| Meno: | Dávid |
|---|---|
| Priezvisko: | Pitoňák |
| Názov: | Sorting plastic waste with AI |
| Vedúci: | doc. RNDr. Zuzana Černeková, PhD. |
| Rok: | 2026 |
| Kµúčové slová: | hyperspectral imaging, plastic waste classification, spectral band reduction |
| Abstrakt: | Reliable and efficient automated sorting of plastic waste is important for recycling, since manual sorting is demanding and difficult to keep consistent at a larger scale. Hyperspectral imaging in the near-infrared range offers a way to distinguish plastic materials from their spectral response and can support more objective material recognition. This diploma thesis studies the classification of five common types of plastic waste from hyperspectral data in the near-infrared range. The aim is to examine what can be achieved under limited data and incomplete annotations by using several machine-learning approaches. The results show that some of the implemented approaches can use a reduced number of spectral bands without a major loss of classification performance. They also show that data quality and data quantity play a major role in reliable evaluation, and under these limitations visual inspection of the classified images still reveals local errors. The results further show that models trained on one dataset can also be applied to another dataset, but their accuracy is sensitive to differences in how the data are captured. They also show that better object separation in the patch-based approach leads to stronger classification accuracy. |
Súbory diplomovej práce:
Autor nedal súhlas so zverejnením svojej diplomovej práce.
Súbory prezentácie na obhajobe: