·1128· 智能系统学报 第15卷 表3图像预处理与残差连接对网络模型影响（德国数据库） Table 3 Influence of image preprocessing and residual connection on a network model(GTSRB) 原始数据 归一化与数据增强 预处理数据 模型名称 损失值 准确率% 损失值 准确率/% 损失值 准确率% 0.092814 83.9382 0.008746 85.9271 0.009894 86.2814 0.008988 85.5620 0.011276 83.5247 0.008892 87.0185 LeNet-5 0.010205 84.5821 0.009687 86.3691 0.010583 86.0019 0.010018 84.6411 0.009825 86.1827 0.009571 86.5147 0.012530 82.1582 0.009825 85.2428 0.011759 85.4862 0.002727 94.0121 0.001978 96.2142 0.001872 97.9501 0.002715 94.7283 0.002214 95.5827 0.001378 98.2605 ResNet-50 0.002428 94.7151 0.003014 94.8517 0.001457 97.7568 0.002823 94.2184 0.002013 96.0851 0.001311 98.6647 0.003410 93.9281 0.002401 95.4724 0.001554 98.0258 0.002979 94.0713 0.003478 94.9212 0.001002 99.0013 0.002807 94.8411 0.002814 95.5287 0.001498 98.2605 SENet 0.003101 93.7151 0.002717 95.7547 0.002052 97.7568 0.002914 94.2634 0.003071 95.5898 0.002118 97.6647 0.003075 93.9826 0.002401 96.4801 0.001559 98.0258 0.003248 89.8255 0.003421 90.0785 0.002952 91.8287 0.003180 90.2402 0.003718 90.8287 0.004859 91.2584 AlexNet 0.002861 90.3042 0.004182 89.9387 0.002148 93.0019 0.003225 89.9821 0.003655 90.0072 0.002281 92.180 0.002962 90.3902 0.004557 89.8718 0.002881 91.8851 0.003058 91.2827 0.004272 91.8552 0.003048 94.5284 0.003180 91.3542 0.003229 92.4632 0.002917 93.4812 8层结构 0.003055 91.2102 0.003138 92.0439 0.003017 94.0021 0.004085 90.9087 0.002927 93.3847 0.002837 95.0528 0.003017 91.2921 0.004317 91.9282 0.002827 93.9223 0.002176 96.0020 0.002488 96.2212 0.000972 99.0500 0.002707 95.8123 0.001914 96.5287 0.000198 99.2605 本文模型 0.002801 95.7151 0.002418 95.8557 0.001458 98.7568 0.002018 96.2184 0.002041 96.5853 0.001510 98.6647 0.002675 95.9916 0.001901 96.4827 0.001154 99.0258表 3 图像预处理与残差连接对网络模型影响 (德国数据库) Table 3 Influence of image preprocessing and residual connection on a network model (GTSRB) 模型名称 原始数据 归一化与数据增强 预处理数据 损失值 准确率/% 损失值 准确率/% 损失值 准确率/% LeNet-5 0.092 814 83.9382 0.008 746 85.927 1 0.009894 86.281 4 0.008 988 85.5620 0.011 276 83.524 7 0.008892 87.018 5 0.010 205 84.5821 0.009 687 86.369 1 0.010583 86.001 9 0.010 018 84.6411 0.009 825 86.182 7 0.009571 86.514 7 0.012 530 82.1582 0.009 825 85.242 8 0.011759 85.486 2 ResNet-50 0.002 727 94.0121 0.001 978 96.214 2 0.001872 97.950 1 0.002 715 94.7283 0.002 214 95.582 7 0.001378 98.260 5 0.002 428 94.7151 0.003 014 94.851 7 0.001457 97.756 8 0.002 823 94.2184 0.002 013 96.085 1 0.001311 98.664 7 0.003 410 93.9281 0.002 401 95.472 4 0.001554 98.025 8 SENet 0.002 979 94.0713 0.003 478 94.921 2 0.001002 99.001 3 0.002 807 94.8411 0.002 814 95.528 7 0.001498 98.260 5 0.003 101 93.7151 0.002 717 95.754 7 0.002052 97.756 8 0.002 914 94.2634 0.003 071 95.589 8 0.002118 97.664 7 0.003 075 93.9826 0.002 401 96.480 1 0.001559 98.025 8 AlexNet 0.003 248 89.8255 0.003 421 90.078 5 0.002952 91.828 7 0.003 180 90.2402 0.003 718 90.828 7 0.004859 91.258 4 0.002 861 90.3042 0.004 182 89.938 7 0.002148 93.001 9 0.003 225 89.9821 0.003 655 90.007 2 0.002281 92.180 0.002 962 90.3902 0.004 557 89.871 8 0.002881 91.885 1 8层结构 0.003 058 91.2827 0.004 272 91.855 2 0.003048 94.528 4 0.003 180 91.3542 0.003 229 92.463 2 0.002917 93.481 2 0.003 055 91.2102 0.003 138 92.043 9 0.003017 94.002 1 0.004 085 90.9087 0.002 927 93.384 7 0.002837 95.052 8 0.003 017 91.2921 0.004 317 91.928 2 0.002827 93.922 3 本文模型 0.002 176 96.0020 0.002 488 96.221 2 0.000972 99.050 0 0.002 707 95.8123 0.001 914 96.528 7 0.000198 99.260 5 0.002 801 95.7151 0.002 418 95.855 7 0.001458 98.756 8 0.002 018 96.2184 0.002 041 96.585 3 0.001510 98.664 7 0.002 675 95.9916 0.001 901 96.482 7 0.001154 99.025 8 ·1128· 智 能 系 统 学 报 第 15 卷