Author Affiliations
1School of Science, Chang'an University, Xi'an, Shaanxi 710064, China2School of Information Engineering, Jiangxi University of Science and Technology, Ganzhou, Jiangxi 341000, Chinashow less
Fig. 1. Auto-encoder model
Fig. 2. Stack auto-encoder and classifier
Fig. 3. Hyperspectral remote sensing images. (a) True classification picture; (b) classification result of S-SAE algorithm; (c) spectral curves
Fig. 4. Spatial-spectral feature extraction method based on rotation invariant property
Fig. 5. Hyperspectral neighborhood information. (a) Spatial position; (b) magnified picture; (c) neighborhood information of point E; (d) neighborhood information of point F
Fig. 6. Classification algorithm framework for deep learning combined with spatial-spectral information
Fig. 7. Selection of parameters. (a) Selection of number of principal components; (b) selection of window size
Fig. 8. Classification results of Pavia University dataset obtained by different algorithms. (a) Original image; (b) true classification picture; (c) SVM; (d) CK-SVM; (e) OMP; (f) SOMP; (g) proposed method (unselect); (h) proposed method
Fig. 9. Classification results of Indian Pines dataset obtained by different algorithms. (a) Original image; (b) true classification picture; (c) SVM; (d) CK-SVM; (e) OMP; (f) SOMP; (g) proposed method (unselect); (h) proposed method
Fig. 10. Effect of number of training samples on overall accuracy of different datasets. (a) Pavia University; (b) Indian Pines
Class | Number of samples | Classification accuracy /% |
---|
Train | Test | SVM | CK-SVM | OMP | SOMP | Our method (unselect) | Our method |
---|
Asphalt | 200 | 6431 | 80.50 | 97.90 | 61.20 | 82.11 | 93.11 | 98.10 | Meadows | 200 | 18449 | 84.48 | 98.95 | 79.47 | 95.50 | 96.11 | 97.32 | Gravel | 200 | 1899 | 78.91 | 93.77 | 68.01 | 98.11 | 95.22 | 97.17 | Trees | 200 | 2864 | 96.24 | 98.96 | 91.95 | 96.24 | 93.74 | 99.35 | Painted metal sheets | 200 | 1145 | 99.74 | 100.00 | 99.22 | 99.06 | 100.00 | 100.00 | Bare soil | 200 | 4829 | 83.96 | 97.06 | 69.84 | 98.55 | 96.91 | 99.28 | Bitumen | 200 | 1130 | 91.39 | 99.56 | 84.39 | 98.34 | 98.81 | 99.51 | Self-blocking bricks | 200 | 3482 | 81.27 | 96.45 | 76.52 | 94.90 | 96.39 | 96.43 | Shadows | 200 | 747 | 98.44 | 99.87 | 98.04 | 88.44 | 97.99 | 100.00 | OA /% | | | 84.98 | 98.16 | 76.60 | 93.93 | 95.88 | 97.87 | Kappa | | | 0.80 | 0.98 | 0.70 | 0.92 | 0.94 | 0.97 |
|
Table 1. Experimental data and classification accuracy of the Pavia University dataset
Parameter | Classification algorithm |
---|
SVM | CK-SVM | OMP | SOMP | Our method (unselect) | Our method |
---|
OA /% | 73.01 | 91.36 | 65.87 | 91.46 | 90.18 | 93.99 | Kappa coefficient | 0.69 | 0.90 | 0.61 | 0.90 | 0.89 | 0.93 |
|
Table 2. OA and Kappa coefficient of the Indian Pines dataset obtained by different algorithms