摘要
研究基于深度学习的车型自动识别技术,运用深度神经网络对在各个角度下拍摄的具有复杂背景的汽车图像进行网络训练,从而达到车辆车型的自动识别的目的。采用先进的深度学习框架Caffe和具有强大计算能力的GPU,使用深度神经网络VGG16和Alex Net,分别对汽车图像进行网络训练与测试,并通过与传统的分类算法,K最近邻进行对比研究。实验显示,VGG16网络模型准确率高达97.58%,在汽车车型识别问题上具有很大优势。
Studies the recognition of vehicle types based on deep learning methods. Deep neural network is trained to classify automobile images which are shot from different angles with complex background. Uses the cutting-edge Deep Learning architecture, Caffe and a powerful computational platform, GPU. VGG16 network and Alex Net network are trained and tested for this task respectively. Moreover, applies the classical algorithm, K-Nearest Neighbor for comparison. The result suggests that VGG16 network outperform other methods by a big margin with the accuracy of 97.58%.
基金
四川省科技计划项目(No.2014GZ0005-5)
关键词
深度学习
车型识别
卷积神经网络
Deep Learning
Vehicle Recognition
Convolutional Neural Network
