Chinese Journal of Liquid Crystals and Displays, Volume. 39, Issue 6, 844(2024)
Hyperspectral image classification based on multi-branch spatial-spectral feature enhancement
Fig. 2. Structure of the multi-branch spatial feature extraction module
Fig. 3. Structure of the multi-branch spatial-spectral feature extraction module
Fig. 5. Structure of the improved multi-scale feature extraction and fusion module
Fig. 7. Comparison of classification accuracy for different numbers of principal components
Fig. 8. Comparison of classification accuracy for different network structures
Fig. 9. Comparison of the number of parameters for different network structures
Fig. 10. Visualization results chart for different models on the IP dataset
Fig. 11. Visualization results chart for different models on the PU dataset
Fig. 12. Visualization results chart for different models on the SA dataset
Table 1. Classification accuracies of the model with different learning rates
View tableView in ArticleTable 1. Classification accuracies of the model with different learning rates
学习率 Indian Pines Pavia University Salinas OA AA Kappa OA AA Kappa OA AA Kappa 0.003 98.68 98.48 98.49 99.24 98.45 98.99 99.07 99.06 98.96 0.001 99.07 98.85 98.94 99.75 99.75 99.67 99.47 99.58 99.41 0.000 5 98.95 98.42 98.81 99.64 99.29 99.52 99.02 99.28 98.91 0.000 3 98.45 98.42 98.23 99.54 99.27 99.39 99.01 99.21 98.91 0.000 1 96.88 94.76 96.44 98.65 97.89 98.21 98.01 98.52 97.77
Table 2. Comparison results of different models on the IP dataset
View tableView in ArticleTable 2. Comparison results of different models on the IP dataset
Class 2D - CNNM3D - CNNRes - 3D- CNNHybridSN MSPN MDAN 3DA - FCNNSSFE - MBACNNAlfalfa 59.46 86.49 67.56 97.30 97.30 89.19 97.30 94.60 Corn-n 89.37 94.30 86.60 96.72 96.02 97.93 96.54 98.19 Corn-min 81.54 87.35 83.18 97.62 95.09 95.83 96.43 99.70 Corn 64.58 95.31 100 96.88 92.71 96.88 97.92 100 Grass-p 97.70 93.88 88.78 95.66 95.41 98.21 99.49 99.23 Grass-t 98.81 99.32 92.72 99.66 98.82 99.32 98.48 99.66 Grass-p-m 95.45 86.36 68.18 100 100 100 100 100 Hay-w 98.44 99.74 99.74 100 100 100 100 100 Oats 25.00 75.00 75.00 93.75 100 87.50 93.75 100 Soybean-n 83.25 84.77 90.36 95.30 98.22 96.20 96.45 97.84 Soybean-m 87.03 95.27 97.13 98.39 99.40 99.45 99.65 99.25 Soybean-c 91.48 96.88 98.75 98.34 98.75 99.58 98.75 97.71 Wheat 99.40 96.99 100 96.99 100 100 95.78 99.40 Woods 99.22 99.22 99.90 99.32 99.81 99.22 99.42 100 Buildings-g-t 80.13 92.63 100 100 99.36 99.68 100 100 Stone-s-t 98.68 82.90 89.47 94.74 96.05 97.37 93.42 96.05 OA 89.55 94.23 93.65 97.90 98.10 98.45 98.32 99.07 AA 84.35 91.65 89.83 97.54 97.93 98.23 98.08 98.94 Kappa 88.08 93.40 92.75 97.61 97.83 97.27 97.71 98.85
Table 3. Comparison results of different models on the PU dataset
View tableView in ArticleTable 3. Comparison results of different models on the PU dataset
