基于深度学习的车辆分类模型研究

巢湖学院学报 ›› 2018, Vol. 20 ›› Issue (6): 115-120.

基于深度学习的车辆分类模型研究

提出了一种基于深度学习的车辆分类模型来处理复杂的交通场景。该模型由车辆检测模型和分类模型两部分组成。在车辆检测模型中采用基于区域的卷积神经网络方法R-CNN（Regions with Convolutional Neural Network），从杂乱背景的图像中提取单个车辆图像，该步骤为下一个分类模型提供数据。在车辆分类模型中，将车辆图像导入CNN模型，并生成特征，然后使用联合贝叶斯网络来实现分类过程。实验表明，该方法可以有效识别交通图像中的车辆构造和模型。

1安徽文达信息工程学院，安徽合肥231201 2合肥学院，安徽合肥230601

出版日期:2018-11-25 发布日期:2018-11-25
作者简介:张少巍（1981-），女，河南南阳人。安徽文达信息工程学院计算机工程学院，讲师。研究方向：计算机网络、机器学习。
基金资助:
安徽文达信息工程学院校级重点项目（项目编号：XZR2018A04）；安徽省高校自然科学基金重点项目（项目编
号：KJ2017A649）

RESEARCH ON VEHICLE CLASSIFICATION MODEL BASED ON DEEP LEARNING

A vehicle classification model based on deep learning is proposed to deal with complex traffic scenes. The model consists of two parts: vehicle detection model and classification model. A region-based convolutional neural network （R-CNN）method is used in the vehicle detection model to extract a single vehicle image from a disorderly background image. This step provides data for the next classification model. In the vehicle classification model, an image of a vehicle is included into a CNN model and a feature is produced, and then a joint Bayesian network is used to implement the classification process. Experiments show that the method can effectively identify vehicle structures and models in traffic images.

1Anhui Wenda University of Information Engineering, Hefei Anhui 231201 2Hefei University, Hefei Anhui 230601

Online:2018-11-25 Published:2018-11-25

摘要/Abstract

摘要： 提出了一种基于深度学习的车辆分类模型来处理复杂的交通场景。该模型由车辆检测模型和分类模型两部分组成。在车辆检测模型中采用基于区域的卷积神经网络方法R-CNN（Regions with Convolutional Neural Network），从杂乱背景的图像中提取单个车辆图像，该步骤为下一个分类模型提供数据。在车辆分类模型中，将车辆图像导入CNN模型，并生成特征，然后使用联合贝叶斯网络来实现分类过程。实验表明，该方法可以有效识别交通图像中的车辆构造和模型。

关键词: 分类, 深度学习, 车辆检测

Abstract: A vehicle classification model based on deep learning is proposed to deal with complex traffic scenes. The model consists of two parts: vehicle detection model and classification model. A region-based convolutional neural network （R-CNN）method is used in the vehicle detection model to extract a single vehicle image from a disorderly background image. This step provides data for the next classification model. In the vehicle classification model, an image of a vehicle is included into a CNN model and a feature is produced, and then a joint Bayesian network is used to implement the classification process. Experiments show that the method can effectively identify vehicle structures and models in traffic images.

Key words: classification, deep learning, vehicle detection

中图分类号:

U495

张少巍, 殷荣网. 基于深度学习的车辆分类模型研究[J]. 巢湖学院学报, 2018, 20(6): 115-120.

ZHANG Shao-wei, YIN Rong-wang. RESEARCH ON VEHICLE CLASSIFICATION MODEL BASED ON DEEP LEARNING[J]. Journal of Chaohu University, 2018, 20(6): 115-120.