Technology Introduction:
Built a proprietary image dataset with over 5 million images of locally collected recyclable materials in Taiwan, combined with synthetic data for model training. The system is trained to recognize over 30 categories (such as various types of plastics, metals like iron and aluminum, paper, paper containers and others). The model can perform real-time detailed recognition based on visual characteristics (color, shape, material), food grade, brands, detect residue presence, and determine compliance with government recycling standards. The model achieves an average accuracy rate of over 90%, and in 2024, it was selected for the Nvidia Poster session at the Nvidia Developer Conference in San Francisco to showcase applications of synthetic data.
Industry Applicability:
Our goal is to build an AI-driven smart recycling ecosystem that connects households, sorting
facilities, and brand owners. By integrating smart recycling machines, AI-powered sorting systems,
and synthetic data, we enhance sorting accuracy, reduce labor costs, and increase the value of recycled
materials.
One of our key solutions is a fully automated, AI-powered sorting system designed for recycling
facilities, capable of processing 5,000 to over 30,000 metric tons of waste annually. It achieves twice the
throughput of traditional sorting lines while reducing labor demand by up to 70%.