DEH-YOLO: A Faster and More Efficient Model for Vehicle Damage Detection Task

In the context of the rapid development of the new energy industry and the automotive sector, the market size of the auto insurance industry has expanded significantly. Vehicle damage assessment, a critical component of insurance claims processing, remains highly dependent on manual inspection, leading to high assessment costs, low efficiency, and strong subjectivity. This necessitates the adoption of intelligent technologies for transformation. In recent years, with the rapid advancement of artificial intelligence, deep learning models have been increasingly applied to vehicle damage recognition. However, existing methods face challenges such as low detection accuracy and high computational overhead, limiting their practical application in real-world damage assessment scenarios. To address these issues, this study investigates vehicle damage recognition based on a self-constructed damage dataset and proposes an improved YOLOv8 object detection algorithm tailored to vehicle damage characteristics. The model is optimized through strategies such as enhancing convolutional operations in the backbone network, incorporating a multi-scale attention mechanism, and refining the detection head. Experimental results demonstrate that the proposed method significantly outperforms the baseline model and other object detection models in terms of detection accuracy.