| China in coal mining in the process of monitoring the behavior of mostly on-site monitoring, weak, and the limited scope of information collection, timeliness is poor, a large number of illegal exploitation of behavior can not be timely detection of irreversible damage to environment caused by mining. Remote sensing technology's broad, real-time, periodic and comprehensive sexual and other characteristics, for fast, accurate, and objective opencast mine area offers the possibility of environmental monitoring. With the space remote sensing information technology, rapid development and Global Earth Observation System of remote sensing data, to provide capacity-growing, but because of the comprehensive nature of remote sensing information, the complexity of the mechanism of remote sensing imaging, information processing technology has lagged behind in information access to technology development. In this paper, the existing remote sensing image classification method and support vector machine principle of a comprehensive analysis will be based on support vector machines for remote sensing image classification and recognition method for opencast mine area environmental monitoring problems.To Hegang City, a TM remote sensing image, for example, use two kinds of support vector machine multi-classification method to classify comparison. Support vector machine classifier modeling process, the choice of kernel function has no theoretical guidance, this experimental approach, manually create a two-dimensional sample data sets of five categories, namely, the use of four kinds of commonly used kernel functions were classification comparison, the results show that the use of Gaussian radial basis kernel function of the classification results when the classifier is most desirable; the use of cross-validation grid search algorithm selected the appropriate parameters, so that classifiers have higher classification accuracy. The results show that, due to error accumulation, support vector machine multi-classifier of the classification accuracy is lower than the accuracy of two classifiers; in the settlement of fewer categories of the classification problem, two kinds of SVM multi-classifier of the classification accuracy is satisfactory, respectively, reached 83.67% and 86.75%, compared to 1-v-1SVM has a higher classification accuracy. |
