Optics and Precision Engineering, Volume. 31, Issue 21, 3221(2023)
Hyperspectral images feature extraction and classification based on fractional differentiation
Figures & Tables(26)
Fig. 2. Process of extracting hyperspectral image feature using fractional differentiation
Fig. 3. Grayscale images of Pavia university and its SpaFD feature
Fig. 5. Flow chart of constructing spectral-spatial joint criterion
Fig. 6. Relationship between proposed spectral-spatial joint criterion and order of SpaFD feature
Fig. 7. Classification maps of Indian Pines using 3DCNN under 5×5 fractional differential mask for feature extraction
Fig. 8. Classification maps of Botswana using 3DCNN under 5×5 fractional differential mask for feature extraction
Fig. 9. Classification maps of Pavia University using 3DCNN under 5×5 fractional differential mask for feature extraction
Fig. 10. Classification maps of Salinas using 3DCNN under 5×5 fractional differential mask for feature extraction
Table 1. Convolutional kernel size settings for three networks
View tableView in ArticleTable 1. Convolutional kernel size settings for three networks
网络模型 I1 Conv2 Conv3 Conv4 Conv5 FC6 FC7 FC8 3DCNN 5×5×N 7×7×3 3×3×3 - - √ - - 3DCNNPCA 11×11×NP 5×5×7 3×3×5 3×3×3 3×3×3 √ √ √ HybridSN 25×25×NP 3×3×7 3×3×5 3×3×3 3×3 √ √ √
Table 2. Number of output feature maps for three network models
View tableView in ArticleTable 2. Number of output feature maps for three network models
网络模型 I1 Conv2 Conv3 Conv4 Conv5 FC6 FC7 FC8 3DCNN 1 2 4 - - C - - 3DCNNPCA 1 8 16 32 64 256 128 C HybridSN 1 8 16 32 64 256 128 C
Table 3. Results of classical classifiers with 20% training samples
View tableView in ArticleTable 3. Results of classical classifiers with 20% training samples
Classifier Indian Pines Botswana SVM MD LR SVM MD LR Spe-Spa 75.67±0.89 49.69±1.28 61.45±1.08 73.52±1.14 76.43±1.46 87.31±1.19 SpaFD5✕5 74.21±0.46 47.87±1.63 57.70±0.73 60.80±2.28 67.21±1.49 79.81±1.74 SpaFD-Spe-Spa5✕5 75.27±0.46 48.12±0.86 58.79±1.17 68.45±1.43 69.67±1.35 84.80±1.41 LBP-Spe-Spa 70.40±0.25 48.72±1.94 61.27±1.39 68.24±1.02 76.00±0.97 85.65±1.43 Classifier Pavia University Salinas SVM MD LR SVM MD LR Spe-Spa 79.75±0.07 59.67±0.45 77.91±0.75 82.68±0.78 75.56±0.45 85.01±1.24 SpaFD5✕5 77.92±0.13 58.56±0.47 74.78±0.41 78.98±0.19 73.47±0.18 83.70±0.85 SpaFD-Spe-Spa5✕5 79.82±0.15 58.70±0.42 77.00±0.55 81.51±0.65 73.13±0.41 84.06±1.13 LBP-Spe-Spa 75.83±0.19 59.80±0.60 77.07±0.65 70.94±0.23 74.50±0.34 83.62±1.23
Table 4. Classification results of Indian Pines under 3% training samples
View tableView in ArticleTable 4. Classification results of Indian Pines under 3% training samples
Model 3DCNN
AOA（%）±SD（%）
3DCNNPCA
