pipette
ESEspañol

Sounds Cheesy - Bridging the Texture Gap Between Dairy and Plant-Based cheese with Non-Contact Ultrasound and AI

Wenjing Feng, Thomas A. Hazlehurst, Nik Watson

PreprintReal-world use

In the authors' words

Plant-based cheeses often fail to reproduce the texture of dairy cheese. Closing this gap is slow because the reference methods, texture profile analysis (TPA) and sensory panels, destroy the sample and are poorly suited to screening many formulations in a timely and cost-effective manner. Here we show that non-contact, air-coupled ultrasonic imaging combined with machine learning can characterise cheese texture rapidly and without damaging the sample. Thirteen commercial products (nine dairy, four plant-based; 119 slices, 357 scans) were imaged in through-transmission at 100 kHz to give two-channel amplitude and time-of-flight C-scans, and each slice was then measured by TPA. The C-scans revealed millimetre-scale acoustic domains in ripened dairy cheeses, consistent with their fused-curd microstructure, that were absent from the homogeneous starch and oil matrix of the plant-based products. TPA showed that the texture gap is one of range and failure behaviour rather than average firmness. Mean hardness was similar for the two categories, but dairy products spanned a 7.5-fold range against 1.9-fold for plant-based products, and the plant-based products failed in a brittle, strain-softening manner rather than by ductile yielding. No single ultrasonic parameter tracked texture across products. However, a compact convolutional neural network predicted full true stress-strain loading and unloading curves from the C-scan images of unseen slices across all thirteen products without product-specific calibration. Clustering the predicted curves produced a texture map on which three plant-based slices sat with the Cheddars and one with Emmental, while soft, mozzarella-like textures remained unoccupied. The approach offers a one-minute, non-contact measurement suited to formulation screening and at-line quality control of cheese.

Main resultLimitation the authors admit

In plain words AI summary

Plant-based cheeses usually do not match the texture of dairy cheese. Standard testing methods destroy samples and are too slow for screening many recipes. The authors used non-contact ultrasound imaging and machine learning to measure texture without damage. They tested thirteen commercial cheeses and found dairy products varied more in texture than plant-based ones. A neural network predicted full texture curves from ultrasound images in about one minute.

Written by Kimi K2.5 using only the abstract above. Jev checked that every sentence is supported by the abstract sentences it cites, and removed any that were not. Point at a sentence to highlight its source.

Appeared: Thursday, October 15. arXiv. Preprint, not yet peer-reviewed.