Dissertation > Excellent graduate degree dissertation topics show

Classification of Remotely Sensed Imagery Based on LVQ Hierarchical Model

Author: GuoYangYao
Tutor: LiuYong
School: Lanzhou University
Course: Cartography and Geographic Information Systems
Keywords: Remote sensing Land Use / Cover Artificial Neural Networks Learning vector quantization Classification Tengger Desert
CLC: P237
Type: Master's thesis
Year: 2008
Downloads: 244
Quote: 5
Read: Download Dissertation

Abstract

Remote sensing is developed in 1960s Earth Observation technology. With its rapid development, remote sensing technology has strong timeliness, covering a wide range of rich and objective information, etc. are gradually being recognized, it has become one of the most effective means of people's access to the earth's surface information. A variety of remote sensing systems, active and passive, aerospace and aviation, visible light, infrared and microwave, etc. to provide people with more and more remote sensing information. In contrast, the level of processing of the remote sensing information has lagged far behind in the remote sensing information acquisition level, which severely restricted the application of remote sensing technology. Remote sensing image classification as a key technology for remote sensing information processing, has been widely used in land use / land cover classification, and achieved good social and economic benefits. In some complex surface area, such as the study area where the southern edge of the Tengger Desert, the surface features of various types, the terrain changes significantly, covering a wide range of features, leading synonyms spectrum and the same spectrum of foreign body very serious problem, The simple use of remote sensing image of the spectral characteristics of land use / land cover classification is often less than ideal precision direct impact on the level of application of remote sensing images and practical value. Therefore, to improve the accuracy of remote sensing image classification is a key issue in the application of remote sensing. Integrated spectral features and additional features to improve the classification accuracy has become a trend in the classification process. There are two ways you can assist features integrated remote sensing image classification process: first auxiliary features as the classification of logical channels; second is the use of auxiliary characteristics of the image geographical districts. For the first method, widely used in the remote sensing image classification based statistical classification algorithm is based on the premise of parameter estimation and statistical assumptions requires classification data subject to a certain distribution, and the distribution of the assist features of each class of data is difficult to satisfy this condition, The resulting classification accuracy is not high. The second method requires access to knowledge from the auxiliary features to classify Unfortunately, this kind of knowledge due to the defects of expression is often subjective and not integrity. As a classification algorithm, without parameters estimated advantages of artificial neural networks and statistical assumptions in the integrated processing spectral features and additional features. This paper describes a learning vector quantization neural network classifier for the establishment of a hierarchical classification model based on this model to the southern edge of the Tengger Desert TM images for basic data and knowledge under the guidance of the shape, position, texture and so assist feature for classification. In contrast, while the simple use of the spectral characteristics of the maximum likelihood classification. In this paper, the main research work and achievements as follows: (1) a systematic review of the principles and limitations of a variety of commonly used remote sensing image classification algorithm to study the basic principles of learning vector quantization neural network and its applications in remote sensing image classification, elaborated the algorithm advantages it has, and classify it as a hierarchical classification model. (2) based on fieldwork and knowledge reasoning, identified several additional features indicative of ground distribution, through the analysis of the distribution of data in each of the categories in the feature space, the assist features in remote sensing image classification, these auxiliary The characteristics of the quantization process, added to the classification process. (3) the use of neural network toolbox of MATLAB software design LVQ neural network classifier TM images of the study area, spectral data and auxiliary data as classification feature regional land use / cover classification. (4) the accuracy of the classification results evaluation: This model-based classification accuracy of 79.5%, higher precision computer automatic classification. In contrast, the simultaneous use of the spectral characteristics of the maximum likelihood classification accuracy of 73.2%. This indicates that the model for other surface complex regional land use / cover remote sensing classification provide reference.

Related Dissertations

CLC: > Astronomy,Earth Sciences > Surveying and Mapping > Photogrammetry and Surveying, Mapping and Remote Sensing > Surveying, Mapping and Remote Sensing technology
© 2012 www.DissertationTopic.Net Mobile