Wetland Extraction in Sanjiang Plain Based on Self-Organized Feature Map Neural Network Clustering Model

Abstract

By analyzing spectral characteristics of MODIS remote sensing data in Sanjiang Plain, we extract the wetland in this area based on a set of multi-temporal and multi-spectral MODIS data. We first transform the selected data by a Minimum Noise Fraction (MNF) rotation and take the first two components of the transformed data as the experimental data. To improve the accuracy of classification of wetland, we use a Self-organization Feature Map (SOM) neural network model. SOM has a good performance in resisting noise and can be implemented with parallel processing technique. It is capable of keeping the topological structure of the original data. As a result, it may achieve a better classification compared with other clustering models. After clustering, we perform a discrete wavelet transform (DWT) to smooth the data and eliminate noise from the data. The result shows that the SOM model is effective and the clustering result has been improved.