AI-powered parameter extraction process
The system automatically extracts key data, such as material composition and usage history, from unstructured user-generated content like product photos of tags and descriptions on flea market apps. This data is then linked with greenhouse gas (GHG) emission factor databases to calculate the carbon footprint for each individual second-hand garment.
A research group from the University of Tokyo and Mercari, Inc. has established a new framework that uses AI to automatically calculate the environmental impact of individual second-hand garments listed on flea market apps. By accurately extracting product-specific data from user posts, the system overcomes a major barrier in product-level environmental assessment and is expected to accelerate information transparency in the circular economy.
The findings could provide valuable insights for automating the laborious data collection process for environmental impact assessments, a key step toward promoting a more sustainable society.
The practice of calculating a product’s environmental impact across its entire life cycle, from raw material sourcing to disposal, is known as Life Cycle Assessment (LCA). While crucial for both businesses and consumers, LCA has long been hampered by the immense time and cost required to collect detailed primary data for each product. This “data collection wall” is even more pronounced for second-hand items, as each one is unique in its condition and history, making individual assessment seem impractical.
In the current study, the research team, led by Professor Yoshihiro Kawahara at the University of Tokyo’s Graduate School of Engineering, leveraged a vision-language model (VLM) to break through this barrier. A VLM is an AI model that can understand and process information from both images and text simultaneously.
“The ‘data collection wall’ has been a major bottleneck in accurately assessing the environmental impact of individual products, especially for unique second-hand items,” said Kusahata Sumiaki, the study’s lead author and a master’s student in Kawahara’s laboratory. “Our AI-powered approach automates this painstaking process, unlocking the potential to visualize the environmental value of each item circulating in the reuse market.”
The team aimed to verify how accurately AI could extract the primary data needed for professional LCA calculations from the kind of unstructured, and often ambiguous, information that users post on flea market apps. They built a system that uses OpenAI’s GPT-4o to automatically extract four key parameters that significantly influence a product’s GHG emissions: material composition, size, washing method, and estimated usage frequency.
Automated calculation of environmental impact
For instance, even if a user does not explicitly state the material in the description, the AI can read the text on the garment’s tag from a photo. It can also infer the item’s past usage frequency from descriptive phrases like “only tried on” or “worn a few times.” The extracted information is then standardized and linked with existing GHG emission factor databases to calculate the carbon footprint.
To validate the system’s accuracy, the researchers tested it on 3,500 “tops” category items from Mercari. The results were reviewed by human evaluators, who confirmed that the AI achieved high precision, successfully identifying material composition in 81.6% of cases and size in 92.3% of cases. The accuracy for the washing method, however, remained a challenge at 56.6%.
Through their analysis, the team also gained a deeper insight. Previous studies on apparel’s environmental impact often focused on consumer behavior during the “use phase,” such as washing frequency or the use of dryers. However, by analyzing a diverse set of 3,500 real-world items, this study revealed that a garment’s “material composition” has a far greater influence on its total carbon footprint than variations in its estimated usage history.
“Our analysis of real-world market data suggests that the diversity of materials—including various blends—and the associated production impacts are the primary drivers of a garment’s carbon footprint,” said Kawahara. “This finding underscores the critical importance of material selection and design at the very beginning of a product’s life cycle.”
In the future, the team believes this automated assessment framework can be applied to product categories beyond apparel. By streamlining data collection, the method could significantly support corporate environmental disclosures, particularly in calculating Scope 3 emissions, which include the entire value chain. Ultimately, the researchers hope this technology will empower consumers to choose products based not only on price and design but also on a new criterion of “environmental impact,” fostering environmentally conscious consumption and advancing the circular economy.
Conference
Conference: The 12th International Conference on Life Cycle Management, LCM2025
Title: Automated Estimation of Clothing Information for Evaluating Environmental Impact
Presenters: Sumiaki Kusahata, Dami Moon, Yoshihiro Kawahara
Date: September 9th – 12th , 2025
Location : Palermo, Italy
URL : https://www.lcm2025.org/
Unveiling the Amount of CO2 Fixed in Japanese Concrete Structures: Contributing to Carbon Neutrality in the Concrete Sector
Leading the way in nuclear safety research with the world’s first study on neutron-flux impact on radiation-induced expansion of rock-forming mineral
