Wei, Yin (Beihang University), Liu, Wentao (Beihang University), Yu, Guizhen (Beihang University), Chen, Peng (Beihang University)

Leveraging Light-Weight Multi-Modal Large Model for UAV-Based Highway Inspection Systems

Scheduled for presentation during the Video Session "On-Demand Video Presentations" (VP-VP), Saturday, November 22, 2025, 08:00−18:00, On-Demand Platform

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on April 2, 2026

Abstract

While Unmanned Aerial Vehicle (UAV)-based highway inspection has emerged as a promising intelligent detection method, current UAV video analysis still depends on manual identification. Compared to automated multi-modal large model-based detection, manual identification exhibits significant limitations in detection accuracy and operational efficiency. These challenges show the urgent need for advanced multi-modal large models to enable intelligent highway inspection. This paper presents a lightweight multi-modal large model for UAV-based highway inspection systems. The proposed framework comprises two key components: (1) Forward  Kullback-Leibler divergence-based knowledge transfer from a teacher model with 7B parameters to a smaller model with 3B parameters, implementing dynamic weight allocation through adaptive distillation; (2) Parameter-efficient adaptation via Low-Rank Adaptation (LoRA) fine-tuning using highway-specific task matrices. Experimental evaluations conducted the optimized model achieved a BLEU score increase of 11.59% and a ROUGE score increase of 7.89%, while reducing runtime by 2.56 seconds. Overall, by maintaining high precision while reducing computational overhead, the optimized model meets the requirements for practical application scenarios.