Enhanced detection and classification of microplastics in marine environments using deep learning

Microplastics (MPs) pose a growing environmental threat due to their accumulation and ecological impact. This study aimed to overcome the limitations of traditional methods, which are labor-intensive and prone to errors, in order to detect and classify MPs more effectively against marine pollution. We assessed object detection and classification algorithms: YOLOv8x, YOLOv8x (with augmentation), YOLOv8m, YOLOv8m (with augmentation), YOLO-NAS-L, and YOLO-NAS-L (with augmentation), focusing on four MP morphologies: fiber, film, fragment, and pellet. The dataset was divided into 80 % for training (320 images), 20 % for validation (80 images), and a fixed testing set of 200 images. The images were augmented using rotation (+25 degrees and - 25 degrees), resizing (640 x 640 pixels), zooming, auto-orient strips, flipping, and noise application. This expanded the training set by 300 %, resulting in a total of 1400 images. The YOLOv8 models, particularly when augmented, outperformed the YOLONAS-L models in both mAP@0.5 and precision across all categories. Notably, YOLOv8x achieved an exceptional 99.0 % in both precision and mAP@0.5, with an impressive inference time of only 1.2 ms per image. The implementation of augmentation significantly enhanced detection accuracy across various models. With augmentation, YOLOv8x, YOLOv8m, and YOLO-NAS-L consistently achieved precision levels exceeding 99 %. For real-time applications, YOLOv8x was selected for the web application designed to detect and classify MPs, providing a more accurate and efficient solution compared to conventional methods. This model serves as a valuable resource for researchers in MP analysis, improving accuracy and reliability in environmental monitoring.