Chip end face image definition evaluation method and system based on semantic information

By using a semantic information method based on the CLIP model and adaptive prompt words, the problems of inconsistent evaluation scales and interference from local high-frequency features in chip end-face image detection are solved, enabling accurate sharpness judgment of a single image and improving the accuracy and efficiency of evaluation.

CN122453744APending Publication Date: 2026-07-24HUNAN UNIV OF SCI & TECH SANYA RES INST
0 Cites 0 Cited by

Patent Information

Application Number
CN202610556897.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-04-24
Publication Date
2026-07-24

AI Technical Summary

Technical Problem

Existing sharpness evaluation functions suffer from inconsistent evaluation scales and susceptibility to interference from local high-frequency features in chip end-face image detection, leading to inaccurate evaluations.

Method used

We employ a semantic information method based on the CLIP model, and construct an adaptive sharpness evaluation system using adaptive cue words and the k-sigma criterion. This system uses image semantic information to determine the sharpness of a single image, avoiding the influence of different depths of focus and texture differences.

Benefits of technology

This technology enables a unified evaluation scale to be used even when the number, density, and shape of image textures vary across different chip end faces. This improves the accuracy and efficiency of the evaluation, reduces the movement of mechanical structures, and extends the service life.

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

Abstract

The invention provides a semantic information-based chip end face image definition evaluation method and system. The method comprises the steps of obtaining a to-be-evaluated chip end face image; the chip end face image is input to an image degradation feature extraction module, degradation features of the image are extracted, global average pooling processing is carried out, and a guide vector is obtained through MLP network mapping; constructing a self-adaptive cue word pair, processing an image to be evaluated through an image encoder to obtain an image feature vector, and processing the self-adaptive cue word pair through a text encoder to obtain a cue word feature vector; and calculating a definition score of the to-be-evaluated image through a Softmax function, and comparing the definition score with a preset threshold to obtain a judgment result. According to the method, the problem of different picture definition evaluation scales under the condition of different texture numbers and densities is solved, and the evaluation score only depends on the overall definition of the picture, so that the problem of inaccuracy caused by definition band deviation scoring of defects at different focal depths of the chip end face image is avoided.
Need to check novelty before this filing date? Find Prior Art