Precise Recognition of Adulterated Sliced Mutton Using Machine Vision on Mobile Phone Images

Area of Science:

• Food Science
• Computer Science
• Agricultural Technology

Background:

• Growing concerns regarding the authenticity of sliced mutton due to adulteration and processed products.
• Need for cost-effective and non-destructive methods for meat authentication.

Purpose of the Study:

• To develop a low-cost, non-destructive method for authenticating real, processed, and reconstituted sliced mutton.
• To integrate smartphone-based image acquisition with machine learning (ML) and deep learning (DL) techniques.

Main Methods:

• Acquisition of 600 standardized RGB images of sliced mutton.
• Extraction of color (RGB, HSV, Lab) and texture (GLCM) features.
• Development and evaluation of ML models (KNN, LDA, RF, SVM) and DL models (VGG16, ResNet50, InceptionV3).

Main Results:

• Statistical analyses confirmed significant differences between mutton types based on extracted features.
• SVM achieved the best ML performance; VGG16 achieved the highest DL performance (96.42% accuracy).
• Processed sliced mutton showed the highest classification ambiguity; Grad-CAM visualized model focus on muscle/fat regions.

Conclusions:

• Smartphone-acquired images combined with ML/DL offer a promising, low-cost approach for preliminary sliced mutton authentication.
• This method can support market surveillance and supply chain inspection.
• The approach provides interpretability through visualization of model decision-making processes.