ARTIFICIAL INTELLIGENCE-BASED SYSTEM FOR DETECTION AND QUANTIFICATION OF ADULTERATION IN CHILLI POWDER USING IMAGE PROCESSING AND MACHINE LEARNING

Abstract

Food adulteration poses significant health risks and economic losses worldwide, particularly in spices such as chilli powder. This study proposes an artificial intelligence-based system for automated detection and quantification of adulteration levels in chilli powder using image processing and machine learning techniques. A structured dataset comprising 480 images across 12 adulteration levels (0% to 100%) was developed. Feature extraction was performed using color histograms, Local Binary Patterns (LBP), and Gray-Level Co-occurrence Matrix (GLCM), resulting in a 526-dimensional feature vector. Three learning paradigms were explored: binary classification, multiclass classification, and regression-based estimation. The results demonstrate that while binary classification achieved near-perfect accuracy (up to 100%), multiclass classification performance decreased (~78%), highlighting inter-class similarity challenges. Regression-based estimation using XGBoost achieved a Mean Absolute Error (MAE) of 4.42%, demonstrating superior capability in predicting continuous adulteration levels. Analytical evaluation using confusion matrices, error distributions, PCA visualization, and tolerance accuracy curves reveals that regression provides a more robust and practical solution for real-world applications. The proposed system offers a scalable and cost-effective approach for food quality monitoring.