Weak supervision progressive multi-level flexible circuit board registration model and method

Through deep learning and gradual optimization technology, combined with deformation field learning, the deformation problem in flexible circuit board registration is solved, efficient and robust registration effect is achieved, and labeling costs are reduced.

CN118918156APending Publication Date: 2024-11-08苏州深视信息科技有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410996692.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-24
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The existing circuit board registration methods are difficult to effectively solve the deformation problems such as bending and compression of flexible circuit boards, especially methods based on camera calibration, grayscale information, feature matching and transformation domains cannot adapt to the elastic transformation of flexible circuit boards.

Method used

A weakly supervised progressive multi-level flexible circuit board registration model is adopted, including feature extraction module, global coarse matching module, local fine matching module and upsampling module. Image features are extracted through deep learning technology, combining progressive optimization and deformation field Learn to adapt to the twisting and deformation of flexible circuit boards.

Benefits of technology

It realizes efficient registration of flexible circuit boards, improves robustness and robustness, can accurately match images at different resolutions, reduces the workload of manual labeling, and adapts to complex scenes and noise.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118918156A_ABST
    Figure CN118918156A_ABST
Patent Text Reader

Abstract

The invention provides a weak supervision progressive multi-level flexible circuit board registration model and method. The model comprises a feature extraction module, a global rough matching module, a local fine matching module and an up-sampling module. The feature extraction module extracts high-resolution features and low-resolution features in the image; the global rough matching module carries out rough matching on the low-resolution characteristics of the graph A and the graph B; the up-sampling module carries out scale expansion on the deformation field and the confidence output by the global rough matching module; and the local fine matching module performs local matching on the high-resolution characteristics of the image A and the image B. According to the method, a deep learning technology is adopted, image features are extracted through a neural network model, a multi-level progressive learning method is adopted, deformation fields between images and a reference image under different resolutions are output, a weak supervised learning method is adopted, one of a manually-labeled training sample and a training sample without manual labeling is used for training, and the training efficiency is improved. And while the robustness of the model is improved, the investment of manually making samples is reduced.
Need to check novelty before this filing date? Find Prior Art