Application of Image-Based Features and Machine Learning Models to Detect Brick Powder Adulteration in Red Chili Powder

Abstract

This study has introduced a novel methodology for detecting brick powder (BP) adulteration in red chili powder (RCP; Variety: Bullet Lanka-5) by strengthening advanced digital image processing techniques. Specifically, this approach integrated color space histogram and texture features, subsequently refined through Z-score normalization and followed by the infinite latent feature selection (InLFS) method. By combining these innovative image-based techniques with machine learning (ML) algorithms, this research sets a standard for ensuring the safety and authenticity of RCP. The digital image-based dataset consisting of images of pure and adulterated RCP with BP at various concentrations, is used to extract the features for the evaluation of models. Three histograms (i.e., YCbCr, RGB, and Lab) and texture feature models (i.e., GLCM, GLDM, and GLRM) are extracted from each image. Subsequently, the InLFS model is employed to identify the most desirable features for the extracted histogram and texture features, which are further trained on the ML models to evaluate the existence and extent of BP adulteration in RCP. The regression model has given a higher coefficient of determination (R²) of 0.99 when using exponential Gaussian Process Regression (GPR) trained on Lab color space histogram features, with corresponding RMSE, MSE, and MAE values of 2.14, 12.21, and 1.08, respectively. Meanwhile, the subspace KNN classifier, with SF-C-Texture-Lab-hist, has achieved an accuracy of 99.31%. Therefore, the findings of this study underscore the potential applications of digital image-based feature extraction in combination with ML models to ensure the safety and authenticity of RCP.