The Research And Application Of The Improved Adaptive Band Selection Algorithm

As hyperspectral remote sensing image has the characteristics of high-dimension, high resolution, high amount of data, it brings the serious problems. For example, information overlap rate is high, the larger redundant information and correlation, the time of data storage is long and so on. Because of data set covers the rich application satellite remote sensing image information, Choose the most valuable and meaningful information becomes more and more important. So in the whole hyperspectral image processing process, the band selection is an important link of the whole process, it has also become a follow-up processing the necessary condition.This paper mainly studies high light spill image band selection techniques, which is based on the analysis of the existing band selection algorithm, including the following aspects:First, on the item of the large amount of information of hyperspectral remote sensing image, fully considering the enrichment of each band information as well as the similarity of the band, using the improved block adaptive band selection algorithm combining with Euclidean distance to choose the larger information and weaker correlation of the band or the band subset as the similarity characteristics of the block remote sensing image is larger.Second, from the Angle of classification, the characteristic of features category in the chosen subset band is the strongest. So use the K-means algorithm to rough classify for the selected band or band subset. For the select band or band subset which have already been classified, fully considering the object spectrum characteristic and separability, Use spectrum correlation coefficient algorithm to futher select band or band subset.In this paper, PengLai19-3C hyperspectral oil image experiments were carried out, It achieves high-standard requirements that separability of the selected band or band subset is the strongest. Paper proposed band selection method, for subsequent oil spill analysis and image applied to provide the high value of the data.