基于分层引导滤波与最近邻正则化子空间的高光谱图像分类 下载: 621次
Hyperspectral Image Classification Based on Hierarchical Guidance Filtering and Nearest Regularized Subspace
中国矿业大学 信息与控制工程学院, 江苏 徐州 221116
图 & 表
图 1. Comparison of spectral feature before and after filtering
Fig. 1. Comparison of spectral feature before and after filtering
下载图片 查看原文
图 2. The flowchart of HGF-NRS
Fig. 2. The flowchart of HGF-NRS
下载图片 查看原文
图 3. Reconstruction residuals of a test sample
Fig. 3. Reconstruction residuals of a test sample
下载图片 查看原文
图 4. The influence of varying ε and r on OA
Fig. 4. The influence of varying ε and r on OA
下载图片 查看原文
图 5. The influence of varying λ and T on OA
Fig. 5. The influence of varying λ and T on OA
下载图片 查看原文
图 6. Classification results of algorithms on Indian Pines dataset
Fig. 6. Classification results of algorithms on Indian Pines dataset
下载图片 查看原文
图 7. Classification results of algorithms on Salinas dataset
Fig. 7. Classification results of algorithms on Salinas dataset
下载图片 查看原文
图 8. Classification results of algorithms on GRSS_DFC_2013 dataset
Fig. 8. Classification results of algorithms on GRSS_DFC_2013 dataset
下载图片 查看原文
表 1实验参数设置
Table1. Experimental parameters setting
| g | r | T | λ | Indian Pines | 0.01 | 2 | 8 | 0.05 | Salinas | 0.000 5 | 2 | 18 | 0.01 | GRSS_DFC_2013 | 0.000 1 | 1 | 7 | 0.03 |
|
查看原文
表 2分类性能对比(Indian Pines)
Table2. Classification performance comparison (Indian Pines)
Class | Train | Test | NRS[10] | Gabor-NRS[19] | JCR2[12] | HiFi-We[17] | EPF-G-g[16] | HGF-NRS | Alfalfa | 24 | 22 | 95.45 | 100.0 | 16.20 | 100.0 | 100.0 | 100.0 | Corn-N | 90 | 1338 | 78.68 | 90.15 | 95.83 | 93.39 | 94.22 | 97.93 | Corn-M | 80 | 750 | 76.57 | 93.42 | 99.44 | 93.60 | 96.66 | 95.05 | Corn | 68 | 169 | 47.16 | 95.48 | 95.83 | 100.0 | 78.97 | 98.83 | Grass-P | 71 | 412 | 90.87 | 97.54 | 99.50 | 98.78 | 97.53 | 100.0 | Grass-T | 74 | 656 | 97.62 | 98.18 | 97.57 | 99.69 | 99.24 | 99.70 | Grass-P-M | 14 | 14 | 100.0 | 100.0 | 100.0 | 92.85 | 100.0 | 100.0 | Hay-W | 70 | 408 | 99.03 | 100.0 | 99.51 | 100.0 | 100.0 | 100.0 | Oats | 10 | 10 | 81.82 | 100.0 | 100.0 | 100.0 | 100.0 | 100.0 | Soybean-N | 79 | 893 | 75.95 | 98.03 | 97.79 | 92.37 | 85.07 | 98.07 | Soybean-M | 109 | 2346 | 80.48 | 92.52 | 95.75 | 97.62 | 95.84 | 98.81 | Soybean-C | 69 | 524 | 83.16 | 94.14 | 98.53 | 98.09 | 94.45 | 99.05 | Wheat | 68 | 137 | 99.28 | 100.0 | 100.0 | 100.0 | 100.0 | 97.86 | Woods | 85 | 1180 | 95.63 | 100.0 | 100.0 | 99.06 | 98.62 | 100.0 | Buildings-G-T-D | 68 | 318 | 69.95 | 85.99 | 95.56 | 100.0 | 85.31 | 100.0 | Stone-S-T | 46 | 47 | 97.87 | 86.79 | 82.35 | 100.0 | 92.16 | 94.0 | OA | - | - | 82.85 | 94.58 | 96.14 | 96.87 | 94.57 | 98.63 | AA | - | - | 85.59 | 95.77 | 92.12 | 97.84 | 94.88 | 98.71 | Kappa | - | - | 80.27 | 93.74 | 95.55 | 96.39 | 93.75 | 98.43 |
|
查看原文
表 3分类性能对比(Salinas)
