<p>In the domain of traffic sign detection, the refined YOLOv8n algorithm has established itself as a predominant methodology. Nonetheless, under adverse weather conditions, such as fog, rain, nighttime, and within the confines of limited computational resources inherent to vehicular systems, prevailing SOTA models face challenges in attaining an optimal equilibrium between precision and the pursuit of a lightweight framework. In response to these challenges, we have devised LiteGhost-YOLO, a lightweight traffic sign recognition approach tailored for complex environments, underpinned by scale-aware mechanisms. This novel method not only accelerates the inferential proceedings of the model, but also significantly reduces computational complexity. In particular, we have pioneered a Ghost Reparameterized (GR) convolution technique that accelerates the inference process by synthesizing a ghost feature map in concert with a reparameterized tactic. Concurrently, we utilize a Scale-Aware Diffusion (SAD) methodology to effectively reduce model complexity through the amalgamation of multi-scale features. To further refine detection performance, we have incorporated a pruning module to excise channels with diminished weight relevance, thereby further streamlining the model while maintaining its detection fidelity. The efficiency of LiteGhost-YOLO is demonstrated through extensive evaluations on the Haze-CCTSDB2021, CCTSDB2021, and GTSDB datasets, achieving reductions of 85.0% in parameters, 65.4% in GFLOPs, and 76.2% in model size. After deploying the model on the Jetson TX2 mobile device, the frames per second (FPS) increased from 11.5 to 18.6.</p>