Class 2D-CNN M3D-CNN Res-3D-CNN HybridSN MSPN MDAN 3DA-FCNN SSFE-MBACNN Asphalt 98.40 99.10 98.66 99.65 99.58 98.85 99.32 99.55 Meadows 98.75 99.51 99.60 100 99.87 99.99 100 99.99 Gravel 86.27 87.28 91.45 99.58 98.89 99.58 99.68 99.63 Trees 96.92 96.96 96.93 95.44 97.94 99.02 96.85 98.73 Painted metal sheets 100 100 99.59 100 100 99.84 100 99.75 Bare soil 95.79 99.03 98.11 100 100 100 100 100 Bitumen 94.91 90.75 94.50 98.92 99.92 97.58 100 100 Self -Blocking Bricks 94.67 94.88 85.50 98.80 99.58 98.07 96.21 99.28 Shadows 98.25 80.23 99.18 94.85 97.78 97.54 94.62 100 OA 97.16 97.52 97.31 99.35 99.59 99.43 99.21 99.75 AA 96.00 94.19 95.95 98.58 99.28 98.94 98.52 99.66 Kappa 96.24 96.71 96.42 99.13 99.45 99.24 98.95 99.67
Table 4. Comparison results of different models on the SA dataset
View tableView in ArticleTable 4. Comparison results of different models on the SA dataset
Class 2D-CNN M3D-CNN Res-3D-CNN HybridSN MSPN MDAN 3DA-FCNN SSFE-MBACNN Broccoli green weeds 1 100 98.73 100 99.39 100 99.90 100 99.85 Broccoli green weeds 2 99.97 100 100 100 100 100 100 100 Fallow 99.07 98.55 100 100 100 100 97.47 100 Fallow rough plow 99.19 90.92 89.39 99.85 99.05 99.85 99.34 99.27 Fallow smooth 97.90 97.22 98.74 99.54 96.72 98.67 95.28 98.25 Stubble 100 100 100 99.79 100 99.92 100 99.92 Celery 99.91 99.97 99.54 99.40 100 100 97.69 99.97 Grapes untrained 89.42 87.27 93.60 98.85 98.43 99.94 99.25 99.15 Soy vineyard develop 100 99.74 99.82 100 100 99.84 100 99.93 Corn senesced green-w 93.22 97.85 98.35 98.94 97.70 99.81 99.88 98.10 Lettuce romaine 4wk 88.53 96.56 99.05 88.91 100 98.76 99.14 100 Lettuce romaine 5wk 100 94.81 99.68 100 100 97.30 98.89 99.79 Lettuce romaine 6wk 100 86.19 68.60 100 99.78 100 100 100 Lettuce romaine 7wk 96.09 96.09 98.85 99.42 100 99.14 99.33 100 Vineyard untrained 75.58 84.04 90.20 97.03 95.96 95.37 98.51 99.30 Vineyard vertical trellis 96.50 96.95 98.19 100 99.10 100 100 99.66 OA 93.52 93.91 96.22 98.97 98.77 99.12 99.07 99.47 AA 95.96 95.31 95.88 98.82 99.17 99.28 99.05 99.57 Kappa 92.78 93.22 95.79 98.84 98.63 99.02 98.97 99.41
Table 5. Classification performance of SSFE-MBACNN with different training ratios on the IP dataset
View tableView in ArticleTable 5. Classification performance of SSFE-MBACNN with different training ratios on the IP dataset
Rate OA AA Kappa 3 92.39 92.34 91.35 5 96.48 96.67 95.98 7 97.63 96.33 97.26 9 98.70 98.75 98.51 10 99.07 98.85 98.94
Table 6. Classification performance of SSFE-MBACNN with different training ratios on the PU dataset
View tableView in ArticleTable 6. Classification performance of SSFE-MBACNN with different training ratios on the PU dataset
Rate OA AA Kappa 1 98.32 96.76 97.78 2 98.89 97.80 98.53 3 99.27 98.69 99.03 4 99.66 99.23 99.55 5 99.75 99.66 99.67
Table 7. Classification performance of SSFE-MBACNN with different training ratios on the SA dataset
View tableView in ArticleTable 7. Classification performance of SSFE-MBACNN with different training ratios on the SA dataset
Rate OA AA Kappa 1.0 99.47 99.57 99.41 1.5 99.55 99.57 99.51 2.0 99.68 99.70 99.64 2.5 99.79 99.83 99.76 3.0 99.86 99.84 99.83
Get Citation
Copy Citation Text
Tie LI, Wenxu LI, Junguo WANG, Qiaoyu GAO. Hyperspectral image classification based on multi-branch spatial-spectral feature enhancement[J]. Chinese Journal of Liquid Crystals and Displays, 2024, 39(6): 844
Category: Research Articles
Received: Apr. 28, 2023
Accepted: --
Published Online: Jul. 30, 2024
The Author Email: Wenxu LI (q1024099536@163.com)
