🍎 Fruit Quality Detector

A deep learning model that automatically classifies fruits as fresh or rotten using Convolutional Neural Networks (CNN) built with TensorFlow and Keras.

---

📌 Overview

This project uses image classification to detect the quality of 6 common fruits. Given an input image, the model predicts whether the fruit is fresh or rotten — making it useful for automated quality control in food supply chains, grocery stores, or smart refrigerators.

---

🍇 Supported Fruit Classes

The model classifies images into 12 categories (fresh + rotten variants):

Fruit	Fresh Label	Rotten Label
Apple	freshapples	rottenapples
Banana	freshbanana	rottenbanana
Mango	freshmango	rottenmango
Orange	freshoranges	rottenoranges
Pomegranate	freshpomegranate	rottenpomegranate
Watermelon	freshwatermelon	rottenwatermelon

---

🧠 Model Architecture

A custom CNN built with Keras Sequential API:

Input (224×224×3)
    │
    ├── Conv2D(32, 3×3, relu) → MaxPool2D(2×2)      # Block 1
    ├── Conv2D(64, 3×3, relu) → MaxPool2D(2×2)      # Block 2
    ├── Conv2D(128, 3×3, relu) → MaxPool2D(2×2)     # Block 3
    │
    ├── Flatten
    ├── Dense(256, relu)
    ├── Dropout(0.4)
    ├── Dense(128, relu)
    └── Dense(12, softmax)                           # Output

• Optimizer: Adam (lr = 0.001)
• Loss: Sparse Categorical Crossentropy
• Metric: Accuracy
• Early Stopping: Monitors val_accuracy with patience = 2

---

📁 Dataset

• Total Images: ~13,252
• Train Split: 10,600 images
• Test Split: 2,652 images
• Image Size: 224 × 224 pixels (RGB)
• Normalization: Pixel values scaled to [0, 1]

Dataset folder structure expected:

Fruits_New/
├── freshapples/
├── freshbanana/
├── freshmango/
├── freshoranges/
├── freshpomegranate/
├── freshwatermelon/
├── rottenapples/
├── rottenbanana/
├── rottenmango/
├── rottenoranges/
├── rottenpomegranate/
└── rottenwatermelon/

---

🚀 Getting Started

Prerequisites

pip install tensorflow keras opencv-python matplotlib seaborn scikit-learn numpy

Training the Model

1. Update the path variable in the notebook to point to your dataset directory.
2. Run all cells in Fruit_Quality_Detector.ipynb.
3. The trained model will be saved as:
   • food_freshness_or_rotten_model.keras
   • food_freshness_or_rotten_model.h5

Running a Single Prediction

import cv2
import numpy as np
from tensorflow.keras.models import load_model

cate = ['freshapples', 'freshbanana', 'freshpomegranate', 'freshwatermelon',
        'freshmango', 'freshoranges', 'rottenapples', 'rottenmango',
        'rottenpomegranate', 'rottenbanana', 'rottenwatermelon', 'rottenoranges']

model = load_model('food_freshness_or_rotten_model.keras')

img = cv2.imread('your_fruit_image.jpg')
img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
img_resized = cv2.resize(img_rgb, (224, 224)) / 255.0
img_input = np.expand_dims(img_resized, axis=0)

pred = model.predict(img_input)[0]
print(f"Prediction: {cate[pred.argmax()]} ({pred.max()*100:.1f}% confidence)")

---

📊 Evaluation

After training, the notebook generates:

• Confusion Matrix (heatmap with Seaborn)
• Classification Report (precision, recall, F1-score per class)
• Training/Validation Accuracy & Loss curves
• Sample Predictions grid (12 random test images with true vs. predicted labels)

---

🛠 Tech Stack

Tool	Purpose
TensorFlow / Keras	Model building & training
OpenCV	Image loading & preprocessing
NumPy	Array operations
Matplotlib	Visualization
Seaborn	Confusion matrix heatmap
Scikit-learn	Metrics & evaluation

---

📂 Project Structure

├── Fruit_Quality_Detector.ipynb       # Main notebook
├── food_freshness_or_rotten_model.keras  # Saved model (Keras format)
├── food_freshness_or_rotten_model.h5    # Saved model (HDF5 format)
└── README.md

---

📄 License

This project is open-source and available under the MIT License.