PIPE-CovNet: Automatic In-Pipe Wastewater Infrastructure Surface Abnormality Detection Using Convolutional Neural Network

Abstract

Regular inspection of multi-billion dollar wastewater pipe infrastructure is crucial to any city around the globe. Traditional processes of inspection are laborious, time-consuming, and prone to human errors, such as the manual assessment of video and image sources obtained by closed-circuit television (CCTV). These limitations can be circumvented through the utilization of novel deep learning techniques. In this letter, we propose the PIPE-CovNet model, leveraging a convolutional neural network for automatic pipe surface abnormality detection. The proposed deep learning framework was trained and evaluated on a publicly accessible data set. Evaluation results indicate the PIPE-CovNet achieves 82 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> accuracy and F1-score 0.82. In addition, the PIPE-CovNet outperformed other comparable deep learning models in terms of accuracy by at least 5 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> and F1-score by at minimum 8 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> .