Classification And Segmentation In Remotely Sensed Imagery

Abstract

Remote sensing technology and it’s applications are rapidly advancing. The algorithms and techniques for processing of remotely sensed images has thus become increasingly important and is an area of active research. Recently, a lot of research has been conducted in the domain of classification techniques of remotely sensed imagery. Classification techniques extract useful features from the remotely sensed data and then categorize it into different categories. This thesis proposes classification techniques for different applications in remotely sensed imagery. The first technique is a novel method for pixel classification. The proposed method exploits the spatial information of image pixels using morphological profiles produced by structuring elements of different sizes and shapes. Morphological profiles produced by multiple structuring elements are combined into a single feature by decimal coding. The advantage of proposed feature is that it can effectively utilize the potential of multiple morphological profiles without increasing the complexity of feature space. The second technique deals with the classification of image patches. The work is presented in the context of image retrieval framework of multispectral image patches. The proposed retrieval method is based on the combination of sparse coding and global image features. The third technique is for segmentation and change classification of built-up area in high resolution imagery using logistic regression. The research was conducted on multi spectral multi temporal images covering the 2010 floods in Pakistan. Segmentation was performed to extract the built up area from the satellite images and then change detection was performed to find the damaged built up area. The damaged area was classified into three categories basing on the extent of damage. The results of change classification were compared and found consistent with the manual assessment report produced by experts of United Nations using Worldview 1 satellite imagery with sub meter resolution. The fourth and the last technique is for regularized classification of changes using elastic net and high dimensional change feature vector comprising spectral, textural and structural changes. The proposed schemes were tested with simulated as well as real life multispectral and hyperspectral remotely sensed datasets. The multispectral dataset comprised of high resolution images with ground resolution of 2.5 meter. The performance was validated using authentic and publicly available ground truth data using standard performance measures. Qualitative and quantitative comparisons have been drawn with state of the art classification schemes and significant improvement is reported