VLM-PI: Power Inspection System Based on Visual Large Models

The inspection of large-scale power transmission towers is crucial for ensuring the safe, stable, and reliable operation of the power grid. Traditional manual inspections pose significant risks. To address this, recent research has explored the use of unmanned aerial vehicles (UAVs) to optimize the inspection process. The emergence of interaction methods between large models and UAVs has greatly enhanced the intelligence of UAV frameworks. Furthermore, the advent of large vision models has simultaneously tackled the challenge of lacking visual input in UAV control interactions and the demand for training resources in visual recognition, making fully automated UAV inspections feasible. This paper proposes a power infrastructure inspection system based on large vision models. The system leverages large vision models to provide primary control strategies for UAVs, complemented by traditional large models and deep learning networks to achieve UAV inspection control. Additionally, we propose a deep learning-based training strategy to improve the efficiency of training large vision models.