Band selection method for hyperspectral images based on self-representation and local similarity protection
A hyperspectral image and band selection technology, applied to computer components, character and pattern recognition, instruments, etc., can solve problems such as lack of interpretability, poor selection of bands, and failure to retain semantic information of original data, etc., to achieve The effect of overcoming poor interpretability and improving classification accuracy
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0031] The band data in the hyperspectral image is large, and there is a high correlation between the bands, and there are many redundant bands. The band selection method can eliminate redundant bands in hyperspectral images and reduce the amount of data. Existing band selection techniques, such as the SC method proposed by V.Kumar et al., implement band selection on the original hyperspectral image data based on the idea of conversion, which cannot preserve the semantic information of the original image and lacks interpretability.
[0032] In order to select the most discriminative band from the original hyperspectral image, the present invention proposes a hyperspectral image band selection method based on self-expression and local similarity protection through research and practice, see figure 1 , including the following steps:
[0033] 1. Input the original hyperspectral image data matrix, assuming that the input original hyperspectral image data matrix is the Indian ...
Embodiment 2
[0048] The hyperspectral image band selection method based on self-expression and local similarity protection is the same as embodiment 1, and the specific steps of the sparse reconstruction matrix and diagonal matrix initialization of the normalized hyperspectral image data matrix described in step 3 include:
[0049] 3.1 Initialize the sparse reconstruction matrix of the normalized hyperspectral image data matrix into an m×m all-1 matrix by using the all-one matrix method, where m represents the total number of bands of the normalized hyperspectral image data matrix;
[0050] 3.2 Use the identity matrix method to initialize the diagonal matrix of the normalized hyperspectral image data matrix into an m×m identity matrix.
[0051] The present invention first continuously updates and iterates the sparse reconstruction matrix of the normalized hyperspectral image data matrix to obtain a band selection matrix, then calculates an evaluation vector through the band selection matrix...
Embodiment 3
[0053] The hyperspectral image band selection method based on self-expression and local similarity protection is the same as embodiment 1-2, and the specific steps of calculating the Laplacian similarity matrix described in step 4 include:
[0054] 4.1 Choose a pixel, and calculate the Euclidean distance between it and all remaining pixels according to the following formula:
[0055]
[0056] Among them, d ij Indicates the Euclidean distance between the i-th pixel and the j-th pixel, x ik Indicates the l-th band of the i-th pixel, ∑ means summation operation, √ means square root operation;
[0057] 4.2 Construct the k-nearest neighbor graph of the normalized hyperspectral image data matrix and calculate the similarity weight matrix of the normalized hyperspectral image data matrix according to the following formula:
[0058]
[0059] Among them, w ij Represents the jth column element of the i-th row in the similarity weight matrix W of the normalized hyperspectral ima...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More