基于多示例深度学习与损失函数优化的交通标志识别算法
|
摘要
为了解决当前交通标志种类繁多和所处环境多变,导致智能识别正确率不高的问题,提出基于多示例深度学习的交通标志识别算法。根据样本图像块与其对应的标签设计一个包含颜色、几何、区域特征的训练集,得到样本特征与标签的对应规律;根据权重修正反馈,推导包与标签的逻辑关系,建立多示例训练学习算子,准确分类交通标志。进行训练集损失函数计算,通过最优分类器响应减少训练数据损失。最后,基于大数据样本驱动形成背景约束,从而去除示例中模棱两可的训练数据,完成交通标志的准确识别。基于QT平台,开发相应的识别软件。实验测试结果显示,与当前交通标志识别技术相比,所提算法拥有更高的识别正确性与鲁棒性,且对各类交通标志具有较高的识别准确率,在智能汽车、自动交通监控等领域具有一定的应用价值。
Since the current traffic signs recognition algorithm has low intelligent recognition accuracy due to its various types of traffic signs and changeable environments,a traffic sign recognition algorithm based on multi-instance deep learning is proposed. According to the image block of samples and its corresponding label,a training set including color,geometry and regional characteristics is designed to obtain the correspondence rule between the sample characteristic and tag. On the basis of feedback of weight correction,the logical relation between package and label is derived,and the learning operator of multiinstance training is established to classify the traffic signs accurately. The loss function of training set is calculated by means of the optimal classifier response to reduce the loss of training data. The background constraint is formed on the basis of large data sample driver,so as to eliminate the ambiguous training data in the instance and accomplish the accurate recognition of traffic signs. The corresponding recognition software was developed with QT platform. The experimental results show that, in comparison with the current traffic signs identification technology,the proposed algorithm has higher recognition accuracy and robustness. The algorithm has high recognition accuracy for various traffic signs,and a certain application value in the fields of intelligent vehicle and automatic traffic monitoring.
Since the current traffic signs recognition algorithm has low intelligent recognition accuracy due to its various types of traffic signs and changeable environments,a traffic sign recognition algorithm based on multi-instance deep learning is proposed. According to the image block of samples and its corresponding label,a training set including color,geometry and regional characteristics is designed to obtain the correspondence rule between the sample characteristic and tag. On the basis of feedback of weight correction,the logical relation between package and label is derived,and the learning operator of multiinstance training is established to classify the traffic signs accurately. The loss function of training set is calculated by means of the optimal classifier response to reduce the loss of training data. The background constraint is formed on the basis of large data sample driver,so as to eliminate the ambiguous training data in the instance and accomplish the accurate recognition of traffic signs. The corresponding recognition software was developed with QT platform. The experimental results show that, in comparison with the current traffic signs identification technology,the proposed algorithm has higher recognition accuracy and robustness. The algorithm has high recognition accuracy for various traffic signs,and a certain application value in the fields of intelligent vehicle and automatic traffic monitoring.
引文
[1]谢锦,蔡自兴,邓海涛,等.基于图像不变特征深度学习的交通标志分类[J].计算机辅助设计与图形学学报,2017,29(4):632-640.XIE Jin,CAI Zixing,DENG Haitao,et al.Classification of traffic signs based on image invariant feature depth learning[J].Journal of computer aided design and computer graphics,2017,29(4):632-640.
[2]KR?áK E,TOTH?.Traffic sign recognition and localization for databases of traffic signs[J].Acta electrotechnica et informatica,2011,11(4):31-35.
[3]王晓斌,黄金杰,刘文举.基于优化卷积神经网络结构的交通标志识别[J].计算机应用,2017,37(2):530-534.WANG Xiaobin,HUANG Jinjie,LIU Wenju.Traffic sign recognition based on optimized convolutional neural network[J].Computer applications,2017,37(2):530-534.
