https://github.com/sayanjit082805/natural-scene-classifier
Image classifier for classifying natural scenes/images using a CNN
https://github.com/sayanjit082805/natural-scene-classifier
convolutional-neural-networks deep-learning tensorflow
Last synced: 3 months ago
JSON representation
Image classifier for classifying natural scenes/images using a CNN
- Host: GitHub
- URL: https://github.com/sayanjit082805/natural-scene-classifier
- Owner: sayanjit082805
- License: unlicense
- Created: 2025-06-21T06:34:44.000Z (about 1 year ago)
- Default Branch: main
- Last Pushed: 2025-06-21T08:18:43.000Z (about 1 year ago)
- Last Synced: 2025-06-21T08:31:00.312Z (about 1 year ago)
- Topics: convolutional-neural-networks, deep-learning, tensorflow
- Language: Jupyter Notebook
- Homepage:
- Size: 72.3 MB
- Stars: 1
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Natural-Scene Classifier
A deep learning image classifier built with Convolutional Neural Networks (CNN) to identify and classify natural scenes from the Intel Image Classification dataset.
# Overview
This project aims to classify natural scenes/images into six categories :
- **Buildings**
- **Forests**
- **Glaciers**
- **Mountains**
- **Seas**
- **Streets**
# Dataset
The dataset has been taken from [Kaggle](https://www.kaggle.com/datasets/puneet6060/intel-image-classification). It contains around 25,000 images of natural scenes from around the world.
**Dataset Statistics** :
- **Total Images**: ~25,000
- **Training Set**: ~14,000 images
- **Test Set**: ~3,000 images
- **Validation Set**: ~7,000 images
- **Classes**: 6 (Buildings, Forest, Glacier, Mountain, Sea, Street)
# Installation
1. Clone the repository:
```bash
git clone https://github.com/sayanjit082805/Natural-Scene-Classifier.git
cd Natural-Scene-Classifier
```
2. Create a virtual environment (recommended):
```bash
python -m venv nature_classification_env
source nature_classification_env/bin/activate # On Windows: nature_classification_env\Scripts\activate
```
3. Install required dependencies:
```bash
pip install -r requirements.txt
```
4. Download the dataset
- Download from [Kaggle](https://www.kaggle.com/datasets/puneet6060/intel-image-classification)
- Extract to `data/` directory
- Ensure folder structure matches:
```
data/
├── seg_train/
│ ├── buildings/
│ ├── forest/
│ ├── glacier/
│ ├── mountain/
│ ├── sea/
│ └── street/
├── seg_test/
└── seg_pred/
```
For local testing, update the `train_data_path` and `test_data_path` in the notebook. Also, make sure to not run the first cell.
I do not recommend installing locally, instead, you can just check out the [Kaggle notebook](https://www.kaggle.com/code/sayanjit082805/notebookef76f9189d).
Alternatively, you can just visit the site.
# Neural Network Architecture
The CNN consists of :
- **Input Layer** : 256x256x3 RGB images.
- **Rescaling Layer** : Rescales the input to be in the [0, 1] range.
- **Four Convolution Layers** :
- First CONV2D layer with 32 filters and a filter size of 3x3, followed by BatchNormalization and MaxPooling.
- Second CONV2D layer with 64 filters and a filter size of 3x3, followed by BatchNormalization and MaxPooling.
- Third CONV2D layer with 128 filters and a filter size of 3x3, followed by BatchNormalization and MaxPooling.
- Fourth CONV2D layer with 256 filters and a filer size of 3x3, followed by BatchNormalization and MaxPooling.
- **Flattening Layer** - Flattens the output of the last convolution layer into a 1D array.
- **Dense Layer** - Consists of 128 neurons and the activation function used is the RelU activation function.
- **Output Layer** : 6 neurons, with softmax activation function for multi-class classification.
The optimiser used is the [Adam](https://keras.io/api/optimizers/adam/) optimiser, and the loss function is the sparse categorical cross-entropy.
# Metrics
The model achieves an overall accuracy of ~83%. For more detailed metrics, please consult the classification report in the jupyter notebook.
Due to the use of Early fitting, the model convergences at 66 epochs.
# License
This project is licensed under The Unlicense License, see the LICENSE file for details.
>[!NOTE]
> The License does not cover the dataset. It has been taken from Kaggle.