Car Condition Detection using CNN Algorithm

Mettu, Keerthana Reddy (2025) Car Condition Detection using CNN Algorithm. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (501kB) | Preview

Abstract

The paper explores the implementation of damage detection in automated vehicles represented by Convolutional Neural Networks (CNNs) and assesses three different architectures including, Custom CNN, ResNet18, and VGG16, used under standardized preprocessing conditions. The goal was to examine the effects of both the architecture depth and transfer learning on the accuracy and robustness of classification. A collection containing pictures of cars, labeled according to the type of damages, was normalised, resized, and augmented to promote generalization. The difference in performance was found to be significant as a result of experimental tests. The Custom CNN performed at 47% accuracy, meaning that it is not effective to deal with the subtle and complex patterns of damages. ResNet18 increased accuracy to 69%, and it was again good clear damage, but inconsistent borderline classes. VGG16 proved to be better than the two with 82% accuracy and showed to be balanced in F1-scores and to be highly adaptable when conditions change. These results point to the efficacy of deep, and pre-trained CNNs in visual damage evaluating duties. In future, segmentation methods, ensemble modeling, and bigger and more varied datasets will be incorporated into the work in order to increase precision and applicability in a practical scenario. Moreover, the explainable AI will lead to better interpretation and confidence in automations. This study will add to already existing practical, AI-based insurance claims, fleet and vehicle inspections, and vehicular maintenance supporting more significant efficiency with precise and faster assessment of the vehicle condition at a lower cost.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Singh, Jaswinder UNSPECIFIED
Subjects:	Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence H Social Sciences > HG Finance > Insurance H Social Sciences > HD Industries. Land use. Labor > Specific Industries > Motor Industry
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	26 Aug 2026 08:59
Last Modified:	26 Aug 2026 08:59
URI:	https://norma.ncirl.ie/id/eprint/9645

Actions (login required)

View Item