University of Melbourne

Plastic pollution is a critical global issue impacting the environment, wildlife, and human health. Efficient recycling requires accurate identification of different plastic types like HDPE, LDPE, PET, PP, and PS. Hyperspectral imaging offers superior material detection compared to traditional imaging by capturing both spatial and spectral data, but its high computational demand limits practical use. Standard RGB imaging is faster but lacks the spectral depth needed for precise classification. This study addresses the underexplored area of optimizing minimal-band hyperspectral composites for object-level plastic detection. 

This study is trying to optimize spectral band selection by creating pseudo-RGB composites from hyperspectral data, enabling efficient and accurate object-level detection of key plastic types (HDPE, LDPE, PET, PP, PS) using advanced object detection models like YOLOv11, RetinaNet, RTDETR, and Faster RCNN.