基于多特征和改进自编码的高光谱图像分类 下载: 1090次
Hyperspectral Image Classification Based on Multiple Features and an Improved Autoencoder
1 长安大学理学院, 陕西 西安 710064
2 陕西师范大学计算机科学学院, 陕西 西安 710000
图 & 表
图 1. 自编码网络结构
Fig. 1. Autoencoder network structure
下载图片 查看原文
图 2. SSAE算法模型图
Fig. 2. SSAE algorithm model diagram
下载图片 查看原文
图 3. PCA与LargeVis算法对比图。(a) PCA-Indian pines;(b) PCA-Pavia U;(c) LargeVis-Indian pines;(d) LargeVis-Pavia U
Fig. 3. Comparison of PCA and LargeVis algorithm. (a) PCA-Indian pines; (b) PCA-Pavia U; (c) LargeVis-Indian pines; (d) LargeVis-Pavia U
下载图片 查看原文
图 4. EAP结构示意图
Fig. 4. EAP structure diagram
下载图片 查看原文
图 5. 批量主动学习采样流程图
Fig. 5. Batch-mode active learning sampling strategy flow chart
下载图片 查看原文
图 6. MF-AL-SSAE算法模型图
Fig. 6. MF-AL-SSAE algorithm model diagram
下载图片 查看原文
图 7. Indian pines数据集的6种算法分类效果图。(a)原始图像;(b)实际地物;(c) SSAE算法;(d) SVM算法;(e) CK-SVM算法;(f) CLBP-SSAE算法;(g) EMAP-SSAE算法;(h) MF-AL-SSAE算法
Fig. 7. Classification renderings of six algorithms on the Indian pines dataset. (a) Original image; (b) real ground; (c) SSAE algorithm; (d) SVM algorithm; (e) CK-SVM algorithm; (f) CLBP-SSAE algorithm; (g) EMAP-SSAE algorithm; (h) MF-AL-SSAE algorithm
下载图片 查看原文
图 8. Pavia U数据集的6种算法分类效果图。(a)原始图像;(b)实际地物;(c) SSAE算法;(d) SVM算法;(e) CK-SVM算法;(f) CLBP-SSAE算法;(g) EMAP-SSAE算法;(h) MF-AL-SSAE算法
Fig. 8. Classification renderings of six algorithms on the Pavia U dataset. (a) Original image; (b) real ground; (c) SSAE algorithm; (d) SVM algorithm; (e) CK-SVM algorithm; (f) CLBP-SSAE algorithm; (g) EMAP-SSAE algorithm; (h) MF-AL-SSAE algorithm
下载图片 查看原文
图 9. 不同数据集的OA随训练样本个数的变化。(a) Indian pines;(b) Pavia U
Fig. 9. Variation in OA of different datasets with the number of training samples. (a) Indian pines; (b) Pavia U
下载图片 查看原文
表 1Indian pines数据集的实验数据和分类精度
Table1. Experimental data and classification accuracies of the Indian pines dataset
Class | Sample | Classification accuracy /% |
---|
Training | Test | SSAE | SVM | CK-SVM | CLBP-SSAE | EMAP-SSAE | MF-AL-SSAE |
---|
Alfalfa | 5 | 41 | 53.42 | 57.32 | 93.91 | 92.12 | 94.26 | 96.88 | Corn-notill | 143 | 1285 | 76.53 | 78.98 | 95.49 | 96.31 | 94.68 | 98.38 | Corn-mintill | 83 | 747 | 46.17 | 67.67 | 95.87 | 97.48 | 96.50 | 98.66 | Corn | 23 | 214 | 52.33 | 51.62 | 94.38 | 96.65 | 96.22 | 96.92 | Grass-pasture | 50 | 433 | 83.65 | 85.21 | 94.27 | 94.25 | 95.98 | 96.65 | Grass-trees | 75 | 655 | 92.19 | 93.83 | 97.65 | 97.21 | 96.71 | 98.12 | Grass-pasture-mowed | 3 | 25 | 81.85 | 80.21 | 98.80 | 96.73 | 96.05 | 97.64 | Hay-windrowed | 49 | 429 | 93.58 | 94.68 | 98.95 | 97.24 | 96.58 | 97.93 | Oats | 2 | 18 | 42.78 | 37.78 | 67.80 | 73.31 | 76.63 | 94.25 | Soybean-notill | 97 | 875 | 67.49 | 69.71 | 93.34 | 92.87 | 94.43 | 96.28 | Soybean-mintill | 247 | 2208 | 68.12 | 74.56 | 96.89 | 96.26 | 96.52 | 98.83 | Soybean-clean | 61 | 532 | 37.91 | 64.71 | 95.33 | 96.64 | 94.49 | 97.11 | Wheat | 21 | 184 | 92.76 | 94.32 | 99.89 | 92.87 | 93.32 | 98.87 | Woods | 129 | 1136 | 93.45 | 91.68 | 95.08 | 98.35 | 97.68 | 100.00 | Bidg-grass-trees-drives | 38 | 348 | 31.03 | 54.39 | 93.65 | 95.79 | 94.94 | 97.83 | Stone-steel-towers | 10 | 83 | 90.80 | 86.36 | 97.63 | 95.51 | 94.75 | 97.91 | OA /% | 76.65 | 77.53 | 94.86 | 95.42 | 96.63 | 98.14 | Kappa | 0.74 | 0.75 | 0.94 | 0.95 | 0.96 | 0.97 |
|
查看原文
表 2Pavia U数据集的实验数据和分类精度
Table2. Experimental data and classification accuracies of the Pavia U dataset
Class | Sample | Classification accuracy /% |
---|
Training | Test | SSAE | SVM | CK-SVM | CLBP-SSAE | EMAP-SSAE | MF-AL-SSAE |
---|
Asphalt | 200 | 6431 | 56.86 | 57.32 | 96.91 | 90.54 | 92.56 | 94.88 | Meadows | 200 | 18389 | 77.68 | 78.98 | 96.49 | 92.36 | 95.68 | 96.58 | Grave | 200 | 1899 | 65.17 | 67.67 | 95.87 | 95.48 | 96.50 | 97.66 | Trees | 200 | 2864 | 60.83 | 51.62 | 97.34 | 96.65 | 96.22 | 96.84 | Painted metal sheets | 200 | 1145 | 50.84 | 90.21 | 98.27 | 94.25 | 92.98 | 96.14 | Baresoil | 200 | 4829 | 92.19 | 94.83 | 96.65 | 94.21 | 95.11 | 97.56 | Bitumen | 200 | 1130 | 73.85 | 69.21 | 96.80 | 96.73 | 97.05 | 97.64 | Self-blocking bricks | 200 | 3842 | 94.58 | 96.68 | 95.25 | 94.55 | 94.58 | 93.95 | Shadows | 200 | 747 | 42.78 | 57.78 | 98.37 | 97.31 | 97.63 | 98.45 | OA /% | 76.87 | 78.03 | 97.86 | 95.78 | 95.98 | 97.24 | Kappa | 0.75 | 0.76 | 0.97 | 0.94 | 0.95 | 0.96 |
|
查看原文
张倩, 董安国, 宋睿. 基于多特征和改进自编码的高光谱图像分类[J]. 激光与光电子学进展, 2020, 57(8): 081010. Qian Zhang, Anguo Dong, Rui Song. Hyperspectral Image Classification Based on Multiple Features and an Improved Autoencoder[J]. Laser & Optoelectronics Progress, 2020, 57(8): 081010.