Identifying the quality characteristics of pork floss structure based on deep learning framework

Area of Science:

• Food Science and Technology
• Computer Vision
• Artificial Intelligence

Background:

• Pork floss, a traditional Chinese food, is recognized as a leisure food product.
• Its unique texture results from a specific processing method applied to pork muscle fibers.
• Assessing pork floss quality traditionally relies on subjective sensory evaluation.

Purpose of the Study:

• To develop and validate a deep learning-based method for detecting pork floss quality characteristics.
• To objectively evaluate the structural features of commercially available pork floss using machine vision.
• To compare the performance of the proposed model against traditional sensory assessments.

Main Methods:

• Collected 8000 images from eight commercial pork floss brands.
• Processed images using sharpening, grayscale conversion, shading correction, and binarization.
• Employed a deep learning framework, coupling residual enhancement mask and region-based convolutional neural network (CRE-MRCNN), for image segmentation.

Main Results:

• The CRE-MRCNN framework successfully identified knot and pore features in pork floss images.
• The model demonstrated capability in differentiating quality characteristics across various brands.
• Machine vision results aligned with sensory test outcomes, indicating TC brand as the highest quality.

Conclusions:

• Deep learning and machine vision offer a viable objective method for assessing pork floss quality.
• The CRE-MRCNN model effectively analyzes structural features relevant to pork floss quality.
• This technology has the potential to supplement or replace traditional sensory evaluations in food quality control.