Automated Marine Debris Detection and Classification Using Deep Learning

Detection of aquatic debris is important for environmental surveillance and pollution control, being a bulwark to ecosystems, biodiversity, and climate stability. A deep-learning-based environmental framework is proposed to integrate two ResNet structures, ResNet-34 and ResNet-101, for debris detection and its corresponding localization from satellite images. Preprocessing and data augmentation techniques are thus applied to make the model more robust and generalizable. The model sets up a dual-branch structure aimed at the concurrent processing and mutual optimization of classification and localization tasks. Training was performed with AdamW optimizer and BCE Loss. The model can achieve a precision of 0.91 and an F1-score of 0.88, showing strong detection ability. The geo relocation-based automated alarm system further harnesses the coordinates from detections to notifying local authorities via email, setting the ground for swift intervention and clean-up. This scalable and automated solution takes environmental surveillance many miles further, prepares the ground for policy implementation, fast-track responses, and is an important milestone toward sustainable ocean management and data-informed conservation efforts.