Drone-Based Inspection of Broken and Defected Pipes on Metal Roofs

Abstract

The roofs of large industrial complexes are subject to various types of damage caused by weather, moisture, corrosion, and vibrations. The detection of such damage, particularly long and thin pipe structures, is challenging due to the high number of lines and edges present in the image. Traditional image processing methods and object detection models have limited success in identifying pipe defects due to weak appearance clues. Our approach addresses this issue by using Spatial CNN (SCNN) to capture the spatial relationships of pixels across the image, enabling the detection of long continuous shape structures. We applied our approach to a dataset of drone imagery of industrial roofs and achieved promising results in detecting pipe defects. Our methodology outperforms traditional image processing methods by a large margin. In conclusion, our approach provides a promising solution for automated pipe defect detection on industrial roofs, which can be extended to other similar scenarios.