Segment, measure, flag: objective visual QA for meat-in-sauce products

Overview

This offering is a proof-of-concept software pipeline that converts routine visual checks of meat-in-gravy and meat-in-jelly products into objective, recorded measurements. Using a standard photo of a tray or portion, the pipeline segments each meat piece from the background and reports per-piece size, shape, and colour, background colour, meat-to-background ratio, and the size and shape distributions. Every result carries a confidence score and a sampling band, so production teams know how much to trust a single image. Deviations from an agreed reference standard are flagged automatically.

The core value is replacing operator-to-operator variance with consistent, image-based measurement. The same image always yields the same numbers, which can be stored with the image and compared across shifts and sites. This supports more honest, traceable, and data-driven quality assurance on food production lines.

Technical specifications

Key features:

• Segments meat pieces from gravy or jelly backgrounds in standard photos
• Reports per-piece size, shape, and colour, plus background colour and meat-to-background ratio
• Provides size and shape distributions with confidence scores and sampling bands
• Flags deviations from an agreed reference standard
• Runs in under one second on a phone for plant macro images with hundreds of small objects per image
• Requires no sample preparation and no operator training beyond taking a photo

Underlying capability:

The pipeline is already in daily commercial use on plant macro images, where it classifies colour per object and computes distributions with bootstrap intervals. This proof of concept transfers that measurement layer to meat-in-sauce products and validates it against the sponsor's own QA panel.

Technology readiness level

The pipeline is in daily commercial use for plant macro imaging, demonstrating a high baseline maturity. This specific application to meat-in-gravy and meat-in-jelly products is at proof-of-concept stage. The proposed 11-week sponsored study would validate accuracy against a customer's QA panel, test on held-out images, and produce an inline-readiness specification covering camera, lighting, and throughput. The approach includes an explicit decision gate: if agreement is below target, that is reported directly, and the customer decides whether to stop or pursue a second iteration.