DETECTION AND IFENTIFICATION OF PILLS USING MACHINE LEARNING

Abstract: In the medical field, precise pill identification and detection are essential for avoiding prescription mistakes and guaranteeing patient safety. Machine learning must be used to automate the process because traditional manual approaches are tedious and susceptible to human mistake. Due to changes in illumination, background noise, and picture quality, conventional rule-based image processing methods—which depend on texture, colour, and shape—frequently have accuracy and robustness issues. Using Convolutional neural networks, more commonly trained on a dataset of pill pictures with data augmentation for improved generalisation, this study suggests a deep learning-based method to overcome these drawbacks. Multiple convolutional, maximal pooling, and layer dropouts are included into the model to improve feature extraction and lessen overfitting. Accuracy under various circumstances is ensured by validating performance on a distinct dataset.