[4]戈侠,于凤芹,陈莹.基于分块自适应融合特征的交通标志识别[J].计算机工程与应用,2017,53(3):188-192.GE Xia,YU Fengqin,CHEN Ying.Traffic sign recognition based on block adaptive fusion feature[J].Computer engineering and applications,2017,53(3):188-192.140
[5]ZHOU Shusen,ZOU Hailin,LIU Chanjuan.Deep extractive networks for supervised learning[J].Optik:international journal for light and electron optics,2016,127(20):9008-9019.
[6]FARIA A W C,COELHO F G F,SILVA A M.A new approach for multiple instance learning based on positive instance selection and kernel density estimation[J].Engineering applications of artificial intelligence,2017,59(12):196-204.
[7]刘占文,赵祥模,李强,等.基于图模型与卷积神经网络的交通标志识别方法[J].交通运输工程学报,2016,16(5):122-131.LIU Zhanwen,ZHAO Xiangmo,LI Qiang,et al.A traffic sign recognition method based on graph model and convolutional neural network[J].Journal of traffic and transportation engineering,2016,16(5):122-131.
[8]ZHONG Jingjing,TSE P W,WANG Dong.Novel Bayesian inference on optimal parameters of support vector machines and its application to industrial survey data classification[J].Neurocomputing,2016,211:159-171.
[9]YU Yongtao.Bag-of-visual-phrases and hierarchical deep models for traffic sign detection and recognition in mobile laser scanning data[J].ISPRS journal of photogrammetry and remote sensing,2016,11(7):36-39.
[10]OUERHANI Y.Advanced driver assistance system:road sign identification using VIAPIX system and a correlation technique[J].Optics and lasers in engineering,2016,12(2):71-76.
[2]KR?áK E,TOTH?.Traffic sign recognition and localization for databases of traffic signs[J].Acta electrotechnica et informatica,2011,11(4):31-35.
[3]王晓斌,黄金杰,刘文举.基于优化卷积神经网络结构的交通标志识别[J].计算机应用,2017,37(2):530-534.WANG Xiaobin,HUANG Jinjie,LIU Wenju.Traffic sign recognition based on optimized convolutional neural network[J].Computer applications,2017,37(2):530-534.
[4]戈侠,于凤芹,陈莹.基于分块自适应融合特征的交通标志识别[J].计算机工程与应用,2017,53(3):188-192.GE Xia,YU Fengqin,CHEN Ying.Traffic sign recognition based on block adaptive fusion feature[J].Computer engineering and applications,2017,53(3):188-192.140
[5]ZHOU Shusen,ZOU Hailin,LIU Chanjuan.Deep extractive networks for supervised learning[J].Optik:international journal for light and electron optics,2016,127(20):9008-9019.
[6]FARIA A W C,COELHO F G F,SILVA A M.A new approach for multiple instance learning based on positive instance selection and kernel density estimation[J].Engineering applications of artificial intelligence,2017,59(12):196-204.
[7]刘占文,赵祥模,李强,等.基于图模型与卷积神经网络的交通标志识别方法[J].交通运输工程学报,2016,16(5):122-131.LIU Zhanwen,ZHAO Xiangmo,LI Qiang,et al.A traffic sign recognition method based on graph model and convolutional neural network[J].Journal of traffic and transportation engineering,2016,16(5):122-131.
[8]ZHONG Jingjing,TSE P W,WANG Dong.Novel Bayesian inference on optimal parameters of support vector machines and its application to industrial survey data classification[J].Neurocomputing,2016,211:159-171.
[9]YU Yongtao.Bag-of-visual-phrases and hierarchical deep models for traffic sign detection and recognition in mobile laser scanning data[J].ISPRS journal of photogrammetry and remote sensing,2016,11(7):36-39.
[10]OUERHANI Y.Advanced driver assistance system:road sign identification using VIAPIX system and a correlation technique[J].Optics and lasers in engineering,2016,12(2):71-76.