Microalgae-enriched biscuit discrimination using image texture features and Artificial Intelligence (AI) algorithms

Abstract

The aim of this study was to design a method for classifying biscuits enriched with various types and concentrations of Chlorella microalgae based on texture features extracted from digital images, using artificial intelligence techniques. The dataset included seven biscuit variants: a control sample without microalgae addition and biscuit samples containing White, Yellow, or Green Chlorella vulgaris at 2.5% and 5% addition levels. Images acquired with the rear camera of a mobile phone were analyzed using MaZda software, which enabled the extraction of 2172 texture descriptors across 12 color channels. A feature selection process identified the 70 most relevant parameters for classification. Both deep learning (Dl4jMlpClassifier) and traditional machine learning models (Random Forest, LogitBoost, Subspace Discriminant (Ensemble), WiSARD, Medium Gaussian SVM, and Wide Neural Network) were developed using WEKA and MATLAB environments. The highest classification correctness, reaching 97.86%, was obtained with the Subspace Discriminant algorithm. Biscuits enriched with Green Chlorella were most effectively distinguished, while the greatest confusion occurred between Yellow Chlorella samples at 2.5% and 5%. This study is the first to combine image texture analysis with AI-based modeling for the non-destructive discrimination of microalgae-enriched biscuits. The developed approach demonstrates a significant advancement in applying computer vision and machine learning for food quality control and could be implemented in practice as a rapid, objective tool for monitoring formulation differences in functional bakery products.