Method for Improving Reliability of Artificial Intelligence based Object Recognition by using Collective Intelligence based Mutual Verification

摘要

According to the method for improving the reliability of AI-based object recognition using mutual verification based on collective intelligence of the present invention, in a method executed through a server interworking with an AI module, the formalization of an object to be recognized through the AI module Learning by injecting M(M≥2) structured data corresponding to an image into the artificial intelligence module, and learning N(N≥2) including one or more images of a structured image and an unstructured image for the object to be recognized Prepare data, recognize one or more object areas included in the N learning data, extract N object area recognition data that set the recognized object area on the N learning data, and U(U≥2 ), i(1≤i≤U) user, i(1≤i≤U) user terminal, which provides ni(1≤ni≤N) object area recognition data according to a specified order. A procedure for receiving ni object area selection data selected from at least one effective object area corresponding to the object to be recognized among one or more object areas included in the object area recognition data, and j(1≤j≤U, i ≠j) One or more objects included in the nj object area recognition data from the jth user terminal by providing nj (1≤nj≤N) object area recognition data in the specified order to the jth user terminal used by the user. A procedure for receiving nj object region selection data receiving at least one valid object region corresponding to the object to be recognized among the regions is performed for a specified time for a specified number of u (1 ≤ u ≤ U) users, and the Specified among the v(1≤v≤u) users who select the effective object area for each object area selection data by comparing and analyzing the object area selection data for each of the u users received from u user terminals for the same object area selection data Performs a collective intelligence-based cross-validation procedure that selects n (1 ≤ n ≤ N) object region selection data where a user with a ratio or higher selects the same effective object region In order to improve the reliability of the AI module, n learning data to be trained by injecting into the AI module are determined, and the determined n learning data is injected into the AI module to be trained.