Laser & Optoelectronics Progress, Volume. 58, Issue 8, 0810010(2021)
Hyperspectral Images Classification Based on Multi-Feature Fusion and Hybrid Convolutional Neural Networks
Figures & Tables(13)
Fig. 1. Comparison of R-HybridSN and M-HybridSN modules. (a) Multi-scale convolutional layer of the first layer of R-HybridSN; (b) non-identical residual connection of the R-HybridSN; (c) multi-feature fusion module of the M-HybridSN
Fig. 6. Comparative experiment results under different conditions of the non-identical residual connection
Table 1. Distribution situation of the data set IP
View tableTable 1. Distribution situation of the data set IP
No. Category Labeled sample Training Validation Testing 1 alfalfa 46 2 3 41 2 corn-notill 1428 71 72 1285 3 corn-mintill 830 42 41 747 4 corn 237 12 12 213 5 grass-pasture 483 24 24 435 6 grass-trees 730 36 37 657 7 grass-pasture-mowed 28 2 1 25 8 hay-windrowed 478 24 24 430 9 oats 20 1 1 18 10 soybean-notill 972 48 49 875 11 soybean-mintill 2455 123 122 2210 12 soybean-clean 593 30 29 534 13 wheat 205 10 10 185 14 woods 1265 63 63 1139 15 buildings-grass-trees-drives 386 19 20 347 16 stone-steel-towers 93 5 4 84 Total 10249 512 512 9225
Table 2. Distribution situation of the data set SA
View tableTable 2. Distribution situation of the data set SA
No. Category Labeled sample Training Validation Testing 1 brocoli_green_weeds_1 2009 20 20 1969 2 brocoli_green_weeds_2 3726 37 37 3652 3 fallow 1976 20 20 1936 4 fallow_rough_plow 1394 14 14 1366 5 fallow_smooth 2678 27 27 2624 6 stubble 3959 39 40 3880 7 celery 3579 36 36 3507 8 grapes_untrained 11271 113 112 11046 9 soil_vinyard_develop 6203 62 62 6079 10 corn_senesced_green_weeds 3278 33 33 3212 11 lettuce_romaine_4wk 1068 11 10 1047 12 lettuce_romaine_5wk 1927 19 20 1888 13 lettuce_romaine_6wk 916 9 9 898 14 lettuce_romaine_7wk 1070 11 10 1049 15 vinyard_untrained 7268 72 73 7123 16 vinyard_vertical_trellis 1807 18 18 1771 Total 54129 541 541 53047
Table 3. Distribution situation of the data set PU
View tableTable 3. Distribution situation of the data set PU
No. Category Labeled sample Training Validation Testing 1 asphalt 6631 66 66 6499 2 meadows 18649 186 186 18277 3 gravel 2099 21 21 2057 4 trees 3064 30 31 3003 5 painted metal sheets 1345 14 13 1318 6 bare Soil 5029 50 50 4929 7 bitumen 1330 14 13 1303 8 self-blocking bricks 3682 37 37 3608 9 shadows 947 9 10 928 Total 42776 427 427 41922
Table 4. Parameter number and input data scale of different models
View tableTable 4. Parameter number and input data scale of different models
Model Res-2D-CNN Res-3D-CNN HybridSN R-HybridSN M-HybridSN Parameter number 1065360 231184 5122176 719112 659296 Input data scale 5×5×200 9×9×200 25×25×30 15×15×16 15×15×16
Table 5. Classification results of the data set IP by different models unit: %
View tableTable 5. Classification results of the data set IP by different models unit: %