AOA（%）±SD（%）
HybridSN
AOA（%）±SD（%）
Spe-Spa 73.91±5.02 82.71±1.04 86.65±1.10 LBP 60.58±4.33 71.20±2.11 75.52±1.02 LBP-Spe-Spa 71.75±2.64 76.36±1.05 80.46±0.88 SpaFD3✕3 77.78±1.56 84.10±0.62 87.81±0.73 SpaFD-Spe-Spa3✕3 77.81 ±2.29 85.80±0.59 88.94 ±0.67 SpaFD5✕5 74.62±2.87 83.48±0.85 87.62±0.86 SpaFD-Spe-Spa5✕5 75.44±2.64 86.11 ±1.22 88.89±0.83 SpaFD7✕7 74.44±3.64 82.97±0.62 87.30±0.71 SpaFD-Spe-Spa7✕7 74.83±2.93 85.59±0.80 88.52±0.73
Table 5. Classification results of Indian Pines under 5% training samples
View tableView in ArticleTable 5. Classification results of Indian Pines under 5% training samples
Model 3DCNN
AOA（%）±SD（%）
3DCNNPCA
AOA（%）±SD（%）
HybridSN
AOA（%）±SD（%）
Spe-Spa 83.78±1.73 89.61±0.57 92.87±0.98 LBP 72.44±1.89 78.55±1.03 82.71±0.96 LBP-Spe-Spa 79.52±1.46 83.41±0.89 85.33±0.71 SpaFD3✕3 84.47±1.13 90.75±0.77 93.44±0.86 SpaFD-Spe-Spa3✕3 84.79±1.22 92.04±0.44 94.34±0.55 SpaFD5✕5 84.43±1.96 89.92±0.52 93.46±0.72 SpaFD-Spe-Spa5✕5 84.69±1.75 90.42±0.41 94.50 ±0.64 SpaFD7✕7 83.87±1.01 89.96±0.64 93.39±0.78 SpaFD-Spe-Spa7✕7 85.05 ±1.14 92.32 ±0.38 94.31±0.59
Table 6. Classification results of Indian Pines under 10% training samples
View tableView in ArticleTable 6. Classification results of Indian Pines under 10% training samples
Model 3DCNN
AOA（%）±SD（%）
3DCNNPCA
AOA（%）±SD（%）
HybridSN
AOA（%）±SD（%）
Spe-Spa 88.93±0.88 93.76±0.68 97.99±0.45 LBP 79.03±0.92 83.63±0.97 87.76±0.50 LBP-Spe-Spa 84.89±1.01 88.73±0.54 89.98±0.39 SpaFD3✕3 89.35±0.77 93.87±0.24 98.12±0.37 SpaFD-Spe-Spa3✕3 89.49±0.85 95.35 ±0.35 98.86±0.40 SpaFD5✕5 89.43±0.85 94.17±0.41 98.30±0.35 SpaFD-Spe-Spa5✕5 89.50 ±1.17 95.08±0.21 99.01 ±0.42 SpaFD7✕7 88.97±0.92 93.99±0.23 98.15±0.37 SpaFD-Spe-Spa7✕7 89.39±0.70 95.32±0.28 98.89±0.41
Table 7. Classification results of Botswana dataset under 3% training samples
View tableView in ArticleTable 7. Classification results of Botswana dataset under 3% training samples
Model 3DCNN
AOA（%）±SD（%）
3DCNNPCA
AOA（%）±SD（%）
HybridSN
AOA（%）±SD（%）
Spe-Spa 87.33±0.95 90.69±0.83 93.01±1.13 LBP 68.21±1.01 80.32±1.31 83.79±1.25 LBP-Spe-Spa 76.55±0.96 84.65±0.98 88.46±0.99 SpaFD3✕3 87.85±0.94 92.11±1.49 94.08±0.98 SpaFD-Spe-Spa3✕3 87.97±0.87 96.29±0.97 94.82±0.64 SpaFD5✕5 87.89±1.19 91.79±0.69 94.11±1.05 SpaFD-Spe-Spa5✕5 88.04±0.71 96.31 ±0.68 95.01 ±0.57 SpaFD7✕7 88.01±1.54 92.08±1.31 93.85±1.03 SpaFD-Spe-Spa7✕7 88.07 ±0.89 96.01±0.81 94.90±0.67
Table 8. Classification results of Botswana dataset under 5% training samples
View tableView in ArticleTable 8. Classification results of Botswana dataset under 5% training samples
Model 3DCNN
AOA（%）±SD（%）
3DCNNPCA
AOA（%）±SD（%）
HybridSN
AOA（%）±SD（%）
