Microbiological Quality Estimation of Meat Using Deep CNNs on Embedded Hardware Systems

Area of Science:

• Food science and technology
• Computer science and engineering
• Spectroscopy and imaging

Background:

• Assessing microbial populations in meat is crucial for food safety and quality control.
• Traditional methods for microbial analysis are often time-consuming and labor-intensive.
• Emerging technologies like spectroscopic imaging and deep learning offer potential for rapid, non-destructive quality assessment.

Purpose of the Study:

• To develop and evaluate an architecture for estimating microbial populations in meat samples using multispectral imaging.
• To implement deep learning models on embedded platforms for real-time, on-site food quality analysis.
• To compare the performance of different deep learning models and embedded hardware for this application.

Main Methods:

• Multispectral imaging was used to capture data from meat samples stored under various conditions.
• Deep convolutional neural networks (CNNs) were trained and deployed on embedded platforms (e.g., XavierNX, Nano, RP4).
• Performance metrics including latency, throughput, efficiency, and value were evaluated for different hardware and data processing setups.

Main Results:

• The XavierNX platform demonstrated superior performance in terms of low latency and high throughput for microbial population estimation.
• The Nano and RP4 platforms showed advantages in terms of energy efficiency and cost-effectiveness, respectively.
• The developed system effectively estimated microbial populations, indicating its potential for practical food quality assessment.

Conclusions:

• Embedded deep learning systems utilizing multispectral imaging are viable for on-site microbial quality assessment of meat.
• Hardware selection (XavierNX, Nano, RP4) depends on specific application requirements regarding speed, efficiency, and cost.
• This approach offers a promising alternative to traditional methods for ensuring food safety and quality.