EVF-YOLO: A Lightweight Network for License Plate Detection Under Severe Weather Conditions

License plate detection is crucial for intelligent supervision systems, particularly in port vehicle control and public safety maintenance. However, extreme weather conditions and harsh environments, especially at night, pose challenges to target detection and supervision tasks. To address this, a lightweight target detection algorithm, EVF-YOLO, based on improved YOLOv8, is proposed. This algorithm can be deployed in resource-constrained camera supervision environments. Firstly, foggy low-light images are synthesized in this paper using a distance-centered fogging method and superimposed Gaussian noise to enhance dataset generalization. Secondly, a low-light feature enhancement module is introduced, which combines attention mechanisms with C2f convolution to improve image feature extraction and suppress noise features. Moreover, an efficient feature fusion structure, V-GFPN, is proposed to remove redundant connections without compromising accuracy. Additionally, an EC-head lightweight detection head is introduced to enhance accuracy and reduce complexity of the algorithm. Experimental evaluation on the PT-IMG dataset demonstrates that the improved EVF-YOLO algorithm achieves better precision, recall, and average accuracy rate (98.5%) compared with baseline YOLOv8 and classical algorithmic models with reduced model size and parameter count, and offers promising possibilities for practical applications as well.