2026 ASEE Annual Conference & Exposition

The health impact of microplastics has raised growing concerns recently. Plastic has been found in human brains, lungs, placentas, and plaque of clogged arteries. A significant portion of the microplastic issue consists of plastic bags and films. Each year, up to one trillion plastic shopping bags are produced globally, according to an article published by the United Nations Environment Programme in December 2021. Only one-tenth of the 4.20 million tons of plastic bags, sacks, and wraps generated were recycled, according to the Environmental Protection Agency’s (EPA) 2018 data. To encourage more people to recycle plastic bags, it is essential to develop technology that helps the public correctly and efficiently identify recyclable bags.

Automating plastic bag recognition is challenging, though. Plastic bags often lack resin identification codes or symbols, such as type 2 (HDPE) or 4 (LDPE), making it difficult to determine if they are recyclable. Visual features, such as shape and color, although necessary for object recognition, are not sufficient to correctly classify plastic bags and films. Bags and films can be transparent and have various shapes. The objects enclosed in transparent bags and films could be recognized instead.

Our goal is to find a low-cost, easily scalable solution to the critical problem of plastic bag and film recycling. In this research, we

• Investigated and compared the capability of two current AI tools, OpenAI API and Gemini API, to recognize plastic bags and wraps from images;
• Designed a new approach for recyclable plastic bag and film recognition, leveraging modern AI’s capability to process both image and text inputs;
• Implemented this approach using both OpenAI and Gemini APIs, tested them respectively, and compared the results; and
• Developed a model for academic institutions to partner with community organizations in educating youth through project-based learning, with AI tools to assist in ideation, prototyping, and programming.

Conclusions: Our data show that the Gemini API (gemini-2.5-flash) performed better than the OpenAI API (gpt-4o) across most bag categories we tested, achieving 92.17% overall accuracy vs 71.93%, respectively. Cost-wise, Gemini-2.5-flash offers a free tier of service, making it a better option. Through exploration of fast software prototyping with AI APIs and chatbots, we found that they not only reduced software development time but also enabled us to engage high school students in real-world computing projects and to teach them research and software engineering with minimal resources.