Consistency Analysis of Multi-Source Remotely Sensed Images for Land Cover Classification

The importance of accurately describing the nature of land cover resources is increasing. With the aim to analyze the consistency of remotely sensed images from different sensors for land cover classification, three medium spatial resolution optical image sources in Xuzhou city were classified in the study, including CBERS, ETM, and ASTER. Land cover classification was conducted by Maximum Likelihood Classification (MLC), Support Vector Machines (SVM) and Decision Tree (DT). By comparing the classification results, SVM performed best and the results of SVM classifier were used for consistency analysis. The results we obtained suggested that different images obtained around the same time can lead to dissimilar classification results. Consistency analysis was carried through according to the experimental results of two groups of data. Apart from the individual data source, the two types of image data in each group were combined to form a mixed dataset of multi-source data and then used as the input of SVM classifier. It proved that the mixed dataset consisting of multi-source data could improve the classification performance of singe image so the collaborative use of multi-source data would be feasible for land cover classification.