Skip to content

Latest commit

 

History

History
121 lines (86 loc) · 3.45 KB

README.md

File metadata and controls

121 lines (86 loc) · 3.45 KB

🪟 Image Classification Using Neural Networks

This project demonstrates image classification using Artificial Neural Networks (ANN) and Convolutional Neural Networks (CNN) with the CIFAR-10 dataset. It includes dataset preprocessing, model building, training, and evaluation.

✨ Table of Contents

  1. Introduction
  2. Dataset
  3. Technologies Used
  4. Project Structure
  5. How to Run
  6. Code Explanation
  7. Results
  8. Conclusion

📝 Introduction

Image classification is the process of identifying and categorizing objects in an image. In this project, we use two types of neural networks:

  • ANN (Artificial Neural Network): A basic model for image classification.
  • CNN (Convolutional Neural Network): A more advanced model that captures image-specific features for better accuracy.

💾 Dataset

We use the CIFAR-10 dataset, which contains:

  • 60,000 images divided into 10 categories: airplane, automobile, bird, cat, deer, dog, frog, horse, ship, truck.
  • Images are 32x32 pixels with 3 color channels (RGB).

💻 Technologies Used

  • Python: Programming language.
  • TensorFlow/Keras: Framework for building and training neural networks.
  • Matplotlib: Library for data visualization.
  • NumPy: Library for numerical computations.

🗂️ Project Structure

.
├── README.md       # Documentation
└── Image_Classification_(Cifar_10).ipynb         # Main script to execute the project

⚡ How to Run

  1. Clone the repository:

    git clone https://github.com/DevSumanP/ImageClassifier101.git
cd ImageClassifier101

  2. Install dependencies:

    pip install tensorflow matplotlib numpy

  3. Run the script:

    python main.py

  4. Expected Output:

    • Model training logs.
    • Accuracy metrics for both ANN and CNN.
    • Visualizations of predictions.

⚡ Code Explanation

  1. Data Preprocessing:

    • Normalize pixel values to range [0, 1] for faster and stable training.
    • Reshape labels to make them compatible with the models.

  2. ANN Model:

    • A simple architecture with fully connected layers.
    • Output layer uses a softmax activation function to predict probabilities for 10 classes.

  3. CNN Model:

    • Includes convolutional and pooling layers to extract features from images.
    • Improves classification accuracy compared to ANN.

  4. Evaluation:

    • Evaluate models using test data and generate classification reports.

😃 Results

Model Accuracy
ANN ~49%
CNN ~70%

Conclusion

  • ANN is suitable for simple datasets but lacks accuracy for images.
  • CNN performs better due to its ability to extract spatial hierarchies (e.g., edges, textures).
  • Image classification is an essential task in AI, with applications in self-driving cars, medical diagnostics, and more.

Future Work

  • Use data augmentation to improve accuracy.
  • Experiment with deeper networks or transfer learning using pre-trained models.
  • Apply this framework to a custom dataset for real-world applications.

🗒️ License

This project is open-source under the MIT License.