UAV and AI-Driven Approaches for Accurate Species Classification in Railway Trackside Vegetation Management

Railway trackside vegetation management is vital for safe and reliable operations. We explore integrating UAV and AI technologies to enhance this practice. Our study developed two AI models: a Mask R-CNN-based segmentation model and a feature classifier-based classification model, both using a ResNet50 backbone. Trained on extensive UAV imagery datasets annotated by domain experts, the models achieved promising results. The segmentation model achieved 78% average accuracy while validating with the ground truth, providing insights into vegetation density. In contrast, the classification model excelled with 96% average precision with the ground truth data, particularly in identifying prevalent vegetation species. Analysis reveals strong performance for specific species despite overall segment accuracy being lower. Ground truth data validation ensured robustness and accuracy. This research demonstrates the transformative potential of UAV and AI technologies in railway vegetation management, empowering operators with advanced tools for efficiency, safety, and sustainability.