Applying deep learning methods for the bridge-based monitoring of floating macroplastics on rivers

In this study we propose a framework to automatically monitor macroplastic loads in rivers using Deep Learning. The approach was evaluated with a measurement campaign at the Rhine river (in Koblenz, Germany). An RGB camera was installed on a bridge and captured images of the objects in the river. Various plastic and vegetation objects in different degradation states were introduced upstream and recollected downstream. Our dataset consists of about 800 images with objects labelled in three categories (plastic bottles, plastic litter, vegetation). The pre-trained YOLOv5 network showed promising validation results with a mean average precision (mAP@0.5) of about 94 %.