Trash screens are used to prevent debris from entering critical parts of rivers. However, debris can accumulate on the screen and generate floods. This makes their monitoring critical both for maintenance and flood modeling purposes (e.g., local forecasts may change because the trash screen is blocked). We developed three novel deep learning methods for trash screen maintenance management consisting of automatically detecting trash screen blockage using cameras: a method based on image classification, a method based on image similarity matching, and a method based on anomaly detection. To facilitate their use by end users, these methods are designed so that they can be directly applied to any new trash screen camera installed by the end users. We have built a new dataset of labeled trash screen images to train and evaluate the efficiency of our methods, in terms of both accuracy and implications for end users. This dataset consists of 80,452 trash screen images from 54 cameras installed by the Environment Agency (UK). This work demonstrates that trash screen blockage detection can be automated using trash screen cameras and deep learning, which could have an impact on both trash screen management and flood modeling.

  • Development and analysis of three deep learning-based trash screen monitoring methods using cameras, tailored for various investment levels.

  • Binary classification: ready-to-use, no additional labeling or retraining required.

  • Image similarity matching: requires minimal user annotation for improved accuracy.

  • Anomaly detection: able to use unlabeled datasets.

  • Network and dataset provided to support further research.

This content is only available as a PDF.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY 4.0), which permits copying, adaptation and redistribution, provided the original work is properly cited (http://creativecommons.org/licenses/by/4.0/).

Supplementary data