Reed Wetland Extraction in the Yellow River Delta Nature Reserve Based on Knowledge Inference Technology

Based on Landsat TM images and 155 field survey points in 2001, reed wetland is derived using knowledge inference technology. Six types of wetland are determined using supervised classification method. On account of the confusions between reed and other types of wetland, rules are established using methods as following steps: (1) reed wetland is extracted from mudflat wetland, rearing and shrimp pond and water body according to the value of normalized digital vegetation index (NDVI) (NDVI > 110 in the image of January and NDVI > 95 in the image of August); (2) the reed is distinguished from paddy field only if the value of texture mean (based on August image) is less than 10; (3) the reed wetland is separated from the Chinese tamarisk, providing the value of band3 based on principal component analysis (KL3) is more than 100. All these classification rules are built using knowledge engineer based on Erdas Image software; then classification map is obtained by neighbor analysis technology. The accuracy estimation shows that knowledge - based classification gets total accuracy of 89. 02% and the kappa coefficient of 0. 893,2, which is more effective than the traditional supervised classification (the total accuracy is 81. 60% and the kappa coefficient is 0. 793).