Research Article
A Wearable Real-Time Character Recognition System Based on Edge Computing-Enabled Deep Learning for Air-Writing
Table 2
CNN parameter configuration and experimental results.
| Number | Convolution kernel size | Number of convolution filters | Number of convolution blocks | Dropout probability | Accuracy |
| N-1 | | 3 | 32 | 0.1 | 90.37% | N-2 | | 3 | 32 | 0.1 | 91.14% | N-3 | | 3 | 32 | 0.2 | 86.09% | N-4 | | 3 | 32 | 0.2 | 89.35% | N-5 | | 4 | 48 | 0.1 | 92.19% | N-6 | | 4 | 48 | 0.1 | 95.81% | N-7 | | 4 | 48 | 0.2 | 90.74% | N-8 | | 4 | 48 | 0.2 | 91.67% | N-9 | | 5 | 80 | 0.1 | 94.20% | N-10 | | 5 | 80 | 0.1 | 97.95% | N-11 | | 5 | 80 | 0.2 | 92.65% | N-12 | | 5 | 80 | 0.2 | 94.06% |
|
|