Research On Target Classification Based On Hyperspectral Image

More and more scholars pay attention on the hyper-spectral image classification with the development of hyper-spectral processing technology. Compared with the traditional remote sensing, hyper-spectral remote sensing image contains abundant spectral information called "image combined with spectra", which determine it has enormous potential to fine classify the features, at the same time, the large dimension of hyper-spectral image also brings a huge challenge to hyper-spectral image classification.The method of traditional remote sensing image classification is no longer suitable for hyper-spectral image. Therefore, in order to make full use of the hyper-spectral image information, to study the new classification algorithm of hyper-spectral image will be very important.Hyper-spectral image has a phenomenon of "same spectrum with different features and same feature of different spectrum", which brings a lot of difficulties to hyper-spectral image classification. This paper presents a hyper-spectral image classification supervised method combined spectral information and spatial information,which is based on combining subspace model and Markov models classification method,and make a improvement to the method. Subspace model can solve the hyper-spectral image big data problem, and Markov models can turn the geometric space continuous distribution of pixels of hyper-spectral image into the pixel and its neighborhood relationships. So we can take full advantage of spatial information of hyper-spectral images, and solve the phenomenon of "same spectrum with different features and same feature of different spectrum" mentioned above. The classification method process is as follows: First, through the subspace model let the hyper-spectral image projected onto a low-dimensional subspace, and we will have an initial classification results. Then, we use the principal component analysis original hyper-spectral image to gain the first pricipal component, which we can extracted gradient information to construct edge coefficient.Finally, we use Markov model again to classify the initial classification results, and finally obtain the classification results. By designing experiments and simulation, the experimental results show that the proposed method has a better classification performance.