Enhancing waste recognition with vision-language models: A prompt engineering approach for a scalable solution

Area of Science:

• Computer Vision
• Artificial Intelligence
• Waste Management

Background:

• Traditional computer vision models struggle with diverse waste images in material recovery facilities.
• Frequent fine-tuning and data augmentation are resource-intensive and unsustainable for scalable waste classification.

Purpose of the Study:

• To implement and evaluate state-of-the-art Vision-Language Models (VLMs) for waste image classification.
• To explore VLM adaptability in data-scarce settings using zero-shot, few-shot, and fully supervised learning.
• To investigate the impact of prompt engineering on VLM performance.

Main Methods:

• Utilized multimodal image classification with Vision-Language Models (VLMs).
• Assessed zero-shot, few-shot, and fully supervised learning capabilities.
• Employed targeted prompt engineering to optimize VLM performance.
• Evaluated models based on accuracy and inference speed trade-offs.

Main Results:

• Zero-shot waste image classification accuracy improved from 82.71% to 90.48% with prompt engineering.
• Fully supervised classification accuracy reached 97.18% with the optimal VLM.
• VLMs demonstrated adaptability in data-scarce scenarios and scalability.

Conclusions:

• Vision-Language Models present a promising, scalable solution for waste image classification.
• Targeted prompt engineering is an effective strategy for enhancing VLM performance in waste management applications.
• This research advances waste management through AI-driven image classification.