Remote.Sensing-master
所属分类:图形图像处理
开发工具:matlab
文件大小:33KB
下载次数:7
上传日期:2019-04-22 18:51:37
上 传 者:
小面包啦啦啦
说明: 高光谱影像混合像元分解的一些常用经典代码
(hyperspectral unmixing)
文件列表:
conjugate.m (1333, 2018-01-16)
create4.m (478, 2018-01-16)
expNoiseComp.m (14120, 2018-01-16)
expRandomInitComp.m (16420, 2018-01-16)
expRealData.m (5571, 2018-01-16)
expSimplexVisual.m (693, 2018-01-16)
expSynData.m (4456, 2018-01-16)
fNorm.m (229, 2018-01-16)
getSynData.m (1573, 2018-01-16)
hyperNmfASCL1.m (3340, 2018-01-16)
hyperNmfASCL1_2.m (3504, 2018-01-16)
hyperNmfMDC.m (2334, 2018-01-16)
hyperNmfMVC.m (4609, 2018-01-16)
hyperNmfMdcAscl1_2.m (3960, 2018-01-16)
hyperVca.m (2743, 2018-01-16)
nFindr.m (2895, 2018-01-16)
nmfAbundance.m (1757, 2018-01-16)
sad.m (123, 2018-01-16)
sadEms.m (501, 2018-01-16)
sparsityAnalysis.m (479, 2018-01-16)
steepdescent.m (2100, 2018-01-16)
testMdcMvc.m (4877, 2018-01-16)
testNfindr.m (724, 2018-01-16)
testNmfASCL1.m (2738, 2018-01-16)
testNmfASCL1_2.m (2672, 2018-01-16)
testNmfMDC.m (2477, 2018-01-16)
testNmfMVC.m (1879, 2018-01-16)
testNmfMdcAscl1_2.m (2728, 2018-01-16)
testVCA.m (507, 2018-01-16)
testVCANfindr.m (658, 2018-01-16)
usgsMatch.m (4017, 2018-01-16)
usgsSampleVisual.m (1494, 2018-01-16)
# Remote.Sensing
Hyperspectral image unmixing.
## Contributor
My.PC HengBao.Desktop
## create4.m getSynData.m
Code for generating spectral samples with dirichlet distribution.
## hyperVca.m
VCA implementation. testVCA.m is the corresponding test code.
## Nfindr.m
Nfindr implementation. testNfindr.m is the corresponding test code.
## hyperNmfMDC.m
NMF with minimum distance constraint on endmembers.
## hyperNmfMVC.m
NMF with minimum volume constraint on endmembers.
## hyperNmfASCL1.m
NMF with L1 sparsity constraint on abundance matrix.
## hyperNmfASCL1_2.m
NMF with L1_2 sparsity constraint on abundance matrix. L1_2 sparsity constraint seems inconsistent with reconstruction error.
During the test, if constraint of sum-to-one constraint is set to be small (less than 1), L1_2 constraint will get very small abundance matrix (sum of coefficient at one pixel is far smaller than 1). However, if constraint of sum-to-one is set to me large (5~10 in synthetic data), resultant endmembers will be close to true value.
Another thing is: when the value of reconstruction error and L1_2 constraint is in the similar level, reconstruction error will first decrease, while L1_2 constraint normally keep unchanged or slightly increase. when reconstruction error go down to a value far smaller than L1_2 constraint, L1_2 constraint will be the dominant factor in the following iterations. Namely, decrease of reconstruction error and L1_2 constraint are not in the same cycles. In some situations in which constraint of sum-to-one is weak, reconstruction error will go up again during decreasing of L1_2 constraint.
## test*
test codes for algorithms.
## exp*
detailed experiments.
近期下载者:
相关文件:
收藏者: