第6期 何锐波，等：一种改进的深度学习的道路交通标识识别算法 ·1127· 表2图像预处理与残差连接对网络模型影响（比利时数据库） Table 2 Influence of image preprocessing and residual connection on a network model(BelgiumTSC) 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% 0.012691 80.9767 0.012205 82.3215 0.008716 85.4107 0.011784 80.8957 0.010177 83.0012 0.009872 85.1258 LeNet-5 0.010335 81.3458 0.017484 81.9824 0.011178 84.8424 0.016348 80.8816 0.014532 82.1958 0.009867 85.0191 0.009871 81.2183 0.012031 82.6507 0.009915 84.9438 0.003014 93.2184 0.003717 93.4281 0.003928 95.9938 0.002958 93.1048 0.002458 93.5214 0.003215 96.1741 ResNet-50 0.003234 92.9437 0.002714 94.1697 0.003001 96.3481 0.002917 93.4414 0.002848 94.1548 0.003128 96.1749 0.003418 92.2184 0.003171 93.6281 0.002714 97.1284 0.003478 92.6481 0.002917 94.2483 0.003288 96.2042 0.002951 93.1008 0.003058 94.1231 0.002114 97.1425 SENet 0.003104 92.9861 0.003014 94.1927 0.003018 96.7657 0.002719 93.4266 0.004048 93.6335 0.003141 96.5283 0.002814 93.1528 0.004175 92.6281 0.002101 97.2177 0.004827 88.0827 0.003527 90.2180 0.004282 91.8472 0.004897 87.3413 0.003208 90.1382 0.003019 92.4582 AlexNet 0.005316 88.8632 0.004083 89.6472 0.002318 92.0069 0.003037 90.0012 0.002544 91.0401 0.001827 93.3101 0.004096 89.0825 0.004218 89.5438 0.004148 91.9811 0.003028 90.1058 0.005487 90.3284 0.004281 91.5287 0.004017 90.9521 0.005284 90.6248 0.005414 90.1857 8层结构 0.004857 88.7632 0.004018 91.5187 0.001217 94.0010 0.004247 89.8211 0.003528 92.5284 0.002108 93.8284 0.003848 89.3844 0.004281 91.2584 0.003031 93.3226 0.002561 94.1011 0.002017 95.0807 0.001727 97.9615 0.003151 93.8217 0.003093 93.8468 0.001958 97.7668 本文模型 0.003039 93.0124 0.003442 93.7649 0.002018 97.5821 0.003096 92.7192 0.003031 93.5246 0.003001 96.9918 0.002189 94.2158 0.002515 94.1218 0.002364 98.0837表 2 图像预处理与残差连接对网络模型影响 (比利时数据库) Table 2 Influence of image preprocessing and residual connection on a network model (BelgiumTSC) 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 0.012 691 80.9767 0.012 205 82.321 5 0.008716 85.410 7 0.011 784 80.8957 0.010 177 83.001 2 0.009872 85.125 8 0.010 335 81.3458 0.017 484 81.982 4 0.011178 84.842 4 0.016 348 80.8816 0.014 532 82.195 8 0.009867 85.019 1 0.009 871 81.2183 0.012 031 82.650 7 0.009915 84.943 8 ResNet-50 0.003 014 93.2184 0.003 717 93.428 1 0.003928 95.993 8 0.002 958 93.1048 0.002 458 93.521 4 0.003215 96.174 1 0.003 234 92.9437 0.002 714 94.169 7 0.003001 96.348 1 0.002 917 93.4414 0.002 848 94.154 8 0.003128 96.174 9 0.003 418 92.2184 0.003 171 93.628 1 0.002714 97.128 4 SENet 0.003 478 92.6481 0.002 917 94.248 3 0.003288 96.204 2 0.002 951 93.1008 0.003 058 94.123 1 0.002114 97.142 5 0.003 104 92.9861 0.003 014 94.192 7 0.003018 96.765 7 0.002 719 93.4266 0.004 048 93.633 5 0.003141 96.528 3 0.002 814 93.1528 0.004 175 92.628 1 0.002101 97.217 7 AlexNet 0.004 827 88.0827 0.003 527 90.218 0 0.004282 91.847 2 0.004 897 87.3413 0.003 208 90.138 2 0.003019 92.458 2 0.005 316 88.8632 0.004 083 89.647 2 0.002318 92.006 9 0.003 037 90.0012 0.002 544 91.040 1 0.001827 93.310 1 0.004 096 89.0825 0.004 218 89.543 8 0.004148 91.981 1 8层结构 0.003 028 90.1058 0.005 487 90.328 4 0.004281 91.528 7 0.004 017 90.9521 0.005 284 90.624 8 0.005414 90.185 7 0.004 857 88.7632 0.004 018 91.518 7 0.001217 94.001 0 0.004 247 89.8211 0.003 528 92.528 4 0.002108 93.828 4 0.003 848 89.3844 0.004 281 91.258 4 0.003031 93.322 6 本文模型 0.002 561 94.1011 0.002 017 95.080 7 0.001727 97.961 5 0.003 151 93.8217 0.003 093 93.846 8 0.001958 97.766 8 0.003 039 93.0124 0.003 442 93.764 9 0.002018 97.582 1 0.003 096 92.7192 0.003 031 93.524 6 0.003001 96.991 8 0.002 189 94.2158 0.002 515 94.121 8 0.002364 98.083 7 第 6 期 何锐波，等：一种改进的深度学习的道路交通标识识别算法 ·1127·