With global municipal solid waste generation projected to rise by 70% by 2050 [1], effective waste management has become a growing challenge, particularly due to the inefficiencies of traditional sorting methods. Our study proposes an AI-driven framework that integrates computer vision and generative large language models specifically, a fine-tuned Llama 3.2 Vision model to automate the identification, classification and quantification of waste. The goal of our project is to distinguish waste into categories such as organic, recyclable and non-recyclable, while also measuring visual properties like composition and volume, thus supporting optimized waste-to-energy processes. Unlike conventional models limited to classification, this approach addresses key challenges including heterogeneous waste types, imbalanced datasets, and lack of quantification. Through prompt engineering and optimization techniques, our model achieved 91.67% classification accuracy for organic waste, with precision and recall evaluated to highlight performance strengths and weaknesses across categories. The results demonstrate the potential of vision-language models to transform waste management by reducing human intervention, increasing sorting efficiency and contributing to sustainable engineering practices, ultimately advancing the state of knowledge in AI-powered environmental systems.
References:
[1] Kaza, S., Yao, L., Bhada-Tata, P., & Van Woerden, F. (2018). What a waste 2.0: A global snapshot of solid waste management to 2050. World Bank Publications.
