Classification of wetland based on object-oriented method in coal mining subsidence area using GF-1 remote sensing image
Cheng Linlin,Dong Xuemei,Zhan Jiaqi,Zhang Ye
DOI: https://doi.org/10.11975/j.issn.1002-6819.2018.09.030
2018-01-01
Abstract:Wetland plays an important role in the diversity maintenance of species, supplements of groundwater, regulation of climate, purification and degradation of pollutants, and so on. In recent years, due to the irrational exploitation and utilization of wetland, wetland resources have been gradually dried up, and how to develop and utilize wetlands has become an important issue. Classification of wetland has also become a hot topic for scholars both at home and abroad. Because of the special geographical position, it is difficult to obtain and monitor wetland by traditional surveying techniques. Therefore, remote sensing has been applied broadly in the classification of wetland. But most researches mainly focus on classification of the natural wetlands, classification of wetland in mining subsidence area, a kind of artificial wetland, is rarely involved. Therefore, this paper discussed classification of wetland in mining subsidence area using GF-1 remote sensing images and object-oriented method with Yanzhou coal field as an example.In order to improve the accuracy of classification, segmentation and classification were based on vector data. The vectors of river wetlands and landscape wetlands (Area 1) were obtained through the map of present land use, and the vectors of other areas (Area 2) were obtained by the erasure of the study area with river wetland and landscape wetland (Area 1). Images of river wetland and landscape wetland were treated with chessboard segmentation based on the vector of Area 1 and the nearest neighbour method was used for classification. Ratio of length to width was selected as the image feature and the threshold value was set to differentiate the 2 types of wetlands. Images of aquaculture wetland, vegetation type wetland, non-reclaimed perennially waterlogged wetlands and non-reclaimed seasonal water-accumulation wetland were processed with multi-scale segmentation based on the vector of Area 2. The relative optimal segmentation parameters were obtained by repeated tests. The main parameters for segmentation were segmentation scale and weights of heterogeneity factors. The segmentation scale determined for vegetation type wetland was 180, the weight of the shape factor was 0.2, and the weight of degree of compact was 0.1. Segmentation scale of non-reclaimed perennially waterlogged wetland was 250, the weight of the shape factor was 0.6, and the weight of degree of compact was 0.7. For non-reclaimed seasonal water-accumulation wetland and aquaculture wetland, the 3 parameters were 170, 0.2, 0.1 and 230, 0.7, 0.9 respectively. Local optimal segmentation parameters were further determined through comparison and analysis of segmentation parameters in several typical regions. After analyzing the segmentation parameters corresponding to different types of wetlands, 3 classification levels were established. The non-reclaimed perennially waterlogged wetland was deemed as L1 layer, and the aquaculture wetland was L2 layer. The vegetation type wetland and the non-reclaimed seasonal water-accumulation wetland with similar segmentation parameters were at the same classification level as the L3 layer. Bottom-up rules were used for classification, which meant it started from the L3 layer with a small segmentation scale, then L2 layer, and finally the L1 layer with the largest segmentation scale. The image features selected for aquaculture wetland were the normalized difference vegetation index (NDVI), the ratio of length to width, and area, those for non-reclaimed perennially waterlogged wetland were NDVI and area, those for vegetation type wetland were the mean of gray level co-occurrence matrix (GLCM mean), the mean of near infrared band and NDVI, and those for non-reclaimed seasonal water-accumulation wetland were spectral brightness and GLCM mean. The classification rules were established according to the particularly selected image features. The fuzzy classification method was adopted to classify the wetland in Area 2. Accuracy evaluation was carried out by field survey and confusion matrix. There were 141 sample spots selected to verify the results. It indicated that the classification accuracy of river wetland and landscape wetland was 100%, total accuracy was 96.95%, and Kappa index was 0.958 4. Therefore, the classification method put forward in this paper is suitable for classification of wetland in mining subsidence area. The study provides a reference for other similar areas, and provides a scientific basis for reclamation, planning and management of wetland in coal mining subsidence area.