Publikation
On the Development of a Pixel-Wise Plastic Waste Identification System for Multispectral Remote Sensing Applications
Christoph Tholen; Eike Rodenbäck; Lars Nolle; Robert Rettig; Frederic Stahl
In: Max Bramer; Frederic Stahl (Hrsg.). Artificial Intelligence XLI. SGAI International Conference on Artificial Intelligence (AI-2024), Cham, Pages 47-60, ISBN 978-3-031-77915-2, Springer Nature Switzerland, 2025.
Zusammenfassung
This paper presents the development of a pixel-wise plastic waste identification system for multispectral remote sensing data, based on artificial intelligence methods. The system will be used as part of a two stage approach to identify and quantify plastic waste in waterways and onshore utilizing airborne based remote sensing and Artificial Intelligence. This work investigates the performance and generalization capabilities of Artificial Neural Networks (ANN), Random Forests (RF), Support Vector Machines (SVM), Logistic Regression (LR), and Decision Trees (DT) on two different multispectral datasets. All models are trained and tested on a dataset with artificial plastic waste targets, covering three different undergrounds, i.e. sand, grass and water. To investigate the generalization capabilities of the models, further tests on a dataset from a real landfill are conducted without retraining. On dataset #1, ANN and RF demonstrated superior performance, both achieving 98.4% accuracy, followed closely by DT at 97.4%. SVM and LR showed lower but comparable accuracies of 87.7% and 87.4%, respectively. RF exhibited the best generalization with 90.4% accuracy, while SVM showed improved relative performance at 88.1% on dataset #2. Furthermore, it was shown that an ensemble of all five methods achieved 91.3% accuracy on dataset #2 without retraining, demonstrating a clear trade-off between false positives and false negatives.