An Innovative Pipeline For Corrosion Segmentation in Telecommunication Infrastructure

The detection of corrosion on metallic structures plays a crucial role in timely maintenance as corrosion can lead to major safety risks. However, developing an own corrosion detection application faces some challenges in datasets: the lack of available datasets, or unsuitable with specific problems. In this work, we proposed a quick, efficient, easy-to-use pipeline to segment datasets based on color filters. After obtaining a dataset consisting of 612 images annotated with mask corrosion, we experimented with this dataset on different segmentation models and achieved an mIoU score of 0.8664 with OCRNet.