Solid waste detection in context: Merging Datasets and using Open-Set Vision Models

Charisis, Alexandros Sokratis (2024) Solid waste detection in context: Merging Datasets and using Open-Set Vision Models. Master's Thesis / Essay, Artificial Intelligence.

Preview	Text mAI2024CharisisAS.pdf Download (56MB) | Preview
	Text toestemming.pdf Restricted to Registered users only Download (467kB)

Abstract

Waste management gets harder the more cities grow and plastic waste degradation leads to unwanted results like ending up in the top of the food chain, with tremendous impact on living organisms and aquatic ecosystems. Efficient waste detection, collection, classification, segregation, and recycling are all important for reducing the amount of waste ending up in landfills and tackle the issue of degra- dation of plastic waste ending up in our ecosystems. Existing publicly available datasets of 2D solid waste images are small in size and not diverse enough to cover all natural settings. This thesis proposes the harmonious merging of three public solid waste datasets to increase the quantity and variability of the training data. It compares various computer vision models like YOLO-NAS, Grounding DINO, YOLO-World and Segment Anything Model on solid waste detection through 2D image data. Qualitative analysis showed a robust combined multipurpose solid waste dataset, poor performance for YOLO-NAS, medium to high performance for YOLO-World/Grounding DINO respectively and high performance for Segment Anything model. Findings showed that open-set vision language mod- els can accelerate image annotation and as a result the automation of solid waste detection.

Item Type:	Thesis (Master's Thesis / Essay)
Supervisor name:	Valdenegro Toro, M.A. and Sabatelli, M.
Degree programme:	Artificial Intelligence
Thesis type:	Master's Thesis / Essay
Language:	English
Date Deposited:	19 Jul 2024 11:37
Last Modified:	19 Jul 2024 11:37
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/33537

Actions (login required)

View Item