High Resolution Remotely Sensed Imagery Classification For Urban Environment Monitoring Based On Support Vector Machine

With the advancement of high resolution remote sensing sensors, their uses to environmental monitoring and management are quite promising. But some new issues in high resolution image processing should be addressed to promote environmental applications, among which the development of novel classifiers, fusion of spatial and spectral information and link of remote sensing data with environmental analysis are quite important, so they are chosen as the topic of this paper. By considering the structure of environmental systems, environmental functions of different land covers, ecological and environmental impacts of land use, and the scale relationship of high resolution remote sensing image and environmental processes along with the natural and geographical conditions of the study area, a specific land cover category is developed for environmental analysis and mapping. Therefore, land cover classification methods via high resolution QuickBird image are discussed in detail. Some classical classification methods, including Maximum Likelihood Classifier, Minimum Distance Classifier, along with Support Vector Machine (SVM) classifier, the most effective statistical learning algorithm which is based on structural risk minimization (SRM) other than empirical risk minimization (EMR), are experimented to remote sensing imagery classification. By the overall analysis, SVM classifiers outperform traditional classifiers and can be used for environmental remote sensing effectively.