Deep_Learning_Based_Fruit_Classification

---

Deep Learning-Based Classification of Fresh and Rotten Fruits 🍎🍌🍊

Overview

This repository contains the implementation of the project “Deep Learning-based Classification of Fresh and Rotten Fruits using Convolutional Neural Networks”, developed as part of an academic project exploring deep learning and real-world image classification applications.

The project leverages Convolutional Neural Networks (CNNs) and transfer learning to automate fruit quality inspection, classifying fruit images into fresh or rotten categories. The system aims to reduce human subjectivity, minimize food waste, and support agricultural and retail industries with scalable AI-driven solutions.

Key Features

• 📊 Dataset: Kaggle dataset with ~10,901 images across six classes:
  • Fresh Apples, Rotten Apples
  • Fresh Bananas, Rotten Bananas
  • Fresh Oranges, Rotten Oranges
• 🧹 Preprocessing: Image resizing, normalization, and augmentation for robust training.
• 🧠 Models Implemented:
  • Normal CNN
  • Custom CNN
  • VGG16 (transfer learning)
  • MobileNetV2 (transfer learning, best performing model)
• ⚙️ Performance:
  • MobileNetV2 achieved 98.41% validation accuracy with strong precision, recall, and F1-score.
• 🌐 Deployment:
  • Flask/FastAPI backend serving predictions via API.
  • Web-based frontend (React.js) for intuitive image upload and classification.

MobileNetV2 emerged as the best-performing model, balancing accuracy and efficiency, making it suitable for real-world deployment.

Applications

• ✅ Automated fruit quality inspection in warehouses and supermarkets.
• ✅ Consumer-facing apps for freshness detection.
• ✅ Agricultural supply chain monitoring to reduce waste.

Future Scope

• Multi-class severity detection (e.g., overripe, bruised).
• Deployment on mobile devices using TensorFlow Lite.
• IoT integration for real-time monitoring in storage/transport.
• Cloud deployment for scalability and accessibility.
• Real-time video-based fruit sorting in industrial settings.