An Improved Residual Network for Pork Freshness Detection Using Near-Infrared Spectroscopy.

Area of Science:

• Food Science
• Spectroscopy
• Artificial Intelligence

Background:

• Pork quality monitoring relies on subjective sensory evaluation or time-consuming physicochemical analysis.
• Developing rapid, objective methods for assessing pork freshness is crucial for food safety.

Purpose of the Study:

• To develop a fast and non-destructive method for determining pork freshness using near-infrared spectroscopy (NIRS).
• To apply a deep learning model, the one-dimensional squeeze-and-excitation residual network (1D-SE-ResNet), for accurate pork freshness classification.

Main Methods:

• Utilized near-infrared spectroscopy (NIRS) for data acquisition.
• Developed and implemented a one-dimensional squeeze-and-excitation residual network (1D-SE-ResNet) for spectral analysis.
• Compared the 1D-SE-ResNet model against five other popular classification models.

Main Results:

• The 1D-SE-ResNet model achieved a high classification accuracy of 93.72% for pork freshness.
• The deep learning approach automatically extracted spectral features, simplifying the modeling process.
• The proposed model outperformed traditional methods and other classification models.

Conclusions:

• Near-infrared spectroscopy combined with deep learning offers a promising tool for objective and rapid pork freshness detection.
• This technique can significantly contribute to ensuring food safety in the pork industry.
• The 1D-SE-ResNet model provides an effective end-to-end solution for NIRS-based pork quality assessment.