Research On The Classification Method Of Hyper-spectral Image

The ability to recognize the world is getting more and more powerful accom-pany with hyper-spectral remote sensing. The hyper-spectral sensing technology, which provide not only spacial information but spectral information as well, made it possible to get the richest information of material, which can be used to identify component might be involved in it. Therefore, the application based on hyper-spectral information is getting more and more attention in many field, such as precision agriculture, atmospheric research, environment monitoring, medicine, etc. The main object in the above research is that the classification of hyper-spectral image, which is the important branch in the hyper-spectral image process field. The three aspects are discussed in this paper:the hyper-spectral image of preprocessing; the image of feature extraction and feature selection; the hyper-spectral image of classification method. Especially, for the classification method, the supervised and un-supervised are discussed separately. The profiles of three main content were described in detail in following:First of all, the pre-processing methods, include image profile and spectral profile, are discussed in detail and the examples are illustrated as well. Based on the hyper-spectral image pre-processing method, the quality of information will be enhanced, and it will provide more reliable and accurate data source for the following analysis.Secondly, we also discuss the method of feature extraction and feature selec-tion from hyper-spectral images in detail from two aspects:spectrum and image, and show the corresponding application examples. The feature extraction and feature selection from hyper-spectral image will condense the data greatly and meanwhile provide the simply input variables for classification model..Finally, classification method of hyper-spectral images, including supervised and un-supervised, are studied. The spectral information on pixel in image are mined and used to constructed classification model, which will achieve more ac-curate pattern recognition.