Fig. 1. Model architecture of CLCT-Net
Fig. 2. Schematic diagram of shared feature extraction network
Fig. 3. Schematic diagram of HSI encoder
Fig. 4. Schematic diagram of LiDAR encoder
Fig. 5. Pseudo color map and ground-truth map of Houston2013 dataset
Fig. 6. Pseudo color map and ground-truth map of Trento dataset
Fig. 7. Feature visualizations of Houston2013 dataset
Fig. 8. Feature visualizations of Trento dataset
Fig. 9. Classification results of different methods on Houston2013 dataset
Fig. 10. Classification results of different methods on Trento dataset
Class name | Train num | Test num | Color |
---|
Healthy grass | 198 | 1 053 | | Stressed grass | 190 | 1 064 | | Synthetic grass | 192 | 505 | | Trees | 188 | 1 056 | | Soil | 186 | 1 056 | | Water | 182 | 143 | | Residential | 196 | 1 072 | | Commercial | 191 | 1 053 | | Road | 193 | 1 059 | | Highway | 191 | 1 036 | | Railway | 181 | 1 054 | | Parking Lot1 | 192 | 1 041 | | Parking Lot2 | 184 | 285 | | Tennis court | 181 | 247 | | Running track | 187 | 473 | | Total | 2 832 | 12 197 | |
|
Table 1. Land class details in Houston2013 dataset
Class name | Train num | Test num | Color |
---|
Apples | 129 | 3 905 | | Buildings | 15 | 2 778 | | Ground | 105 | 374 | | Woods | 154 | 8 969 | | Wineyard | 184 | 10 317 | | Roads | 122 | 3 052 | | Total | 819 | 29 395 | |
|
Table 2. Land class details in Trento dataset
Class | Two-Branch | EndNet | MDL-Middle | MAHiDFNet | Spectrum-LiDAR | TB-HSI | CLCT-Net |
---|
C1 C2 C3 C4 C5 C6 C7 C8 C9 C10 C11 C12 C13 C14 C15 | 83.10 84.10 100.00 93.09 100.00 99.30 92.82 82.34 84.70 65.44 88.24 89.53 92.28 96.76 99.79 | 81.58 83.65 100.00 93.09 99.91 95.10 82.65 81.29 88.29 89.00 83.78 90.39 82.46 100.00 98.10 | 83.10 85.06 99.60 91.57 98.86 100.00 97.64 88.13 85.93 74.42 84.54 95.39 87.37 95.14 100.00 | 98.53 92.87 91.11 98.10 98.38 98.58 99.15 80.94 98.04 72.81 72.71 76.80 95.80 99.53 100.53 | 49.17 35.93 72.12 65.08 59.63 24.54 75.61 75.23 74.74 62.37 85.37 50.13 41.26 28.46 70.24 | 100.0 98.04 30.64 99.28 99.62 93.38 85.66 92.76 94.46 88.41 96.16 84.15 96.30 100.00 90.27 | 87.07 98.05 96.67 94.78 99.34 77.14 89.50 83.31 94.33 92.84 95.79 86.26 87.37 100.00 94.22 | OA | 87.98 | 88.52 | 89.55 | 89.58 | 60.81 | 85.22 | 92.01 | AA | 90.11 | 89.95 | 91.05 | 91.36 | 57.99 | 89.96 | 91.78 | K×100 | 86.98 | 87.59 | 87.59 | 88.74 | 57.67 | 84.02 | 91.33 |
|
Table 3. Comparison of classification accuracy of different methods on Houston2013 dataset
Class | Two-Branch | EndNet | MDL-Middle | MAHiDFNet | Spectrum-LiDAR | TB-HSI | CLCT-Net |
---|
C1 C2 C3 C4 C5 C6 | 99.78 97.93 99.93 99.46 98.96 91.68 | 88.19 98.49 95.19 99.30 91.96 90.14 | 99.50 97.55 99.10 99.90 99.71 92.25 | 99.91 88.92 97.53 99.98 99.90 99.78 | 74.00 62.45 26.00 99.54 98.45 88.94 | 99.19 81.24 63.46 99.93 97.35 94.52 | 99.30 97.49 96.45 99.29 99.70 96.28 | OA | 98.36 | 94.17 | 98.73 | 98.59 | 84.94 | 95.42 | 98.90 | AA | 97.96 | 93.88 | 98.00 | 97.55 | 74.90 | 89.28 | 98.10 | K×100 | 97.83 | 92.22 | 98.32 | 98.12 | 80.56 | 93.89 | 98.54 |
|
Table 4. Comparison of classification accuracy of different methods on Trento dataset
Method | #param./M | FLOPs/M |
---|
Two-Branch | 12 | 25 | EndNet | 0.07 | 0.49 | MDL-Middle MAHiDFNet | 0.25 77 | 4.7 155 | CLCT-Net | 5.1 | 384 |
|
Table 5. FLOPs and parameters of different classification models