A Remote Sensing Study Of Land Cover Classification Based On BP Artificial Neural Network

Currently, using RS (Remote Sensing) technology to capture information of land cover has become one of the main research on LUCC (Land use and land cover change), while classification is an important part when using RS to obtain information of land cover. RS,as products and tools of information age, with the advantages of short period, informative data, efficient to get access, can reflect the dynamic sequence using various means countered by few limitations, which makes it more popular amongst the scholars. Recently, with the increasing of various remote sensing platforms and image resolution, and the rapid development of computer technology, remote sensing has played an imperative role in investigation of regional land cover types and quantitative extraction of land cover information.In recent years, accompany with the theoretic development of Artificial Neural Network, neural network technology is increasingly becoming an important means of remote sensing in land cover classification which higher classification accuracy than that of traditional, but also a broader range of applications. This thesis tries on the software, ENVI, the remote sensing processing platform, using BP artificial neural network classification method; combined with ideas of stratified classification and analysis of texture features, adopting SPOT5, the remote sensing images, to conduct land cover classification in Hanchuan City, Hubei Province, which is of great significance to improve the classification accuracy.This thesis can be divided into five parts;The thesis begins with the introduction, which gives an overview of research background, purports of topics and recent research status at home and abroad. The first chapter analyses the existing method of land cover classification, and studies at the structure design and parameters choice of BP artificial neural network, which is used in land cover classification.In the second chapter, the classification theory of Remote sensing image has been analyzed. It is mainly to introduce the principles of image classification in remote sensing and the theories, ideas of stratified classification, analytic method of texture features, BP artificial neural network, involved in land cover classification. Additionally, the characteristics, structure and algorithm of BP neural network are highlighted to prepare a theoretical foundation for the following sections. Within the next chapter, the paper gives an overview on the basic information of geographical location,natural and socio-economic conditions in research area.Then, the conditions of data collection and processing of research area is described in the next segment. This part, using ENVI as a platform, through analysis of principal components, images fusion, tries on preprocessing SPOT5,the remote sensing images used in the second rural land survey in Hanchuan City, Hubei Province, broadening the difference between image features, improving the capacity of image interpretation, to determine the pilot area more accurately, and improve the classification accuracy, which are prepared for the image classification of remote sensing in the following step.The fifth part is not only the highlight but also the core of this paper. Firstly, the system and interpretation signs of Land cover classification have been studied, to determine the types of Land cover classification. And then, with the ideas of stratified classification, the bodies of water are extracted from the original image using the mask technology. Additionally, through calculating gray level cooccurrencematrix, texture features will be extracted and classified from the SPOT images, and useful texture information will be added to the images which have been shielded off the water bodies before. After that, according to the research purposes and image features, this thesis will accurately design the BP network structure and training and carry out BP neural network classification in images, which have been added by useful texture information. This method that combines the extracted texture information with the spectral information of multi-spectral image itself, can contribute to improve the classification accuracy. At the end, through confusion matrix, the accuracy of classification result will be evaluated. The results show that BP neural network classification is an effective method of land cover classification, which, when compared with other traditional methods, has better abilities of self-learning and self-configuring, to improve the classification accuracy.The final part of this thesis is the conclusion and outlook. On the basis of research conclusion, the paper proposes the main innovations:(1)With effective use of Remote sensing processing software, combining the foundation of analyzing the space size, shape and spectral characteristics of different objects, and using the internationally advanced classification method, the BP artificial neural network, to classify land cover by the SPOT5 images of Hanchuan City, Hubei Province is beneficial for the improving of classification accuracy. (2) During the process of land cover classification by BP artificial neural network, taking full advantage of the stratified classification ideas, and filtering the texture feature information through calculating the gray level cooccurrencematrix, can not only contribute to the extension of Band information of remote sensing images, but also advance the classification accuracy.