Author Affiliations
1Jilin Key Laboratory of Satellite Remote Sensing Application Technology, Chang Guang Satellite Technology Co., Ltd., Changchun, Jilin 130012, China2Changchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences, Changchun, Jilin 130033, China3Jilin Institute of Land Survey & Planning, Changchun, Jilin 130061, China;show less
Fig. 1. Flow chart of MPCNet classification algorithm
Fig. 2. Multispectral remote sensing image classification model based on MPCNet
Fig. 3. Introduction of experimental data. (a) True color image of Jilin-1GP01; (b) true color image of Landsat8; (c) true color image of Sentinel-2A; (d) true color image of HJ-1A; (e) high resolution image of Google Earth; (f) land cover product of FROM-GLC-2017
Fig. 4. Reflectivity curves of samples in Nashik research area of Jilin-1GP01. (a) Uncultivated land; (b) cultivated land; (c) building; (d) grassland; (e) bare land; (f) road; (g) forest land; (h) water; (i) cloud; (j) shadow
Fig. 5. Classification results using MPCNet algorithm. (a) Jilin-1GP01 image; (b) Landsat8 image; (c) Sentinel-2A image; (d) HJ-1A image
Fig. 6. Classification detail results of surface features at local Nashik area. (a) Pseudo-color composite image; (b) local enlargement image; (c) classification result by Jilin-1GP01; (d) classification result by Sentinel-2A; (e) remote sensing image superimposed vector data
Fig. 7. Classification results of different algorithms on Jilin-1GP01 images in Nashik. (a) Local image; (b) SVM; (c) LGBM-GBDT; (d) shallow CNN; (e) MPCNet
Fig. 8. Classification results of different algorithms on Jilin-1GP02 images in Xintai city. (a) Local image; (b) SVM; (c) LGBM-GBDT; (d) shallow CNN; (e) MPCNet
Satellite type | Available/selective band number | Wavelength range /nm | Spatial resolution /m | Shooting time |
---|
Jilin-1GP01 | 26/10 | 400-13500 | 5 | 2019-01-22 | Landsat8 | 11/11 | 430-12510 | 30 | 2019-01-13 | Sentinel-2A | 13/13 | 443-2190 | 10 | 2019-01-20 | HJ-1A | 4/4 | 430-900 | 30 | 2019-01-14 |
|
Table 1. Band selection, spatial resolution, and shooting time of multispectral satellite
Satellite type | Sensor | Scene ID | Available/selective band number | Wavelength range /nm | Spatial resolution /m | Shooting time |
---|
Jilin-1GP02 | PMS1PMS1PMS2PMS2 | 0001000200010002 | 26/10 | 400-13500 | 5 | 2019-03-15 |
|
Table 2. Band selection, spatial resolution, and shooting time of Jilin-1GP02
Area type | Jilin-1GP01 | Landsat8 | Sentinel-2A | HJ-1A | | | |
---|
P/R | F1 | P/R | F1 | P/R | F1 | P/R | F1 |
---|
