RESUMEN
With the development of remote sensing technology and imaging spectrometer, the resolution of hyperspectral remote sensing image has been continually improved, its vast amount of data not only improves the ability of the remote sensing detection but also brings great difficulties for analyzing and processing at the same time. Band selection of hyperspectral imagery can effectively reduce data redundancy and improve classification accuracy and efficiency. So how to select the optimum band combination from hundreds of bands of hyperspectral images is a key issue. In order to solve these problems, we use spectral clustering algorithm based on graph theory. Firstly, taking of the original hyperspectral image bands as data points to be clustered , mutual information between every two bands is calculated to generate the similarity matrix. Then according to the graph partition theory, spectral decomposition of the non-normalized Laplacian matrix generated by the similarity matrix is used to get the clusters, which the similarity between is small and the similarity within is large. In order to achieve the purpose of dimensionality reduction, the inter-class separability factor of feature types on each band is calculated, which is as the reference index to choose the representative bands in the clusters furthermore. Finally, the support vector machine and minimum distance classification methods are employed to classify the hyperspectral image after band selection. The method in this paper is different from the traditional unsupervised clustering method, we employ spectral clustering algorithm based on graph theory and compute the interclass separability factor based on a priori knowledge to select bands. Comparing with traditional adaptive band selection algorithm and band index based on automatically subspace divided algorithm, the two sets of experiments results show that the overall accuracy of SVM is about 94. 08% and 94. 24% and the overall accuracy of MDC is about 87. 98% and 89. 09%, when the band selection achieves a relatively optimal number of clusters using the method propoesd in this paper. It effectively remains spectral information and improves the classification accuracy.
RESUMEN
As the rotation speed of ground based hyperspectral imaging system is too fast in the image collection process, which exceeds the speed limitation, there is data missed in the rectified image, it shows as the_black lines. At the same time, there is serious distortion in the collected raw images, which effects the feature information classification and identification. To solve these problems, in this paper, we introduce the each component of the ground based hyperspectral imaging system at first, and give the general process of data collection. The rotation speed is controlled in data collection process, according to the image cover area of each frame and the image collection speed of the ground based hyperspectral imaging system, And then the spatial orientation model is deduced in detail combining with the star scanning angle, stop scanning angle and the minimum distance between the sensor and the scanned object etc. The oriented image is divided into grids and resampled with new spectral. The general flow of distortion image corrected is presented in this paper. Since the image spatial resolution is different between the adjacent frames, and in order to keep the highest image resolution of corrected image, the minimum ground sampling distance is employed as the grid unit to divide the geo-referenced image. Taking the spectral distortion into account caused by direct sampling method when the new uniform grids and the old uneven grids are superimposed to take the pixel value, the precise spectral sampling method based on the position distribution is proposed. The distortion image collected in Lao Si Cheng ruin which is in the Zhang Jiajie town Hunan province is corrected through the algorithm proposed on above. The features keep the original geometric characteristics. It verifies the validity of the algorithm. And we extract the spectral of different features to compute the correlation coefficient. The results show that the improved spectral sampling method is better than the direct sampling method. It provides the reference for the similar product used on the ground.
RESUMEN
Aiming at the spectral distortion produced in PCA fusion process, the present paper proposes an improved low spectral distortion PCA fusion method. This method uses NCUT (normalized cut) image segmentation algorithm to make a complex hyperspectral remote sensing image into multiple sub-images for increasing the separability of samples, which can weaken the spectral distortions of traditional PCA fusion; Pixels similarity weighting matrix and masks were produced by using graph theory and clustering theory. These masks are used to cut the hyperspectral image and high-resolution image into some sub-region objects. All corresponding sub-region objects between the hyperspectral image and high-resolution image are fused by using PCA method, and all sub-regional integration results are spliced together to produce a new image. In the experiment, Hyperion hyperspectral data and Rapid Eye data were used. And the experiment result shows that the proposed method has the same ability to enhance spatial resolution and greater ability to improve spectral fidelity performance.