SDTP-ViT: Vision Transformers based obstruction classification in telecommunication pipes

Telecommunication pipes play a critical role in maintaining global communication infrastructure. These pipes are of small diameter, and obstructions within them can lead to significant service disruptions. Existing inspection methods are predominantly manual, labor-intensive, and time-consuming. To address these challenges, we are developing a robotic solution equipped with cameras and deep-learning algorithms for autonomous obstruction detection. However, there is currently no publicly available dataset for obstruction classification in telecommunication pipe environments. In this context, this paper presents the development of the SDTP-ViT model, a Vision Transformer (ViT)-based framework for obstruction classification in small-diameter telecommunication pipes (SDTP). We introduce the SDTP dataset, comprising 69,710 images categorized into five classes as a benchmark dataset. We use this dataset to train and evaluate the SDTP-ViT model with other state-of-the-art defect detection algorithms. The results show the SDTP-ViT model has performances comparable with the state-of-the-art with significant improvement in running time.