In a beverage factory in Thailand, old, used glass bottles move on a conveyer belt, heading toward their fate. They are on their way to be examined and put into specific categories based on whether or not they can be reused and recycled. Some of the bottles are perfect and pristine, some are chipped and scratched, while others fall somewhere in between.
The moving bottles meet a system equipped with four cameras that snap images of the bottles’ neck, bottom, and two sides. By the time they reach the end of the conveyer belt, the system has used AI to select a category for the bottle. This is the work of a collaborative project in partnership with CMKL University to developing an effective and efficient artificial intelligence (AI) system for accurate bottle screening for reuse and recycling, an important issue for industrial plants.
In the U.S., states with container deposit legislation (also known as “bottle bills”) have an average beverage container recycling rate of around 60 percent, while non-deposit states only reach about 24 percent, according to the Container Recycling Institute. But in Thailand, there is no container deposit legislation. ThaiBev, Thailand’s largest beverage company, has set a goal to achieve a recycling rate of 80 percent. To do this, they need to streamline the recycling process, and this project will help increase efficiency.
ThaiBev has used different systems to segregate good quality bottles from unusable ones. However, a previous single-camera system proved to be inaccurate, leading to false negatives. In turn, manual filtering, while accurate, was a slow process that led to diminished productivity. To address this problem, researchers partnered with ThaiBev to automate the screening process with AI.
What really excites me is taking the basic research of how we design pattern recognition and AI algorithms to look for defects in bottles.Marios Savvides, Professor, Electr
“Bottles are one of the wastes that come from business, creating a lot of trash,” said Orathai Sangpetch, vice president of CMKL University. “But not all bottles can be reused—some are broken, something inside or on the bottle can’t be washed, and they need to be thrown away.”
Sangpetch reached out to Marios Savvides, professor of electrical and computer engineering and director of the CyLab Biometrics Center. Savvides, who has a strong research record in the application of pattern recognition and AI for face and iris biometrics and more recently building AI for robots in retail environments, began working with his lab on an algorithm for automatic bottle screening.
“Automated bottle screening for reusing is a challenging task, where we need to consider different components like bottle type, type of substance present on and in the bottle, along with other external factors like lighting variation, data capturing speed, and image quality,” said Sreena Nallamothu, project manager and researcher at CyLab Biometrics Center. “We mainly focus on utilizing the multi-view information about the texture and minor details of the bottle while being invariant to external factors to filter it into ‘Reusable,’ ‘Clean before reuse,’ and ‘Non-reusable’ categories.”
On the ground in ThaiBev’s factory, researchers gathered data on a wide variety of bottles from each of the categories. The data was sent to Savvides and his team to train the AI algorithm. The data is not constant; it changes as ThaiBev’s needs and priorities change. With the various datasets, the team in Pittsburgh has trained the models, and with easily available on-the-market cameras, the system analyzes the images taken as a bottle moves along the conveyer belt and makes a quick decision on which category to put it in. The project is still in progress and undergoing changes and revisions, but so far, the results have been positive.
For Savvides, this unique project as not only an opportunity to collaborate with CMKL University, but also a chance to engineer a solution to a scientific problem in real time. “I’m excited about how we’re applying AI to solve a real-world practical problem that ThaiBev is facing in Thailand,” he said. “What really excites me even further is taking the basic research of how we design pattern recognition and AI algorithms to look for defects in bottles.”
The AI system technology could potentially be adapted for different products in more factories. Currently, it is trained to decipher specific characteristics found in a specific type of ThaiBev bottle, but can be trained with other kinds of data with new priorities and characteristics. Beyond the work on the conveyer belt, Sangpetch sees this project as a technology that can truly help the world.
“There’s a lot of trash out of there, so if we can reuse and recycle some of it, we can really help this world,” she said.