YU Zhengyang,CHEN Jun,JIANG Minghua,et al.Research on Intelligent Extraction Method of Spatial Objects from Online High-Resolution Remote Sensing Images based on Tile-overlapping Strategy[J].Journal of Chengdu University of Information Technology,2022,37(05):520-526.[doi:10.16836/j.cnki.jcuit.2022.05.006]
基于瓦片重叠法的在线高分遥感图像目标智能提取方法研究
- Title:
- Research on Intelligent Extraction Method of Spatial Objects from Online High-Resolution Remote Sensing Images based on Tile-overlapping Strategy
- 文章编号:
- 2096-1618(2022)05-0520-07
- 关键词:
- 人工智能; Mask R-CNN; 目标提取; 在线遥感影像; 地表机器智能感知
- Keywords:
- artificial intelligence; Mask R-CNN; object extraction; online remote sensing images; machine intelligent perception of earth surface
- 分类号:
- TP751
- 文献标志码:
- A
- 摘要:
- 在线遥感影像上目标智能提取是地表机器智能感知的重要途径。然而，如何利用对单个图像大小限制的Mask R-CNN模型在线遥感影像上指定空间范围内有效提取空间目标，是亟待解决的关键问题。为此，提出了一种瓦片重叠法以实现在线高分遥感影像上大范围目标进行智能提取。首先，依据高分遥感影像上空间目标大小，设定检测单元的宽度和高度均为512个像素，并按一定的步长将相邻瓦片拼接成一系列相互重叠的检测单元; 然后，利用Mask R-CNN提取局部范围的空间目标，并将其转换为空间坐标; 最后，通过空间重叠融合相邻检测单元的重复目标。实验发现，当重叠率为50%时，能以最优的性能达到较高的提取精度。以在线地图上的运动场、城市路口、跨河大桥为例，开展了中国部分城市的在线高分遥感影像目标提取实验，平均目标提取率为80%左右，为地表机器智能感知提供了新途径。
- Abstract:
- The intelligent extraction of spatial objects from online remote sensing images is an important approach for the machine intelligence perception of earth surface. However, how to use the Mask R-CNN model which limits the size of a single image to effectively extract the spatial objects in the specified spatial range on the online remote sensing images is a key problem to be solved. In this paper, an extraction method of spatial objects based on tile-overlapping strategy is proposed to realize the intelligent extraction of spatial objects from online high-resolution remote sensing images. Firstly, according to the size of the spatial objects on the high-resolution remote sensing images, the width and height of the detection unit are set to 512 pixels, and the adjacent tiles are spliced into a series of overlapping detection units according to a certain overlap rate. Then, Mask R-CNN is used to extract the spatial objects from each of the detection units. The local coordinates of each extracted spatial object are converted to spatial coordinates through the coordinates and spatial resolution of the upper left corner of the prediction box. Finally, the repeated objects of adjacent detection units are fused through spatial overlap analysis to obtain the final spatial object sequence. Taking part of Chengdu urban area(with an area of 33.4 km2)as the experimental area, set different overlap rates for spatial object extraction, and evaluate the performance and accuracy of different tile overlap rates. The experimental result indicated that the higher the overlap rate, the more opportunities there are to repeatedly identify the same object from different "fragments" at the same location, so as to reduce the leakage. On the other hand, repeated identification of the same location also increases the number of false identification. When the overlap rate is 50%, it can achieve high extraction accuracy with the best performance. Taking the playground, urban intersection and river crossing bridge on the online map as examples, the spatial objects of some cities in China were extracted from online remote sensing images with the the overlap rate set to 50%. The results show that the average object extraction rate is about 80%, which provides a new way for the intelligent perception of earth surface.
参考文献/References:
[1] Kulasekara D,Harshanath S.Detecting changed areas in images from different view points[C].International Conference on Computer Science & Education.IEEE,2013:525-528.
[2] Marmanis D,Datcu M,Esch T,et al.Deep Learning Earth Observation Classification Using ImageNet Pretrained Networks[J].IEEE Geoscience & Remote Sensing Letters,2016,13(1):105-109.
