Abstract: Soils in subtropical hilly lands were surveyed using TM image, and according to the environmental characteristics in this region, linear spectral mixture unmixing, KT transformation, texture extracting were used to enhance soil information and eliminate noises. The results showed that the use of TM image in the accuracy of soil classification survey had an increase of more than 5 percent when the noiseeliminated TM6 was added, and the characteristic images extracted by other several image processing methods discriminated some soil classes very well. The accuracy of paddy soils classfied using the images processed by linear spectral mixture unmixing and their texture images were generally higher than that of using raw images, and waterlogged and percogenic paddy soils had the highest classification accuracy using KT transformation. However, the characteristic images extracted by these image processing methods were not better than the raw images on the increase of the total accuracy of soil classification, and all the characteristic images except that of processed by linear spectral mixture unmixing needed to add the raw images for improving the total accuracy.

Key words: remote sensing, soil surveying, image processing, supervised classification