Data cleaning method, device and equipment and storage medium

By using the confidence prediction model to clean the picture data of PCB board components during the SMT production process, the problems of low manual discrimination efficiency and low accuracy are solved, efficient and accurate data cleaning is achieved, and data set quality and detection accuracy are improved.

CN119669506AActive Publication Date: 2025-03-21BEIJING XIAOMI MOBILE SOFTWARE CO LTD
10 Cites 0 Cited by

Patent Information

Application Number
CN202311227476.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-09-21
Publication Date
2025-03-21
Estimated Expiration
2043-09-21

AI Technical Summary

Technical Problem

During the SMT production process, the error problem of image data labels of PCB board components mainly relies on manual judgment, resulting in low discrimination efficiency, large-scale impact on detection results, poor discrimination consistency, wasted a lot of manpower and low detection accuracy.

Method used

By collecting component picture data sets, the confidence of the first tag is determined, and the second tag is predicted based on the confidence, and the cleaning of the first tag is performed. The specific steps include inputting component picture data into the prediction model for classification, cross-verification determines the category probability as confidence, and retaining the first label if the confidence is greater than the threshold, otherwise cleaning the first label.

Benefits of technology

It realizes efficient and accurate data cleaning, improves the quality of data sets, reduces labor costs, improves the speed and accuracy of discrimination, and adapts to the needs of industrial production.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention relates to a data cleaning method, device and equipment and a storage medium, the data cleaning method comprises the steps that a component picture data set is collected, the component picture data set comprises a plurality of component pictures, each component picture in the plurality of component pictures is marked with a first label, and the first label is a first label; the first label is used for marking qualified mounting or mounting defects of the component on the printed circuit board; determining the confidence coefficient of the first tag, and predicting the tag of each component picture in the component picture data set based on the confidence coefficient of the first tag to obtain a second tag; and cleaning the first label based on the second label. According to the data cleaning method and device, data cleaning which is high in efficiency, low in cost, high in accuracy, high in judgment speed, short in re-judgment time and capable of being popularized and reused can be achieved.
Need to check novelty before this filing date? Find Prior Art