Table 4.
Results of MobileViT-xs training using G1 and G2 data.
| Sample Size | DA Method | BLACK (%) | DBS (%) | QC (%) | SLH (%) | XBS (%) | AIA (%) |
|---|---|---|---|---|---|---|---|
| N | None | 86.2 ± 2.67 | 86.23 ± 1.41 | 90.78 ± 2.76 | 91.94 ± 1.77 | 87.48 ± 1.63 | / |
| 2N | Erasing × 1 | 92.33 ± 0.85 | 89.15 ± 0.74 | 96.36 ± 0.68 | 94.98 ± 0.32 | 92.18 ± 1.46 | 4.47 |
| Noise × 1 | 91.32 ± 0.9 | 87.68 ± 1.36 | 95.36 ± 1.99 | 94.06 ± 1.04 | 91.23 ± 1.76 | 3.4 | |
| Rotation × 1 | 93.21 ± 0.6 | 89.72 ± 1.35 | 97.59 ± 0.29 | 95.65 ± 0.64 | 94.17 ± 0.96 | 5.54 | |
| DSM × 1 | 92.81 ± 0.89 | 90.38 ± 1.61 | 97.87 ± 0.58 | 96.0 ± 0.4 | 95.08 ± 1.58 | 5.9 |
N, the data amount of the original training sample; 2N, the sample size after once augmentation; DA, data augmentation; None, original data; Erasing, random erasing; Noise, random noise; Rotation, rotation; DSM, difference of spectral mean; BLACK, Black peanut; DBS, Dabaisha; QC, Qicai; SLH, Silihong; XBS, Xiaobaisha; AIA, the average improved accuracy.