摘要
在真实路况行驶过程中发现并准确识别周围的交通标志是自动驾驶系统和辅助驾驶系统的研究重点。为解决经典的VGG网络在训练过程中需要消耗大量计算资源和参数量巨大的问题,提出了使用深度可分离卷积模块,用其代替传统的卷积层减少了近9倍参数量且获得了更多局部感受野,使用平均池化层取代全连接层进一步压缩了参数量。改进的模型在真实场景下的交通标志图像数据集GTSRB的准确率达到98.38%。实验结果表明,改进的模型提高了识别准确率的同时减少模型参数量,具有实际意义。
Finding and identifying the traffic signs in the real driving process is the research focus of the automatic driving system and the auxiliary driving system.To address the problem that the classical VGG network needs a lot of computational resources and parameters in the training process,a deep separable convolution module is proposed to replace the traditional convolution layer,which reduces the parameters by about 9 times and obtains multiple local receptive fields.The average pooling layer is used instead of the full connection layer to further reduce the parameters.The accuracy of GTSRB traffic sign image data set in real scene is 98.38%.The experimental results show that the improved model improves the recognition accuracy and reduces the number of model parameters,which is of practical significance.
作者
张璟
ZHANG Jing(School of Computer and Information Technology,Shanxi University,Taiyuan 030006,China)
出处
《电脑知识与技术》
2019年第7X期195-197,共3页
Computer Knowledge and Technology
关键词
卷积神经网络
交通标志牌检测
交通标志牌识别
交通标志牌分类
深度可分离卷积
convolutional neural network
traffic sign detection
traffic sign recognition
traffic sign classification
Separable convolution layer
