Development of an AI-Based Model for Identifying Doping Drugs Through Pill Image Analysis

초록

The purpose of this study is to develop a model for identifying prohibited doping drugs through AI-based pill image analysis. To achieve this objective, the study utilized the oral medication image dataset provided by AI Hub. The dataset was divided into prohibited and allowed drugs, and a total of 160,056 images of 500 types of oral medications were used. The performance of the prediction model was evaluated using five metrics: Recall, Precision, Sensitivity, Specificity, and Accuracy. A CNN (Convolutional Neural Network) model and the SIFT (Scale-Invariant Feature Transform) algorithm were applied to analyze the external features of the drugs, such as size, color, and shape. The model demonstrated a high accuracy of 96.5% in distinguishing between prohibited and allowed drugs. Moreover, it achieved superior performance in key metrics, including Recall/Sensitivity (95.2%), Precision (98.6%), and Specificity (98.2%).

키워드