外技术, 2020, 42(5): 426-433. http://hwjs.nvir.cn/article/id/hwjs202005003YANG Tao, DAI Jun, WU Zhongjian, et al. Target recognition of infrared ship based on deep learning[J]. Infrared Technology, 2020, 42(5): 426-433. http://hwjs.nvir.cn/article/id/hwjs202005003 [12] 黄洁, 姜志国, 张浩鹏, 等. 基于卷积神经网络的遥感图像舰船目标检测[J]. 北京航空航天大学学报, 2017(9): 132-139. https://www.cnki.com.cn/Article/CJFDTOTAL-BJHK201709016.htmHUANG Jie, JIANG Zhiguo, ZHANG Haopeng, et al. Ship target detection in remote sensing image based on convolutional neural network[J]. Journal of Beijing University of Aeronautics and Astronautics, 2017(9): 132-139. https://www.cnki.com.cn/Article/CJFDTOTAL-BJHK201709016.htm [13] 顾佼佼, 李炳臻, 刘克, 等. 基于改进Faster R-CNN的红外舰船目标检测算法[J]. 红外技术, 2021, 43(2): 170-178. http://hwjs.nvir.cn/article/id/6dc47229-7cdb-4d62-ae05-6b6909db45b9GU Jiaojiao, LI Bingzhen, LIU Ke, et al. Infrared ship target detection algorithm based on improved faster R-CNN[J]. Infrared Technology, 2021, 43(2): 170-178. http://hwjs.nvir.cn/article/id/6dc47229-7cdb-4d62-ae05-6b6909db45b9 [14] GUO Haoyuan, YANG Xi, WANG Nannan, et al. A CenterNet++ model for ship detection in SAR images[J]. Pattern Recognition, 2021, 112(88): 25-34. [15] FAN S, ZHU F, CHEN S, et al. FII-CenterNet: an anchor-free detector with foreground attention for traffic object detection[J]. IEEE Transactions on Vehicular Technology, 2021, 120(99): 1-14. [16] Barrios J M, Bustos B. Competitive content-based video copy detection using global descriptors[J]. Multimedia Tools and Applications, 2013, 62(1): 75-110. [17] ZHANG Xiang, YANG Wei, TANG Xiaolin, et al. A fast learning method for accurate and robust lane detection using two-stage feature extraction with YOLOv3[J]. Sensors, 2018, 18(12): 4308-4315. [18] LIN G, Milan A, SHEN C, et al. RefineNet: multi-path refinement networks for high-resolution semantic segmentation[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017: 1289-1301. [19] TIAN Z, SHEN C, CHEN H, et al. FCOS: fully convolutional one-stage object detection[C]//2019 IEEE/CVF International Conference on Computer Vision (ICCV), 2020: 1099e3233. [20] LIU Z, ZHANG X, JIANG T, et al. Infrared salient object detection based on global guided lightweight non-local deep features[J]. Infrared Physics & Technology, 2021, 12(3): 2309-2315.