Acta Photonica Sinica, Volume. 52, Issue 11, 1110002(2023)
Multi-scale Remote Sensing Image Classification Based on Weighted Feature Fusion
Figures & Tables(14)
Fig. 4. The network structure diagram of the three branches in the training phase
Fig. 5. Classification results for the IP dataset using 10% of the available labeled data
Fig. 6. Classification results for the UP dataset using 10% of the available labeled data
Fig. 7. Classification results for the SV dataset using 10% of the available labeled data
Fig. 8. The relationship between the proportion of the available labeled data and classification results
Fig. 9. Airborne hyperspectral dataset and its classification result map
Table 1. Sample distribution of hyperspectral remote sensing dataset
View tableView in ArticleTable 1. Sample distribution of hyperspectral remote sensing dataset
Indian Pines Pavia University Salinas Class Samples Class Samples Class Samples Asphalt 46 Asphalt 6 631 Brocoli-green-weeds-1 2 009 Corn-notill 1 428 Meadows 18 649 Brocoli-green-weeds-2 3 726 Corn-mintill 830 Gravel 2 099 Fallow 1 976 Corn 237 Trees-painted 3 064 Fallow-rough-plow 1 394 Grass-pasture 483 Metal-sheets 1 345 Fallow-smooth 2 678 Grass-trees 730 Bare soil 5 029 Stubble 3 959 Grass-pasture-mowed 28 Bitumen 1 330 Celery 3 579 Hay-windrowed 478 Blocking-bricks 3 682 Grapes-untrained 11 271 Oats 20 Shadows 947 Soil-vinyard-develop 6 203 Soybean-notill 972 Corn-senesced-green-weeds 3 278 Soybean-mintill 2 455 Lettuce-romaine-4wk 1 068 Soybean-clean 593 Lettuce-romaine-5wk 1 927 Wheat 205 Lettuce-romaine-6wk 916 Woods 1 265 Lettuce-romaine-7wk 1 070 Building-grass-Trees-drives 386 Vinyard-untrained 7 268 Stone-steel-towers 93 Vinyard-vertical-trellis 1 807 Total 10 249 Total 42 776 Total 54 129
Table 2. Classification results for the IP dataset using 10% of the available labeled data
View tableView in ArticleTable 2. Classification results for the IP dataset using 10% of the available labeled data
Class SVM/% LSTM/% GRU/% CDCNN/% DBDA/% DBMA/% Ours/% Asphalt 73.17 46.34 51.21 51.75 60.00 93.82 95.37 Corn-notill 78.37 81.78 78.52 77.64 95.10 97.03 97.66 Corn-mintill 70.55 71.35 61.98 77.13 96.57 97.11 99.95 Corn 72.30 61.03 72.30 78.92 98.98 98.58 99.23 Grass-pasture 90.80 81.14 88.04 91.86 96.83 97.03 94.93 Grass-trees 90.11 95.89 89.64 95.45 97.60 98.66 99.26 Grass-pasture-mowed 68.00 80.00 68.00 54.20 58.82 80.01 97.40 Hay-windrowed 97.67 99.53 97.67 90.46 96.71 99.90 100.00 Oats 61.11 77.77 50.00 63.27 40.00 81.54 97.22 Soybean-notill 71.31 75.88 73.82 81.34 88.24 95.30 98.37 Soybean-mintill 83.30 74.78 78.81 79.97 96.66 94.06 98.73 Soybean-clean 73.78 30.14 80.71 64.98 90.72 95.49 98.56 Wheat 95.68 97.83 98.37 97.89 100.00 98.46 97.89 Woods 95.43 96.13 94.29 93.99 96.72 97.36 99.73 Building-grass-trees-drivers 63.11 45.53 58.50 83.31 96.56 93.64 98.17 Stone-steel-towers 88.10 88.09 91.66 92.39 93.23 96.89 95.89 OA 82.05 77.65 80.43 81.35 95.07 96.02 98.60 AA 79.55 75.20 77.10 79.66 87.67 94.68 98.02 Kappa 79.47 74.51 77.66 78.67 94.38 95.55 98.41
