| 刘玉民,张雨虹.基于迁移学习的交通标志识别系统设计[J].唐山学院学报,2023,36(6):1-4 |
| 基于迁移学习的交通标志识别系统设计 |
| Design of Traffic Sign Recognition System Based on Transfer Learning |
| |
| DOI：10.16160/j.cnki.tsxyxb.2023.06.001 |
| 中文关键词: 迁移学习 交通标志 识别系统 自动驾驶 |
| 英文关键词: transfer learning traffic sign recognition system autonomous driving |
| 基金项目: |
|
| 摘要点击次数: 1878 |
| 全文下载次数: 1464 |
| 中文摘要: |
| 针对自动驾驶领域对交通标志识别的需求，设计了一种基于迁移学习的交通标志识别系统。该系统采用了预训练的MobileNetV3（去掉输出层）作为特征提取网络，然后添加两个自定义的全连接层以实现信号的分类和输出。由于采用迁移学习方法，深度学习网络中需训练的模型参数大幅减少，训练所需时间大为缩短。该系统使用经典的中国交通标志数据库（CTSDB）中的数据作为交通标志的训练数据和测试数据，训练结果表明，损耗低至0.024 3，准确率高达99.88%；测试结果表明，可以对58类交通标志进行识别，准确率为55.3%。 |
| 英文摘要: |
| To meet the demand for traffic sign recognition in the field of autonomous driving, a traffic sign recognition system based on transfer learning has been designed, which uses pre-trained MobileNetV3 (excluding the output layer) as the feature extraction network, and then adds two self-defined fully connected layers for signal classification and output. Due to the adoption of transfer learning, the number of model parameters to be trained in the deep learning network are significantly reduced, resulting in a considerable reduction in training time. The system uses data from the classic Chinese Traffic Sign Database (CTSDB) as training and testing data for traffic signs. The training results indicate a low loss of 0.0243 and a high accuracy of 99.88%, while the testing results show that this traffic sign recognition system can recognize 58 different classes of traffic signs,with an accuracy rate of 55.3%. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |