Remote Sensing Image Classification Method Based On Convolutional Neural Networks

How to classify the remote sensing image accurately and efficiently, has always been an important research content in the field of remote sensing. In recent years, with the development of technology of artificial intelligence, neural network has become an effective way of dealing with remote sensing image classification. Compared with the traditional statistical classification methods, neural network has its own characteristics,such as the learning capability,fault-tolerant ability and without assumed probability model. Convolutional Neural Networks is a technique which combine the artificial neural network and deep learning, it has characteristics of local perception area, hierarchical, feature extraction and classification process combined with the global training, its core idea is to combine the local receptive field, the weights of sharing and the time or space sampling to achieve some degree of displacement and scale invariance and deformation.In this thesis, we reviewed the domestic and foreign research achievements in artificial neural network and the convolutional neural network, especially the Convolutional Neural Networks. We selected the classical convolution neural network model as a classifier and exercised classification experiment. The main work and conclusions in this thesis can be summarized as,(1) Convolutional algorithm of neural network is established, and the programming based on MATLAB programming language is implemented.(2) The feasibility of remote sensing image classification with Convolutional Neural Networks is obtained by experiment. We also analyzed the advantages and disadvantages of the Convolutional Neural Networks model, with the comparison of maximum likelihood classification, support vector machine and BP neural network recognition method.Our results show that the Convolutional Neural Networks classification method is suitable for the shape obvious features, such as water, buildings, etc., which is related to the displacement of the invariance of Convolutional Neural Networks itself, it is also limited by the nature of network, with the comparison of the traditional classification methods, its classification line is not nuance, which may influence the image classification expression.