Uncultivated land | 1.00/0.93 | 0.96 | 0.91/0.83 | 0.87 | 0.96/0.89 | 0.92 | 0.76/0.74 | 0.75 | Cultivated land | 0.94/0.89 | 0.91 | 0.81/0.88 | 0.84 | 0.83/0.98 | 0.90 | 0.52/0.70 | 0.60 | Building | 0.96/0.99 | 0.97 | 0.94/0.79 | 0.86 | 0.97/0.94 | 0.96 | 0.85/0.61 | 0.71 | Grassland | 1.00/0.88 | 0.94 | 1.00/0.73 | 0.85 | 1.00/0.82 | 0.90 | 0.92/0.69 | 0.79 | Bare land | 0.96/1.00 | 0.98 | 0.96/0.93 | 0.94 | 0.94/0.99 | 0.96 | 0.75/0.86 | 0.80 | Road | 0.95/0.80 | 0.87 | 0.43/0.80 | 0.56 | 0.85/0.83 | 0.84 | 0.14/0.23 | 0.18 | Forest land | 0.71/1.00 | 0.83 | 0.62/1.00 | 0.76 | 0.69/1.00 | 0.82 | 0.27/0.68 | 0.38 | Water | 1.00/1.00 | 1.00 | 1.00/0.97 | 0.98 | 1.00/0.93 | 0.96 | 0.91/0.60 | 0.72 | K | 0.948 | 0.828 | 0.920 | 0.595 | POA | 0.958 | 0.859 | 0.935 | 0.667 |
|
Table 3. Classification accuracy evaluation index of different satellite data based on MPCNet algorithm
Area type | SVM | LGBM-GBDT | Shallow CNN | MPCNet | | | |
---|
P/R | F1 | P/R | F1 | P/R | F1 | P/R | F1 |
---|
Uncultivated land | 0.91/0.70 | 0.78 | 0.90/0.83 | 0.86 | 0.91/0.85 | 0.88 | 1.00/0.93 | 0.96 | Cultivated land | 0.90/0.89 | 0.89 | 0.91/0.90 | 0.91 | 0.98/0.90 | 0.94 | 0.94/0.89 | 0.91 | Building | 0.86/1.00 | 0.92 | 0.93/0.99 | 0.96 | 0.91/1.00 | 0.95 | 0.96/0.99 | 0.97 | Grassland | 0.99/0.79 | 0.88 | 1.00/0.69 | 0.82 | 1.00/0.89 | 0.94 | 1.00/0.88 | 0.94 | Bare land | 0.91/0.99 | 0.95 | 0.90/0.99 | 0.94 | 0.92/1.00 | 0.96 | 0.96/1.00 | 0.98 | Road | 0.64/0.28 | 0.39 | 0.88/0.62 | 0.73 | 0.76/0.40 | 0.53 | 0.95/0.80 | 0.87 | Forest land | 0.60/0.85 | 0.71 | 0.65/1.00 | 0.79 | 0.67/1.00 | 0.80 | 0.71/1.00 | 0.83 | Water | 1.00/0.96 | 0.98 | 1.00/0.97 | 0.99 | 1.00/0.96 | 0.98 | 1.00/1.00 | 1.00 | K | 0.857 | 0.899 | 0.901 | 0.948 | POA | 0.886 | 0.919 | 0.921 | 0.958 |
|
Table 4. Classification accuracy evaluation index of different algorithms on Jilin-1GP01 images
Area type | SVM | LGBM-GBDT | Shallow CNN | MPCNet | | | |
---|
P/R | F1 | P/R | F1 | P/R | F1 | P/R | F1 |
---|
Uncultivated land | 0.56/0.66 | 0.61 | 0.99/0.92 | 0.95 | 0.96/0.99 | 0.98 | 0.99/0.98 | 0.99 | Cultivated land | 0.97/0.98 | 0.97 | 0.71/0.67 | 0.69 | 0.88/0.85 | 0.86 | 0.96/0.94 | 0.95 | Forest land | 0.95/0.87 | 0.90 | 0.85/0.85 | 0.85 | 0.97/0.97 | 0.97 | 0.97/0.98 | 0.97 | Shrub | 0.56/0.85 | 0.67 | 0.44/0.87 | 0.59 | 0.75/0.98 | 0.85 | 0.88/0.97 | 0.92 | Water | 1.00/0.78 | 0.88 | 0.75/0.68 | 0.71 | 0.92/0.78 | 0.84 | 0.95/0.93 | 0.94 | Building | 0.62/0.69 | 0.65 | 0.80/0.76 | 0.78 | 0.83/0.97 | 0.90 | 0.92/0.94 | 0.93 | Bare land | 0.57/0.49 | 0.53 | 0.70/0.70 | 0.70 | 0.85/0.78 | 0.81 | 0.83/0.89 | 0.86 | K | 0.731 | 0.724 | 0.877 | 0.932 | POA | 0.763 | 0.756 | 0.891 | 0.940 |
|
Table 5. Classification accuracy evaluation index of different algorithms on Jilin-1GP02 images
Algorithm | Image size | Image storage /Mbit | Feature extraction time /s | Model training time /min | Inference time /min | Total process time /min |
---|
SVMLGBM-GBDTShallow CNNMPCNet | 5368×4565 | 888.04 | 713.424.338.037.5 | 47.451.3414.9210.80 | 63.427.422.963.56 | 122.769.1718.5114.98 |
|
Table 6. Processing efficiency of different algorithms on Jilin-1GP01 images