A Comparative Analysis Based on Metamorphic Testing of CNN Algorithms for Classification of Vehicle Dashboards

This paper proposes an effective classification system for vehicle dashboards using various pretrained convolutional neural network (CNN) models and presents a comparative analysis of their performance. Focusing on five types of dashboards that are visually indistinguishable, the CNNs are used to extract detailed features and develop an automated classification system. Nine pretrained CNN models are evaluated based on accuracy, loss rate, and F1-score, with training enhanced by emphasizing regions of interest. In particular, the study validates the robustness and consistency of the models through metamorphic testing, which demonstrates that the CNN model works reliably on deformed data in a real environment and can maintain high classification accuracy. The results highlight the significant improvements of the proposed system over visual inspection in reliability and efficiency. This research offers a pathway for automating material classification in manufacturing sites and illustrates the practical applicability of AI-based classification technologies.