Image-Based Detection of Adulterants in Milk Using Convolutional Neural Network.

Area of Science:

• Food Science and Technology
• Analytical Chemistry
• Machine Learning Applications

Background:

• Milk adulteration is a significant public health concern.
• Conventional adulterant detection methods are often costly and impractical for rural settings.
• Need for accessible and reliable milk quality assessment techniques.

Purpose of the Study:

• To investigate the potential of machine learning for detecting milk adulterants.
• To develop a deep learning model using evaporative milk deposit patterns.
• To assess the efficacy of different regularization techniques for model accuracy.

Main Methods:

• Collected image datasets of evaporative milk deposit patterns from adulterated milk samples.
• Developed a Convolutional Neural Network (CNN) model for pattern classification.
• Applied implicit (data augmentation) and explicit regularization techniques for optimization.

Main Results:

• CNN successfully classified distinct evaporation patterns corresponding to different adulterants and concentrations.
• The method detected specific minimum concentrations: 5% urea, 2.4% ammonium sulfate, and 2% oil.
• Implicit regularization via data augmentation yielded the highest testing accuracy of 98%.

Conclusions:

• Machine learning, specifically CNNs, can effectively detect common milk adulterants using evaporative patterns.
• The developed technique offers a promising, low-cost alternative for milk quality control.
• Data augmentation significantly enhances the accuracy of adulterant detection models.