US 12,444,161 B2
Hyperspectral image distributed restoration method and system based on graph signal processing and superpixel segmentation
Junzheng Jiang, Guilin (CN); Wanyuan Cai, Wenzhou (CN); and Jiang Qian, Chengdu (CN)
Assigned to Hangzhou Institute of Technology, Xidian University, Hangzhou (CN); Guilin University of Electronic Technology, Guilin (CN); and Yangtze Delta Region Institute (Huzhou), University of Electronic Science and Technology of China, Huzhou (CN)
Filed by Hangzhou Institute of Technology, Xidian University, Hangzhou (CN); Guilin University of Electronic Technology, Guilin (CN); and Yangtze Delta Region Institute (Huzhou), University of Electronic Science and Technology of China, Huzhou (CN)
Filed on Jan. 12, 2023, as Appl. No. 18/096,054.
Claims priority of application No. 2022109338569 (CN), filed on Aug. 4, 2022.
Prior Publication US 2024/0046602 A1, Feb. 8, 2024
Int. Cl. G06V 10/426 (2022.01); G06T 5/70 (2024.01); G06T 7/11 (2017.01); G06V 10/80 (2022.01)
CPC G06V 10/426 (2022.01) [G06T 5/70 (2024.01); G06T 7/11 (2017.01); G06V 10/803 (2022.01); G06T 2207/10036 (2013.01)] 9 Claims
OG exemplary drawing
 
1. A hyperspectral image distributed restoration method based on graph signal processing and superpixel segmentation, comprising:
step 1, constructing an input signal model, to linearly normalize a hyperspectral image to obtain a normalized hyperspectral image;
step 2, selecting an image corresponding to a target-band from the normalized hyperspectral image, and pre-denoising the image to obtain a pre-denoised image;
step 3, segmenting the pre-denoised image into a plurality of superpixel regions using simple linear iterative cluster (SLIC) superpixel segmentation algorithm;
step 4, constructing a skeleton graph based on the plurality of superpixel regions;
step 5, constructing a local graph for all pixels of each of the plurality of superpixel regions, to obtain a plurality of local graphs respectively corresponding to the plurality of superpixel regions;
step 6, constructing sub-graphs based on the skeleton graph and the plurality of local graphs;
step 7, applying an exchange distributed mode to each of the sub-graphs, to perform distributed denoising on the hyperspectral image;
step 8, establishing an optimization model of each of the sub-graphs, based on the each of the sub-graphs; and
step 9, repeating step 7 and step 8 through a plurality of iterations, wherein each of the plurality of iterations adopts the exchange distributed mode in the step 7 to iteratively solve the optimization model in the step 8; and in a situation that each of the sub-graphs being meeting an iterative convergence condition, the plurality of iterations are stopped, and a corresponding sub-graph of the sub-graphs meeting the iterative convergence condition is no longer updated while other sub-graphs of the sub-graphs not meeting the iterative convergence condition continue to participate the plurality of iterations.