Table 3. Classification results for the UP dataset using 10% of the available labeled data
View tableView in ArticleTable 3. Classification results for the UP dataset using 10% of the available labeled data
Class SVM/% LSTM/% GRU/% CDCNN/% DBDA/% DBMA/% Ours/% Asphalt 93.37 93.12 92.69 95.48 99.04 99.60 99.91 Meadows 97.81 97.54 97.11 98.77 99.86 99.86 100.00 Gravel 78.00 75.91 79.17 95.73 96.19 99.45 99.08 Trees-painted 95.32 92.93 95.30 98.85 99.01 99.04 99.28 Metal-sheets 99.42 99.63 99.75 99.83 99.93 99.61 99.88 Bare soil 90.37 86.86 88.17 95.25 99.93 99.90 100.00 Bitumen 87.97 83.96 87.05 93.30 100.00 100.00 99.83 Blocking-bricks 88.61 87.24 85.95 89.37 98.27 97.64 99.66 Shadows 99.94 99.88 99.71 99.07 99.08 99.42 99.53 OA 94.09 93.01 93.23 96.57 99.33 99.53 99.83 AA 92.31 90.79 91.65 96.18 99.03 99.39 99.68 Kappa 92.16 90.71 91.02 95.46 99.11 99.38 99.78
Table 4. Classification results for the SV dataset using 10% of the available labeled data
View tableView in ArticleTable 4. Classification results for the SV dataset using 10% of the available labeled data
Class SVM/% LSTM/% GRU/% CDCNN/% DBDA/% DBMA/% Ours/% Brocoli_g_w_1 99.56 99.69 99.86 79.94 100.00 99.23 100.00 Brocoli_g_w_2 99.94 99.63 99.88 92.88 99.98 99.99 100.00 Fallow 99.49 99.69 99.71 98.60 99.92 98.52 100.00 Fallow_r_p 99.52 99.56 99.60 98.74 99.35 98.84 99.92 Fallow_s 99.59 98.28 99.04 89.00 99.44 99.55 99.96 Stubble 99.92 99.84 99.89 99.87 99.92 100.00 100.00 Celery 99.66 99.78 99.76 99.72 99.92 99.94 100.00 Grapes_u 90.20 87.56 87.11 85.78 96.09 99.80 99.95 Soil_v_d 99.91 99.02 99.67 99.79 99.94 99.84 100.00 Corn_s_g_w 97.69 95.74 97.67 97.62 99.80 98.84 99.97 Lettuce_r_4wk 99.06 98.02 98.59 93.85 100.00 99.19 100.00 Lettuce_r_5wk 99.77 96.86 99.42 99.45 99.97 99.73 100.00 Lettuce_r_6wk 99.39 98.24 99.21 98.79 100.00 99.27 100.00 Lettuce_r_7wk 97.92 96.62 97.82 97.76 99.77 99.74 99.83 Vinyard_u 71.24 82.15 82.55 83.56 97.86 94.59 99.96 Vinyard_v_t 99.20 98.83 98.86 93.05 86.77 99.96 99.80 OA 93.74 94.18 94.55 91.99 97.45 98.89 99.97 AA 97.00 96.85 97.42 94.27 98.67 99.18 99.96 Kappa 93.02 93.53 93.93 91.09 97.16 98.77 99.96
Table 5. The overall classification accuracy of each algorithm when using 2% labeled samples for training
View tableView in ArticleTable 5. The overall classification accuracy of each algorithm when using 2% labeled samples for training
Algorithm IP/% UP/% SV/% DBDA 87.29 97.81 96.28 DBMA 78.35 97.85 98.14 OURS 89.68 99.29 99.86
Tools
Get Citation
Copy Citation Text
Yinzhu CHENG, Song LIU, Nan WANG, Yuetian SHI, Geng ZHANG. Multi-scale Remote Sensing Image Classification Based on Weighted Feature Fusion[J]. Acta Photonica Sinica, 2023, 52(11): 1110002
Paper Information
Category:
Received: Apr. 4, 2023
Accepted: May. 22, 2023
Published Online: Dec. 22, 2023
The Author Email: ZHANG Geng (gzhang@opt.ac.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20