🌰 Pistachio Denoising Autoencoder (Keras)

This project builds a denoising autoencoder using Keras to reconstruct pistachio images.
The model learns to remove noise from input images while preserving key visual structures.

---

🎥 Analysis Video

Watch the analysis walkthrough here:
👉 https://drive.google.com/file/d/13tzPI8Nu54--qUF7t2QlANgjQZ82gDCa/view?usp=sharing

---

📊 Dataset

The dataset contains pistachio images with a consistent black background.
All images are in .jpg format with uniform size.

You can access the dataset here: https://drive.google.com/drive/folders/1KXWgrZw1nJY6u8xo9grzWbSVbxEFuaCt?usp=sharing

Preprocessing

• Images split into train/test sets using train_test_split.
• Synthetic noise (e.g., Gaussian noise) is added to simulate noisy inputs.
• Normalization: pixel values scaled to [0, 1].

---

✨ Highlights

• Model Architecture

  • Autoencoder with convolutional encoder and decoder.
  • Encoder: stacked Conv2D + MaxPooling layers.
  • Decoder: Conv2DTranspose (upsampling) layers.

• Training Setup

  • Loss: Mean Squared Error (MSE)
  • Optimizer: Adam
  • Callback: EarlyStopping to prevent overfitting.

• Evaluation

  • Quantitative metrics: PSNR (Peak Signal-to-Noise Ratio), SSIM (Structural Similarity Index), MSE.
  • Qualitative results: comparison of original → noisy → reconstructed images.

• Hyperparameter Tuning

  • Implemented with Keras Tuner for hidden size and learning rate search.

---

📂 Repository Layout

pistachio-autoencoder-keras/
├─ notebook/
│ └─ pistachio-autoencoder.ipynb
├─ README.md
├─ requirements.txt
├─ LICENSE
└─ .gitignore

---

🛠 Environment

Install dependencies:

pip install -r requirements.txt

🚀 Run the Notebook

jupyter notebook notebook/pistachio-autoencoder.ipynb