Detecting district heating leaks in thermal imagery: Comparison of anomaly detection methods

Abstract

District heating systems offer means to transport heat to end-energy users through underground pipelines. When leakages occur, a lack of reliable monitoring makes pinpointing their locations a difficult and costly task for network operators. In recent years, aerial thermography has emerged as a means to find leakages as hot-spots, with several papers proposing image analysis algorithms for their detection. While all publications boast high performance metrics, the methods are constructed around very different datasets, making a true comparison impossible.

Using a new set of aerial thermal images from two German cities, this paper implements, improves, and evaluates three anomaly detection methods for leakage detection: triangle-histogram-thresholding, saliency mapping, and local thresholding with filter kernels. The approaches are integrated into a software pipeline with globally applicable pre- and postprocessing, including vignetting correction. While all methods reliably detect thermal anomalies and are suitable for automated leakage detection, triangle-histogram-thresholding is the most robust.