Vehicle Classification in Intelligent Transportation Systems: Enhanced CNN Approach and Data Set Expansion

Abstract

Vehicle classification systems are a crucial component of intelligent transportation systems (ITS). Improving the accuracy and efficiency of image recognition for vehicle classification based on the number of axles is vital for intelligent toll systems on highways and for pavement structure design. However, current research faces challenges in the segmentation of trucks due to the limited range of vehicle types covered in existing data sets, which affects the potential of ITS to enhance transportation efficiency and achieve precise traffic management. To address this issue, a new data set comprising 21,240 images covering 10 different vehicle types was created, and an improved vehicle classification method based on convolutional neural networks (CNN) was proposed. The model’s classification performance was evaluated using metrics such as accuracy, precision, recall, and F1 score, while also analyzing the sensitivity of model performance to hyperparameters like learning rate, epochs, iterations, and optimizers. Results showed that the improved model achieved a classification accuracy of 98.8%, with both precision and recall at 97.3%, and an F1 score of 97.2%. Furthermore, when using the Adam optimizer, especially with a learning rate of 1.0×10−7, 70 epochs, and 1,500 iterations, the model performed best in classifying various vehicle types and truck axles. Compared with other CNN models such as visual geometry group (VGG)-16 and GoogLeNet, as well as existing vehicle classification models, the improved AlexNet model achieved a classification accuracy of 98.8%. Accurate axle number recognition not only ensures fair and reasonable toll charges, enhancing highway operational efficiency and economic returns, but also plays a crucial role in assessing pavement load-bearing capacity and ensuring the scientific design and durability of road surfaces. Therefore, the proposed model not only improves the efficiency and accuracy of toll collection and pavement design but also offers significant economic and social benefits for related applications. This study presents a significant advancement in vehicle classification systems with a focus on accurate axle-based recognition. The improved classification model, developed from a data set of 21,240 images across 10 vehicle types, achieves an impressive accuracy of 98.8%. The primary applications of this model lie in intelligent toll systems, where precise axle number recognition can ensure equitable toll charges and enhance highway operational efficiency. By replacing traditional data collection methods, which often lack accuracy and efficiency, our model offers a reliable solution for real-time vehicle classification at toll stations. Additionally, the model plays a crucial role in pavement structure design by accurately assessing the load-bearing capacity based on vehicle axle counts, thereby informing the scientific design and durability of road surfaces. This not only extends the life span of infrastructure but also reduces maintenance costs. In summary, the proposed vehicle classification model streamlines toll collection and improves transportation infrastructure planning, providing significant economic and social benefits. Future enhancements will further broaden its applicability across various contexts.