-
Notifications
You must be signed in to change notification settings - Fork 13
/
main_montecarlo.m
223 lines (169 loc) · 6.93 KB
/
main_montecarlo.m
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
% =========================================================================
% This code compares different specral unmixing algorithms that consider
% spectral variability using realistic synthetically generated endmember
% data.
% *** This code follows the same structure of main.m, but uses a Monte
% Carlo simulation to evaluate the algorithms more reliably ***
%
% The code is provided as part of the following publication:
% Spectral Variability in Hyperspectral Data Unmixing: A
% Comprehensive Review
%
% =========================================================================
clear all
close all
clc
addpath(genpath('synthetic_data_generation'))
addpath(genpath('library_extraction'))
addpath(genpath('utils'))
addpath('methods')
% number of Monte Carlo realizations
num_runs = 10;
% initialize the vectors to store the quantitative metrics
RMSEA(1) = nan;
RMSEY(1) = nan;
RMSEM(1) = nan;
SAMM(1) = nan;
TIMES(1) = nan;
for mc_run = 1:num_runs
rng(mc_run,'twister');
% select the signal to noise ratio:
SNR = 30;
% generate synthetic data (abundance, endmember signatures, mixed image)
[Y,Yim,A,A_cube,Mth,M_avg]=generate_image(SNR);
P = size(M_avg,2);
[nr,nc,L] = size(Yim);
% extract reference endmember and spectral library from the image
M0 = vca(Y,'Endmembers',P);
M0 = sort_endmembers_to_ref(M_avg,M0);
% spectral library extraction
bundle_nbr = 5; % number of VCA runs
percent = 20; % percentage of pixels considered in each run
Lib = extractbundles_batchVCA(Y, M0, bundle_nbr, percent);
% *********************************************************************** %
% perform unmixing with the different methods
% *********************************************************************** %
clear algnames
% ----------------------------------------
[A_FCLS,M_FCLS,time_fcls,Yhat_FCLS] = adaptor_FCLS(Yim,M0,Lib,[]);
% saves the performance metrics:
algnames{1} = 'FCLS';
RMSEA(end+1) = norm((A-A_FCLS)/numel(A),'fro');
RMSEY(end+1) = norm((Y-Yhat_FCLS)/numel(Y),'fro');
RMSEM(end+1) = nan;
SAMM(end+1) = nan;
TIMES(end+1) = time_fcls;
% ----------------------------------------
% Set the initialization of some of the other algorithms:
A_init = A_FCLS;
% ----------------------------------------
[A_MESMA,M_MESMA,time_MESMA,Yhat_MESMA] = adaptor_MESMA(Yim,M0,Lib,A_init);
% saves the performance metrics:
algnames{end+1} = 'MESMA';
RMSEA(end+1) = norm((A-A_MESMA)/numel(A),'fro');
RMSEY(end+1) = norm((Y-Yhat_MESMA)/numel(Y),'fro');
RMSEM(end+1) = norm((Mth(:)-M_MESMA(:))/numel(Mth),'fro');
temp = 0; for i=1:nr*nc, for j=1:P, temp = temp + subspace(Mth(:,j,i),M_MESMA(:,j,i))/(nr*nc); end, end; SAMM(end+1) = temp;
TIMES(end+1) = time_MESMA;
% ----------------------------------------
opt_SocialSparseU.fraction = 1/10;
opt_SocialSparseU.lambda = 0.1;
[A_SocialSparseU,M_SocialSparseU,time_socialSparseU,Yhat_SocialSparseU] = adaptor_SocialSparseU(Yim,M0,Lib,A_init,opt_SocialSparseU);
% saves the performance metrics:
algnames{end+1} = 'Fractional SU';
RMSEA(end+1) = norm((A-A_SocialSparseU)/numel(A),'fro');
RMSEY(end+1) = norm((Y-Yhat_SocialSparseU)/numel(Y),'fro');
RMSEM(end+1) = norm((Mth(:)-M_SocialSparseU(:))/numel(Mth),'fro');
temp = 0; for i=1:nr*nc, for j=1:P, temp = temp + subspace(Mth(:,j,i),M_SocialSparseU(:,j,i))/(nr*nc); end, end; SAMM(end+1) = temp;
TIMES(end+1) = time_socialSparseU;
% ----------------------------------------
opt_elmm.lambda_s = 1;
