基于局部重构Fisher分析的高光谱遥感影像分类

刘嘉敏; 杨松; 黄鸿

doi:doi:10.3788/CJL202047.0710001

中国激光, 2020, 47 (7): 0710001, 网络出版: 2020-07-10

基于局部重构Fisher分析的高光谱遥感影像分类下载： 877次

Hyperspectral Remote Sensing Image Classification Based on Local Reconstruction Fisher Analysis

刘嘉敏杨松黄鸿

作者单位

重庆大学光电工程学院光电技术及系统教育部重点实验室, 重庆 400044

图 & 表

图 1. LRFA算法流程图

Fig. 1. Flowchart of the proposed LRFA method

下载图片查看原文

图 2. Pavia University数据集的遥感影像。(a)假彩色图;(b)真实地物标记图

Fig. 2. Pavia University hyperspectral image. (a) False-color image; (b) ground-truth map

下载图片查看原文

图 3. Urban数据集遥感影像。(a)假彩色图;(b)真实地物标记图

Fig. 3. Urban hyperspectral image. (a) False-color image; (b) ground-truth map

下载图片查看原文

图 4. 在Pavia University数据集上,本文算法在不同参数k和k_p下的总体分类精度

Fig. 4. Overall accuracy of LRFA at different parameters (k and k_p) in Pavia University dataset

下载图片查看原文

图 5. 在Pavia University数据集上,各算法在不同嵌入维度下的总体分类精度

Fig. 5. Overall accuracy of each algorithm at different dimensions in Pavia University dataset

下载图片查看原文

图 6. 各算法在Pavia University数据集上的地物分类图。(a)真实地物;(b) RAW(OA:78.95%);(c) PCA(OA:78.98%);(d) LPP(OA:80.55%);(e) NPE(OA:80.98%);(f) LDA(OA:76.50%);(g) MMC(OA:75.11%);(h) MFA(OA: 82.62%);(i) LGSFA(OA:78.23%);(j) LRFA(OA:86.07%)

Fig. 6. Classification maps of each algorithm on Pavia University dataset. (a) Ground-truth map; (b) RAW(OA:78.95%); (c) PCA(OA:78.98%); (d) LPP(OA:80.55%); (e) NPE(OA:80.98%); (f) LDA(OA:76.50%); (g) MMC(OA:75.11%); (h) MFA(OA:82.62%); (i) LGSFA(OA:78.23%); (j) LRFA(OA:86.07%)

下载图片查看原文

图 7. 本文算法在Urban数据集上不同参数k和k_p下的总体分类精度

Fig. 7. Overall accuracy of LRFA at different parameters (k and k_p) in Urban dataset

下载图片查看原文

图 8. 在Urban数据集上,各算法在不同嵌入维度下的总体分类精度

Fig. 8. Overall accuracy of LRFA at different dimensions in Urban dataset

下载图片查看原文

图 9. 各算法在Urban数据集上的地物分类图。(a)真实地物;(b)RAW(OA:80.86%);(c) PCA(OA:80.79%);(d) LPP(OA:80.66%);(e) NPE(OA:81.89%);(f) LDA(OA:82.60%);(g) MMC(OA:81.40%);(h) MFA(OA:82.32%); (i) LGSFA(OA:82.50%);(j) LRFA(OA:83.77%)

Fig. 9. Classification maps of each algorithm on Urban dataset. (a) Ground-truth image; (b) RAW(OA:80.86%); (c) PCA(OA:80.79%); (d) LPP(OA:80.66%); (e) NPE(OA:81.89%); (f) LDA(OA:82.60%); (g) MMC(OA: 81.40%); (h) MFA(OA:82.32%); (i) LGSFA(OA:82.50%); (j) LRFA(OA:83.77%)

下载图片查看原文

图 10. 各算法在Pavia University数据集上的二维嵌入特征分布。(a) PCA;(b) LPP;(c) NPE;(d) LDA;(e) MMC;(f) MFA;(g) LGSFA;(h) LRFA

Fig. 10. Two-dimensional embedding distribution of each algorithm on Pavia University dataset. (a) PCA; (b) LPP; (c) NPE; (d) LDA; (e) MMC; (f) MFA; (g) LGSFA; (h) LRFA

下载图片查看原文

表 1不同训练样本下各算法在Pavia University数据集上的分类结果

Table1. Classification results of each algorithm at different training sample sizes in Pavia University dataset

