---

🏞️ Natural Scene Image Classifier

---

This project is a deep learning–based image classification system that identifies natural scenes such as buildings, forests, glaciers, mountains, seas, and streets. It uses MobileNetV2 with transfer learning for model training and a Streamlit web app for interactive predictions.

---

📌 Features

• ✅ Train a CNN using MobileNetV2 on Intel’s Natural Scene dataset.
• ✅ Use data augmentation to improve generalization.
• ✅ Save and load the trained model (scene_classifier_model.h5).
• ✅ Deploy a Streamlit app to classify uploaded images.
• ✅ Outputs the predicted class along with confidence percentage.

---

📂 Project Structure

.
├── app.py                   # Streamlit web application
├── train_model.py           # Model training script
├── scene_classifier_model.h5 # Saved trained model (generated after training)
├── data/                    # Dataset folder (download from Kaggle)
└── README.md                # Project documentation

---

📊 Dataset

The model is trained on the Intel Image Classification Dataset from Kaggle:

🔗 Intel Image Classification Dataset

• Classes (6 total):

  • 🏙️ Buildings
  • 🌲 Forest
  • 🧊 Glacier
  • ⛰️ Mountain
  • 🌊 Sea
  • 🚦 Street

---

⚙️ Installation & Setup

1. Clone the repository

   git clone https://github.com/mithun-27/natural-scene-classifier.git
   cd natural-scene-classifier

2. Install dependencies

   pip install -r requirements.txt

   Example requirements.txt:

   streamlit
   tensorflow
   pillow
   numpy
   

3. Download the dataset

   • Download from Kaggle: Intel Image Classification Dataset
   • Extract it into the data/ folder.

---

🏋️‍♂️ Train the Model

Run the training script:

python train_model.py

This will:

• Load and preprocess the dataset.
• Train a MobileNetV2-based classifier.
• Save the trained model as scene_classifier_model.h5.

---

🚀 Run the Streamlit App

Start the app:

streamlit run app.py

Then, open the provided local URL in your browser.

• Upload an image (jpg, jpeg, or png).
• Click Classify Image.
• See the predicted class and confidence score.

---

🖼️ Screenshots

Below are example screenshots of the app in action:

🔹 Upload Screen

🔹 Prediction Screen

---

📈 Model Details

• Base Model: MobileNetV2 (pretrained on ImageNet).
• Input Size: 150×150 RGB images.
• Optimizer: Adam (lr = 0.001).
• Loss Function: Categorical Crossentropy.
• Training Epochs: 5 (can be increased for better accuracy).

---

🛠️ Future Improvements

• Increase training epochs for higher accuracy.
• Fine-tune MobileNetV2 layers.
• Deploy as a web service (Flask/FastAPI backend).
• Add support for more scene categories.

---

👨‍💻 Author

Developed by Mithun 🚀

• 📧 Email: kvl202014@gmail.com
• 💼 LinkedIn
• 💻 GitHub

---