Multi-Modal AI Framework for Predicting Lemon Juice Stability Using ML, Deep Learning, and Computer Vision

Abstract

Lemon juice undergoes noticeable chemical and physical transformations during storage, making its stability and safety a crucial concern in food monitoring. A multi-modal artificial intelligence framework is developed, integrating Machine Learning (ML), Deep Learning (DL), and Computer Vision (CV) methods to predict lemon juice stability across different aspects. The approach is divided into three modules: (i) a Light Gradient Boosting Machine (LightGBM) model for classifying drinkability based on physicochemical and observational features, (ii) a Gated Recurrent Unit (GRU) network for forecasting future pH values to capture temporal quality changes, and (iii) an EfficientNet-based Convolutional Neural Network (CNN) for detecting spoilage visually from juice color patches. Experimental results show high classification accuracy, effective pH trend prediction, and robust visual stage identification. The hybrid framework supports early spoilage detection and enhances decision-making for food quality assurance.