Improved ECOC SVMs And Its Application In LUCC Study Of Mining Area

Land Use/Cover Change(LUCC) monitoring of mining area is important for regional resources exploitation and environmental sustainable development. Postclassification Comparison Method, one of the most commonly used approach for LUCC study, has a high demand for the accuracy of classification algorithm due to its error accumulation characteristic. Therefore, it is nessary to carry out research on classifiers, e.g. Support Vector Machine(SVM), which can sovle the problem of high dimensionality and sparse samples. The existing multi-classification methods of SVM have shortcomings in classifier redundancy, uncertain regions and error accumulation. There still has great potential to continue optimizing the efficiency and accuracy of these methods. Due to the unique error correcting mechanism, ECOC SVMs can correct the wrong sub-results of SVM. Using information of clustering and separability for optimizaating the ECOC SVMs can improve its accuracy, which also will be of great importance in the RS image classification and LUCC study.This study constructed an improved error-correcting output codes support vector machine classification algorithm(SW-ECOC SVMs) based on separabiliy measure, and modified it from two aspects: code word distributon and weighted decoding method. Expriments of multispectral and hyperspectral data shows that SW-ECOC SVMs can effectively and efficiently improve the accuracy of multi-classification SVM. The experiments of FLC1, AVIRIS KSC datasets all show that the accuracy rise for around 2%. Based on the improved algorithm, this study analysed the Huainan Mining Area combining with RS and GIS. Referencing to the LUCC and landscape knowlodge of mining area, this study analysed local land use and landscape change from the aspects of structure, speed, extent and various of landscape indexs. The results showed that land reclamation work in Huainan Mining Area could effectively slow down the growing trend of subsided ponding land, thus protecting regional land use structure.