Spe-Spa 90.77±1.68 91.90±1.50 94.05±1.16 LBP 71.36±1.43 82.98±1.46 85.01±0.89 LBP-Spe-Spa 83.49±0.97 86.02±0.88 89.32±0.92 SpaFD3✕3 91.73±1.01 92.64±1.44 95.02±1.01 SpaFD-Spe-Spa3✕3 92.10±0.97 95.98±0.62 97.90±0.76 SpaFD5✕5 91.82±0.90 92.34±1.79 94.98±1.05 SpaFD-Spe-Spa5✕5 92.08±0.49 96.21 ±0.73 97.77±0.80 SpaFD7✕7 91.24±1.41 91.92±0.77 95.11±1.22 SpaFD-Spe-Spa7✕7 92.18 ±0.98 95.99±0.39 98.05 ±0.94
Table 9. Classification results of Botswana dataset under 10% training samples
View tableView in ArticleTable 9. Classification results of Botswana dataset under 10% training samples
Model 3DCNN
AOA（%）±SD（%）
3DCNNPCA
AOA（%）±SD（%）
HybridSN
AOA（%）±SD（%）
Spe-Spa 93.35±2.76 97.77±0.49 98.89±0.38 LBP 79.88±1.36 85.46±0.51 88.13±0.47 LBP-Spe-Spa 86.25±0.97 90.04±0.37 92.22±0.36 SpaFD3✕3 94.97±0.79 97.82±0.45 99.00±0.31 SpaFD-Spe-Spa3✕3 95.09 ±0.89 99.35±0.22 99.61±0.25 SpaFD5✕5 94.53±0.63 97.90±0.38 98.98±0.29 SpaFD-Spe-Spa5✕5 94.89±0.59 99.30±0.29 99.50±0.21 SpaFD7✕7 94.46±0.73 97.80±0.27 99.02±0.25 SpaFD-Spe-Spa7✕7 94.84±0.60 99.49 ±0.14 99.87 ±0.11
Table 10. Classification results of Pavia University under 3% training samples
View tableView in ArticleTable 10. Classification results of Pavia University under 3% training samples
Model 3DCNN
AOA（%）±SD（%）
3DCNNPCA
AOA（%）±SD（%）
HybridSN
AOA（%）±SD（%）
Spe-Spa 88.51±3.47 97.88±0.29 98.42±0.52 LBP 76.73±3.58 83.28±0.40 87.56±0.58 LBP-Spe-Spa 80.61±0.89 85.01±0.33 90.31±0.55 SpaFD3✕3 90.77±1.09 98.00±0.25 98.56±0.36 SpaFD-Spe-Spa3✕3 91.46±0.64 98.26±0.36 98.88±0.26 SpaFD5✕5 90.92±0.61 97.96±0.30 98.67±0.43 SpaFD-Spe-Spa5✕5 91.86 ±0.73 98.27 ±0.45 98.93±0.25 SpaFD7✕7 90.90±1.49 98.04±0.52 98.69±0.45 SpaFD-Spe-Spa7✕7 91.75±0.58 98.25±0.43 99.01 ±0.37
Table 11. Classification results of Pavia University under 5% training samples
View tableView in ArticleTable 11. Classification results of Pavia University under 5% training samples
Model 3DCNN
AOA（%）±SD（%）
3DCNNPCA
AOA（%）±SD（%）
HybridSN
AOA（%）±SD（%）
Spe-Spa 92.71±0.91 98.47±0.25 99.08±0.15 LBP 83.32±0.95 87.12±0.29 91.71±0.33 LBP-Spe-Spa 86.41±0.63 89.90±0.31 92.89±0.16 SpaFD3✕3 93.46±0.49 98.89±0.12 99.14±0.20 SpaFD-Spe-Spa3✕3 93.83±0.35 98.97±0.30 99.35±0.13 SpaFD5✕5 93.06±0.84 98.49±0.24 99.17±0.21 SpaFD-Spe-Spa5✕5 93.90 ±0.42 98.79±0.22 99.44±0.10 SpaFD7✕7 92.88±1.20 98.75±0.28 99.18±0.20 SpaFD-Spe-Spa7✕7 93.19±1.01 99.06 ±0.10 99.46 ±0.11
Table 12. Classification results of Pavia University under 10% training samples
View tableView in ArticleTable 12. Classification results of Pavia University under 10% training samples
Model 3DCNN
AOA（%）±SD（%）
3DCNNPCA
AOA（%）±SD（%）
HybridSN
AOA（%）±SD（%）
