摘要
提出了一种基于融合特征稀疏编码模型的车辆品牌识别方法,该方法首先提取车脸图像的方向梯度直方图特征作为融合特征稀疏编码模型的一级特征向量,然后将车脸图像的一级特征向量作为过完备字典中训练样本集的线性组合,并构建非负性约束稀疏编码模型,最后采用重构误差最小原则对车辆品牌进行识别。基于东南大学的车脸数据库进行了试验,结果表明,基于融合特征稀疏编码模型的车辆品牌识别方法优于HOG+SVM、传统稀疏表示和字典学习稀疏表示的车辆品牌识别方法,其平均识别率达到96.16%。理论分析和试验结果表明,基于融合特征稀疏编码模型的车辆品牌识别方法具有较强的鲁棒性和适用性。
A novel recognition method of vehicle brands based on sparse coding model of fused features was proposed.Firstly,histogram of oriented gradient(HOG)of vehicle face image was extracted as the first-level feature vector of sparse coding model of fused features.Secondly,the first-level feature vector of vehicle face image was used as the linear combination of training samples in over-complete dictionary,and a sparse coding model with non-negative constraints was constructed.Finally,the principle of minimum reconstruction was used to identify vehicle brands.The comparative experiments based on vehicle face database of Southeast University were carried out.The results show that the proposed recognition method of vehicle brands based on sparse coding model of fused features is superior to HOG+SVM,traditional sparse representation and dictionary learning sparse representation,and the average recognition rate is 96.16%.Theoretical analysis and experimental results show that the novel recognition method of vehicle brands based on sparse coding model of fused features has strong robustness and applicability.
作者
石鑫
赵池航
张小琴
李彦伟
薛善光
毛迎兵
SHI Xin;ZHAO Chi-hang;ZHANG Xiao-qin;LI Yan-wei;XUE Shan-guang;MAO Ying-bing(Department of Civil Engineering,Hebei Jiaotong Vocational and Technical College,Shijiazhuang 050011,Hebei,China;School of Transportation,Southeast University,Nanjing 211189,Jiangsu,China;Research and Development Center of Transport Industry for Technologies,Materials and Equipment of Highway Construction and Maintenance,Hebei Provincial Communications Planning and Design Institute,Shijiazhuang 050011,Hebei,China)
出处
《筑路机械与施工机械化》
2020年第3期59-63,共5页
Road Machinery & Construction Mechanization
基金
河北省重点研发计划项目(19270802D)。
关键词
融合特征
稀疏编码模型
车脸图像
鲁棒性
fused feature
sparse coding model
vehicle face image
robustness
