Advanced Deep Learning for Pharmaceutical Pill Defect Detection
Disciplines
Artificial Intelligence and Robotics | Engineering | Industrial Engineering | Manufacturing | Mechanical Engineering | Operations Research, Systems Engineering and Industrial Engineering | Pharmacy Administration, Policy and Regulation | Pharmacy and Pharmaceutical Sciences | Quality Improvement
Abstract (300 words maximum)
Quality assurance in pharmaceutical manufacturing represents pharmaceutical production quality assurance is an important patient and regulatory safeguard. Here, a new application of Detectron2, a cutting-edge deep learning framework that enables instance segmentation, is introduced to streamline the pill defect inspection process. Compared to human examination or general computer vision algorithms, the technique applied is more accurate in identifying various kinds of defects like cracks, chips, coloration, and impurities. The study employed a data set that comprised of 780 high-quality images of pharmacy pills and consisted of 400 training images, 250 validation images, and 130 test images from Kaggle (PudPawat). Defect annotation using the Makes Sense AI tool was performed to have standardized and proper labeling of defect morphologies. The Detectron2 model was selected due to its better ability to perform instance-level segmentation. Performance analysis revealed remarkable figures on different fronts. The model revealed a confidence rate of 99% in detecting defects with a total accuracy of 98.5% and an average Mean Average Precision (mAP) of 97.2%. Performance of bounding box detection provided an Average Precision (AP) of 41.95 with AP50 and AP75 scores of 68.58 and 44.06 respectively. Most significantly, segmentation performance achieved an AP of 45.14, with AP50 of 68.58 and AP75 of 64.64, confirming the model's capability to accurately define defect boundaries. By eliminating human subjectivity, the system reduces the likelihood of defective products reaching consumers. This research demonstrates that instance segmentation using deep learning is a revolutionary technology for pharmaceutical quality control. The high precision, real-time detection provided by Detectron2 provides manufacturers with an efficient way of keeping high regulatory compliance while achieving high production throughput. This research paves the way for increased AI integration in pharmaceutical production with potential extension into other quality factors and intricate pharmaceutical formulations to create automated quality control systems.
Use of AI Disclaimer
no
Academic department under which the project should be listed
SPCEET – Mechanical Engineering
Primary Investigator (PI) Name
Sathish Gurupatham
Advanced Deep Learning for Pharmaceutical Pill Defect Detection
Quality assurance in pharmaceutical manufacturing represents pharmaceutical production quality assurance is an important patient and regulatory safeguard. Here, a new application of Detectron2, a cutting-edge deep learning framework that enables instance segmentation, is introduced to streamline the pill defect inspection process. Compared to human examination or general computer vision algorithms, the technique applied is more accurate in identifying various kinds of defects like cracks, chips, coloration, and impurities. The study employed a data set that comprised of 780 high-quality images of pharmacy pills and consisted of 400 training images, 250 validation images, and 130 test images from Kaggle (PudPawat). Defect annotation using the Makes Sense AI tool was performed to have standardized and proper labeling of defect morphologies. The Detectron2 model was selected due to its better ability to perform instance-level segmentation. Performance analysis revealed remarkable figures on different fronts. The model revealed a confidence rate of 99% in detecting defects with a total accuracy of 98.5% and an average Mean Average Precision (mAP) of 97.2%. Performance of bounding box detection provided an Average Precision (AP) of 41.95 with AP50 and AP75 scores of 68.58 and 44.06 respectively. Most significantly, segmentation performance achieved an AP of 45.14, with AP50 of 68.58 and AP75 of 64.64, confirming the model's capability to accurately define defect boundaries. By eliminating human subjectivity, the system reduces the likelihood of defective products reaching consumers. This research demonstrates that instance segmentation using deep learning is a revolutionary technology for pharmaceutical quality control. The high precision, real-time detection provided by Detectron2 provides manufacturers with an efficient way of keeping high regulatory compliance while achieving high production throughput. This research paves the way for increased AI integration in pharmaceutical production with potential extension into other quality factors and intricate pharmaceutical formulations to create automated quality control systems.