A Machine Learning-Based Approach for Counting Blister Cards Within Drug Packages

This repository implements the methodology described in our paper:

Bahaghighat, Mahdi, Leila Akbari, and Qin Xin. "A machine learning-based approach for counting blister cards within drug packages." IEEE Access 7 (2019): 83785-83796.

Overview

Counting blister cards within drug packages is a crucial task in pharmaceutical industries to ensure packaging accuracy and efficiency. Our approach leverages machine learning and image processing techniques to provide a reliable and scalable solution for this task.

Key Features:

• Automated Counting: Uses a machine learning model to accurately count blister cards in complex scenarios.
• Scalable Design: Capable of handling various types and shapes of drug packages.
• Open Source: Freely available for research and educational purposes.

Methodology

The workflow consists of the following steps:

1. Image Acquisition: Capturing high-resolution images of the drug packages.
2. Preprocessing: Enhancing image quality and reducing noise.
3. Feature Extraction: Extracting features such as edges and contours to identify individual blister cards.
4. Model Training: Training a machine learning model to distinguish and count blister cards.
5. Validation: Testing the model on unseen data to ensure accuracy.

System Workflow

Below is the block diagram illustrating the process:

graph TD
    A[Image Acquisition] --> B[Preprocessing]
    B --> C[Feature Extraction]
    C --> D[Model Training]
    D --> E[Counting and Validation]

Loading

Installation

To set up the environment and run the project:

1. Clone this repository:

   git clone [email protected]:leila4793/MyPaper.git
   cd blister-card-counting

2. Install dependencies:

   pip install -r requirements.txt

3. Download the dataset (link provided in the paper or repository).

Usage

To perform blister card counting, run the following command:

python count_blister_cards.py --image_path /path/to/image

block diagram

Example

Input:

Output:

Results

The proposed method achieves high accuracy on diverse datasets, as detailed in the paper:

• Accuracy: 98.5%
• Processing Time: Average 0.5 seconds per image

Citation

If you find this repository useful in your research or work, please cite the following paper:

@article{bahaghighat2019machine,
  title={A machine learning-based approach for counting blister cards within drug packages},
  author={Bahaghighat, Mahdi and Akbari, Leila and Xin, Qin},
  journal={IEEE Access},
  volume={7},
  pages={83785--83796},
  year={2019},
  publisher={IEEE}
}

License

This project is licensed under the MIT License. See the LICENSE file for details.

Contact

For questions or collaborations, please contact:

• Leila Akbari: [email protected]
• GitHub: (https://github.com/leila-akbari)

---

Thank you for your interest in our work!