Abstract:
Unmanned Aerial Vehicle (UAV) remote sensing can obtain high-resolution images at low cost and high efficiency. However, there is rare research on the use of UAV remote sensing images to extract information from impervious surfaces. The difficulty of the research lies in that for the high-resolution UAV red-green-blue images, there is currently not an exclusive impervious surface index that can be applied to the extraction of impervious surface information. To address this problem, this study established the green-blue spectral feature space in the blue and green bands of the visible-bands. Under this spectral feature space, the green-blue impervious surface index was proposed to effectively separate soil-vegetation pixels and impervious surface pixels. The process of constructing the index was as follows. Firstly, the feature points were divided into impervious surface points and pervious surface points in the green-blue spectral feature space. Secondly, the least-squares fitting was conducted on the impervious surface points and the previous surface points. Then, the impervious surface line and the soil line were obtained, and a reference line was constructed between the two straight lines. Finally, the distance from the feature point to the reference line was used as the expression of the index. To verify the accuracy of the difference between the green-blue impervious surface index and other impervious surface indexes applied to satellite remote sensing images, comparison and analysis were performed on the extraction of the impervious surface by perpendicular impervious index, ratio resident-area index and green-blue impervious surface index. At the same time, the experiment was conducted using the UAV orthophoto image data of Hongya county in Meishan city to extract the impervious surface. The analysis was conducted to investigate the effect of the green-blue impervious surface index on the extraction of the impervious surface information in the UAV remote sensing image. The experimental results showed that: 1) The green-blue impervious surface index constructed the blue and green bands as the feature space had the same accuracy as the perpendicular impervious index and ratio resident-area index based on the blue and near-infrared bands in terms of the impervious surface extraction, and the overall accuracy reached over 94%. 2) The green-blue impervious surface index showed strong applicability and replaced the indexes like perpendicular impervious index, ratio resident-area index and biophysical composition index in the images lacking near-infrared, mid-infrared, and thermal-infrared bands. The green-blue impervious surface index was used as a remote sensing index to extract visible-bands on the impervious surfaces. 3) As a visible light wave impervious surface extraction index used in UAV remote sensing, the green-blue impervious surface index could not only effectively distinguish between soil and impervious surfaces, but also utilized the characteristics of easy-to-obtain terrain features by UAV remote sensing. For the problem of misclassification, the overall accuracy of the extraction results by the green-blue impervious surface index reached 96.95%, and the Kappa coefficient was 0.936 1. The green-blue impervious surface index constructed based on the green-blue spectral feature space effectively separated soil pixels and extracted high-precision urban impervious surfaces from UAV remote sensing images. For satellite imagery, the existing impervious surface index had a good performance in extracting the surfaces, while the proposed green-blue impervious surface index was more suitable for UAV remote sensing. At present, extracting urban impervious surfaces from UAV remote sensing images has gradually become an important application.