Please wait a minute...
img

官方微信

遥感技术与应用  2021, Vol. 36 Issue (2): 256-264    DOI: 10.11873/j.issn.1004-0323.2021.2.0256
CNN 专栏     
结合VGGNet与Mask R-CNN的高分辨率遥感影像建设用地检测
陈敏1,2(),潘佳威2,李江杰2,徐璐2,刘加敏1,2,韩健3(),陈奕云2,4
1.广州市城市规划勘测设计研究院,广东 广州 510060
2.武汉大学资源与环境科学学院,湖北 武汉 430079
3.广西华遥空间信息科技有限公司,广西 南宁 530031
4.土壤与农业可持续发展国家重点实验室,江苏 南京 210008
High Resolution Remote Sensing Image Construction Land Detection Combined with VGGNet and Mask R-CNN
Min Chen1,2(),Jiawei Pan2,Jiangjie Li2,Lu Xu2,Jiamin Liu1,2,Jian Han3(),Yiyun Chen2,4
1.Guangzhou Urban Planning Survey & Design Survey Research Institute,Guangzhou 510060,China
2.School of Resource and Environmental Science,Wuhan University,Wuhan 430079,China
3.Guangxi Huayao Space Information Technology co. LTD,Nanning 530031,China
4.State Key Laboratory of Soil and Sustainable Agriculture,Chinese Academy of Sciences,Nanjing 210008,China
 全文: PDF(17103 KB)   HTML
摘要:

针对当前多数深度学习模型只能对高分辨率遥感影像裁剪图片进行土地利用类型判别的问题,结合VGGNet与Mask R-CNN开展了智能化建设用地目标检测研究。在建立研究区4类土地利用类型遥感影像数据集的基础上,对比了VGGNet、ResNet和DenseNet 3种卷积神经网络模型的分类精度,选取分类效果最优的神经网络模型VGGNet与Mask R-CNN实现建设用地目标检测智能化。结果表明:①VGGNet、ResNet和DenseNet 3种卷积神经网络模型的分类精度分别为:97.44%、93.75%和95.13%,且VGG16模型迭代次数最少,训练时间相对较少;②Mask R-CNN阈值设置对目标检测精度有重要的影响,当阈值设定为0.3时,VGG16结合Mask R-CNN的联合模型对建设用地检测的标定框精度最高。同时联合模型比单一使用Mask R-CNN模型对建设用地检测有更高的准确率,并且表现出了更强的适应性和鲁棒性。

关键词: 卷积神经网络(CNN)目标检测影像分类高分辨率遥感影像建设用地    
Abstract:

To address the problem that most current deep learning models can only discriminate land use types for cropped images of high-resolution remote sensing images, this paper combines VGGNet and Mask R-CNN to carry out a study on intelligent construction land target detection. On the basis of establishing remote sensing image datasets of four types of land use types in the study area, we compare the classification accuracy of three convolutional neural network models, VGGNet, ResNet and DenseNet, and select the neural network model with the best classification effect, VGGNet and Mask R-CNN, to achieve intelligent construction land target detection. The results show that: (1) the classification accuracies of the three convolutional neural network models VGGNet, ResNet and DenseNet are 97.44%, 93.75% and 95.13%, respectively, and the VGG16 model has the least number of iterations and relatively less training time; (2) the Mask R-CNN threshold setting has an important influence on the target detection accuracy, when the threshold is set to is 0.3, the joint model of VGG16 combined with Mask R-CNN has the highest calibration frame accuracy for construction land detection. Also the joint model has higher accuracy than the single use of Mask R-CNN model for construction land detection, and shows more adaptability and robustness.

Key words: Convolutional Neural Network(CNN)    Target detection    Image classification    High resolution remote sensing image    Construction land
收稿日期: 2019-10-14 出版日期: 2021-05-24
ZTFLH:  TP79  
基金资助: 国家重点研发计划“绿色宜居村镇技术创新”重点专项项目子课题“村镇发展潜力因子识别与指标信息快速获取技术”(2018YFD1100801-01)
通讯作者: 韩健     E-mail: 15927214729@163.com;hanjianspace@126.com
作者简介: 陈敏(1994-),男,湖北咸宁人,硕士研究生,主要从事土地变化科学与可持续发展研究。E?mail:15927214729@163.com
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  
陈敏
潘佳威
李江杰
徐璐
刘加敏
韩健
陈奕云

引用本文:

陈敏,潘佳威,李江杰,徐璐,刘加敏,韩健,陈奕云. 结合VGGNet与Mask R-CNN的高分辨率遥感影像建设用地检测[J]. 遥感技术与应用, 2021, 36(2): 256-264.

Min Chen,Jiawei Pan,Jiangjie Li,Lu Xu,Jiamin Liu,Jian Han,Yiyun Chen. High Resolution Remote Sensing Image Construction Land Detection Combined with VGGNet and Mask R-CNN. Remote Sensing Technology and Application, 2021, 36(2): 256-264.

