Research On Classification Of Hyperspectral Image Based On Spatial-spectral Structural Excavation

With the rapid development of hyperspectral imaging technology, the image data’s spatial resolution and spectral resolution have been significantly improved, while the commonly-used classification algorithms on natural images cannot get good results under the circumstance that hyperspectral data are huge and spectrum information are way too many. So the classification algorithm which is designed according to hyperspectral data itself has became a focus in recent years. In this paper, treating combined spatial-spectral information as the core, we put forward some different hyperspectral image classification algorithms based on spatial-spectral information. The main work is summarized as follows:1.A sparsity spatial-spectral based hyperspectral image classification algorithm is proposed, based on hyperspectral data sparsity analysis, making the full use of hyperspectral data spectral and spatial information, with the help of superpixel theory, we can betterly ensure the accuracy of classification with a significant reduction in the number of samples. Firstly, we do the AP clustering on the basis of the spatial structure of the samples, and obtain superpixel results, combined with the spatial relationship between superpixels, we use joint sparse coding. And then use a simulation method frequently used on actual hyperspectral data sets, compared with some classical methods by sparse representation, we can see the sparsity spatial-spectral based hyperspectral image classification algorithm presented in this chapter has obvious advantages.2.A hyperspectral image classification algorithm based on spatial-spectral nucleared low-rank graph is proposed. Based on semi-supervised method, joining the knowledge of the nuclear mapping and spatial constraints, and using low rank representation to mine the global information data. Firstly, we do the superpixel block on the nuclear map which has been done AP clustering to reduce the difficulty caused by the nonlinear bad data points, then we construct nuclear low-rank showing graph among the structure of the data. At the same time, we make the spatial constraint graph by using hyperspectral’s spatial information. Finally, get samples’ classification results with the idea of semi-supervised sparse graph,and simulate them in real hyperspectral data sets. By comparing with some of the classic classification algorithm, the algorithm proposed in this chapter has obvious advantages;3.A hyperspectral image classification algorithm based on sparse tensor spatial-spectral is proposed. On the basis of sparse representation theory, taking into account the spatial coherence of hyperspectral images at the same time, we combine the spatial information data and spectral information data together, promoting them to the hyperspectral data tensor representation. Then we propose an tensor sparse representation algorithm based on spatial-spectral, and using it for hyperspectral data classification. By the simulation experiments did on the hyperspectral data sets, it shows that the method proposed in this chapter can get high recognition rate feature.