A quantum-inspired YOLOv8 vision framework for high-precision marine debris detection using enhanced Quantum ResNet50 feature encoding

Marine debris continues to pose a serious threat to marine ecosystems, biodiversity, and coastal economies, creating an urgent need for reliable and real-time detection mechanisms. Despite recent progress, conventional deep learning approaches often exhibit reduced performance in complex marine environments characterized by low visibility, occlusions, and inconsistent illumination. In this work, a hybrid quantum-inspired deep learning framework is proposed to address these challenges by integrating a Quantum ResNet-based feature encoding strategy with the YOLOv8 object detection architecture. The proposed model enhances feature representation by combining classical convolutional features with quantum-inspired transformations, implemented through angle encoding and parameterized quantum circuits. The framework was trained and evaluated on a dataset comprising 3,000 annotated images of marine debris, categorized into plastic, metal, glass, and fishing nets. Experimental results demonstrate that the proposed approach achieves a precision of 93.7%, a recall of 89.2%, and a mAP@0.5 of 91.3%, consistently outperforming the baseline YOLOv8n model by approximately 3%–4% in detection performance. Further analysis under challenging conditions, including Gaussian noise, motion blur, and underwater haze, reveals improved robustness, indicating that the incorporation of quantum-inspired feature transformations enhances the model’s discriminative capability. Overall, the findings suggest that integrating quantum-inspired representations with deep learning models provides a promising pathway for improving both accuracy and resilience in marine debris detection, thereby supporting the development of more effective intelligent environmental monitoring systems.