Research On The Classification Method Of Remote Sensing Images Based On Spectral And Texture Features Fusion

The classification methods of remote sensing image based on spectral and texturefeatures fusion is proposed. The remote sensing image classification is realizedrespectively based on DS evidence theory and fuzzy reasoning algorithm similarity.In the classification method that based on DS evidence theory, the spectral andtexture features are extracted in the first stage, and then the features fusion andprobability distribution are taken. Secondly, the combination of the probability offour-channel are taken.The classification based on the maximum probability and theresults evaluated are made at last.Compared with the images classification results thatbefore and after the fusion. For the three classes remote sensing image, the classificationaccuracy of the pre-fusion image is 0.934, and the afte-fusion image's is 0.959. For thefive classes remote sensing image, the classification accuracy of the pre-fusion image is0.765, and the afte-fusion image's is 0.812.In the classification method that based on Similarity fuzzy reasoning,the spectraland texture features are extracted in the first stage, and then take the features fusion infour channels.The classification based on fuzzy reasoning and the results evaluated aremade at last.Compared with the images classification results that before and after thefusion, for the three classes remote sensing image, the classification accuracy of thepre-fusion image is 0.964,and the afte-fusion image's is 0.994.for the five classes remotesensing image,the classification accuracy of the pre-fusion image is 0.922,and theafte-fusion image's is 0.939.The Results show that the multi-feature classification method is better than thesingle feature classification method.The classification accuracy of the method using theimage fusion can effectively improve.