AI-Powered Maintanence

AI-powered maintenance refers to the use of AI techniques and algorithms to optimize and automate the maintenance processes to enable proactive and efficient maintenance practices. Our study involves these areas:

1. Road crack detection: Road crack detection involves classifying each pixel as either a crack or not, which can be formulated as a semantic segmentation problem. We propose a novel model that can simultaneously address aerial road extraction, blood vessel segmentation, and road crack detection, unlike most studies that treat these tasks individually. By leveraging a single model, we demonstrate the potential to detect curved lines and continuous edges, introducing the concept of curvilinear structure segmentation (check our paper at ICCV21 and code).
2. Metallic corrosion detection: Metallic corrosion detection involves patch-wise classification with three labels: corrosion, non-corrosion, and background. We recognize metallic corrosion detection as learning with noisy data problem, since annotating corrosion patches is challenging due to diverse corrosion levels, regions, and vague edges. To address this, we have created a new benchmark comprising ~10k patches annotated with corrosion, non-corrosion, and background labels (check the details).