Measuring water holding capacity in pork meat images using deep learning.

Area of Science:

• Food Science
• Computer Vision
• Artificial Intelligence

Background:

• Water holding capacity (WHC) is crucial for high-quality pork.
• Traditional WHC estimation involves pressing meat and measuring absorbed water on filter paper.
• Accurate WHC assessment is vital for the meat industry.

Purpose of the Study:

• To evaluate the U-Net deep learning architecture for estimating pork WHC from filter paper images.
• To assess U-Net's capability in segmenting distinct regions within WHC images.
• To determine the impact of varying input image sizes on U-Net's performance.

Main Methods:

• Utilized the U-Net deep learning architecture.
• Applied the U-Net model to segment filter paper images obtained from the pork press method.
• Investigated U-Net performance with different input image dimensions.

Main Results:

• U-Net demonstrated high precision in segmenting external and internal areas of WHC images.
• The model successfully identified subtle visual differences in these regions.
• Performance was evaluated across various input sizes, indicating robustness.

Conclusions:

• Deep learning, specifically U-Net, is a viable tool for estimating pork WHC from image data.
• U-Net segmentation provides precise analysis of WHC-related image features.
• This approach offers a non-destructive and potentially more efficient method for meat quality evaluation.