Automated Vial and Pre-Filled Syringe Counting in the Pharmaceutical Industry Using YOLO and SAHI Techniques

In the pharmaceutical industry, manual counting of vials and pre-filled syringes (PFS) is a time- consuming process prone to human error, which can lead to inventory discrepancies and operational inefficiencies. This research addresses these challenges by automating the counting process using state-of-the-art deep learning techniques. We employ the YOLO (You Only Look Once) architecture from the Ultralytics library, renowned for its real-time object detection capabilities. Our study compares three versions of the YOLO models (v8, v9, v10) to determine the most accurate and efficient model for this application and designed to handle both images and videos. In this study, we applied the Slicing Algorithms for Hyper Inference (SAHI) technique to enhance object detection by efficiently handling smaller objects within larger images, thereby aiming to improve the overall accuracy and robustness of the model. However, our experimental results did not show a significant improvement over existing methods which highlights the potential limitations of the SAHI technique in certain contexts, suggesting the need for further investigation into its effectiveness and adaptability across diverse applications. Using more than 6000 images, the model were trained with a result of high mean average precision of 0.969 showcasing their high detection precision. With a counting accuracy of more than 95%, the proposed model offers an effective solution by eliminating the need for manual counting, thus reducing the potential for human error inherent in traditional methods. Additionally, the developed system seamlessly integrates the counting values with existing inventory management platforms, ensuring up-to-date stock levels and enhancing inventory accuracy. This integration offers substantial time and cost savings for the pharmaceutical and healthcare industries.