Classification of Camouflages Using Hyperspectral Images Combined With Fusing Adaptive Sparse Representation and Correlation Coefficient

Bing ZHOU; Bing-xuan LI; Xuan HE; He-xiong LIU; Fa-zhen WANG

doi:10.3964/j.issn.1000-0593(2021)12-3851-06

Journals >Spectroscopy and Spectral Analysis >Volume 41 >Issue 12 >Page 3851 > Article

Spectroscopy and Spectral Analysis
Vol. 41, Issue 12, 3851 (2021)

Classification of Camouflages Using Hyperspectral Images Combined With Fusing Adaptive Sparse Representation and Correlation Coefficient

Bing ZHOU, Bing-xuan LI^*;, Xuan HE, He-xiong LIU, and Fa-zhen WANG

Author Affiliations

Department of Opto-electronics, Army Engineering University of PLA, Shijiazhuang 050000, China

show less

DOI: 10.3964/j.issn.1000-0593(2021)12-3851-06 Cite this Article

Bing ZHOU, Bing-xuan LI, Xuan HE, He-xiong LIU, Fa-zhen WANG. Classification of Camouflages Using Hyperspectral Images Combined With Fusing Adaptive Sparse Representation and Correlation Coefficient[J]. Spectroscopy and Spectral Analysis, 2021, 41(12): 3851 Copy Citation Text

show less

Fig. 1. Gray scale and pseudo colour images of experimental data 1

Download full size | View in the Article

Fig. 2. Gray scale and pseudo colour images of experimental data 2

Download full size | View in the Article

Fig. 3. The classification accuracies of the three algorithm are affected by the parameter setting
(a): The number of neighborhoods and KNN classification accuracy; (b): K and SRC classification accuracy;(c): Regularization parameter λ and CRC classification accuracy

Download full size | View in the Article

Fig. 4. The classification accuracy of CCASRC algorithm is impacted by the parameters λ and N

Download full size | View in the Article

Fig. 5. Classification results of various classification methods in the background of green space
(a): K-means; (b): KNN; (c): SRC; (d): CRC; (e): ASRC; (f): CCASRC

Download full size | View in the Article

Fig. 6. Matrix histogram of various classification method OA

Download full size | View in the Article

Fig. 7. Classification results of various methods in desert background
(a): SRC; (b): CRC; (c): ASRC; (d): CCASRC

Download full size | View in the Article

输入: 高光谱图像数据x, 样本字典D^t

循环: 遍历高光谱图像的所有像元
步骤1: 确定待测样本。
步骤2: 构造样本字典, 针对待测样本x, 构造局部背景样本字典D^h, 联合样本字典与局部背景样本字典D=[D^t, D^h]。
步骤3: 对待测样本进行自适应稀疏编码, 计算重构误差

\begin{array}{l} {\hat{α}}^{i} = argmin ‖ x - D α^{i} ‖_{*} \\ s.t. ‖ α^{i} ‖_{2} \leq η \\ r^{i} (x) = ‖ x - D^{t} {\hat{α}}^{i} ‖_{2} \end{array}

步骤4: 根据式(10)计算待测样本与训练样本之间的CC值
步骤5: 根据提出的融合策略来融合CC值与残差值。

Class (x) = argmin (r^{i} (x) + λ \times (1 - c c^{i} (y)))

输出: 高光谱图像分类图

Table 1. CCASR algorithm flow chart

View in the Article

类别	方法
类别	训练样本	测试样本	SRC	CRC	ASRC	CCASRC
A	670	6 783	5 865(0.864)	4 703(0.69)	6 327(0.93)	6 437(0.94)
B	31	344	302(0.877)	344(1)	344(1.)	323(0.93)
C	37	376	324(0.86)	331(0.88)	297(0.789)	281(0.747)
D	231	2 314	1 770(0.76)	2 124(0.91)	2 075(0.89)	2 145(0.92)
E	491	49 142	37 897(0.77)	37 642(0.77)	42 082(0.85)	44 385(0.89)
F	830	83 080	69 038(0.83)	66 116(0.79)	67 234(0.81)	70 450(0.84)
G	245	21 633	19 196(0.88)	15 939(0.73)	20 306(0.93)	20 834(0.96)
OA			0.839	0.794	0.866	0.905
KC			0.684	0.642	0.757	0.891

Table 2. Classification of various features and backgrounds OA and Kappa

Download Citation

Save the article for my favorites

Paper Information