Abstract: Micro landform classification of grid DEM is the foundation of digital landform refinement application. The micro landform classification method of grid DEM based on regular knowledge has problems such as low degree of automation and incomplete classification. With the advantages of BP neural network, an artificial intelligence method and implementation approach for micro landform classification of grid DEM are constructed. The experimental results show that the BP neural network method of micro landform classification of grid DEM has the advantages over the existing method of overlay analysis by landform factors. The BP neural network method of micro landform classification of grid DEM not only avoids the complicated data overlay analysis process, but also effectively improves the completeness and misclassification. Among the six kinds of micro landform classified from the hill-position, the alluvium has the strongest adaptability to this method, with accuracy of 100% and the weakest accuracy of 89.23% for the back-slope.

Key words: grid DEM, landform classification, BP neural network, hill-position