Machine learning on multisensor data from airborne remote sensing to monitor plastic litter in oceans and rivers (PlasticObs+)Christoph Tholen; Mattis Wolf; Carolin Leluschko; Oliver Zielinski
In: Proceedings of OCEANS 2023. OCEANS MTS/IEEE Conference (OCEANS-2023), June 5-8, Limerick, Ireland, Pages 1-7, ISBN 979-8-3503-3226-1, IEEE, 2023.
This paper presents the main ideas and initial findings of the PlasticObs+ project. The long-term goal of the project is to develop an airborne based method for monitoring plastic waste on the water surface. For this project, aircrafts usually applied for oil spill detection are used. Plastic waste detection and analysis is achieved using artificial intelligence (AI) within four different AI-systems. Furthermore, results from field tests were used to determine the limits of detectability of plastic waste from different altitudes. It was shown that both color and size of the items have an influence on the detectability. In addition, the underground plays an important role. A binary classifier, based on a Convolutional Neural Network (CNN) was trained to distinguish between images containing plastic and those not polluted. The accuracy of the CNN was 93.3 % while the accuracy of the labels generated by humans was 92.6 %.
PlasticObs_plus - Verbund - KI: PlasticObs_plus - Maschinelles Lernen auf Multisensordaten der flugzeuggestützten Fernerkundung zur Bekämpfung von Plastikmüll in Meeren und Flüssen