Study On The Land Cover Based On The Comprehensive Utilization Of SAR Images And TM Images

The classification of remote sensing image is an important means of information auto-extraction from remote sensing image.The accuracy and completeness of classification results will directly affect the completion of the information extraction.Because the forming mechanisms of SAR image and TM image are different,the two images of the same area are complementary,land cover classification accuracy can be improved by comprehensively applying these two images to land cover classification.This paper described the fusion technology of multi-source remote sensing images,the remote sensing principle of SAR,the image characteristics of TM and the classification and evaluation methods.In addition,this paper studied the multi-source remote sensing image classification based on the fusion of SAR and TM image.The theory of feature fusion and decision fusion,SVM,H/?-Wishart together with maximum likelihood classification method were systematically described;support vector machine was used in feature fusion classification with unipolar SAR and TM images,the overall classification accuracy was 90.80%,which was improved by 22.40% and 6.20% compared to SAR and TM image classification respectively;decision fusion classification method with rule-based polarimetric SAR and TM image was proposed,H/?-Wishart preliminary classification results of polarimetric SAR image and maximum likelihood classification results of TM image were analyzed.Two seriously confused map samples in the classification were selected,and then decision fusion rules were constructed based on the complementary relationship between the advantages of multi-source remote sensing images,the result of classification was fused according to the established rules,and the overall accuracy was 91.80%,which was improved by13.80% and 6.40% compared to SAR and TM image classification respectively.Besides,woodland and building land classification accuracy improved by 95.15% and 13.38% than that of the SAR image.The results showed that: the method used in this paper could effectively extract the arable land,forest,water,construction land and unused land these five kinds of land cover types in the study area,seriously confused woodlands,construction land and water could be well separated so as to increase the classification accuracy of the image.