Multi-color space feature engineering combined with deep learning for rapid non-destructive detection of added water in pomegranate juice

Abstract Ensuring the authenticity of fruit juices is a critical challenge in modern agriculture and the food supply chain, where economically motivated adulteration compromises product quality, consumer safety, and market transparency. Although spectroscopic and laboratory-based techniques have been widely used for juice authentication, they are often expensive, time-consuming, and require specialized equipment. Therefore, this study proposes a novel machine vision-based feature extraction framework that integrates multi-color space feature extraction combined with deep learning classification for rapid and non-destructive detection of added water in pomegranate juice using image-derived color and textural features extracted from visible images. A dataset consisting of 625 pomegranate juice samples representing five levels of water adulteration was acquired under controlled imaging conditions. To enhance model robustness and generalization, data augmentation techniques were applied, expanding the dataset sixteen-fold. The acquired images were transformed into 19 color channels to capture comprehensive color information associated with different adulteration levels. An enhanced convolutional neural network architecture incorporating Conv1D, MaxPooling, BatchNormalization, and Dense layers was developed to learn discriminative patterns from image-derived feature vectors. The convolutional neural network model achieved a classification accuracy of 99.45%, with Precision, Recall, and F1-score values approaching 1.00, outperforming conventional machine learning models including multilayer perceptron (87.75%), support vector machine (96.95%), and random forest (92.45%). The results demonstrate that the combination of multi-color space analysis and deep learning provides a highly accurate and robust framework for pomegranate juice adulteration detection. Compared with conventional spectroscopic approaches, the used method offers a rapid, cost-effective, non-destructive, and automated solution with strong potential for integration into real-time machine vision systems for fruit juice quality control and authenticity assessment.