Tyson Foods Improves Efficiency with Computer Vision and Machine Learning Using AWS Services

As one of the world’s largest food companies and a recognized leader in protein, Tyson Foods Inc. (Tyson Foods) processes millions of pounds of food each week. Because of its scale of production, Tyson Foods needs its facilities to be efficient while maintaining high food quality. Manual processes, like inventory counting and machine inspections, cost valuable employee time and don’t offer near real-time insights at scale.

Tyson Foods was already using computer vision (CV) solutions to augment time-consuming processes, but the company wanted to incorporate CV powered by machine learning (ML) to reduce the cost and complexity of implementing CV while improving operational efficiency. Tyson Foods looked to Amazon Web Services (AWS) to quickly add ML to its CV solutions on production lines to increase efficiency and realize cost savings in its facilities.

Tyson Foods produces beef, pork, chicken, and prepared foods in over 100 facilities worldwide. In the United States, an estimated 20 percent of the country’s chicken, beef, and pork came from Tyson Foods facilities in 2021.

Tyson Foods started a cloud migration from data centers to AWS in 2018. During this cloud migration, Tyson Foods saw how the Amazon Go store was automating the checkout and retail experiences using cameras and CV. CV is a process that involves capturing, processing, and analyzing images and videos so that machines can extract meaningful, contextual information from the physical world. This technology at the Amazon Go store inspired the company’s emerging technology team to pursue similar CV solutions to address challenges and increase efficiency in its production processes. Due to the scale of production at Tyson Foods facilities, manual inspection processes can be time consuming and create bottlenecks. Tyson Foods successfully developed an initial CV solution to augment these manual inspection processes but knew that implementing ML would increase efficiency and decrease complexity even further. The company approached AWS for support with implementing CV solutions powered by ML for inventory management and product carrier failure identification.

Tyson Foods has the capacity to process 40 million chickens per week, and the company relies on accurate inventory measurements in the facilities to fulfill customer orders. Due to the scale of production, manual techniques for counting chicken trays that pass quality assurance measures aren’t accurate enough. Alternate strategies like monitoring the hourly total weight of production per rack don’t provide data right away, preventing team members from taking action in near real time. In 2021, Tyson Foods collaborated with the Amazon Machine Learning Solutions Lab (Amazon ML Solutions Lab), which pairs an organization’s team with ML experts, to train an object detection model using Amazon SageMaker with fully managed infrastructure, tools, and workflows to build, train, and deploy ML models for any use case. This model automatically detects and counts chicken trays on video streams from production lines as employees load them onto carts. Using AWS Panorama, a collection of ML devices and a software development kit that brings CV to on-premises cameras, the company was able to deploy this model at the edge to analyze video in milliseconds. With this CV solution, poultry production supervisors receive near-real-time insights into production quantity, avoiding both underproduction and overproduction during the shift.

To improve another use case with CV powered by ML, Tyson Foods developed a solution to identify faulty plastic pins that hold product carriers in place in its poultry production facilities. Employees previously needed to manually inspect nearly 8,000 pins per line every shift because safety issues or unplanned downtime could occur if a pin fell out of place. This inspection process required attention to detail and valuable operator time. To automate the process, Tyson Foods turned to Amazon Lookout for Vision, an ML service that uses CV to spot product defects in objects at scale. Using Lookout for Vision, the company created a custom ML model to analyze images and detect anomalies, without needing ML expertise. Tyson Foods deployed the model at the edge on an AWS Panorama Appliance, which organizations can use to connect cameras and process multiple CV applications on multiple video streams simultaneously, so that its employees are notified right away that a product carrier needs maintenance when the model identifies anomalies. With this solution, team members no longer need to spend an estimated 1 hour per shift per line inspecting product carriers, which can save the company 15,000 hours of skilled labor annually in a single facility. 

Tyson Foods plans to continue using the same foundational processes to develop CV solutions powered by ML that address production needs and automate more of the business. Using AWS services, the company is now developing solutions faster and continuing to optimize processes. “These solutions help us use exactly what we need by understanding the true demand and optimizing inventory so that we can effectively plan and reduce waste,” says Barret Miller, senior manager of the emerging technology team at Tyson Foods.