Experimental Study on Precise Recognition of Settlements in Mountainous Areas Based on UAV Image and LIDAR Point Cloud

• Published in: Journal of the Indian Society of Remote Sensing Vol. 50; no. 10; pp. 1827 - 1840
• Main Authors: Gao, Sha, Yuan, Xiping, Gan, Shu, Yang, Yafu, Yuan, Xinyue
• Format: Journal Article
• Language: English
• Published: New Delhi Springer India 01.10.2022; Springer Nature B.V
• Subjects: Airborne sensing; Classification; data collection; Data sources; Earth and Environmental Science; Earth Sciences; Feature extraction; High definition; Image classification; Lidar; Mountain regions; Mountainous areas; Mountains; Object recognition; Qualitative analysis; Remote sensing; Remote Sensing/Photogrammetry; Review Article; Root-mean-square errors; Spatial data; texture; Unmanned aerial vehicles; Mountain settlements; LiDAR data; Feature recognition; UAV images
• ISSN: 0255-660X; 0974-3006
• DOI: 10.1007/s12524-022-01548-1

As the main type of ground objects in mountainous environment, mountain settlement is also an important subject to monitor for the prevention of geological disasters. The accurate and fast identification of mountain settlement is essential for disaster monitoring and rural planning. The fusion of multi-source heterogeneous data has attracted widespread attention for remote sensing. Equipped with high-definition cameras, UAV (unmanned aerial vehicle, UAV) can capture the image data with plenty of information about spectral texture, while the airborne LiDAR (light detection and ranging, LiDAR) equipped with high precision sensors is capable to locate objects precisely. Combining the advantages shown by the two kinds of data features enables the integration of two kinds of data, thus supplementing the advantages of spatial information and optical information. This paper aims at addressing the inaccuracy of boundary extraction in the existing methods of mountain settlement extraction and the limitation on the use of manual features to express image information. A mountainous settlement classification method is proposed by combining the feature information of LiDAR data and UAV image data sources. According to the qualitative and quantitative analyses of the experimental results, when the point cloud data with fused spectral information is oriented to the precise identification of mountainous areas, the RMSE (root-mean-square error, RMSE) value of building boundaries is relatively stable, ranging between 0 and 0.2 m through comparison of the calculation results with the actual data. The overall accuracy evaluation index of the extracted building area is determined to exceed 80%, which is higher than the other two single data sources extracted from the mountain settlement information. Besides, the extracted contour lines are made clearer, which suggests the classification advantages of the point cloud data fused with image information. It is demonstrated that the classification method is conducive to the accurate identification of mountain settlements, which indicates its practical value in the accurate identification of mountain settlements.