Scholarlink

<p>Yoghurt fermentation is a complex biochemical process governed by microbial activity and acidification dynamics, directly influencing product quality. Traditionally, monitoring this process relies on invasive sampling and pH measurement, which are labour-intensive, pose contamination risks, and hinder continuous quality control. Currently, no established non-invasive system exists which integrates deep learning models for real-time prediction of key fermentation parameters such as pH and time. Addressing this gap, the present study introduces a computer vision system employing deep learning to non-destructively assess yoghurt fermentation quality using surface chromatic features (L*, a*, b*). Comprehensive modeling was conducted using multiple deep learning architectures like Feed forward neural network, Long short-term memory, Gated recurrent unit and Transformer with FFNN4 (256, 128, 64) emerging as the most accurate model. It achieved high prediction accuracy for fermentation time (R²: 0.993 ± 0.004, RMSE: 8.03 ± 1.85, MAE: 6.54 ± 1.75 and MBE: 0.11 ± 0.55) and pH (R²: 0.996 ± 0.001, RMSE: 0.06 ± 0.01, MAE: 0.04 ± 0.005 and MBE: − 0.001 ± 0.01). A Python-based framework was developed using FFNN4 for real-time prediction, integrating a GUI, live camera feed, ROI-based image capture, and data logging. Calibration analysis showed excellent agreement with actual standards (R²: 0.9998, RMSE: 0.5799 and MAE: 0.4527), validating chromatic data reliability. The experimental yoghurt conformed to the commercial samples in physicochemical, sensory, texture, and microbiological properties, with no significant differences (<i>p</i> &gt; 0.05). This non-invasive monitoring framework can be effectively deployed for industrial purpose, enhancing microbial safety, enabling continuous quality control, and reducing manual dependency, thereby supporting scalable, automated yoghurt production in smart dairy environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep learning-enabled framework for real-time non-invasive measurement of yoghurt fermentation kinetics

  • 期刊论文

  • 发表日期：

    2025年11月18日

查看全文
  • Arijit Ray,
  • Chitranayak Sinha,
  • J. K. Dabas

摘要

Yoghurt fermentation is a complex biochemical process governed by microbial activity and acidification dynamics, directly influencing product quality. Traditionally, monitoring this process relies on invasive sampling and pH measurement, which are labour-intensive, pose contamination risks, and hinder continuous quality control. Currently, no established non-invasive system exists which integrates deep learning models for real-time prediction of key fermentation parameters such as pH and time. Addressing this gap, the present study introduces a computer vision system employing deep learning to non-destructively assess yoghurt fermentation quality using surface chromatic features (L*, a*, b*). Comprehensive modeling was conducted using multiple deep learning architectures like Feed forward neural network, Long short-term memory, Gated recurrent unit and Transformer with FFNN4 (256, 128, 64) emerging as the most accurate model. It achieved high prediction accuracy for fermentation time (R²: 0.993 ± 0.004, RMSE: 8.03 ± 1.85, MAE: 6.54 ± 1.75 and MBE: 0.11 ± 0.55) and pH (R²: 0.996 ± 0.001, RMSE: 0.06 ± 0.01, MAE: 0.04 ± 0.005 and MBE: − 0.001 ± 0.01). A Python-based framework was developed using FFNN4 for real-time prediction, integrating a GUI, live camera feed, ROI-based image capture, and data logging. Calibration analysis showed excellent agreement with actual standards (R²: 0.9998, RMSE: 0.5799 and MAE: 0.4527), validating chromatic data reliability. The experimental yoghurt conformed to the commercial samples in physicochemical, sensory, texture, and microbiological properties, with no significant differences (p > 0.05). This non-invasive monitoring framework can be effectively deployed for industrial purpose, enhancing microbial safety, enabling continuous quality control, and reducing manual dependency, thereby supporting scalable, automated yoghurt production in smart dairy environments.