HydroDetectNet: A Hybrid Deep Learning System for Robust Aquatic Macroplastic Monitoring

Macroplastic pollution is a critical threat to marine ecosystems, severely affecting their health and functionality. The ability of deep learning-powered object detection algorithms to identify macroplastics has been demonstrated. The primary objective of this model is to evaluate modern deep learning techniques specifically designed to tackle the challenges of automated floating macroplastic detection, providing a reliable, data-driven approach for environmental monitoring. Additionally, provides a constructive comparison for macroplastic detection, focusing on SSD, Faster R-CNN, and the proposed model. The purpose is to highlight the advantages and disadvantages of each approach while focusing on the ongoing development of effective detection methods. This paper decisively outlines the practical applications of various macroplastic detection models and represents the significant potential of convolutional autoencoders for enhancing environmental monitoring efforts. These approaches offer several advantages, such as automated feature learning, adaptability to diverse debris types, and high scalability. The results indicate that the proposed approach achieves an improvement of 5.6% mAP over SSD and 3.2% mAP over Faster R-CNN. Additionally, the convolutional autoencoder shows enhanced resilience to limited labeled data.