Abstract: Taking the northeast part of Sanjiang Plain as a case, the information extraction method for typical freshwater marsh wetland was approached. By using TM images and based on semi-variograms and Z-test, different scales texture features of typical vegetations in study area were comparatively studied, and the optimum window size, texture features, and their derivative spectral bands were selected to maximize the structural separation of the vegetations. The quick, unbiased, and efficient statistical tree (QUEST) algorithm was used to build the decision tree model of wetland information extraction, integrating the spectral and texture features with assistant geographical data. The classification results based on QUEST algorithm were examined by confusion matrix accuracy assessment using field GPS samples, and the validation showed that the total classification accuracy was 8458%, and the Kappa coefficient was 0816. It was suggested that the accuracy of classification based on QUEST algorithm was higher than that based on maximum likelihood classification (MLC) supervised method, being proved to be an effective means to extract inland freshwater marsh wetland information.

Key words: Weedy rice, Luolijing (Oryza sativa), Biology, Stress-resistance