AI on the Edge: A Novel Approach to Detect Waste on Water Bodies

Marine environments are being increasingly burdened by waste from heavy industries and the large scale disposal of electronics and plastics into waterways that flow into the oceans. Recent advancement in technologies like real time image processing, machine learning and edge computing has made it feasible to develop commercially viable solutions for automated identification of garbage on water bodies. This paper explores image processing as a targeted approach for the on-field identification of water pollutants. A unique aspect of this approach was to demonstrate the effectiveness of its implementation on standalone edge devices viz. ones that could be used in remote locations with limited compute capabilities and no wireless connectivity. The study demonstrates this use-case with TensorFlow as an AI framework and Raspberry Pi 4B as an edge device. The model achieved a mean Average Precision (mAP) of 51.20%, validating its performance in identifying water pollutants.