Originality determination of vehicle body paints with deep learning approaches

The automotive industry acknowledges that paint plays a crucial role in the production of high-quality and reliable vehicles. Automotive body paint is not merely an aesthetic component but also serves the essential function of protecting the vehicle's exterior surface against environmental factors. Therefore, the originality and quality of automotive body paint hold significant importance for both manufacturers and consumers.Today, the process of detecting the originality of automotive body paint is generally performed during vehicle inspections using micron gauges that measure paint thickness, or through manual methods relying on the skill and experience of body repair experts. However, these methods are both time-consuming and prone to human error. With the rapid advancement of technology, the automotive sector faces a growing demand for more innovative and reliable methods in quality control processes.In this context, deep learning technologies have emerged as a promising solution for determining the originality of automotive body paint. Thanks to its ability to learn from large datasets, deep learning has the capacity to identify complex patterns and relationships. When combined with image processing techniques, deep learning models can analyze the originality and quality of automotive body paint with high accuracy.In this thesis, the effectiveness of a deep learning–based system in detecting the originality of automotive body paint is examined. The study aims to determine whether automotive body paint originates from its primary source through image analysis. In doing so, it seeks to provide an alternative approach to the commonly used methods for detecting paint originality.By employing image processing and deep learning methods, an ESA model was developed and trained on a dataset of 2,000 images obtained from factory-painted and repainted vehicles. The dataset was divided into 80% training and 20% validation sets, and the ESA models ResNet50, VGG16, InceptionV3, EfficientNetB0, and MobileNetV2 were tested. The highest accuracy rate, 98%, was achieved with InceptionV3. These results indicate that deep learning can be effectively used to determine the originality of automotive paint; however, larger datasets and different model architectures may be required to further improve accuracy.