AI-driven non-destructive detection of meat freshness using a multi-indicator sensor array and smartphone technology

Abstract

This study presents the development of a sensor array for classifying meat samples (buffalo, lamb, and beef) based on their Total Volatile Basic Nitrogen (TVB-N) levels, a key indicator of freshness. The sensor array was created by depositing solutions of seven pH and redox indicators: Aniline blue, Nile blue, Alizarin red S, Cresol red, Methyl violet, Methyl orange, and Chlorophenol red. Classification was performed using Linear Discriminant Analysis-Principal Component Analysis (LDA-PCA) models and genetic programming (GP). The GP analysis identified Chlorophenol red, Cresol red, and Methyl violet as the most significant indicators, selected frequently over 1,000 iterations. The resulting mathematical models were implemented in a smartphone application, which achieved high classification accuracy, reporting strong performance metrics (aka., precision, accuracy, F1-score, specificity, sensitivity, error rate, and kappa coefficient) for all three meat types. The LDA-PCA model demonstrated discrimination accuracies of 96 % for buffalo, 81 % for lamb, and 88 % for beef, with corresponding precision values of 90 %, 84 %, and 77 %. These results highlight the potential of this method for real-time, reliable assessment of meat freshness.

Keywords

• Meat quality assessment
• Sensor array
• Genetic programming
• LDA-PCA Model