Method for generating multi-target data training set based on verification picture

CN116051405BActive Publication Date: 2026-02-27ZHEJIANG UNIV OF TECH
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202211700289.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-12-28
Publication Date
2026-02-27
Estimated Expiration
2042-12-28

AI Technical Summary

Technical Problem

Existing deep learning datasets have limited categories, making it difficult to effectively train multi-object detection models. Furthermore, manually labeling data is labor-intensive and inefficient, hindering the development of artificial intelligence in interdisciplinary fields.

Method used

By preprocessing unlabeled images, preliminary segmentation is performed using a pre-trained object detection network model. Verification images are generated by combining credibility calculation and push strategies during the user login verification process. Multi-object data training sets are generated using user annotations, and image data is labeled on a large scale using a crowdsourcing approach.

Benefits of technology

It achieves efficient and automated generation of multi-target data training sets, reduces the workload of manual annotation, improves the generalization ability and recognition accuracy of training data, and supports the training of multi-target detection models.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116051405B_ABST
    Figure CN116051405B_ABST
Patent Text Reader

Abstract

The application discloses a multi-target data training set generation method based on verification pictures, adopts a pre-trained target detection network model to recognize preprocessed unlabeled pictures, splits out a to-be-detected target region according to a recognition result, splits the rest of the unlabeled picture according to a preset splitting strategy to obtain a split unknown split picture, splices the unknown split picture and a known split picture pushed according to a picture pushing strategy to generate a verification picture pushed to a user, and finally realizes labeling of the unlabeled picture and puts the unlabeled picture into a training set after obtaining a labeling result of the user on the unknown split picture and the known split picture in the verification picture. The application can solve the difficulty of low efficiency and heavy workload in the manual production process of the training set, can provide a corresponding training set for model training more quickly, and provides help for further development of the field of artificial intelligence and machine learning.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Network verification system and method, client and server

    CN105306603A

  • Login verification method and device, equipment and storage medium

    CN113194079A