Algorithm	(OA±std) /%Kappa
Algorithm	20 samples	40 samples	60 samples	100 samples	200 samples
RAW	67.70±2.180.597	71.62±1.350.641	74.35±0.940.673	76.28±1.020.696	79.01±0.710.728
PCA	67.68±2.170.597	71.59±1.350.641	74.33±0.940.673	76.28±1.010.696	78.93±0.690.727
LPP	69.41±2.560.618	73.64±0.960.666	76.53±0.950.700	78.31±1.280.722	81.82±0.510.763
NPE	69.69±2.910.621	73.57±0.940.665	76.60±1.090.701	79.63±1.280.737	82.63±0.790.773
LDA	60.25±2.010.504	71.58±1.790.640	75.59±1.670.687	79.28±1.000.732	82.94±1.070.776
MMC	66.32±2.190.581	69.11±1.430.612	70.76±0.940.631	72.22±0.910.647	74.05±0.870.668
MFA	74.76±3.120.680	78.79±1.840.728	81.52±1.230.761	83.48±1.740.786	86.36±0.900.821
LGSFA^[20]	66.24±2.010.575	73.46±2.100.662	76.08±1.810.694	79.87±1.690.739	82.36±1.220.769
LRFA	75.69±3.320.691	80.25±1.400.745	82.18±1.110.769	84.55±1.520.798	86.80±0.550.826

查看原文

表 2各算法在Pavia University数据集每类地物上的分类性能

Table2. Classification performance of each algorithm on class samples in Pavia University dataset

Class	Classification accuracy /%
Class	RAW	PCA	LPP	NPE	LDA	MMC	MFA	LGSFA^[20]	LRFA
1	79.94	80.09	81.04	82.01	81.65	83.53	84.36	77.76	84.63
2	91.45	91.46	93.01	92.92	89.81	84.21	98.28	98.09	97.77
3	45.52	45.43	50.67	65.16	52.94	40.66	42.69	36.28	59.72
4	66.37	66.34	68.08	70.33	73.69	69.11	80.15	80.51	79.92
5	99.02	99.02	99.32	99.62	99.55	99.10	99.55	99.62	99.55
6	44.33	44.33	46.37	52.80	51.34	44.43	42.16	35.77	60.80
7	82.54	82.69	85.50	64.16	25.44	62.11	58.92	36.67	80.71
8	76.63	76.65	77.15	69.82	60.88	70.04	78.38	65.60	77.53
9	99.79	99.79	99.68	98.29	73.13	99.47	99.04	78.68	99.79
OA /%	78.95	78.98	80.55	80.98	76.50	75.11	82.62	78.23	86.07
AA /%	76.18	76.20	77.87	77.24	67.60	72.52	75.95	67.67	82.27
Kappa	0.715	0.715	0.736	0.743	0.685	0.667	0.762	0.701	0.811
Time /s	-	0.032	0.089	0.119	0.030	0.146	0.339	0.590	0.414

查看原文

表 3不同训练样本量下各算法在Urban数据集上的分类结果

Table3. Classification results of each algorithm at different training sample sizes in Urban dataset

Algorithm	(OA±std) /%Kappa
Algorithm	20 samples	40 samples	60 samples	100 samples	200 samples
RAW	71.61±3.220.57	74.81±1.70.611	75.34±1.290.618	75.96±1.060.627	76.78±0.990.638
PCA	71.6±3.220.57	74.8±1.690.611	75.35±1.270.618	75.95±1.040.627	76.79±0.970.638
LPP	71.97±3.160.575	74.23±2.120.603	74.85±1.120.612	75.08±1.130.615	76.23±1.190.63
NPE	71.04±3.420.561	74.69±2.20.611	75.7±1.170.625	76.08±1.030.63	77.78±1.170.653
LDA	58.77±3.940.414	68.24±2.570.529	76.76±0.970.64	78.75±0.970.667	79.14±0.790.672
MMC	70.61±1.870.553	71.45±3.40.567	71.87±2.180.57	72.51±1.540.577	72.7±1.440.582
MFA	74.77±1.670.617	75.4±1.410.621	76.26±2.210.635	76.82±1.650.642	77.83±1.180.655
LGSFA^[20]	74.82±2.60.612	76.61±1.640.640	78.3±1.190.662	78.54±1.140.664	79.28±1.080.674
LRFA	76.64±2.670.64	78.12±1.710.66	78.82±1.240.67	79.28±0.990.676	80.00±0.780.685

查看原文

表 4各算法在Urban数据集每类地物上的分类性能

Table4. Classification results of each algorithm on class samples in Urban dataset

Class	Classification accuracy /%
Class	RAW	PCA	LPP	NPE	LDA	MMC	MFA	LGSFA^[20]	LRFA
1	76.85	76.95	76.87	77.48	80.00	77.24	75.05	76.46	80.70
2	43.28	43.00	39.71	48.49	43.14	39.92	40.95	46.64	47.46
3	81.74	81.71	83.52	90.09	82.65	81.14	88.31	83.74	90.03
4	79.81	79.81	83.32	85.96	75.02	81.30	68.87	85.48	90.48
5	87.74	87.62	87.72	88.35	89.22	88.42	90.13	92.37	90.89
6	66.39	66.30	64.94	66.15	69.08	67.18	67.76	59.39	65.56
OA /%	80.86	80.79	80.66	81.89	82.60	81.40	82.32	82.50	83.77
AA /%	72.63	72.57	72.68	76.08	73.18	72.53	71.84	74.01	77.52
Kappa	0.677	0.675	0.673	0.694	0.706	0.686	0.698	0.695	0.723
Time /s	-	0.063	0.119	0.221	0.047	0.559	0.975	1.459	1.194