Table3. Classification performance comparison (Salinas)
Class | Train | Test | NRS[10] | Gabor-NRS[19] | JCR2[12] | HiFi-We[17] | EPF-G-g[16] | HGF-NRS | Brocoli-G-W-1 | 30 | 1979 | 99.95 | 100.0 | 100.0 | 99.24 | 100.0 | 100.0 | Brocoli-G-W-2 | 30 | 3696 | 99.81 | 99.11 | 99.70 | 99.86 | 99.95 | 99.92 | Fallow | 30 | 1946 | 95.15 | 96.38 | 100.0 | 99.79 | 94.95 | 100.0 | Fallow-R-P | 30 | 1364 | 97.63 | 95.29 | 96.51 | 98.97 | 97.36 | 96.06 | Fallow-S | 30 | 2648 | 99.77 | 99.06 | 99.58 | 99.16 | 99.89 | 100.0 | Stubble | 30 | 3929 | 99.95 | 100.0 | 100.0 | 99.23 | 99.92 | 100.0 | Celery | 30 | 3549 | 99.30 | 99.66 | 100.0 | 99.04 | 100.0 | 99.94 | Grapes-U | 30 | 11241 | 81.48 | 94.50 | 92.30 | 82.93 | 90.56 | 99.94 | Soil-V-D | 30 | 6173 | 99.44 | 98.88 | 99.80 | 99.96 | 99.13 | 99.97 | Corn-S-G-W | 30 | 3248 | 90.82 | 95.79 | 97.99 | 89.28 | 91.44 | 99.16 | Lettuce-R-4 | 30 | 1038 | 94.86 | 88.19 | 99.81 | 100.0 | 94.70 | 100.0 | Lettuce-R-5 | 30 | 1897 | 98.91 | 99.95 | 100.0 | 100.0 | 100.0 | 100.0 | Lettuce-R-6 | 30 | 886 | 99.66 | 99.44 | 100.0 | 98.87 | 100.0 | 100.0 | Lettuce-R-7 | 30 | 1040 | 98.04 | 90.43 | 98.38 | 96.82 | 98.47 | 93.88 | Vinyard-U | 30 | 7238 | 58.04 | 79.50 | 80.98 | 88.31 | 77.03 | 95.98 | Vinyard-V-T | 30 | 1777 | 95.81 | 98.88 | 100.0 | 99.38 | 99.33 | 99.83 | OA | - | - | 88.11 | 94.48 | 95.36 | 93.86 | 93.67 | 99.13 | AA | - | - | 94.29 | 95.94 | 97.82 | 96.93 | 96.42 | 99.04 | Kappa | - | - | 86.81 | 93.87 | 94.83 | 93.17 | 92.96 | 99.03 |
|
查看原文
表 4分类性能对比(GRSS_DFC_2013)
Table4. Classification performance comparison (GRSS_DFC_2013)
Class | Train | Test | NRS[10] | Gabor-NRS[19] | JCR2[12] | HiFi-We[17] | EPF-G-g[16] | HGF-NRS | Healthy grass | 99 | 1152 | 99.20 | 99.65 | 97.04 | 92.10 | 99.03 | 99.91 | Stressed grass | 95 | 1159 | 98.29 | 97.22 | 99.91 | 95.85 | 96.71 | 99.91 | Synthetic grass | 96 | 601 | 100.0 | 100.0 | 100.0 | 98.83 | 100.0 | 100.0 | Trees | 94 | 1150 | 97.87 | 98.83 | 99.74 | 88.08 | 99.48 | 100.0 | Soil | 93 | 1149 | 97.60 | 99.91 | 100.0 | 99.65 | 98.63 | 100.0 | Water | 91 | 234 | 100.0 | 87.64 | 89.65 | 97.86 | 100.0 | 100.0 | Residential | 98 | 1170 | 84.82 | 93.19 | 93.66 | 82.64 | 97.04 | 97.73 | Commercial | 95 | 1149 | 91.59 | 96.82 | 97.40 | 62.83 | 97.37 | 99.56 | Road | 96 | 1156 | 77.34 | 87.53 | 93.58 | 88.06 | 96.84 | 97.22 | Highway | 95 | 1132 | 95.96 | 98.57 | 99.55 | 96.11 | 97.66 | 100.0 | Railway | 90 | 1145 | 93.23 | 97.82 | 98.06 | 89.95 | 93.59 | 99.13 | Parking lot 1 | 96 | 1137 | 93.05 | 99.27 | 100.0 | 86.10 | 94.59 | 100.0 | Parking lot 2 | 92 | 377 | 57.96 | 93.70 | 94.68 | 93.37 | 87.61 | 99.47 | Tennis court | 90 | 338 | 99.40 | 100.0 | 94.41 | 100.0 | 95.48 | 100.0 | Running track | 93 | 567 | 100.0 | 96.76 | 100.0 | 100.0 | 100.0 | 100.0 | OA | - | - | 91.86 | 96.68 | 97.67 | 89.70 | 97.10 | 99.42 | AA | - | - | 92.42 | 96.46 | 97.18 | 91.43 | 96.94 | 99.53 | Kappa | - | - | 91.19 | 96.41 | 97.48 | 88.85 | 96.86 | 99.37 |