opt_elmm.lambda_a = 0.05;
opt_elmm.lambda_psi = 0.01;
[A_ELMM,M_ELMM,time_elmm,Yhat_ELMM] = adaptor_ELMM(Yim,M0,Lib,A_init,opt_elmm);
% saves the performance metrics:
algnames{end+1} = 'ELMM';
RMSEA(end+1) = norm((A-A_ELMM)/numel(A),'fro');
RMSEY(end+1) = norm((Y-Yhat_ELMM)/numel(Y),'fro');
RMSEM(end+1) = norm((Mth(:)-M_ELMM(:))/numel(Mth),'fro');
temp = 0; for i=1:nr*nc, for j=1:P, temp = temp + subspace(Mth(:,j,i),M_ELMM(:,j,i))/(nr*nc); end, end; SAMM(end+1) = temp;
TIMES(end+1) = time_elmm;
% ----------------------------------------
opt_DeepGUn.dimAut = 2;
opt_DeepGUn.lambda_zref = 0.001;
opt_DeepGUn.lambda_a = 0.01;
[A_DeepGUn,M_DeepGUn,time_DeepGUn,Yhat_DeepGUn] = adaptor_DeepGUn(Yim,M0,Lib,A_init,opt_DeepGUn);
% saves the performance metrics:
algnames{end+1} = 'DeepGUn';
RMSEA(end+1) = norm((A-A_DeepGUn)/numel(A),'fro');
RMSEY(end+1) = norm((Y-Yhat_DeepGUn)/numel(Y),'fro');
RMSEM(end+1) = norm((Mth(:)-M_DeepGUn(:))/numel(Mth),'fro');
temp = 0; for i=1:nr*nc, for j=1:P, temp = temp + subspace(Mth(:,j,i),M_DeepGUn(:,j,i))/(nr*nc); end, end; SAMM(end+1) = temp;
TIMES(end+1) = time_DeepGUn;
% ----------------------------------------
opt_RUSAL.tau = 0.001;
opt_RUSAL.tau2 = 0.001;
[A_RUSAL,M_RUSAL,time_RUSAL,Yhat_RUSAL] = adaptor_RUSAL(Yim,M0,Lib,A_init,opt_RUSAL);
% saves the performance metrics:
algnames{end+1} = 'RUSAL';
RMSEA(end+1) = norm((A-A_RUSAL)/numel(A),'fro');
RMSEY(end+1) = norm((Y-Yhat_RUSAL)/numel(Y),'fro');
RMSEM(end+1) = nan;
SAMM(end+1) = nan;
TIMES(end+1) = time_RUSAL;
% ----------------------------------------
opt_NCM = [];
[A_NCM,M_NCM,time_NCM,Yhat_NCM] = adaptor_NCM(Yim,M0,Lib,A_init,opt_NCM);
% saves the performance metrics:
algnames{end+1} = 'NCM';
RMSEA(end+1) = norm((A-A_NCM)/numel(A),'fro');
RMSEY(end+1) = norm((Y-Yhat_NCM)/numel(Y),'fro');
RMSEM(end+1) = nan;
SAMM(end+1) = nan;
TIMES(end+1) = time_NCM;
% ----------------------------------------
[A_BCM,M_BCM,time_BCM,Yhat_BCM] = adaptor_BCM(Yim,M0,Lib,A_init);
% saves the performance metrics:
algnames{end+1} = 'BCM';
RMSEA(end+1) = norm((A-A_BCM)/numel(A),'fro');
RMSEY(end+1) = norm((Y-Yhat_BCM)/numel(Y),'fro');
RMSEM(end+1) = nan;
SAMM(end+1) = nan;
TIMES(end+1) = time_BCM;
end
% remove the 'nan' initialization from the vectors
RMSEA = RMSEA(2:end);
RMSEY = RMSEY(2:end);
RMSEM = RMSEM(2:end);
SAMM = SAMM(2:end);
TIMES = TIMES(2:end);
% reorder
RMSEA = reshape(RMSEA,[],num_runs);
RMSEY = reshape(RMSEY,[],num_runs);
RMSEM = reshape(RMSEM,[],num_runs);
SAMM = reshape(SAMM, [],num_runs);
TIMES = reshape(TIMES,[],num_runs);
% *********************************************************************** %
% show average results
% *********************************************************************** %
scaleV = 10000;
fprintf('\n\n Abundance estimation results: \n')
for i=1:length(algnames)
fprintf([pad(algnames{i},15,'right','.') ': %f +-%f \n'], scaleV*mean(RMSEA(i,:)), scaleV*std(RMSEA(i,:)) )
end
fprintf('\n\n RMSEM results: \n')
for i=1:length(algnames)
fprintf([pad(algnames{i},15,'right','.') ': %f +-%f \n'], scaleV*mean(RMSEM(i,:)), scaleV*std(RMSEM(i,:)) )
end
fprintf('\n\n SAMM results: \n')
for i=1:length(algnames)
fprintf([pad(algnames{i},15,'right','.') ': %f +-%f \n'], scaleV*mean(SAMM(i,:)), scaleV*std(SAMM(i,:)) )
end
fprintf('\n\n RMSEY results: \n')
for i=1:length(algnames)
fprintf([pad(algnames{i},15,'right','.') ': %f +-%f \n'], scaleV*mean(RMSEY(i,:)), scaleV*std(RMSEY(i,:)) )
end
fprintf('\n\n Times results: \n')
for i=1:length(algnames)
fprintf([pad(algnames{i},15,'right','.') ': %f +-%f \n'], mean(TIMES(i,:)), std(TIMES(i,:)) )
end