查看原文

刘嘉敏, 杨松, 黄鸿. 基于局部重构Fisher分析的高光谱遥感影像分类[J]. 中国激光, 2020, 47(7): 0710001. Liu Jiamin, Yang Song, Huang Hong. Hyperspectral Remote Sensing Image Classification Based on Local Reconstruction Fisher Analysis[J]. Chinese Journal of Lasers, 2020, 47(7): 0710001.

基于局部重构Fisher分析的高光谱遥感影像分类下载： 877次

图 1. LRFA算法流程图

Fig. 1. Flowchart of the proposed LRFA method

图 2. Pavia University数据集的遥感影像。(a)假彩色图;(b)真实地物标记图

Fig. 2. Pavia University hyperspectral image. (a) False-color image; (b) ground-truth map

图 3. Urban数据集遥感影像。(a)假彩色图;(b)真实地物标记图

Fig. 3. Urban hyperspectral image. (a) False-color image; (b) ground-truth map

图 4. 在Pavia University数据集上,本文算法在不同参数k和k_p下的总体分类精度

Fig. 4. Overall accuracy of LRFA at different parameters (k and k_p) in Pavia University dataset

图 5. 在Pavia University数据集上,各算法在不同嵌入维度下的总体分类精度

Fig. 5. Overall accuracy of each algorithm at different dimensions in Pavia University dataset

图 6. 各算法在Pavia University数据集上的地物分类图。(a)真实地物;(b) RAW(OA:78.95%);(c) PCA(OA:78.98%);(d) LPP(OA:80.55%);(e) NPE(OA:80.98%);(f) LDA(OA:76.50%);(g) MMC(OA:75.11%);(h) MFA(OA: 82.62%);(i) LGSFA(OA:78.23%);(j) LRFA(OA:86.07%)

Fig. 6. Classification maps of each algorithm on Pavia University dataset. (a) Ground-truth map; (b) RAW(OA:78.95%); (c) PCA(OA:78.98%); (d) LPP(OA:80.55%); (e) NPE(OA:80.98%); (f) LDA(OA:76.50%); (g) MMC(OA:75.11%); (h) MFA(OA:82.62%); (i) LGSFA(OA:78.23%); (j) LRFA(OA:86.07%)

图 7. 本文算法在Urban数据集上不同参数k和k_p下的总体分类精度

Fig. 7. Overall accuracy of LRFA at different parameters (k and k_p) in Urban dataset

图 8. 在Urban数据集上,各算法在不同嵌入维度下的总体分类精度

Fig. 8. Overall accuracy of LRFA at different dimensions in Urban dataset

图 9. 各算法在Urban数据集上的地物分类图。(a)真实地物;(b)RAW(OA:80.86%);(c) PCA(OA:80.79%);(d) LPP(OA:80.66%);(e) NPE(OA:81.89%);(f) LDA(OA:82.60%);(g) MMC(OA:81.40%);(h) MFA(OA:82.32%); (i) LGSFA(OA:82.50%);(j) LRFA(OA:83.77%)

Fig. 9. Classification maps of each algorithm on Urban dataset. (a) Ground-truth image; (b) RAW(OA:80.86%); (c) PCA(OA:80.79%); (d) LPP(OA:80.66%); (e) NPE(OA:81.89%); (f) LDA(OA:82.60%); (g) MMC(OA: 81.40%); (h) MFA(OA:82.32%); (i) LGSFA(OA:82.50%); (j) LRFA(OA:83.77%)

图 10. 各算法在Pavia University数据集上的二维嵌入特征分布。(a) PCA;(b) LPP;(c) NPE;(d) LDA;(e) MMC;(f) MFA;(g) LGSFA;(h) LRFA

Fig. 10. Two-dimensional embedding distribution of each algorithm on Pavia University dataset. (a) PCA; (b) LPP; (c) NPE; (d) LDA; (e) MMC; (f) MFA; (g) LGSFA; (h) LRFA

表 1不同训练样本下各算法在Pavia University数据集上的分类结果

Table1. Classification results of each algorithm at different training sample sizes in Pavia University dataset

表 2各算法在Pavia University数据集每类地物上的分类性能

Table2. Classification performance of each algorithm on class samples in Pavia University dataset

表 3不同训练样本量下各算法在Urban数据集上的分类结果

Table3. Classification results of each algorithm at different training sample sizes in Urban dataset

