摘要：Information of land use plays a key role in the ecological systems. Most studies focus on the land use studies in large cities or large areas, but rarely carry out a land system in vulnerable areas. In this study, the size of 1000×1000 pixels in Hegang coal mining area was used as the experimental area. Based on Landsat TM image of September 8, 2010, principal component analysis (PCA) and optimum index factor (OIF) were used to select the best band combination of images, and the texture statistics, texture features and spectral information of the homogeneity, contrast, entropy and angular second moments of the remote sensing image were extracted by using the gray level co-occurrence matrix texture feature. A sample of 600 pixels was selected, of which 400 pixels were used as training samples and 200 pixels as test samples. The results show that the support vector machine (SVM) and neural network (NN) classification technique are used to classify land use in coal mining area. The overall accuracy obtained was 92.40% and the kappa statistics 0.9126 for SVM, 90.90% and 0.8930 for NN, respectively. This study provided a comprehensive extraction of samples to improve the accuracy method. The SVM and NN classification results show that SVM classification method is superior to NN classification method, and it can effectively be utilized for Landsat TM images to identify land use types in coal mining areas.
关键词：Coal mining area;Land use classification;Support vector machine;Neural network;Texture feature
