A Model for Infrastructure Detection along Highways Based on Remote Sensing Images from UAVs

Area of Science:

• Civil Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Highway infrastructure requires efficient monitoring due to outdoor exposure and damage.
• Existing drone-based detection models struggle with variable conditions and small targets.
• A lack of comprehensive UAV-captured highway infrastructure datasets hinders practical application.

Purpose of the Study:

• To develop an advanced drone-based model for precise highway infrastructure detection, classification, and localization.
• To address challenges posed by image quality, complex backgrounds, and small targets in UAV imagery.
• To create a valuable dataset for training and evaluating highway infrastructure detection models.

Main Methods:

• Proposed a multi-classification infrastructure detection model integrating multi-scale feature fusion and an attention mechanism.
• Replaced the CenterNet backbone with ResNet50 for enhanced fine-grained feature generation.
• Developed a novel highway infrastructure dataset by filtering and annotating laboratory-captured images.

Main Results:

• The proposed model achieved a mean Average Precision (mAP) of 86.7%, a 3.1 percentage point improvement over the baseline.
• The model demonstrated superior performance compared to other detection models in experimental evaluations.
• Enhanced detection of small targets and improved focus on relevant infrastructure features were observed.

Conclusions:

• The developed model significantly enhances the accuracy and efficiency of highway infrastructure monitoring using drones.
• The integration of multi-scale feature fusion and attention mechanisms is effective for complex visual detection tasks.
• The creation of a dedicated dataset facilitates future research and development in intelligent road infrastructure management.