ScholarWorks@성균관대학교

초록

The demand for technology that can rapidly and precisely detect drugs in the field is increasing because of the growing number of criminal incidents involving the tampering of beverages with drugs. This paper presents the development and validation of a strip-type drug test kit that is user-friendly and portable, reduces the consumption of reagents that are hazardous to humans, and produces less waste. A drawback of strip-type drug test kits is that they depend on visual interpretation of the results, which can vary depending on user skill, lighting, and slight color changes. To address this limitation, this study implemented an analytical method that minimizes uncertainty and facilitates quantitative evaluation by utilizing images of color reaction outcomes from the newly developed strip-type drug test kit. Specifically, we developed an automatic discrimination algorithm using artificial intelligence (AI) image analysis techniques. Using a collection of 7,295 kit images, we evaluated the efficacy of several deep-learning models, including Convolutional Neural Network (CNN), Vision Transformer (ViT), and CNN-GRU (a hybrid of CNN and Gated Recurrent Unit). The findings indicated that the CNN-GRU model showed superior performance across all parameters, including accuracy, precision, recall, F1-score, and Cohen's kappa, achieving a perfect score of 1.0000. The model achieved perfect accuracy, surpassing the average performance of 32 police officers in a comparative experiment. This study demonstrated that AI-based precision reading algorithms can significantly improve the automation and reliability of strip-type drug test kits, highlighting their potential applicability and effectiveness in investigative and forensic settings.

키워드