<![CDATA[This study aims to automatically detect overdimension trucks using a lightweight and efficient deep learning model based on MobileNet. Overdimension trucks pose serious threats to road infrastructure, traffic safety, and contribute to increased economic costs due to road damage and congestion. The developed model utilizes MobileNet as a feature extractor without the standard fully connected layers, and is equipped with additional layers including Flatten, Batch Normalization, Dense with Leaky ReLU activation, and Dropout to enhance training stability and prevent overfitting. The dataset consists of two classes—normal trucks and overdimension trucks—with images sized 128×128 pixels, collected from internet sources and field photos. The training process employs binary crossentropy loss, the Adam optimizer with an initial learning rate of 0.0001, and an Early Stopping mechanism. Fine-tuning is performed by unfreezing layers from the 100th layer upward and lowering the learning rate to 0.00001. Evaluation results show an accuracy of 97.92%, with consistent loss and accuracy visualization, demonstrating the model's capability in classifying overdimension trucks to support automatic traffic monitoring systems. This model has the potential to be implemented in toll gate systems to automatically deny access to overdimension vehicles. Furthermore, integration with roadside CCTV allows real-time monitoring of vehicle dimension violations across various traffic checkpoints.]]>
