Please wait a minute...
 
自然资源遥感  2021, Vol. 33 Issue (3): 97-106    DOI: 10.6046/zrzyyg.2020372
  技术方法 本期目录 | 过刊浏览 | 高级检索 |
多尺度特征增强的遥感图像舰船目标检测
刘万军(), 高健康(), 曲海成, 姜文涛
辽宁工程技术大学软件学院,葫芦岛 125105
Ship detection based on multi-scale feature enhancement of remote sensing images
LIU Wanjun(), GAO Jiankang(), QU Haicheng, JIANG Wentao
College of Software, Liaoning Technical University, Huludao 125105, China
全文: PDF(5105 KB)   HTML  
输出: BibTeX | EndNote (RIS)      
摘要 

针对背景复杂的遥感图像中,舰船方向任意、密集排列造成的漏检问题,基于旋转区域检测网络,提出多尺度特征增强的遥感图像舰船目标检测算法。在特征提取阶段,利用密集连接感受野模块改进特征金字塔网络,选用不同空洞率的卷积获取多尺度感受野特征,增强高层语义信息的表达; 为了抑制噪声并突出目标特征,在特征提取后设计基于注意力机制的特征融合结构,根据各层在空间上的权重值融合所有层,得到兼顾语义信息和位置信息的特征层,再对该层特征进行注意力增强,将增强后的特征融入原金字塔特征层; 在分类和回归损失基础上,增加注意力损失,优化注意力网络,给予目标位置更多关注。在DOTA遥感数据集上的实验结果表明,该算法平均检测精度可以达到71.61%,优于最新的遥感图像舰船目标检测算法,有效地解决了目标漏检问题。

服务
把本文推荐给朋友
加入引用管理器
E-mail Alert
RSS
作者相关文章
刘万军
高健康
曲海成
姜文涛
关键词 卷积神经网络多尺度特征融合注意力机制遥感图像舰船目标检测    
Abstract

Aiming at the omission in the ship target detection from remote sensing images with complex background caused by the arbitrary and dense arrangement of ships, this study, based on the rotation region generation network, proposes a ship target detection algorithm using the multi-scale feature enhancement of remote sensing images. The detailed steps are as follows. Firstly, improve the feature pyramid network using the receptive field module with dense connection at the feature extraction stage. Then obtain the characteristics of multi-scale receptive fields using the convolution of different dilate rates. In this way, the expression of high-level semantic information can be enhanced. Then design a feature fusion structure based on attention mechanisms to restrain noise and highlight the target characteristics. Afterward, fuse all layers according to the spatial weight value of each layer to obtain a feature layer that takes both semantic and position information into account. Then conduct attention enhancement to the features of this layer, and integrate the enhanced features into the original feature layer in the pyramid network. Consequently, pay more attention to target locations by increasing attention loss and optimizing the attention network according to the classification and regression loss. As indicated by the experiment results of DOTA remote sensing dataset, the average precision of this algorithm is as high as 71.61%, which is higher than the latest ship target detection algorithm based on remote sensing images. In this manner, the omission in ship target detection can be effectively solved.

