Hyper-Spectral Imaging for Overlapping Plastic Flakes Segmentation

TL;DR

This paper explores hyper-spectral imaging for segmenting overlapping plastic flakes, introducing a bitfield encoding method that enhances multi-label prediction capabilities in plastic sorting applications.

Contribution

It proposes a simple multi-label encoding approach, bitfield encoding, to improve segmentation of overlapping plastics in hyper-spectral images, even when trained on non-overlapping data.

Findings

Bitfield encoding outperforms baseline single-label methods.

The approach predicts multiple labels for overlaps with only non-overlapping training data.

Demonstrates potential for improved plastic sorting accuracy.

Abstract

Given the hyper-spectral imaging unique potentials in grasping the polymer characteristics of different materials, it is commonly used in sorting procedures. In a practical plastic sorting scenario, multiple plastic flakes may overlap which depending on their characteristics, the overlap can be reflected in their spectral signature. In this work, we use hyper-spectral imaging for the segmentation of three types of plastic flakes and their possible overlapping combinations. We propose an intuitive and simple multi-label encoding approach, bitfield encoding, to account for the overlapping regions. With our experiments, we show that the bitfield encoding improves over the baseline single-label approach and we further demonstrate its potential in predicting multiple labels for overlapping classes even when the model is only trained with non-overlapping classes.