The Improved Algorithm Research On Band Selection For Hyperspectral Imagery By Feng Shuo

In the 1980s, a new earth observation technology----hyperspectral technology had produced. The progress of remote sensing technique and spectral imager had promoted the development of hyperspectral technology. Hyperspectral image is made up of hundreds continuous narrowband cartographic feature by hyperspectral imager, therefore,it has very high resolution. At the same time, it could reflect the spatial information and spectral information of the target zone, so the ground-object identification becomes easier. But, the hyperspectral image is also a kind of high dimensional remote sensing image, the huge amount of data brought difficulties to image processing. For example, hyperspectral data has a great redundancy, the data memory space is large, computing speed is slow, and easy to produce Hughes phenomenon,led to a decline in classification accuracy. The high-resolution of hyperspectral images through the price by increased data dimension and the correlation between data. In order to achieve the purpose of dimension reduction, hyperspectral images need to select the band before practical application. So, how to rapidly filtering the band of better recognition capability from dozens or even hundreds bands in the hyperspectral image has become a pressing problem. Aiming at this problem, designing a more efficient algorithm will greatly improve the classification speed and classification accuracy of remote sensing data.This paper tries to explore a kind of improved algorithm of hyperspectral images band selection. This algorithm can maximize the effective information of the original data, and reduce the data dimension. By this improved algorithm will improve precision and computing speed of remote sensing image classification. Through serious algorithm learning and research with a large number of literatures reading, the author proposed to select band by combining two algorithms and several selection criteria,then searched and selected the optimal sample data. With a combination of both methods could make the most advantages of each algorithm. Specific content is as follows:1、Through studying and researching a variety of algorithms in the application of hyperspectral images band selection problem, and analyzed the advantages and disadvantages in all kinds of the algorithm.Design a new improved algorithm to solve the problem of hyperspectral images band selection. This new algorithm had high accuracy and short computing time.2、At the beginning, according to the matrix of pixel gray value between two bands, we compartmentalized the hyperspectral band. Through such operation, try to reduce the pertinence of bands.Using the rapid global optimization of genetic algorithm to select band.By this way, we could get the pheromone list. Using this list we could optimize the ant colony algorithm. At least, select the bands twice by ant colony algorithm’s high accuracy. Then output the excellent band combination.3、Using AVIRIS image to test the improved algorithm. The experimental results show that the improved method made the high classification accuracy of remote sensing images and computation time also had a good effect.4、Compared with the experimental results of improved algorithm and others to verify the superiority of the improved algorithm.5、Analyzing andO Discussing the experimental results, then got the final conclusion:the improved algorithm are better than other algorithms in classification precision and computing time.