高光谱分类算法

相关运算

软阈值

2015 Supervised-Spectral-spatial-Hyperspectral-Image-Classification-with-Weighted-Markov-Random-Fields

(1)

(2)使用ADMM算法处理优化问题

高光谱分类算法_第1张图片

2017 Dissimilarity-Weighted Sparse Representation for Hyperspectral Image Classification

通过考虑基于待分类像素与字典的像素局部信息,提出了DWSRC(dissimilarity-weighted sparse representation-based classifier)分类器。

Algorithm 1 Proposed DWSRC for HSIs
Input: A initial dictionary of training samples D =[D_1, D_2,\cdots, D_C]\in R^{R\times N} for C class; a test pixel set Y = \{y^m\}_{m=1,\cdots,M}; sparsity regularization \lambda
For each test pixel y^m do
    1) Calculate the weight matrix w_i betweeny^m and each dictionary atom d_i via Gaussian kernel distance with a suitable distance metric;
    2) Generate the locality-constrained dictionary set  D'=[D'_1, \cdots, D'_c, \cdots, D'_C] \in R^{B\times N}, where D'_c = [w_{c1}d_{c1}, w_{c2}d_{c2},\cdots , w_{cN_c} d_{cN_c}]
    3) Sparsely code y^m over the new weighted dictionary D' via 1-norm minimization according to

\hat{\alpha} = \arg\min_\alpha \|y-D'\alpha\|_2 + \lambda\|\alpha\|_1
    4) Identify the final class of  y^m with the minimal residual rule according to
\text{class}(y) = \arg\min_{c=1,\cdots,C} \|y -D'_c\delta _c(\alpha)\|_2^2
end for
Output: An 2-D map that records the labels of the test sample set Y

2015-Extreme-Learning-Machine-With-Composite-Kernels-for-Hyperspectral-Image-Classification

(1)提取每一个像素的空间特征x^s和光谱特征x^\omega

The spectral feature vector x^\omega_i is the original x_i, which consists of spectral reflection values across all bands.
The spatial feature vector x^s_i i is extracted from the local spatial neighborhood of pixel x_i and defined as the mean of pixels in the spatial neighborhood of x_i in this paper.

(2)计算ELM隐含层输出矩阵H_sH_\omega

K_s(x_i, x_j) = \exp\left(-\frac{\|x_i^s-x_j^s\|^2}{2\delta_s^2}\right)

K_\omega(x_i, x_j) = \exp\left(-\frac{\|x_i^\omega-x_j^\omega\|^2}{2\delta_\omega^2}\right)

K = \mu Ks + (1 - \mu)K\omega

(3)输出分类结果

计算系数\alpha

\alpha = \left(\frac{I}{C}+K\right)^{-1}Y

f(x) = K_x\alpha = [ f_(x), \cdots, f_m(x)]

Local adaptive joint sparse representation for hyperspectral image classification

2019

此文提出了LAJSR算法.

高光谱分类算法_第2张图片


 

你可能感兴趣的:(遥感,分类,机器学习,算法)