Rough Sets Approach To Remote Sensing Image Processing And Classification

Rough sets theory is a new mathematic approach to uncertain and vague data analysis. It is, no doubt, one of the most challenging areas of modern computer applications nowadays and a new very important and rapidly growing area of research and applications. The application of rough sets for knowledge discovery, data reduction decision support, pattern recognition and others have proved to be a very effective new mathematic approach. The theory found many, interesting real-life application in medicine, banking, engineering and others.The uncertainty is a key issue for Remote Sensing theory and application, especially in classification. Evaluating and processing the uncertainty of the Remote Sensing information is an important task for RS application. As a result of that, the theme of "Rough Sets Approach to Remote Sensing Image Processing and Classification" is selected as the main research topic of this dissertation. The detailed research work and suggestions can be sum up as the following:1. Investigates the uncertainty mechanism in Remote Sensing data obtaining, information processing, observation scale and rpresentation of production. And the theoretical framework of uncertainty in Remote Sensing information processing via Rough Sets theory is analysed in detail.2. Presents an image pre-processing algorithm that combines human visual properties with rough sets. According to rough sets theory, an image is segmented to different sub-images according to condition attribute on human visual properties in the algorithm, and the contrast-response of the sub-images are transformed respectively. It can realize filtering noise when image enhanced.3. Propose a new method of discretization of continuous attributes based on dynamic layer cluster. A unified framework of the rough set theory to deal with discrete and continuous attributes is suggested.4. Based on the investigation of the concept and algorithm of significance of attributes, core, attribute reduction and values reduction, investigates the methodology of features selection, classification rule induction and rough uncertainty assessment in classification decision table. Employing the MIBARK, method of attribute reduction, and heuristic method of value reduction induce the classification rule from decisiontable in a remote sensing classification experiment.5. A new approach of Remote Sensing image classification based on rough BP neural networks (RBPNN), which promise to overcome some problems encountered in a conventional BP neural network (BPNN) is presented. The novelty of this network lies in applying rough sets for extracting classification rules directly from training dataset, then no extra parameters have to be set for the network. The architecture and training method of this network are presented. The survey and analysis of the RBPNN for the classification of remote sensing image is also discussed. The successfully application of this network in a land cover classification illustrates the simple computation and exact accuracy of the new neural network and the flexibility and practicality of this new approach.6. Discribes the semantic expression of rough sets under the meaning of set-valued measure and establishes a RBFNN modal based on rough sets. A learning mechanism of RBFNN is constructed by means of rough logical. These survey and analysis of the RBFNN based on rough sets for the classification of Remot Sensing multi-spectral image is presented. The proposed method was successfully applied in a classification of land cover with results confirming the flexibility and practicality of this rough approach.7. Error matrix, a fairy tool for assessing a classifier's quality, can't meet the need for the uncertainty because it ignores the spatial information during calculation and the local uncertainty in the image is discarded. A new classification uncertainty assessment strategy together with its computational implementation via Rough Sets is proposed. This methodology assesses classification uncertainty concerned spatial information on the pixel scale. Consequently, it can be visualized via a surface model and is helpful to understand the spatial structure of uncertainty.