This study explores the application of deep convolutional neural networks (CNNs) for accurate pill identification, addressing the limitations of traditional human-based methods. Using a dataset of 1,250 images across 10 household remedy drugs, various CNN architectures, including YOLO models, were tested under different conditions. Results showed:

• Natural lighting was optimal for imprinted pills, while a lightbox improved detection for plain pills.
• The YOLOv5-tiny model demonstrated the best detection accuracy.
• EfficientNet_b0 achieved the highest classification performance.

While the model showed strong results, its generalization is limited by sample size and drug variability. Nonetheless, this approach holds promise for enhancing medication safety and reducing errors in outpatient care.