🧠 Scene Image Classifier using CNN

This project is a Convolutional Neural Network (CNN)-based image classifier that predicts the type of scene in a photograph. It was trained on the Intel Image Classification Dataset and identifies six natural and urban classes.

📌 Classes

• 🏙️ Buildings
• 🌳 Forest
• ❄️ Glacier
• 🏔️ Mountain
• 🌊 Sea
• 🛣️ Street

---

📁 Dataset

Intel Image Classification Dataset
📎 Download from Kaggle

---

🔧 Tech Stack

• Python 🐍
• TensorFlow/Keras 🤖
• Google Colab (GPU Training) ⚙️
• Matplotlib 📊

---

🧠 Model Summary

• CNN with multiple Conv2D and MaxPooling2D layers
• Activation: ReLU & Softmax
• Optimizer: Adam
• Categorical crossentropy loss
• Achieved ~98% training accuracy and ~84% validation accuracy

---

📊 Sample Prediction Output

✅ The model successfully predicted this image as mountain.

---

⚠️ Model Limitation

This model was trained exclusively on the Intel Scene Classification Dataset, which contains only 6 types of natural and urban scenes:

🏙️ Buildings | 🌳 Forest | ❄️ Glacier | 🏔️ Mountain | 🌊 Sea | 🛣️ Street

🔍 Note: This classifier will only work reliably on images that fall into one of these six categories. It is not designed to detect arbitrary objects or scenes outside this dataset (e.g., animals, humans, vehicles, etc.).

---

🚀 How to Use

1. Load the trained model model.h5 using Keras or TensorFlow
2. Resize your image to 150x150 pixels
3. Normalize and preprocess the image
4. Run model.predict() to classify the image
5. Match the output index with the class label

📦 Files Included

• scene_classifier.ipynb — Final Jupyter notebook
• model.h5 — Trained CNN model
• sample_image.jpg — Test input
• sample_output.png — Prediction result

---

🧠 Download Model

Due to GitHub's file size restrictions, the trained model (model.h5) is hosted externally. 📥 Click here to download model.h5 from Google Drive

---

📌 Future Improvements

• Deploy as a Streamlit web app
• Add confusion matrix & metrics
• Fine-tune with transfer learning
• Expand dataset diversity

---

🙋‍♂️ Author

Ankit Kumar Panda
📫 GitHub: ankitpanda030306
🔗 ORCID: 0009-0009-3098-2336

---

⭐ If you like this project, please consider giving it a star!