A Remote Sensing Study Of Land Use Classification Based On Back-Propagation Artificial Neural Network

Remote sensing (RS) image classification is an important part of remote sensing study. How to improve the accuracy of RS classification is an urgent problem in the research of RS application.In the past decade, with the great development of the research on Artificial Neural Networks (ANN), ANN has become an important method for the application of RS image classification. This study first reviewed and analyzed the latest progress of research on RS image classification home and abroad, especially the theory of ANN; and then, based on this, the classification of land use in the area around Summer Palace in Beijing was investigated by using Back-Propagation (BP) ANN with SPOT image. Before the classification, the contrast extension, the principal components analysis, and the image fusion were used in pre-processing the image; and then, the training area was selected with a high precision and on the basis of the extraction of vegetation and water. The network's construction and parameter were designed according to the study purpose and the features of the image. According to the error matrix analysis and sampling analysis, BP neural networks method's overall accuracy was 91.90%, and the Kappa coefficient was 89.84%. Compared with maximum likelihood methods, both of these values had been improved by approximately 8%.This result indicates that the BP neural network method is an effective classification method, and that it can improve the classification accuracy of RS image. Compared with the traditional methods, the BP neural network method has a better efficiency of self-learning and self-adaptation, and was more useful in studying the application of multi-source data.