Automatic UAV-based Airport Pavement Inspection Using Mixed Real and Virtual Scenarios

TL;DR

This work tackles the challenge of UAV-based pavement inspection under data scarcity by generatING a synthetic, hyperrealistic UAV dataset in Unreal Engine 5/AirSim and fusing it with real crack data to train an EfficientNet-B1 with an FPN segmentation head for embedded, real-time crack detection. The approach emphasizes a high-resolution input to preserve small crack details and employs Dice loss to handle severe class imbalance. Experiments on public crack datasets and self-annotated runway imagery show that mixing synthetic with real data improves cross-domain performance, particularly in recall and generalization to new environments, while maintaining feasible on-device latency with TensorRT on Jetson platforms. The work contributes a practical pipeline for synthetic data generation, a lightweight yet effective segmentation model, and insights into domain transfer for airport pavement maintenance.

Abstract

Runway and taxiway pavements are exposed to high stress during their projected lifetime, which inevitably leads to a decrease in their condition over time. To make sure airport pavement condition ensure uninterrupted and resilient operations, it is of utmost importance to monitor their condition and conduct regular inspections. UAV-based inspection is recently gaining importance due to its wide range monitoring capabilities and reduced cost. In this work, we propose a vision-based approach to automatically identify pavement distress using images captured by UAVs. The proposed method is based on Deep Learning (DL) to segment defects in the image. The DL architecture leverages the low computational capacities of embedded systems in UAVs by using an optimised implementation of EfficientNet feature extraction and Feature Pyramid Network segmentation. To deal with the lack of annotated data for training we have developed a synthetic dataset generation methodology to extend available distress datasets. We demonstrate that the use of a mixed dataset composed of synthetic and real training images yields better results when testing the training models in real application scenarios.

Figures (6)

• Figure 1: Designed virtual airport environment.
• Figure 2: Synthetic pavement defect generation process.
• Figure 3: RGB and corresponding annotation samples from the generated synthetic dataset. Samples (\ref{['fig:syn_dataa']}), (\ref{['fig:syn_datab']}) and (\ref{['fig:syn_datac']}) correspond to simulations at dusk, noon rain and at night, respectively.
• Figure 4: EfficientNet-FPN segmentation network architecture.
• Figure 5: Detail of the annotation from the airport captures. (\ref{['fig:dataexamplea']}) and (\ref{['fig:dataexamplec']}) are the images, and (\ref{['fig:dataexampleb']}) and (\ref{['fig:dataexampled']}) are the annotations.