|
查看原文
表 5不同训练样本比例下的全局准确率(Indian Pines)
Table5. Overall classification accuracy in varyingproportion of training samples(Indian Pines)
| Proportion of training samples/% | 1 | 2 | 3 | 4 | 5 | NRS[10] | 58.78(3.57) | 66.71(1.93) | 72.28(2.03) | 74.46(1.10) | 76.40(1.48) | Gabor-NRS[19] | 58.40(3.54) | 71.50(2.11) | 81.13(1.85) | 84.75(2.73) | 88.38(1.97) | JCR2[12] | 68.42(3.98) | 79.45(3.79) | 87.27(1.09) | 89.88(1.34) | 92.49(1.06) | HiFi-We[17] | 74.74(2.31) | 85.66(1.98) | 87.73(3.16) | 91.72(1.11) | 93.40(1.03) | EPF-G-g[16] | 64.34(3.69) | 74.63(4.00) | 83.02(2.96) | 86.30(0.74) | 88.54(1.42) | HGF-NRS | 79.66(4.04) | 89.47(1.60) | 92.75(0.94) | 94.19(1.29) | 96.38(0.66) |
|
查看原文
表 6不同训练样本比例下的全局准确率(Salinas)
Table6. Overall classification accuracy in varying proportion of training samples(Salinas)
| Proportion of training samples/% | 0.4 | 0.5 | 0.6 | 0.7 | 0.8 | NRS[10] | 85.33(1.94) | 87.32(1.74) | 87.18(1.02) | 87.52(1.13) | 87.77(1.23) | Gabor-NRS[19] | 90.20(0.97) | 91.67(0.76) | 92.42(1.33) | 92.77(1.30) | 93.61(0.56) | JCR2[12] | 91.30(1.54) | 92.32(0.61) | 92.90(1.34) | 93.48(1.26) | 94.22(1.36) | HiFi-We[17] | 91.78(1.43) | 91.78(0.94) | 92.40(1.32) | 92.50(1.32) | 93.19(1.14) | EPF-G-g[16] | 89.03(3.39) | 89.23(1.77) | 91.58(2.71) | 91.77(2.70) | 91.34(3.07) | HGF-NRS | 96.28(0.79) | 96.86(0.88) | 98.40(0.51) | 98.42(0.50) | 98.93(0.43) |
|
查看原文
表 7不同训练样本比例下的全局准确率(GRSS_DFC_2013)
Table7. Overall classification accuracy in varying proportion of training samples (GRSS_DFC_2013)
| Proportion of training samples/% | 1 | 2 | 3 | 4 | 5 | NRS[10] | 80.46(1.67) | 86.24(1.23) | 88.70(1.09) | 90.22(0.56) | 90.85(0.58) | Gabor-NRS[19] | 75.19(1.60) | 84.76(1.35) | 88.96(0.85) | 91.21(0.98) | 93.21(0.78) | JCR2[12] | 84.92(1.33) | 91.15(0.95) | 93.85(1.08) | 95.41(0.94) | 96.52(0.69) | HiFi-We[17] | 80.43(2.62) | 85.22(1.71) | 86.71(1.66) | 88.08(1.03) | 88.72(0.63) | EPF-G-g[16] | 78.08(2.75) | 87.00(1.60) | 90.87(2.08) | 93.11(0.66) | 93.74(0.61) | HGF-NRS | 86.14(2.13) | 93.00(1.66) | 95.18(1.03) | 96.71(0.74) | 97.61(0.74) |
|
查看原文
表 8各算法的计算时间(s)
Table8. Computing time of algorithms (s)
| NRS[10] | Gabor-NRS[19] | JCR2[12] | HiFi-We[17] | EPF-G-g[16] | HGF-NRS | Indian Pines | 48.3 | 33.10 | 49.5 | 589.5 | 7.9 | 53.4 | Salinas | 136.6 | 91.32 | 140.2 | 647.26 | 13.7 | 179.9 | GRSS_DFC_2013 | 90.9 | 86.6 | 103.6 | 1 208.4 | 108.5 | 270.4 |
|
查看原文
徐冬冬, 程德强, 陈亮亮, 寇旗旗, 唐守锋. 基于分层引导滤波与最近邻正则化子空间的高光谱图像分类[J]. 光子学报, 2020, 49(4): 0410004. Dong-dong XU, De-qiang CHENG, Liang-liang CHEN, Qi-qi KOU, Shou-feng TANG. Hyperspectral Image Classification Based on Hierarchical Guidance Filtering and Nearest Regularized Subspace[J]. ACTA PHOTONICA SINICA, 2020, 49(4): 0410004.