Remote Sensing Image Classification Based On Neural Network

ENVI is a set of wide application of remote sensing image processing software, can be on the ENVI neural network for remote sensing images, maximum likelihood, minimum distance, ISODATA algorithm, the decision tree classification and object-oriented and so on. Than any other in the classification method, neural network classification method not only long but also for the time involved is slow, because from the mechanism of the neural network algorithm, the in the weight adjustment is a process of iteration convergence and its classified nature very slow. According to the above phenomenon, has decided to adopt the Matlab neural network toolbox and IDL language, in order to programming to realize the optimization algorithm.With the deep research in the theory of neural network, neural network has played a very important position in the classification of the image. This paper adopts two plans to increase the speed of remote sensing image classification, the first solution is through the study of neural network, to analyze the network weights are revised, iterative convergence, and provide neural network toolbox in Matlab software, the weight vector and the BP neural network were analyzed, the mathematical factor in the form of code added to the characteristics of adaptive neural network toolbox, realized the fast convergence rate in the iteration process. The second scheme is integrated on the ENVI IDL language in neural network optimization algorithm of weight adjustment process, has reached the purpose of the neural network classification speed and accuracy and reliable.