基于深度学习的智能治超场景下货车车型识别

Type Recognition of Trucks Based on Deep Learning in Intelligent Overload Management Scenarios

针对智能治超场景下超载车辆自动化检测的需求,在YOLOv5s的基础上从数据、模型和算法3个方面提出了一种改进的货车车型识别算法;在数据层面,使用的数据增强模拟了现实中面对恶劣天气、图像噪声和数据损坏等复杂场景,丰富了训练数据的多样性,提高了模型在复杂场景下的鲁棒性;在模型方面,提出了一种新的注意力机制来综合考虑不同通道的重要性和编码特征的位置信息,提高了模型的识别准确性;在算法层面,针对现有算法的不足,提出了一种更通用的标准来判断货车与轮轴的隶属关系,以适用更复杂的场景;实验结果表明,提出的改进模型对货车和轮轴的识别精度分别达到99.34%和99.22%,对货车车型识别的准确率为98.71%;与经典的YOLOv5s网络相比,货车和轮轴的平均识别精度提高了2.39%,货车车型的识别准确率提高了2.22%;综上,所提出的方法实现了对货车车型自动和准确的识别,可以为智能治超场景下的货车车型识别提供理论支撑。

In response to the automated detection demand of overloaded trucks in intelligent overload management scenarios,based on YOLOv5s,an improved method is proposed to recognize the truck type from three aspects of data,model,and algorithm.At the data level,the dataed data enhancely simulates complex scenarios such as facing severe bad weather conditions,image noise,and data damage in real life,which enriches the diversity of training data and improves the robustness of the model under complex scenarios.In terms of the model,a new attention mechanism is proposed to comprehensively consider the importance of different channels and the positional information of encoding features,which improves the recognition accuracy of the model.At the algorithmic level,in order to overcome the shortcomings of existing algorithms,a more general standard for determining the subordinate relationship between trucks and axles is proposed to be applied in more complex scenarios.The experimental results show that the proposed improved model achieves the recognition accuracy of 99.34%and 99.22%for trucks and axles,respectively,and the recognition accuracy of truck type is 98.71%.Compared with classic YOLOv5s networks,the average recognition accuracy of trucks and axles has increased by 2.39%,and the truck type recognition accuracy by 2.22%.In summary,the proposed method achieves automatic and accurate recognition of truck type,which can provide a theoretical support for truck type recognition in intelligent overload management scenarios.