With the increasing amount of plastic waste being produced worldwide every year, effective recycling of this waste becomes increasingly important. With this research, we aim to introduce Instance Segmentation to this problem. We analyse the resulting Hyper Spectral Images of a Near Infrared Camera using the Mask R-CNN architecture. This was made possible by adding a convolutional layer to the beginning of Mask R-CNN for dimensionality reduction. The Mask R-CNN architecture and modifications have shown to handle dimensionality reduction and processing the Hyper Spectral Images. We found that Mask R-CNN shows promising results in creating instance segmentation masks with an average F1-score of 0.749 across all experiments and polymer types. Improvements could be made to the classification results as these results were most lacking. With visible evidence of the model confusing PE for PS in the second experiment. With this research we have shown promising results and that Instance Segmentation is well worth future research when tackling this subject.
