Deep Learning-Based Automated Cell Detection-Facilitated Meat Quality Evaluation.

Area of Science:

• Food Science
• Computational Biology
• Machine Learning

Background:

• Global meat consumption necessitates robust methods for assessing meat safety and quality.
• Microbiological parameters, specifically indicator cell counts, are crucial for meat quality evaluation.
• Manual cell counting methods are prone to errors and inefficiency.

Purpose of the Study:

• To develop an automated system for identifying and counting stained cells in meat quality assessments.
• To improve the accuracy and efficiency of microbiological analysis in meat.
• To introduce a novel Convolutional Neural Network (CNN) architecture, Detect-Cells-Rapidly-Net (DCRNet), for cell detection.

Main Methods:

• Proposed a Convolutional Neural Network (CNN) with a Detect-Cells-Rapidly-Net (DCRNet) backbone for automatic cell identification and counting.
• Implemented aggregated residual blocks in DCRNet to enhance feature learning with fewer parameters.
• Integrated deformable convolution networks to accommodate the varied shapes of stained animal cells.
• Ensured the CNN model is self-adaptive to different image resolutions.

Main Results:

• The proposed CNN with DCRNet achieved an Average Precision of 81.2%, outperforming traditional neural networks.
• The method demonstrated high accuracy, with results differing by less than 0.5% from manual cell counts.
• DCRNet showed superior performance in identifying and counting stained cells.

Conclusions:

• The DCRNet-based CNN is a highly effective and accurate solution for automated cell detection.
• This technology offers a significant improvement over manual methods for microbiological analysis in meat.
• DCRNet shows promise for integration into future meat quality monitoring systems, enhancing safety and quality assurance.