The Research On Remote Sensing Image Classification Algorithm Based On Texture And Its Applications

Remote sensing image classification, namely assigning each pixel of image to the appropriate category, is one of the important research issues in the remote sensing technology field, and it is essential in monitoring change, extracting information, setting up database, producing thematic map, etc. In recent years, with the improvement of the resolution of remote sensing image, texture and other characteristics are getting clearer and we can get more spectral and textural information. Texture analysis can improve the classification accuracy of remote sensing images.In this paper, remote sensing image textural feature extraction, remote sensing image classification and its application are discussed, and some major classification methods, including the ISODATA algorithm, minimum distance classification, maximum likelihood classification and so on, are studied; mainstream remote sensing image textural feature extraction methods, including gray-level co-occurrence matrix, wavelet transform, fractal model and geo-statistics extraction, are analyzed.The contributions are as follows:1. A remote sensing image unsupervised classification method in combination with textural features is proposed. This method extracts uniform local binary pattern histogram of images as textural features. It yields good performance to distinguish different texture, and it increases the classification accuracy of remote sensing images and can be implemented easily.2. A remote sensing image supervised classification method based on spectral and textural features is proposed. This method combines the color classification function of maximum likelihood method and the texture discriminant of minimum distance method as the final classification function, which has a high accuracy of classification and is a simple algorithm.3. A mining satellite monitoring system is designed and implemented. If image analyst is inexperience, he/she can select unsupervised classification method. Otherwise, he/she select supervised classification method, and then get the location of illegal mining through change detection. The two methods above have high classification accuracy, and can provide more reliable basis for change detection.