Medicine Identification System on Mobile Devices for the Elderly

This research develops an application that helps the elderly to identify medicine from a mobile image, to reduce confusion in taking medication, and thus to reduce the rate of medication errors. The data used in this research are collected from the medicine blister packs for the elderly consisting of 14 types of medicine, which are taken with the smartphone cameras and amounting to a total of 56,000 single medicine blister pack images for image classification model training. For object detection model training, there are a total of 21,000 single medicine blister pack images with added multiple medicine blister pack images amounting to 120 images from the image dataset. Text recognition is used to identify the medicine type using Keras-OCR. For all experimental results in the image classification model experiments reveal that the MobileNet V2 with 14-class detection has the highest accuracy at 93.79 percent. The object detection model is the MobileNet V1 with the highest mAP of 0.875 with the Average Precision with 0.5 IoU and 0.75 IoU at 0.998 and 0.91, respectively.