Application of back-propagation neural network interpolation method supported by lidar data and geomorphic unit classification

Abstract

In tidal flat terrain of the yellow sea radial sand ridges in eastern China, tidal creeks with water are regarded as the "blind area" of LiDAR surveys. These areas are also hard to be surveyed efficiently and cheaply by traditional surveying methods. To solve the problems of high cost and great effort, this paper researches a Back-Propagation neural network interpolation method, supported by LiDAR data and geomorphic unit classification. The interpolation model structure contains 2 hidden layers with 6 neurons in every layer. This research consists of the following steps: (1) geomorphic unit classification by analyzing dynamic geomorphology of tidal creeks, (2) terrain spatial regularity learning by analyzing a large set of LiDAR data, (3) model building based on the Back-Propagation neural network technique, (4) sample data training with similar tidal creek geomorphic unit data, (5) model structure and parameters determination, (6) testing by comparing the results with the survey data. The test results show that the developed methodology is effective in producing the terrain lacking LiDAR DEM in tidal flats.