<p>This paper addresses the challenges of insufficient detection accuracy and limited scene coverage in vehicle high-beam headlight detection under adverse weather conditions such as rain and snow. To tackle these issues, this paper proposes HBD-NanoDetPlus, a lightweight and efficient detection model designed for complex environments. In the backbone, a Multi-scale Convolution Feature Enhancement (MCE) module and a Coordinate Attention (CA) module are designed to improve feature extraction. The MCE module captures features at multiple receptive fields through parallel convolutions and enhances high-beam representations via feature fusion, while the CA module strengthens spatial and channel attention to improve detection accuracy. In the neck, a Squeeze Excitation and Convolution (SEC) module is introduced to further enhance feature representation. Additionally, we have collaborated with traffic management departments and enterprises to construct a medium-scale data set covering diverse weather conditions, including 10,000 continuous videos from 11 monitoring sites across 4 provinces and 8 cities in China. Experimental results demonstrate that HBD-NanoDetPlus achieves 93.5% mAP@0.5 and 64.8% mAP@0.5:0.95 with only 2.47M parameters and 3.58 GFLOPs, outperforming several mainstream detectors. Ablation and generalization experiments further validate the effectiveness and robustness of the proposed method.</p>