链接本文:

http://www.rsta.ac.cn/CN/10.11873/j.issn.1004-0323.2021.2.0256        http://www.rsta.ac.cn/CN/Y2021/V36/I2/256

图1  右江区区位图
图2  影像数据分类实例
图3  数据增强技术
图4  VGGNet网络结构示意图
图5  用于实例分割的Mask R-CNN框架[18]
图6  VGG16+Mask R-CNN目标检测流程
图7  基于不同模型的高分辨率遥感影像分类
模型VGGNetResNetDenseNet
训练时间/s9006401 440
测试集精度/%97.4493.7595.13
迭代次数/次121635
表1  不同模型对高分辨影像分类结果

标定框

精度/%

测试集

数量

测试集

目标数

准确

目标数

准确率

/%

Mask R-CNN76.3430086556465.20
表2  Mask R-CNN训练结果
Mask R-CNN阈值标定框精度/%准确率/%
0.185.3597.46
0.284.3295.61
0.387.9394.34
0.479.6972.32
0.578.5465.2
0.669.8449.94
0.766.3246.59
0.864.8744.74
0.961.5638.84
表3  不同阈值下Mask R-CNN训练结果
图8  不同阈值下Mask R-CNN训练结果
图9  Mask R-CNN与VGG16+Mask R-CNN处理效果
1 Hu Wei, Gao Bochuan, Huang Zhenhang, et al. Tree-structured Convolutional Neural Network Optimization for Semantic Segmentation of Remote Sensing Images in Urban Areas[J].Chinese Journal of Graphics,2020,25(5):197-206.
1 胡伟,高博川,黄振航,等.树形结构卷积神经网络优化的城区遥感图像语义分割[J].中国图象图形学报,2020,25(5):197-206.
2 Deng Zhipeng, Sun Hao, Lei Lin, et al. High-resolution Remote Sensing Image Target Detection based on Multi-scale Deformation Feature Convolution Network[J]. Journal of Surveying and Mapping,2018,47(9):1216-1227.
2 邓志鹏,孙浩,雷琳,等.基于多尺度形变特征卷积网络的高分辨率遥感影像目标检测[J].测绘学报,2018,47(9):1216-1227
3 Ding Xiaohui. Classification of Hyperspectral Remote Sensing Images based on ant Colony Algorithm[D].Changchun: Graduate School of Chinese Academy of Sciences (Northeast Institute of Geography and Agricultural Ecology), 2016.
3 丁小辉. 基于蚁群算法的高光谱遥感影像地物分类研究[D]. 长春:中国科学院研究生院(东北地理与农业生态研究所), 2016.
4 Pan Yifan, Zhang Xianfeng, Yu Hongfeng, et al. Urban Land Cover Classification based on Texture Information of Joint Kuaizhou-1 Image[J]. Remote Sensing Technology and Application,2016,31(1):194-202.
4 潘一凡,张显峰,于泓峰,等.联合快舟一号影像纹理信息的城市土地覆盖分类[J].遥感技术与应用,2016,31(1):194-202.
5 Xianxia. High-resolution Remote Sensing Image Scene Classification based on Multi-feature Deep Learning[D]. Kunming:Yunnan Normal University,2019.
5 鲜霞.基于多特征深度学习的高分辨率遥感影像场景分类[D]. 昆明:云南师范大学,2019.
6 Cao Zhaowei. Classification of Island Remote Sensing Images and Features based on Artificial Neural Network[D]. Shanghai:Shanghai Ocean University, 2016.
6 曹兆伟.基于人工神经网络的海岛遥感影像地物分类研究[D].上海:上海海洋大学,2016.
7 Ali S S , Raza I U R , Kazmi S S A , et al. Comparison of Pixel-based and Object-based Classification for Glacier Chan-ge Detection[C]∥ Second International Workshop on Earth Observation & Remote Sensing Applications.IEEE,2012.
8 Luo Zhiyong, Li Xiaodong. Research on High Performance Tile Map Service based on Database Storage Scheme[J]. Geography and Geographic Information Science, 2013, 29 (3): 52-55,112.
8 罗智勇, 黎小东. 基于数据库存储方案的高性能瓦片地图服务研究[J]. 地理与地理信息科学,2013, 29(3):52-55,112.
9 Li Xin, Tang Wenli, Yang Bo. Semantic Segmentation of High Resolution Remote Sensing Image Using Depth Residual Network[J]. Journal of Applied Sciences, 2019, 37 (2):282-290.
9 李欣, 唐文莉, 杨博. 利用深度残差网络的高分遥感影像语义分割[J]. 应用科学学报, 2019, 37(2):282-290.
