인공지능 기반 장비의 자동탐지 성능인증을 위한 시험법 설계

Purpose: This paper examined the design of a standard test method for certifying the performance of automatic detection equipment using artificial intelligence (AI). Methods: First, indicators for measuring the detection performance were determined and the terms were redefined. Then, the number of iterations required to give a criterion value at a given confidence level for data following the binomial distribution was calculated. As a case study, an aviation security x-ray scanner was tested. Results: If failures occur before the number of allowable times during the trial, the minimum probability of occurrence of true positive (TP) can be demonstrated. This means that F2-Score is satisfied, and the performance can be certified. Conclusion: This method can be used in industries that operate automatic detectors based on AI technology. The method should contribute to the development of a national certification system and improve product quality control.