% % Semi-SAE: Semisupervised stacked autoencoders for hyperspectral image classification DEMO. % Version: 1.0 % Date : May 2021 % % This demo shows the Semi-SAE method for hyperspectral image classification. % % /data ... The folder contains the original hyperspectral data, the selected 10 bands data, the patch features for the IP and PU % data sets. % /LP-BS ... The folder contains LB-PS code for band selection. % /run_IP_codes ... The folder contains the main codes for IP. % /run_PU_codes ... The folder contains the main codes for PU. % /UFLDL-Tutorial-Exercise-master ... The folder contains the tutorial exercise by Andrew Ng. % regionadjacency.m ... The function used to computes adjacency matrix for image of labeled segmented regions. % regiongrowHHUZhou.m ... The function used for regiongrow in a multispectral image. % Main steps (take IP for example): step1_IndianPSample;% Select the initial training and test samples. %step2_LP_BS;% band selection, skip this while using the selected 10 bands: %dataIndianP_10Bands.mat or dataPaviaU_10Bands.mat; %step3_Generate_PatchFeatures; % patch features generation, skip this %while using the generated patch features: dataIndianP_Patch.mat or %dataPaviaU_Patch.mat; step4_IndianP_HSP_PreTraining;% Pre-training based on the spectral information (SAE1) step5_IndianP_SPF_PreTraining;% Pre-training based on the spatial information (SAE2) step6_IndianP_HSP1;% fine-tuning based on the spectral information and initial labeled samples step7_IndianP_regrow_10bands_HSP2;% region grow based on the spectral information and SAE1 step8_IndianP_SPF3;% fine-tuning based on the spatial information and grown labeled samples step9_IndianP_regrow_10bands_SPF4;% region grow based on the spatial information and SAE2 step10_IndianP_HSP5;% fine-tuning based on the spectral information and grown labeled samples (SAE1) step11_IndianP_SPF6;% fine-tuning based on the spatial information and grown labeled samples (SAE2) step12_IndianP_NewProbMerge;%merge the probabilities of SAE1 and SAE2 using MRF % % -------------------------------------- % Note: Required toolbox/functions are covered % -------------------------------------- % 1. LP-BS: Provided by Jenny Du. % 2. Peter Kovesi_seg_functions: https://www.peterkovesi.com/matlabfns/index.html#segmentation (for regionadjacency.m). % 3. UFLDL-Tutorial-Exercise-master: https://github.com/dkyang/UFLDL-Tutorial-Exercise. % -------------------------------------- % Cite: % -------------------------------------- % % [1]S. G. Zhou, Z. H. Xue, P. J. Du. Semisupervised Stacked Autoencoder With Cotraining for Hyperspectral Image Classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2019, 57(6): 3813-3826. % % -------------------------------------- % Copyright & Disclaimer % -------------------------------------- % % The programs contained in this package are granted free of charge for % research and education purposes only. % % Copyright (c) 2021 by Zhaohui Xue % zhaohui.xue@hhu.edu.cn % -------------------------------------- % For full package: % -------------------------------------- % https://sites.google.com/site/zhaohuixuers/
ZhaohuiXue/Semi-SAE-release
A DEMO for "Semisupervised Stacked Autoencoder With Cotraining for Hyperspectral Image Classification" (Xue et al., TGRS, 2019)
MATLAB