No. Res-2D-CNN Res-3D-CNN HybridSN R-HybridSN M-HybridSN 1 9.51 23.78 58.54 58.17 65.61 2 72.39 83.73 93.08 94.98 95.28 3 60.31 76.53 96.57 97.38 97.36 4 37.86 53.47 75.09 92.16 94.51 5 80.14 93.54 94.00 96.68 97.01 6 94.00 96.54 97.19 99.08 98.63 7 34.60 71.20 82.40 94.00 99.80 8 99.13 98.66 98.73 99.81 99.93 No. Res-2D-CNN Res-3D-CNN HybridSN R-HybridSN M-HybridSN 9 3.89 67.50 83.89 63.06 76.67 10 78.42 85.75 94.27 95.81 96.58 11 84.12 90.02 97.93 98.31 98.55 12 54.19 63.40 84.49 92.43 91.97 13 85.16 88.43 92.68 98.46 97.41 14 89.44 97.48 97.96 99.25 99.03 15 52.98 79.35 83.18 92.52 96.80 16 80.54 93.63 83.33 98.21 95.54 Kappa 74.0 ± 2.8 84.5 ± 2.4 93.4 ± 1.2 96.3 ± 0.6 96.7 ± 0.4 OA 77.28 ± 2.33 86.42 ± 2.13 94.26 ± 1.08 96.74 ± 0.52 97.09 ± 0.38 AA 63.54 ± 4.66 78.94 ± 3.22 88.33 ± 2.40 91.90 ± 2.58 93.79 ± 1.99
Table 6. Classification results of the data set SA by different models unit: %
View tableTable 6. Classification results of the data set SA by different models unit: %
No. Res-2D-CNN Res-3D-CNN HybridSN R-HybridSN M-HybridSN 1 66.09 97.13 99.98 100.00 99.92 2 99.36 99.92 99.97 99.96 99.99 3 61.79 93.00 99.82 99.62 99.56 4 99.19 99.09 97.39 98.87 99.22 5 94.62 97.75 98.79 98.83 99.21 6 99.95 99.97 99.78 99.90 99.91 7 97.34 98.24 99.77 99.88 99.91 8 82.99 87.66 99.04 98.33 98.96 9 99.19 99.58 100.00 99.99 99.96 10 85.81 91.16 98.98 98.06 98.89 11 83.73 90.83 98.95 98.62 98.83 12 98.32 99.20 99.09 99.88 99.29 13 95.23 97.88 97.28 92.41 96.99 14 96.07 98.25 96.60 93.96 97.17 15 70.49 77.52 98.57 96.61 98.90 16 91.08 86.44 99.69 99.46 99.56 Kappa 86.1 ± 1.6 91.6 ± 0.8 99.1 ± 0.3 98.5 ± 0.3 99.2 ± 0.3 OA 87.54 ± 1.40 92.48 ± 0.69 99.20 ± 0.27 98.66 ± 0.31 99.30 ± 0.24 AA 88.83 ± 2.64 94.60 ± 0.50 98.98 ± 0.28 98.40 ± 0.43 99.14 ± 0.30
Table 7. Classification results of the data set PA by different models unit: %
View tableTable 7. Classification results of the data set PA by different models unit: %
No. Res-2D-CNN Res-3D-CNN HybridSN R-HybridSN M-HybridSN 1 92.18 90.81 92.15 96.21 95.04 2 97.47 96.63 99.53 99.70 99.88 3 15.33 66.54 90.51 90.93 93.77 4 94.94 96.24 92.50 94.62 92.92 5 99.45 99.86 97.75 99.79 99.57 6 88.04 80.75 99.46 99.25 99.50 7 40.70 68.12 96.25 94.36 94.92 8 86.93 80.01 91.75 94.09 95.68 9 97.40 97.38 75.04 94.23 92.85 Kappa 85.0 ± 1.2 86.9 ± 1.9 94.8 ± 1.3 96.7 ± 0.6 96.8 ± 0.4 OA 88.72 ± 0.85 90.16 ± 1.40 96.07 ± 0.96 97.55 ± 0.48 97.60 ± 0.33 AA 79.16 ± 3.10 86.26 ± 2.10 92.77 ± 2.33 95.91 ± 0.96 96.01 ± 0.60
Tools
Get Citation
Copy Citation Text
Fan Feng, Shuangting Wang, Jin Zhang, Chunyang Wang. Hyperspectral Images Classification Based on Multi-Feature Fusion and Hybrid Convolutional Neural Networks[J]. Laser & Optoelectronics Progress, 2021, 58(8): 0810010
Paper Information
Category: Image Processing
Received: Aug. 5, 2020
Accepted: Sep. 10, 2020
Published Online: Apr. 12, 2021
The Author Email: Chunyang Wang (wcy@hpu.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20