表 4各算法在Urban数据集每类地物上的分类性能

Table4. Classification results of each algorithm on class samples in Urban dataset

关于本站 Cookie 的使用提示

全站搜索

基于局部重构Fisher分析的高光谱遥感影像分类 下载： 877次

图 1. LRFA算法流程图

Fig. 1. Flowchart of the proposed LRFA method

图 2. Pavia University数据集的遥感影像。(a)假彩色图;(b)真实地物标记图

Fig. 2. Pavia University hyperspectral image. (a) False-color image; (b) ground-truth map

图 3. Urban数据集遥感影像。(a)假彩色图;(b)真实地物标记图

Fig. 3. Urban hyperspectral image. (a) False-color image; (b) ground-truth map

图 4. 在Pavia University数据集上,本文算法在不同参数k和kp下的总体分类精度

Fig. 4. Overall accuracy of LRFA at different parameters (k and kp) in Pavia University dataset

图 5. 在Pavia University数据集上,各算法在不同嵌入维度下的总体分类精度

Fig. 5. Overall accuracy of each algorithm at different dimensions in Pavia University dataset

图 6. 各算法在Pavia University数据集上的地物分类图。(a)真实地物;(b) RAW(OA:78.95%);(c) PCA(OA:78.98%);(d) LPP(OA:80.55%);(e) NPE(OA:80.98%);(f) LDA(OA:76.50%);(g) MMC(OA:75.11%);(h) MFA(OA: 82.62%);(i) LGSFA(OA:78.23%);(j) LRFA(OA:86.07%)

Fig. 6. Classification maps of each algorithm on Pavia University dataset. (a) Ground-truth map; (b) RAW(OA:78.95%); (c) PCA(OA:78.98%); (d) LPP(OA:80.55%); (e) NPE(OA:80.98%); (f) LDA(OA:76.50%); (g) MMC(OA:75.11%); (h) MFA(OA:82.62%); (i) LGSFA(OA:78.23%); (j) LRFA(OA:86.07%)

图 7. 本文算法在Urban数据集上不同参数k和kp下的总体分类精度

Fig. 7. Overall accuracy of LRFA at different parameters (k and kp) in Urban dataset

图 8. 在Urban数据集上,各算法在不同嵌入维度下的总体分类精度

Fig. 8. Overall accuracy of LRFA at different dimensions in Urban dataset

图 9. 各算法在Urban数据集上的地物分类图。(a)真实地物;(b)RAW(OA:80.86%);(c) PCA(OA:80.79%);(d) LPP(OA:80.66%);(e) NPE(OA:81.89%);(f) LDA(OA:82.60%);(g) MMC(OA:81.40%);(h) MFA(OA:82.32%); (i) LGSFA(OA:82.50%);(j) LRFA(OA:83.77%)

Fig. 9. Classification maps of each algorithm on Urban dataset. (a) Ground-truth image; (b) RAW(OA:80.86%); (c) PCA(OA:80.79%); (d) LPP(OA:80.66%); (e) NPE(OA:81.89%); (f) LDA(OA:82.60%); (g) MMC(OA: 81.40%); (h) MFA(OA:82.32%); (i) LGSFA(OA:82.50%); (j) LRFA(OA:83.77%)

图 10. 各算法在Pavia University数据集上的二维嵌入特征分布。(a) PCA;(b) LPP;(c) NPE;(d) LDA;(e) MMC;(f) MFA;(g) LGSFA;(h) LRFA

Fig. 10. Two-dimensional embedding distribution of each algorithm on Pavia University dataset. (a) PCA; (b) LPP; (c) NPE; (d) LDA; (e) MMC; (f) MFA; (g) LGSFA; (h) LRFA

表 1不同训练样本下各算法在Pavia University数据集上的分类结果

Table1. Classification results of each algorithm at different training sample sizes in Pavia University dataset

表 2各算法在Pavia University数据集每类地物上的分类性能

Table2. Classification performance of each algorithm on class samples in Pavia University dataset

表 3不同训练样本量下各算法在Urban数据集上的分类结果

Table3. Classification results of each algorithm at different training sample sizes in Urban dataset

表 4各算法在Urban数据集每类地物上的分类性能

Table4. Classification results of each algorithm on class samples in Urban dataset

相关论文

相关资讯

关于本站 Cookie 的使用提示

全站搜索

基于局部重构Fisher分析的高光谱遥感影像分类下载： 877次

图 4. 在Pavia University数据集上,本文算法在不同参数k和k_p下的总体分类精度

Fig. 4. Overall accuracy of LRFA at different parameters (k and k_p) in Pavia University dataset

图 7. 本文算法在Urban数据集上不同参数k和k_p下的总体分类精度

Fig. 7. Overall accuracy of LRFA at different parameters (k and k_p) in Urban dataset