Thresholded Subspace Clustering (TSC) Algorithm Demo on Hyperspectral Datasets
This Live Script lets you:
- Choose a dataset (Pavia, PaviaUni, Salinas, Indian Pines)
- Run TSC on the filtered pixels i.e., foreground pixels
- Visualize the resulting cluster map
Dropdown to select desired dataset
% % Dataset Configuration
this_script_path = matlab.desktop.editor.getActiveFilename;
this_script_folder = fileparts(this_script_path);
proj_root = fileparts(this_script_folder);
addpath(fullfile(proj_root, "TSC_MATLAB"));
clear; clc;
DATASETS = struct( ...
'Pavia', struct( ...
'data_file', "Pavia.mat", ...
'gt_file', "Pavia_gt.mat", ...
'data_key', "pavia", ...
'gt_key', "pavia_gt"), ...
'PaviaUni', struct( ...
'data_file', "PaviaUni.mat", ...
'gt_file', "PaviaU_gt.mat", ...
'data_key', "paviaU", ...
'gt_key', "paviaU_gt"), ...
'Salinas', struct( ...
'data_file', "Salinas_corrected.mat", ...
'gt_file', "Salinas_gt.mat", ...
'data_key', "salinas_corrected", ...
'gt_key', "salinas_gt"), ...
'Indian_pines', struct( ...
'data_file', "Indian_pines.mat", ...
'gt_file', "Indian_pines_gt.mat", ...
'data_key', "indian_pines", ...
'gt_key', "indian_pines_gt") ...
);
ds_name = "Pavia";
Function to run TSC on remote sensing datasets
function run_tsc_hsi(ds_name, DATASETS)
cfg = DATASETS.(ds_name);
% Load Data and Ground Truth
data_struct = load(fullfile("MAT Files", cfg.data_file));
gt_struct = load(fullfile("GT Files", cfg.gt_file));
data_cube = data_struct.(cfg.data_key);
gt_data = gt_struct.(cfg.gt_key);
[H, W, B] = size(data_cube);
% Mask to remove the background pixels
mask_fg = gt_data ~= 0;
mask_vec = mask_fg(:);
% Determine K from unique non-zero labels
gt_labels = unique(gt_data);
gt_labels(gt_labels == 0) = [];
K = numel(gt_labels);
% Reshape the data cube to a matrix (D x N)
% D = Bands; N=pixels
X_full = reshape(data_cube, [], B);
X_full = X_full.';
X = X_full(:, mask_vec);
% Run TSC on the filtered out pixels
[~, ~, ~, c, K_used] = tsc(X, K, 'Verbose', true, 'MaxNZ', 50);
% Map cluster labels back to image grid (H x W)
c_image_vec = zeros(H*W, 1);
c_image_vec(mask_vec) = c; % fill only foreground pixels
c_image = reshape(c_image_vec, H, W);
gt_image = gt_data;
% Visualize clustering result
figure('Color', 'w', 'Position', [100 100 900 800]);
t = tiledlayout(2,2, ...
'TileSpacing','compact', ...
'Padding','compact');
% -------------------------
% Shared colormap
% -------------------------
colors = lines(K);
cmap = [0 0 0; colors];
colormap(cmap);
% -------------------------
% LEFT: Ground Truth
% -------------------------
ax1 = nexttile(1);
imagesc(gt_data);
axis image off;
title('Ground Truth', 'FontWeight','bold');
clim([-0.5 K+0.5]);
% -------------------------
% RIGHT: Clustering
% -------------------------
ax2 = nexttile(2);
imagesc(c_image);
axis image off;
title(sprintf('TSC Clustering (%s, K = %d)', ds_name, K_used), 'FontWeight','bold');
clim([-0.5 K+0.5]);
% -----------
% Colorbars
% -----------
cb1 = colorbar(ax1, 'southoutside');
cb1.Ticks = 0:K;
cb2 = colorbar(ax2, 'southoutside');
cb2.Ticks = 0:K;
end
Run TSC button
run_tsc_hsi(ds_name, DATASETS);
