Regenerative Energie

Biogas

Biomasse

Erneuerbare allgemein

Energieeinsparung

Erdwärme

Grünes Geld

Klimaschutz

Kraft-Wärme-Kopplung

Meeresenergie

Mobilität

Photovoltaik

Solarthermie

Wärmepumpen

Wasserkraft

Windenergie

content-Partner

Comparative Analysis of Transfer and Continual Learning for Vision-Based Particle Classification in Plastics Sorting for Recycling

In this study, we evaluate the effectiveness of transfer and continual learning techniques for vision-based trash particle detection and classification in plastics recycling. This task poses unique challenges for vision-based methods due to the great variety of particles in recycling material flows, their variability over time, and the lack of real recycling industrial datasets available for research.

We use the publicly available computer vision research dataset CIFAR100 to assess model performance, calibrate the model hyperparameters, simulate and explore transfer and continual learning techniques. According to our results, the main factors affecting test accuracy are the following: calibration of hyperparameters, such as the learning rate (up to +13%), bigger batch sizes of only new data (up to +9%) or mixed with previous data with more frequent retraining phases (+4%) and differing continual learning strategies (±3%). Based on these assessments, we provide evaluations on our DWRL dataset, which features plastic trash particles on a moving conveyor belt and provide recommendations for the implementation of continual learning frameworks in the context of plastics recycling.

Copyright:	© Lehrstuhl für Abfallverwertungstechnik und Abfallwirtschaft der Montanuniversität Leoben
Quelle:	Recy & Depotech 2024 (Dezember 2024)
Seiten:	8
Preis:	€ 4,00
Autor:	Somayeh Shami Benedikt Haecker Julian Aberger Jesus Pestana Michael Krisper
	Diesen Fachartikel kaufen... (nach Kauf erscheint Ihr Warenkorb oben links)
	Artikel weiterempfehlen
	Artikel nach Login kommentieren

---