Investigation of Deep Learning Models for Vehicle Damage Classification
No Thumbnail Available
Authors
Rababaah, Aaron Rasheed
Issue Date
2023-01-01
Type
Journal Article
Language
Keywords
Alternative Title
Abstract
This paper presents a study of Deep Learning models of convolution neural networks (CNN) applied to vehicle damage classification (VDC). Number of real-world domains may benefit from the proposed solution such as: car rental, auto dealerships, auto insurance businesses, etc. DL has significant advantages over conventional machine learning (ML) models. The primary advantage of DL models is their ability to learn and extract features automatically as opposed to hand-crafting them as in ML models. The study used MatLab as the development and testing environment. A CNN based architecture was constructed which comprised typical DL layers of: raw image input, convolution, activation, pooling, flattening and fully-connected layers. The study used real world images collected from online sources to conduct the experimental work to validate the proposed model. The results showed that the overall average accuracy of all tested models was 91.8% and the best model produced an impressive accuracy of 99.4%. Furthermore, confusion matrix metrics were used to further validate the best performing model and all metrics such as accuracy, precision, sensitivity, specificity were reliable.