Explainable Deep Anomaly Detection with Sequential Hypothesis Testing for Robotic Sewer Inspection

TL;DR

This paper presents an explainable deep learning system for sewer pipe fault detection that combines spatial anomaly localization with temporal evidence aggregation, improving robustness and interpretability in robotic sewer inspections.

Contribution

It introduces a novel spatiotemporal anomaly detection framework using explainable deep learning and sequential hypothesis testing for automated sewer inspection.

Findings

Enhanced detection accuracy over baseline methods

Improved robustness to noisy data

Provides interpretable localization of anomalies

Abstract

Sewer pipe faults, such as leaks and blockages, can lead to severe consequences including groundwater contamination, property damage, and service disruption. Traditional inspection methods rely heavily on the manual review of CCTV footage collected by mobile robots, which is inefficient and susceptible to human error. To automate this process, we propose a novel system incorporating explainable deep learning anomaly detection combined with sequential probability ratio testing (SPRT). The anomaly detector processes single image frames, providing interpretable spatial localisation of anomalies, whilst the SPRT introduces temporal evidence aggregation, enhancing robustness against noise over sequences of image frames. Experimental results demonstrate improved anomaly detection performance, highlighting the benefits of the combined spatiotemporal analysis system for reliable and robust sewer inspection.