HuYOLO-NAS

This research addresses the challenges posed by the high environmental variability in waste disposal sites and the inherent inaccuracies in manual waste classification by introducing HuYOLO-NAS, an adaptive neural network model designed to enhance the precision of real-time classification of recyclable solid waste. The system integrates the YOLO-NAS architecture with the Hu moments algorithm to optimize object detection and spatial localization. The model was trained on the 'EcoSight' dataset, comprising 8,400 annotated images of paper, cardboard, PET plastic and hard plastic. Performance was quantitatively assessed using metrics such as precision, recall, F1 score, accuracy, and mean Average Precision (mAP), supplemented by confusion matrix analysis. The results underscore HuYOLO-NAS’s potential as an advanced solution for automated waste sorting, reducing the manual labor involved and mitigating sanitation risks, thus providing a robust foundation for future advancements in machine vision for waste management.