Product defect management method and system and computer readable storage medium

Abstract

The invention relates to the technical field of computers, and provides a product defect management method, which comprises the following steps that: an algorithm identification unit obtains a picture of a product, identifies the picture based on a preset defect identification model to obtain an algorithm identification result, and stores the picture and the algorithm identification result into a database; the artificial recognition unit obtains a picture with an algorithm recognition result being defective or unrecognizable, provides the picture and the algorithm recognition result for a user for artificial recognition, and collects an artificial recognition result fed back by the user as a defect recognition result of the product; a sampling inspection unit in the defect identification module obtains an algorithm identification result of a preset proportion from the database, forwards the algorithm identification result to an artificial identification unit for sampling inspection, and collects a sampling inspection identification result fed back by the artificial identification unit; and the monitoring module counts the probability of inconsistency with the algorithm recognition result in the manual recognition result and the spot check recognition result, and if the probability exceeds a first preset threshold value, a preset defect recognition model is optimized.

Description

technical field [0001] The present application relates to the field of computer information technology, and in particular, to a product defect management method, system, and computer-readable storage medium. Background technique [0002] At present, the detection of most online AOI (Automated Optical Inspection, automatic optical inspection) equipment is based on traditional image processing algorithms. Although the real-time performance is good and the import is convenient, it is often reckless when dealing with detection targets with complex patterns / textures / backgrounds. The stick is not good, and it is easy to be over-checked or missed. In order to ensure the quality, the factory can only strictly control the algorithm parameters and improve the defect detection rate, and then arrange manpower to re-evaluate the "defective products" that have a large number of over-inspected products determined by the AOI equipment, so as to recover the over-inspected products, and worke...

AI Technical Summary

Problems solved by technology

[0003] In view of the above problems, in order to reduce the reliance on labor and improve the efficiency of feedback, an embodiment of the present application provides a product defect management method, which is applied to a product defect management system, wherein the system includes a defect identification module and a monitoring module, the method includes the steps: the algorithm identification unit in the defect identification module acquires a picture of the product, identifies the picture based on a preset defect identification model to obtain an algorithm identification result, and combines the picture and The algorithm identification results are stored in the database; the manual identification unit in the defect identification module obtains the pictures whose identification results of the algorithm are defective or unrecognizable, and provides the pictures and the algorithm identification results to the user for further processing. Manual identification, and collect the manual identification results fed back by users as the defect identification results of the product; the sampling unit in the defect identification module obtains the algorithm identification results of a preset ratio from the database, and forwards them to the artificial The identification unit performs random inspection, and collects the random inspection identification results fed back by the artificial identification unit; the monitoring module counts the probability of the artificial identification results and the random inspection identification results being inconsistent with the algorithm identification results, if the probability exceeds the first preset threshold, then optimize the preset defect recognition model

Images