FruitQ-GradeX: Fruit Quality Grading with Explainable AI
Dual-Head CNN for Fruit Classification & Quality Grading with Grad-CAM
Deep learning system that classifies fruit images by both type and quality using a custom multi-headed CNN. Achieves 97.75% fruit classification accuracy and 98.21% quality assessment accuracy. Integrated Grad-CAM for visual explanations and features a real-time Streamlit interface with webcam support.
Multi-Headed Deep CNN with Simultaneous Dual Prediction & Grad-CAM Visual Heatmaps
Multi-class prediction across 4 major fruit varieties (Apple, Banana, Guava, Orange).
Binary classification discerning fresh vs rotten fruit across diverse lighting conditions.
Visualizes gradient activation maps for both classification heads to explain decision rationale.
Rigorous augmentation pipeline with rotation, zoom, shear, and color jittering.
Engineering Highlights
- Dual prediction: fruit type AND quality simultaneously in a single pass
- 97.75% fruit classification accuracy, 98.21% quality accuracy
- Supports 4 fruit types: Apple, Banana, Guava, Orange
- Grad-CAM heatmaps for both output heads enabling model explainability
- Real-time inference via webcam or image upload on Streamlit Cloud