Development of Extendable Open-Source Structural Inspection Datasets

Abstract

Recent infrastructure inspection has used deep-learning models to enhance and augment typical inspection tasks such as detecting and quantifying damage. One of the issues with this trend is that deep-learning models typically require a significant amount of data. In a data domain such as structural inspection, publicly accessible data are difficult to find, making the advancement of this research slower. Therefore, we set out to acquire bridge inspection data by selectively extracting candidate images from hundreds of thousands of bridge inspection reports from the Virginia Department of Transportation. Using this rich source of diverse data, we refined our collected data to develop four high-quality, easily extendable, publicly accessible datasets, tested with state-of-the-art models to support typical bridge inspection tasks. The four datasets: labeled cracks in the wild, 3,817 image sets of semantically segmented concrete cracks taken from diverse scenery; 3,817 image sets of semantically segmented structural inspection materials (concrete, steel, metal decking); 440 images of finely annotated steel corrosion condition state (good, fair, poor, severe); and 1,470 images of fatigue-prone structural steel bridge details (bearings, gusset plates, cover plate terminations, and out-of-plane stiffeners) for object detection. To ensure the extendibility of the datasets, the authors have proposed annotation guidelines to maintain consistent growth through annotation collaboration. Researchers can use these trained models and data for auxiliary inspection tasks such as damage detection, damage forecasting, automatic report generation, and, coupled with the assistance of unmanned aerial systems, for autonomous flight path planning and object avoidance. The procedures, concepts, and repositories provided in this paper will help to set a course for the advancement of better detection models using high-quality accessible and extendable datasets. Researchers can use these datasets and trained models for auxiliary inspection tasks such as damage detection, damage forecasting, automatic report generation, and, coupled with the assistance of unmanned aerial systems, for autonomous flight path planning and object avoidance. The models may also be used as starting points for researchers to extend or begin their own use-case datasets for this domain or similar domain of data. The authors have presented three main reasons for including annotations guidelines for any public dataset: to establish consistent annotations for the training and testing data; to clearly present the expected model prediction; for guiding future collaborators. Annotation guidelines are a practical tool for any researcher procuring a dataset for supervised machine learning.