Key wordsconvolution neural network    multi-scale feature fusion    attention mechanism    remote sensing image    ship target detection
收稿日期: 2020-11-23      出版日期: 2021-09-24
ZTFLH:  TP751.1  
基金资助:国家自然科学基金青年基金项目“面向宽幅高光谱遥感影像的高效压缩方法研究”(41701479);辽宁工程技术大学学科创新团队资助项目“智慧农业遥感监测创新团队”(LNTU20TD-23)
通讯作者: 高健康
作者简介: 刘万军(1959-),男,教授,主要研究方向为数字图像处理、运动目标检测与跟踪。Email: liuwanjun@lntu.edu.cn
引用本文:   
刘万军, 高健康, 曲海成, 姜文涛. 多尺度特征增强的遥感图像舰船目标检测[J]. 自然资源遥感, 2021, 33(3): 97-106.
LIU Wanjun, GAO Jiankang, QU Haicheng, JIANG Wentao. Ship detection based on multi-scale feature enhancement of remote sensing images. Remote Sensing for Natural Resources, 2021, 33(3): 97-106.
链接本文:  
https://www.gtzyyg.com/CN/10.6046/zrzyyg.2020372      或      https://www.gtzyyg.com/CN/Y2021/V33/I3/97
Fig.1  FPN结构
Fig.2  方向包围框的表示
Fig.3  总体结构
Fig.4  密集连接感受野模块
Fig.5  特征融合结构
Fig.6  双重注意力网络
Fig.7  注意力网络可视化
实验方法 召回率/% 精确率/% AP/% 检测时间/s
基础网络 76.56 87.86 67.37 0.21
+DCRF 80.82 85.92 69.52 0.21
+AFF 80.99 85.82 69.66 0.22
本文方法 81.74 87.04 71.61 0.22
Tab.1  不同模块的消融实验结果
类型 示例1 示例2 示例3
基础网络结果图
DCRF结果图
MFEDet结果图
基础网络特征图
MFEDet特征图
Tab.2  不同模块的结果展示
对比方法 召回率 精确率 AP
FR-O 58.53 65.60 39.24
R-DFPN 67.85 87.67 59.78
RRPN 69.35 89.68 63.42
RADet 68.86
本文方法 81.74 87.04 71.61
Tab.3  不同方法的对比结果
方法 训练时间 测试时间
FR-O 0.34 0.10
RRPN 0.85 0.35
R-DFPN 1.15 0.38
Baseline 0.58 0.21
本文方法 0.64 0.22
Tab.4  不同方法的训练时间和测试时间
[1] 王彦情, 马雷, 田原. 光学遥感图像舰船目标检测与识别综述[J]. 自动化学报, 2011, 37(9):1029-1039.
Wang Y Q, Ma L, Tian Y. Overview of ship target detection and recognition based on optical remote sensing image[J]. Acta Automatica Sinica, 2011, 37(9):1029-1039.
[2] 谢奇芳, 姚国清, 张猛. 基于Faster R-CNN的高分辨率图像目标检测技术[J]. 国土资源遥感, 2019, 31(2):38-43.doi: 10.6046/gtzyyg.2019.02.06.
doi: 10.6046/gtzyyg.2019.02.06
Xie Q F, Yao G Q, Zhang M. Research on high resolution image object detection technology based on Faster R-CNN[J]. Remote Sensing for Land and Resources, 2019, 31(2):38-43.doi: 10.6046/gtzyyg.2019.02.06.
doi: 10.6046/gtzyyg.2019.02.06
[3] 史文旭, 江金洪, 鲍胜利. 基于特征融合的遥感图像舰船目标检测方法[J]. 光子学报, 2020, 49(7):57-67.
Shi W X, Jiang J H, Bao S L. Ship target detection in remote sensing image based on feature fusion[J]. Acta Photonica Sinica, 2020, 49(7):57-67.
[4] Szegedy C, et al. Going deeper with convolutions[C]// IEEE Conference on Computer Vision and Pattern Recognition(CVPR),Boston,MA, 2015:1-9.
[5] Redmon J, Divvala S, Girshick R, et al. You only look once:Unified,real-time object detection[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition(CVPR),Las Vegas,NV, 2016:779-788.
[6] Liu W, Anguelov D, Erhan D, et al. Ssd:Single shot multibox detector[C]// European Conference on Computer Vision,Springer,Cham, 2016:21-37.
[7] 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, 2015:91-99.
[8] Lin T Y, Dollár P, Girshick R, et al. Feature pyramid networks for object detection[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017:2117-2125.
[9] He K, Gkioxari G, Dollár P, et al. Mask R-CNN[C]// Proceedings of the IEEE International Conference on Computer Vision, 2017:2961-2969.
[10] Ma J. Arbitrary-oriented scene text detection via rotation proposals[J]. IEEE Transactions on Multimedia, 2018, 20(11):3111-3122.
doi: 10.1109/TMM.2018.2818020
[11] Yang X, Sun H, Fu K, et al. Automatic ship detection in remote sensing images from google earth of complex scenes based on multiscale rotation dense feature pyramid networks[J]. Remote Sensing, 2018, 10(1):132.
