Implementation of CNN Algorithm with ResNet-50 Architecture for Vehicle Image Classification

Vehicles are essential transportation tools in our daily lives, serving a crucial role in facilitating mobility and commerce. However, with the rapid growth of the population and urbanization, Indonesia has witnessed a significant increase in the number of vehicles over the years. This surge has led to challenges in data management for government agencies, making it imperative to develop an efficient vehicle classification system. According to previous research, ResNet-50 has consistently shown superior accuracy compared to other neural network architectures. This study aims to design and implement a vehicle classification system utilizing a Convolutional Neural Network (CNN) with the ResNet-50 architecture. The dataset employed in this research comprises seventeen distinct vehicle types, ensuring a comprehensive evaluation of the model's effectiveness. The implementation was carried out using the Adam optimizer, with various hyperparameter settings tested to optimize performance. The evaluation results revealed that the model without pretrained weights yielded the lowest accuracy, as indicated by numerous prediction errors in the confusion matrix. Conversely, the model leveraging pretrained weights demonstrated significantly higher accuracy. Among the models tested, the third model, which incorporated pretrained weights, achieved the highest accuracy of 91.16%. This was followed by the first model with an accuracy of 87.55%. These results underscore the critical role of pretrained weights in enhancing the performance of classification models, particularly in complex tasks such as vehicle classification.