Gemstone Classification using Transfer Learning with MobileNetV2

Overview

This code implements a gemstone classification model using transfer learning with the MobileNetV2 architecture. The model is trained on a dataset of gemstone images, and its performance is evaluated using various metrics, including accuracy, loss, and a confusion matrix.

Prerequisites

• Python 3.x
• TensorFlow 2.x
• Matplotlib
• Seaborn
• Scikit-learn

Installation

Ensure you have the required dependencies installed using:

pip install tensorflow matplotlib seaborn scikit-learn

Usage

1. Download the gemstone dataset and organize it into train and test directories.

2. Update the train_data_path and test_data_path variables with the correct paths to your train and test datasets.

3. Optionally, set the image_to_predict_path variable to the path of a specific image for prediction.

4. Run the script to train the model, evaluate its performance, and make predictions.

python gemstone_classification.py

Code Structure

• Data Preprocessing:

  • Images are loaded and preprocessed using data augmentation for training (ImageDataGenerator).

• Model Architecture:

  • MobileNetV2 is used as the base model with additional custom top layers for classification.
  • The model is compiled with categorical cross-entropy loss and Adam optimizer.

• Training:

  • The model is trained using the specified learning rate schedule and early stopping.

• Evaluation:

  • The model is evaluated on the test set, and accuracy and loss curves are plotted.

• Prediction:

  • A sample image is loaded, preprocessed, and the model makes predictions.

• Confusion Matrix:

  • Confusion matrix and classification report are generated for evaluating model performance.

Hyperparameters

• input_shape: Input image dimensions (224x224x3).
• num_classes: Number of gemstone classes (87).
• batch_size: Batch size for training and testing (32).
• epochs: Number of training epochs (50).

Customization

• Adjust the hyperparameters to suit your specific dataset and computing resources.
• Modify the learning rate schedule, early stopping parameters, or model architecture as needed.

License

This code is licensed under the MIT License.