Hyperspectral Band Selection Based on Adaptive Neighborhood Grouping and Local Structure Correlation

Algorithm 1

ANG-LSC for band selection.

Input: Hyperspectral image , the number of bands to be selected , the size of the Gaussian window .

Output: The number of selected bands .

1: The original hyperspectral image cube is segmented into subcube by adaptive subspace partition method.

2: Calculate the information entropy of each band.

3: The information entropy matrix is convolved with the Gaussian kernel function, and the shape is the same.

4: According to Equation (9), the similarity matrix between the two bands is calculated.

5: Calculate local density according to Euclidean distance and similarity matrix.

6: By calculating the minimum distance between each band and other high-density bands, the distance factor coefficient of each band is obtained.

7: Take the product of the three factors as the comprehensive weight of each band. After sorting the weights in descending order, select the expected number of optimal bands as the clustering center to construct the desired band subset.

8: Calculate the local structural similarity index of each band in the subcube.

9: A new weight is defined to reevaluate the quality of each band. According to Equation (16), the band with the largest weight is selected from each cluster as the representative band.