Research On Classification Techniques Of The Hyperspectral Images Based On Case-Based Reasoning

Hyperspectral remote sensing as a new remote sensing technology has played an important role in both military and civilian areas in the past 50 years. Compared with the multi-spectral remote sensing, hyperspectral remote sensing data has characteristics such as hundreds of bands, high spectral resolution, a large amount of data, etc. It has an important research value and significance of application for target classification, recognition and tracking. However, the huge amount and high dimension of data make the hyperspectral image classification processing difficult. Hyperspectral image classification often takes many steps as data analysis, feature extraction, then using different classifiers and adjusting the parameters until finding the optimal classification algorithm. Repetitive process makes hyperspectral image classification low efficiency and low intelligence.The dissertation takes good use of the past experience of hyperspectral image classification and achieves efficient and accurate hyperspectral image classification by changing the experience into practical treatment options. The case-based reasoning hyperspectral image classification system is proposed and established.First, the characteristics of hyperspectral image are analyzed to solve the dimension disaster problem. The principal component analysis linear feature extraction method and the kernel-based and manifold learning-based nonlinear methods are studied. The main features of images are effectively retained and extracted. This step makes subsequent classification possible.Secondly, unsupervised classification algorithm, statistical decision theory-based and machine learning-based supervised classification algorithm, small-sample theory-based semi-supervised classification algorithm are studied according to the conditions of prior knowledge. Meanwhile, the classification results of empirical data and the corresponding classifier parameters are recorded, preparing for the establishment of case-based reasoning system.Finally, the case-based reasoning hyperspectral image classification system is constructed based on the above theoretical research and specific implementation. Multi-case, multi-rule and multi-classifier are integrated. Intelligent learning and the choice of optimal classification algorithm of hyperspectral images are achieved depending on the actual application requirements. The efficiency of classification is improved and the classification algorithm is optimized.