Comparison between ETM+ imageries and ICESat-GLAS waveforms for forest classification

Abstract

The paper presents a method for classifying Light Detection And Ranging (LiDAR) full waveform data, using an artificial neural network (ANN) approach. The ANN classifier used was a multilayer perceptron trained through the generalized predefined learning rule functions. Compared with the unsupervised classification based on Landsat 7 ETM+ (Enhanced Thematic Mapper Plus) images, the ANN classifier was suitable to better represent the nonlinearity in the LiDAR waveforms dataset. The multilayer perceptron neural network has proved to be a very effective tool for the classification of waveforms data. The classification results show that forest classification accuracy for broadleaved forest and needleleaved waveforms using ANN classifer is better than the classification accuracy of ETM+ image based-unsupervised classifer. Whereas, the overall classification accuracy of testing datasets using ANN classifier with using waveform data without a prior class probabilities is lower than the unsupervised classifier based-image.