doi: 10.3390/rs10010132
[12] Zhu Y, Mu J, Pu H, et al. FRFB:Integrate receptive field block into feature fusion net for single shot multibox detector[C]// 2018 14th International Conference on Semantics,Knowledge and Grids(SKG),Guangzhou,China, 2018:173-180.
[13] Szegedy C, Vanhoucke V, Ioffe S, et al. Rethinking the inception architecture for computer vision[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition(CVPR),Las Vegas,NV, 2016:2818-2826.
[14] Huang G, Liu Z, Der Maaten L V, et al. Densely connected convolutional networks[C]// Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR),Honolulu,HI, 2017:2261-2269.
[15] Pang J, Chen K, Shi J, et al. Libra R-CNN:Towards balanced learning for object detection[C]// Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR),Long Beach,CA,USA, 2019:821-830.
[16] Wang X, Girshick R, Gupta A, et al. Non-local neural networks[C]// Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition,Salt Lake City,UT, 2018:7794-7803.
[17] Woo S, Park J, Lee J Y, et al. CBAM:Convolutional block attention module[J]. Lecture Notes in Computer Science, 2018:3-19.
[18] Hu J, Shen J, Sun G. Squeeze-and-excitation networks[C]// Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition,Salt Lake City,UT, 2018:7132-7141.
[19] Han J, Zhou P, Zhang D, et al. Efficient,simultaneous detection of multi-class geospatial targets based on visual saliency modeling and discriminative learning of sparse coding[J]. ISPRS Journal of Photogrammetry & Remote Sensing, 2014, 89:37-48.
[20] Xia G, et al. 2018. DOTA:A large-scale dataset for object detection in aerial images[C]// Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition,Salt Lake City,UT, 2018:3974-3983.
[21] Li Y, Huang Q, Pei X, et al. RADet:Refine feature pyramid network and multi-layer attention network for arbitrary-oriented object detection of remote sensing images[J]. Remote Sensing, 2020, 12(3):389.
doi: 10.3390/rs12030389
[1] 牛祥华, 黄微, 黄睿, 蒋斯立. 基于注意力特征融合的高保真遥感图像薄云去除[J]. 自然资源遥感, 2023, 35(3): 116-123.
[2] 王建强, 邹朝晖, 刘荣波, 刘志松. 基于U2-Net深度学习模型的沿海水产养殖塘遥感信息提取[J]. 自然资源遥感, 2023, 35(3): 17-24.
[3] 徐欣钰, 李小军, 赵鹤婷, 盖钧飞. NSCT和PCNN相结合的遥感图像全色锐化算法[J]. 自然资源遥感, 2023, 35(3): 64-70.
[4] 郑宗生, 刘海霞, 王振华, 卢鹏, 沈绪坤, 唐鹏飞. 改进3D-CNN的高光谱图像地物分类方法[J]. 自然资源遥感, 2023, 35(2): 105-111.
[5] 胡建文, 汪泽平, 胡佩. 基于深度学习的空谱遥感图像融合综述[J]. 自然资源遥感, 2023, 35(1): 1-14.
[6] 孙盛, 蒙芝敏, 胡忠文, 余旭. 多尺度轻量化CNN在SAR图像地物分类中的应用[J]. 自然资源遥感, 2023, 35(1): 27-34.
[7] 金远航, 徐茂林, 郑佳媛. 基于改进YOLOv4-tiny的无人机影像枯死树木检测算法[J]. 自然资源遥感, 2023, 35(1): 90-98.
[8] 沈骏翱, 马梦婷, 宋致远, 柳汀洲, 张微. 基于深度学习语义分割模型的高分辨率遥感图像水体提取[J]. 自然资源遥感, 2022, 34(4): 129-135.
[9] 张鹏强, 高奎亮, 刘冰, 谭熊. 联合空谱信息的高光谱影像深度Transformer网络分类[J]. 自然资源遥感, 2022, 34(3): 27-32.
[10] 王艺儒, 王光辉, 杨化超, 刘慧杰. 基于生成对抗网络的遥感影像色彩一致性方法[J]. 自然资源遥感, 2022, 34(3): 65-72.
[11] 马晓剑, 赵法舜, 刘艳宾. 多特征准则融合的遥感图像脉冲噪声的识别处理[J]. 自然资源遥感, 2022, 34(3): 17-26.
[12] 尚晓梅, 李佳田, 吕少云, 杨汝春, 杨超. 用于遥感图像超分辨率重建的残差对偶回归网络[J]. 自然资源遥感, 2022, 34(2): 112-120.
[13] 廖廓, 聂磊, 杨泽宇, 张红艳, 王艳杰, 彭继达, 党皓飞, 冷伟. 基于多维卷积神经网络的多源高分辨率卫星影像茶园分类[J]. 自然资源遥感, 2022, 34(2): 152-161.
[14] 杨昭颖, 韩灵怡, 郑向向, 李文吉, 冯磊, 王轶, 杨永鹏. 基于卷积神经网络的遥感影像及DEM滑坡识别——以黄土滑坡为例[J]. 自然资源遥感, 2022, 34(2): 224-230.
[15] 孔爱玲, 张承明, 李峰, 韩颖娟, 孙焕英, 杜漫飞. 基于知识引导的遥感影像融合方法[J]. 自然资源遥感, 2022, 34(2): 47-55.
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
京ICP备05055290号-2
版权所有 © 2015 《自然资源遥感》编辑部
地址:北京学院路31号中国国土资源航空物探遥感中心 邮编:100083
电话:010-62060291/62060292 E-mail:zrzyyg@163.com
本系统由北京玛格泰克科技发展有限公司设计开发