AquaGuardian: A Deep Learning Approach to Detect Deep Underwater Litter

Abstract

"The issue of underwater sinking litter is a significant concern for both marine ecosystems and human activities. To address this challenge, AquaGuardian introduces an innovative system that utilizes deep learning techniques. The abstract emphasizes the critical need for advanced detection technologies to protect aquatic environments. The methodology revolves around selecting a suitable technology stack, including PyTorch for dynamic neural network architectures, Python as primary programming languages, and the TrashCan-Instance dataset from JAMSTEC for training. Our focus during implementation was on developing a hybrid deep-learning model using pre-trained CNN models RTMDet and YOLOv8. Data preparation involved preprocessing the AquaGuardian dataset with Roboflow, while we fine-tuned model architectures to optimize detection performance. Evaluation metrics like precision, recall, and F1 score were utilized to gauge the effectiveness of the models. Initial results show promising performance from the AquaGuardian model in detecting underwater debris. These early findings suggest that AquaGuardian has the potential to be a reliable tool for underwater debris detection, aiding environmental conservation efforts and enhancing marine safety."