Fuel Consumption Prediction from Car Images Using CNN-Based Deep and Transfer Learning for Energy Efficiency and Sustainability

Improving energy efficiency and promoting sustainable transportation have become critical goals in combating climate change and reducing environmental impact. In this study, we propose a novel approach for estimating car fuel consumption directly from exterior car images by facilitating deep learning and transfer learning techniques. Generating a publicly available car image dataset with associated fuel consumption labels, we develop and evaluate several convolutional neural network (CNN) architectures, including pre-trained models such as VGG16, and MobileNetV2, fine-tuned on the task of fuel consumption prediction. Our approach aims to provide an automated, visionbased solution to estimate vehicle energy efficiency, which could support regulatory enterprises, car manufacturers, and consumers in making informed decisions. Experimental results demonstrate the potential of transfer learning in capturing visual cues correlated with fuel consumption, achieving promising prediction performance across multiple model configurations. This study contributes to the intersection of computer vision and environmental sustainability by offering an alternative data-driven tool for vehicle efficiency assessment.