Remote Sensing Image Study Of Classification Based On BP Neural Network

Remote sensing(RS) has become an important technical of earth observation for its high timeliness and high-integrated. The multi-spectral, high-resolution remote sensing images can explore the law of development of the earth phenomenon effectively, and can predict the development trends. Therefore, the study of remote sensing images has taken more and more attention of researcher, and remote sensing image classification a research focus in the field.Artificial neural network is a kind of network structure, simulation of the human brain function with adaptive, self-organizing, self-learning ability, and can realize the distributed storage, parallel processing of information. Because the artificial neural network is a nonlinear system, it is suitable for solving complex nonlinear problems, such as remote sensing image classification. The BP neural network is a kind of error back propagation neural network, it will feedback the study error to the hidden layer, change the initial weights and threshold value of network structure, and achieve the desired learning objectives. Practice has proved that the BP neural network can improve the classification accuracy greatly.However, there are some problems during the application of BP neural network. For example, it’s usually get local minimum value and the network convergence speed is slow. Meanwhile, the number of neurons in the hidden layer can’t be get, and the problem of “spectrum with different spectrum, with foreign body” can’t be solved properly. According to the above problem, this paper will explore the new methods of image classification based on Landsat-7 images covering Jiangyin, Jingjiang. This paper will combining genetic algorithm with BP algorithm to optimize the initial weights of BP neural network, and obtain the optimal number of neurons in hidden layer. In order to solve the existing the problem of "spectrum with different spectrum, with foreign body", the NDVI index value will be regard as image feature, combing with image texture feature, spectral information of common, and used in image classification. Finally, the improved BP neural network is used in image classification for the data combine the NDVI index, texture characteristics and spectral information, and compared the results with the traditional method. The experimental results show that the result and precision of classification has the obvious improved with the improved BP neural network.