[3] Krizhevsky A,Sutskever I,Hinton G.ImageNet Classification with Deep Convolutional Neural Networks[J].Communications of the ACM,2012,60(6):84-90.
[4] Simonyan K,Zisserman A.Very Deep Convolutional Networks for Large-Scale Image Recognition[C].Computational and Biological Learning Society,2015:1-14.
[5] Szegedy C,Liu W,Jia Y,et al.Going deeper with convolutions[C].2015 IEEE Conference on Computer Vision and Pattern Recognition(CVPR),2015:1-9.
[6] Ronneberger O,Fischer P,Brox T.U-Net: Convolutional Networks for Biomedical Image Segmentation[C].Medical Image Computing and Computer-Assisted Intervention-MICCAI 2015:234-241.
[7] He Kaiming,Georgia Gkioxari,Piotr Dollar,et al.Mask R-CNN[C].Proceedings of the IEEE International Conference on Computer Vision(ICCV),2017,42(2):2961-2969.
[8] Yayla R,Sen B.A New Classification Approach with Deep Mask R-CNN for Synthetic Aperture Radar Image Segmentation[J].Elektronika ir Elektrotechnika,2020,26(6):52-57.
[9] Xu Yingying,Li Dawei,Xie Qian,et al.Automatic defect detection and segmentation of tunnel surface using modified Mask R-CNN[J].Measurement,2021,178(4):109316.
[10] Kassim Y M,Byrne M E,Burch C,et al.Small Object Bird Detection in Infrared Drone Videos Using Mask R-CNN Deep Learning[C].Imaging and Multimedia Analytics in a Web and Mobile World,2021,2020(8):85-1-85-8.
[11] Naufal A,Fatichah C,Suciati N.Preprocessed Mask RCNN for Parking Space Detection in Smart Parking Systems[J].International Journal of Intelligent Engineering and Systems,2020,13(6):255-265.
[12] Wong CC,Yeh LY,Liu CC,et al.Manipulation Planning for Object Re-Orientation Based on Semantic Segmentation Keypoint Detection[J].Sensors(Basel).2021,21(7):2280-2286.
[13] Shi J,Zhou Y,Zhang W X Q.Target Detection Based on Improved Mask Rcnn in Service Robot[C].2019 Chinese Control Conference(CCC),2019:8519-8524.
[14] Lu Chuhan,Kong Yang,Guan Zhaoyong.A mask R-CNN model for reidentifying extratropical cyclones based on quasi-supervised thought[J].Scientific Reports,2020,10(1):15011-15017.
[15] Lv Y,Zhang C,Yun W,et al.The Delineation and Grading of Actual Crop Production Units in Modern Smallholder Areas Using RS Data and Mask R-CNN[J].Remote Sensing,2020,12(7):1074-1080.
[16] Ji Chao,Tang Hong.Number of Building Stories Estimation from Monocular Satellite Image Using a Modified Mask R-CNN[J]. Remote Sensing,2020,12(22):3833-3890.
[17] Mahmoud A,Yushen He,Qiang Xu,et al.Object Detection Using Adaptive Mask RCNN in Optical Remote Sensing Images[J].International Journal of Intelligent Engineering and Systems,2020,13(1):65-76.
相似文献/References:
[1]王 强,李孝杰,陈 俊.基于He-Net的卷积神经网络算法的图像分类研究[J].成都信息工程大学学报,2017,(05):503.[doi:10.16836/j.cnki.jcuit.2017.05.007]
WANG Qing,LI Xiao-jie,CHEN Jun.Research on Image Classification based on HE-Net Convolutional Neural Networks[J].Journal of Chengdu University of Information Technology,2017,(05):503.[doi:10.16836/j.cnki.jcuit.2017.05.007]
备注/Memo
收稿日期:2021-12-24
基金项目:国家重点研发计划地球观测与导航重点专项资助项目(2018YFB0505300); 四川省科技计划项目重点研发资助项目(2020YFG0146)