摘要
提出一种基于卷积神经网络(CNN)的车标识别方法,通过多层的特征学习,能够直接从训练样本中提取特征,最后送入神经网络分类器进行分类。验证集采用5000个从属于10类车标并附有各类变化的车标数据库,该应用算法达到98.28%的平均准确率和每张少于3ms的识别速度(在MATLAB环境下),实验表明,该方法对于车标识别问题具有优异的准确率和鲁棒性,且对于计算资源要求很低。
Proposes a vehicle logo recognition based on Convolutional Neural Networks. With a deep hierarchical feature learning process, the pro- posed method extracts the features from the training samples directly, and trains the classier based on neural network. Applies 5,000 logos belonging to 10 vehicle manufactures for validation. The average accuracy 98.28% for ten classes and fast implementation （less than 3ms for each logo in MATLAB） has demonstrated that the proposed method outperforms than state-of-art with higher accuracy, stronger ro- bustness, and less computational cost.
关键词
智能交通
车标识别
深度学习
卷积神经网络
Intelligent Transportation Systems
Vehicle Logo Recognition
Deep Learning
Convolutional Neural Networks
