EVALUATION OF AREA AND TYPES OF FLOATING MACROPLASTICS IN RIVERS DUE TO DEEP LEARNING

Abstract

 Marine plastic wastes have been mostly originated from inland, and it is important to monitor macroplastics inflow into oceans via rivers. This study aims to develop a new image processing to capture the area and types of macroplastics with deep learning. The learning data for floating debris were collected with the field test under normal and flooding conditions. CNN and YOLO were applied to find the area and types of macroplastics. The results indicated that the CNN and YOLO can capture acceptably the area and types of macroplastics in normal flow condition. It is noted that add of the learning data under flooding conditions can greatly improve the accuracy of distinguishing the types of macroplastics by YOLO.