Spe-Spa 94.82±0.56 99.45±0.11 99.70±0.08 LBP 86.51±0.54 90.46±0.15 93.01±0.11 LBP-Spe-Spa 89.44±0.42 91.68±0.10 94.97±0.08 SpaFD3✕3 95.06±0.25 99.54±0.11 99.73±0.10 SpaFD-Spe-Spa3✕3 95.32 ±0.33 99.60 ±0.12 99.79±0.07 SpaFD5✕5 94.91±1.01 99.49±0.08 99.74±0.08 SpaFD-Spe-Spa5✕5 95.19±0.45 99.56±0.04 99.81 ±0.11 SpaFD7✕7 94.81±0.28 99.50±0.19 99.73±0.07 SpaFD-Spe-Spa7✕7 94.97±0.54 99.50±0.05 99.80±0.06
Table 13. Classification results of Salinas dataset under 3% training samples
View tableView in ArticleTable 13. Classification results of Salinas dataset under 3% training samples
Model 3DCNN
AOA（%）±SD（%）
3DCNNPCA
AOA（%）±SD（%）
HybridSN
AOA（%）±SD（%）
Spe-Spa 89.86±1.63 97.17±0.48 99.71±0.18 LBP 81.28±1.55 88.32±0.50 90.12±0.15 LBP-Spe-Spa 85.96±1.46 89.85±0.41 91.35±0.10 SpaFD3✕3 91.83±1.67 97.95±0.45 99.73±0.20 SpaFD-Spe-Spa3✕3 94.11 ±1.64 98.48 ±0.45 99.74±0.09 SpaFD5✕5 90.25±1.58 97.80±0.32 99.75±0.14 SpaFD-Spe-Spa5✕5 92.66±1.54 98.32±0.36 99.79 ±0.08 SpaFD7✕7 90.75±1.91 97.24±0.25 99.75±0.18 SpaFD-Spe-Spa7✕7 93.34±1.61 97.99±0.22 99.77±0.06
Table 14. Classification results of the Salinas dataset under 5% training samples
View tableView in ArticleTable 14. Classification results of the Salinas dataset under 5% training samples
Model 3DCNN
AOA（%）±SD（%）
3DCNNPCA
AOA（%）±SD（%）
HybridSN
AOA（%）±SD（%）
Spe-Spa 90.81±1.20 97.96±0.30 99.80±0.09 LBP 82.86±1.43 89.93±0.35 91.84±0.10 LBP-Spe-Spa 86.75±0.88 90.72±0.24 93.44±0.07 SpaFD3✕3 93.17±1.56 98.45±0.32 99.83±0.09 SpaFD-Spe-Spa3✕3 95.99 ±0.48 98.95±0.26 99.88±0.07 SpaFD5✕5 91.40±2.03 98.60±0.35 99.85±0.11 SpaFD-Spe-Spa5✕5 95.34±0.94 99.02 ±0.22 99.89 ±0.07 SpaFD7✕7 92.48±0.97 98.58±0.38 99.83±0.14 SpaFD-Spe-Spa7✕7 95.15±0.89 99.00±0.28 99.87±0.10
Table 15. Classification results of Salinas dataset under 10% training samples
View tableView in ArticleTable 15. Classification results of Salinas dataset under 10% training samples
Model 3DCNN
AOA（%）±SD（%）
3DCNNPCA
AOA（%）±SD（%）
HybridSN
AOA（%）±SD（%）
Spe-Spa 92.55±0.97 99.60±0.12 99.95±0.04 LBP 85.27±1.02 94.43±0.24 95.21±0.09 LBP-Spe-Spa 89.46±0.50 95.01±0.14 96.63±0.05 SpaFD3✕3 95.42±1.01 99.64±0.25 99.95±0.09 SpaFD-Spe-Spa3✕3 96.79 ±0.28 99.71 ±0.15 99.97±0.05 SpaFD5✕5 93.76±1.11 99.61±0.27 99.96±0.03 SpaFD-Spe-Spa5✕5 96.61±0.46 99.65±0.17 99.97 ±0.04 SpaFD7✕7 93.82±2.40 99.60±0.36 99.94±0.10 SpaFD-Spe-Spa7✕7 96.72±0.40 99.62±0.20 99.96±0.06
Tools
Get Citation
Copy Citation Text
Jing LIU, Yang LI, Yi LIU. Hyperspectral images feature extraction and classification based on fractional differentiation[J]. Optics and Precision Engineering, 2023, 31(21): 3221
Paper Information
Category:
Received: May. 8, 2023
Accepted: --
Published Online: Jan. 5, 2024
The Author Email: LIU Jing (zyhalj1975@163.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20