FruitQ-GradeX: Determining Fruit Quality and Grading with Explainable Deep Learning

This paper presents a novel multi-task deep learning framework for simultaneous fruit classification and quality assessment using a multi-headed Convolutional Neural Network (CNN). The proposed model achieves state-of-the-art performance on a curated dataset of four Indian fruits (apple, banana, guava, and orange) with two quality classes (good and bad), leveraging aggressive data augmentation and cyclic learning rate scheduling. Proposed Architecture employs a shared feature extractor with task-specific heads, achieving 98% accuracy in fruit classification and 99% accuracy in quality detection on the test set. To enhance interpretability, Gradient-weighted Class Activation Mapping (Grad-CAM) is integrated, providing visual explanations of the model's decision-making process for both tasks. The system is deployed via a user-friendly Streamlit interface, enabling real-time predictions with XAI visualizations. Our experiments demonstrate that the model outperforms existing single-task approaches in computational efficiency and generalization, while the Grad-CAM analysis reveals critical image regions influencing quality judgments. This work bridges the gap between high-accuracy fruit grading and explainable AI for agricultural applications, offering a scalable solution for automated quality control in supply chains.