Ai-powered computer vision system for quantitative food quality assessment
An AI-powered computer vision system that automatically analyzes images of food products, segments individual items such as meat pieces and gravy, quantifies quality characteristics using pixel data, and provides consistent quality scores based on a predefined rubric. Applicable to food products containing meat in gravy or jelly, the system delivers objective, repeatable quality evaluation.
Overview
An AI-powered computer vision system is being developed to quantitatively assess the quality of food products containing meat pieces in gravy or jelly. The system captures images of food products, automatically segments individual items, and analyzes pixel-level characteristics to generate objective quality scores based on a predefined quality rubric.
The solution addresses the need for consistent, automated quality evaluation in food manufacturing. Manual inspection is often subjective and labor-intensive, whereas this system delivers repeatable, quantifiable results that reduce variability and support more efficient quality control workflows.
Technical specifications
Key components:
- Imaging system using a commercial light box and USB or iPhone camera for standardized image capture
- Trained AI segmentation model that identifies meat pieces and other food items with accurate boundaries
- Pixel-level quantitative analysis to characterize product quality attributes
- Classification model that scores product quality against a predefined rubric
- Computer or iPhone application that automates the evaluation workflow and reports results
Development approach:
- Annotated training dataset of 100+ food samples, adjustable based on product complexity
- Iterative model refinement with feedback on annotation accuracy
- Integration of segmentation and classification models into a user-facing application
- System testing and refinement in production environments
Technology readiness level
The technology is at an early development stage, approximately TRL 3-4. Early phases focus on establishing the imaging setup, collecting and annotating sample images, and training the segmentation model to accurately identify meat boundaries. Subsequent phases involve developing the classification model, building the application interface, and conducting system testing and refinement in production environments. The project timeline spans approximately 12 months, with validation requiring at least 100 food samples and iterative improvements based on annotation quality and model performance.
About University of Tennessee, Knoxville
UT offers a partnership model that reduces risk, accelerates time-to-value, and supports long-term growth strategies for industry—providing an attractive platform for companies seeking to formalize collaboration and establish a long-term presence in the East Tennessee region. By aligning institutional investment with regional and national priorities, UT provides a stable and effective platform for industry collaboration, technology deployment, and sustained economic impact.
delivered to your inbox.