Approach study of remote sensing image classification by automatically extracting the optimal bands

In remote sensing image automatic identification and classification of the computer,selecting the best band subset is es- sential to improve classification accuracy.According to statistical theory,in remote sensing images,the eigenvectors of surface features belong to a certain type are subjected to normal distribution.So the normal test to training samples is the key;based on this theory, this paper using TM image data,through the examination of the training area is Normal or not,let the computer automatically select the best combination of bands,and to assess the accuracy of classification.Research shows that the pre-assessment to the classification ac- curacy after the computer automatically selecting the best combination of bands,are more convenient,faster,time-saving and effort than the conventional the accuracy test data.