10 Yu Ximing, Hong Shuo, Yu Jinxiang, et al. Research on the Enhancement Method of Visible Light Remote Sensing Image Ship Target Data[J]. Journal of Instrumentation,2020,41(11):261-269.
10 于希明,洪硕,于金祥,等.可见光遥感图像船舶目标数据增强方法研究[J].仪器仪表学报,2020,41(11):261-269.
11 Xu Chenyu. Research and Improvement of YOLOv3-based Hazardous Goods Detection Model[D]. Lanzhou:Lanzhou University,2019.
11 徐晨玉. 基于YOLOv3的危险物品检测模型研究与改进[D]. 兰州:兰州大学,2019.
12 Simonyan K, Zisserman A.Very Deep Convolutional Networks for Large-scale Image Recognition[C]∥ The 3rd International Conference on Learning Representations (ICLR2015).
13 Zeiler M D, Fergus R. Visualizing and Understanding Convolutional Networks[C]∥ European Conference on Computer Vision. Springer, Cham, 2014:818-833.
14 Srivastava N, Hinton G, Krizhevsky A, et al. Dropout: A Simple Way to Prevent Neural Networks from Overfitting[J]. The Journal of Machine Learning Research, 2014, 15(1): 1929-1958.
15 He K M, Zhang X Y, Ren S Q, et al. Deep Residual Learning for Image Recognition [J]. IEEE Computer Society, 2016:770-778.doi:10.1109/CVPR.2016.90.
doi: 10.1109/CVPR.2016.90
16 Huang G, Liu Z, Van der Maaten L, et al. Densely Connected Convolutional Networks[S/OL]. http://arxi.org/abs/1608. 06993,2017.doi:10.1109/CVPR.2017.243.
doi: 10.1109/CVPR.2017.243
17 Ren S Q, He K, Girshick Ross , et al. Faster R-CNN: Towards Real-time Object Detection with Region Proposal Networks[J].IEEE Transactions on Pattern Analysis & Machine Intelligence,2017,39(6):1137-1149. doi:10.1109/TPAMI.2016.2577031.
doi: 10.1109/TPAMI.2016.2577031
18 He K, Gkioxari G, Dollar P, et al. Mask R-CNN[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018:1-1.doi:10.1109/TPAMI.2018.2844175.
doi: 10.1109/TPAMI.2018.2844175
[1] 柴栋,许夙晖,罗畅,鲁彦辰. 使用贝叶斯优化对遥感影像目标进行精确定位[J]. 遥感技术与应用, 2020, 35(6): 1377-1385.
[2] 刘美,杜国明,于凤荣,匡文慧. 哈尔滨城乡梯度建设用地结构变化及不透水面遥感监测分析[J]. 遥感技术与应用, 2020, 35(5): 1206-1217.
[3] 王卓,闫浩文,禄小敏,冯天文,李亚珍. 一种改进U-Net的高分辨率遥感影像道路提取方法[J]. 遥感技术与应用, 2020, 35(4): 741-748.
[4] 连喜红,祁元,王宏伟,张金龙,杨瑞. 基于面向对象的青海湖环湖区居民地信息自动化提取[J]. 遥感技术与应用, 2020, 35(4): 775-785.
[5] 潘灼坤,胡月明,王广兴,刘吼海,刘江,李波,樊舒迪. 对遥感在城市更新监测应用中的认知和思考[J]. 遥感技术与应用, 2020, 35(4): 911-923.
[6] 张因果,陈芸芝. 一种基于改进土地覆盖更新方法的新增建设用地自动提取[J]. 遥感技术与应用, 2019, 34(5): 1073-1081.
[7] 王颖洁,张荞,张艳梅,蒙印,郭文. 基于深度卷积神经网络的油罐目标检测研究[J]. 遥感技术与应用, 2019, 34(4): 727-735.
[8] 李淑敏, 冯权泷, 梁其椿, 张学庆. 基于深度学习的国产高分遥感影像飞机目标自动检测[J]. 遥感技术与应用, 2018, 33(6): 1095-1102.
[9] 熊伟,徐永力,姚力波,崔亚奇. 基于SVM的高分辨率SAR图像舰船目标检测算法[J]. 遥感技术与应用, 2018, 33(1): 119-127.
[10] 王苏芸,孙中昶,郭华东,申维. 基于面向对象的东营市城乡建设用地信息提取[J]. 遥感技术与应用, 2017, 32(4): 780-786.
[11] 朱济帅,尹作霞,谭琨,王雪,李二珠,杜培军. 基于空间邻域信息的高光谱遥感影像半监督协同训练[J]. 遥感技术与应用, 2016, 31(6): 1122-1130.
[12] 鲍蕊,夏俊士,薛朝辉,杜培军,车美琴. 基于形态学属性剖面的高光谱影像集成分类[J]. 遥感技术与应用, 2016, 31(4): 731-738.
[13] 张文博,覃志豪,刘含海. 基于分形理论的土地利用类型变化研究—以广州番禺为例[J]. 遥感技术与应用, 2015, 30(5): 952-958.
[14] 费鲜芸,王婷,魏雪丽. 基于多尺度分割的遥感影像滨海湿地分类[J]. 遥感技术与应用, 2015, 30(2): 298-303.
[15] 梁治华,刘勇. 近20年来腾格里沙漠南缘土地利用/土地覆盖变化及其成因探讨—以甘肃省古浪县冰草湾地区为例[J]. 遥感技术与应用, 2015, 30(2): 383-390.