Fine-Tuning YOLOv8 for Enhanced Accuracy in Marine Environments

Marine ecosystems, vital for biodiversity, are threatened by plastic pollution. This study enhances underwater debris detection using YOLOv8, addressing poor image quality issues with contrast-enhanced histogram equalization (CLAHE) and gamma correction. The Adaptive Feature Aggregation (AFA) method improves detection by replacing the traditional YOLOv8 feature extractor. Despite pruning the model, transfer learning ensures high performance. Results show 97.8% detection accuracy, 97.4% mAP@50, and superior GPU speed. This research, adhering to ethical standards and plagiarism checks, aims to protect marine ecosystems using advanced ML and DL techniques.