CN-121982028-A - Multi-depth-of-field image fusion multi-layer transparent body defect positioning invasiveness assessment method and defect positioning system

CN121982028ACN 121982028 ACN121982028 ACN 121982028ACN-121982028-A

Abstract

The invention discloses a multi-layer transparent defect positioning invasiveness assessment method and a defect positioning system for multi-depth image fusion, wherein the pre-positioning of an XY coordinate of a suspected defect is completed through global scanning, and then a multi-depth image sequence is acquired along a Z axis by adopting a mixed mode of uniform step scanning and key layer fixed-point shooting; the method comprises the steps of generating an accumulated gray level information image by fusing image sequences, calculating the similarity between an image ROI and the accumulated gray level information image ROI through normalization cross correlation and other algorithms, screening candidate images, calculating definition by adopting a gradient amplitude method and the like, determining the position of a defect Z axis according to the depth corresponding to the maximum value of definition, and quantitatively evaluating the invasiveness degree of the defect to a key layer by combining the definition secondary peak value of the key layer fixed point image and a double-sided detection result, so that the micron-level automatic positioning of the defect is realized, the background interference resistance is high, and the detection efficiency and consistency are greatly improved by a full-automatic process.

Inventors

LIU JIEFENG
SONG XIUZHENG
ZHENG ZILIANG
Feng Tangdong

Assignees

上海帆声图像科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260408

Claims (10)

1. The multi-layer transparent body defect positioning invasiveness assessment method for multi-depth image fusion is characterized by comprising the following steps of: s1, performing global scanning on a multilayer transparent body by adopting a linear array scanning camera, identifying and recording XY coordinates of all suspected defects, moving a detection module to the position above the XY coordinates of the suspected defects, switching to a high-magnification objective lens, and acquiring a multi-depth-of-field image sequence along the Z-axis direction by adopting a mixed mode of step scanning and key layer fixed-point shooting; s2, carrying out gray accumulation or averaging on the images obtained by step scanning to generate an accumulated gray information image, carrying out image preprocessing, defect characteristic enhancement and structure extraction on the accumulated gray information image, and automatically detecting and determining a global region of interest (ROI) of the defect; s3, performing similarity calculation on the ROI area of each frame of image in the original image sequence and the ROI area of the accumulated gray information image, and selecting the previous k images to form a candidate image set according to similarity score sorting; s4, determining the physical depth of the defect by using the Z-axis shooting depth matched with the image corresponding to the maximum value of the definition score, judging the specific level of the defect by combining the material layer thickness information of the multilayer transparent body, analyzing the definition score of the key layer fixed-point shooting image, quantitatively evaluating the invasiveness degree of the defect to the key layer by combining the ratio and the position relation of the main peak value and the secondary peak value of the definition, repeating the steps S1-S4 after the multilayer transparent body is turned over, and verifying and perfecting the three-dimensional space position and the invasiveness evaluation result of the defect by combining the positive and negative detection results.
2. The method according to claim 1, wherein the step-and-scan in the step S1 is implemented by setting the total thickness of the multilayer transparent body in the Z-axis direction as n, setting the depth of field step as deltaz, driving a camera or a sample stage to move along the Z-axis direction from the upper surface to the lower surface of the multilayer transparent body by deltaz as a step, continuously acquiring m images, wherein m is the number of shooting layers and m=n/deltaz, and performing fixed-point focusing shooting on the surface or the interface depth of each key Layer according to the known structural information of the multilayer transparent body, wherein the key Layer comprises at least one of a Cover Layer core glass Layer and a UTG ultrathin glass Layer.
3. The method of claim 1, wherein the image preprocessing in step S2 adopts one or more combination modes of median filtering, bilateral filtering, gaussian filtering, guided filtering and adaptive filtering to realize noise reduction and smoothing processing of the accumulated gray information map, and the defect feature enhancement and structure extraction adopts one or more combination modes of edge detection operators, gradient analysis, morphological processing and multi-scale feature extraction methods to extract outline or structure information of defects and complete defect region judgment and segmentation.
4. The method according to claim 1, further comprising the step of expanding or correcting the boundary of the ROI after the global region of interest ROI is determined in step S2, so as to form a stable and consistent subsequent analysis region.
5. The method of claim 1, wherein the similarity calculation in step S3 uses at least one algorithm selected from the group consisting of normalized cross-correlation, structural similarity index, and mutual information, wherein the normalized cross-correlation is used to measure the linear correlation degree of two image regions, the structural similarity index comprehensively evaluates the similarity of the image regions from three aspects of brightness, contrast, and structure, and the mutual information is used to measure the statistical correlation of the two image regions from the perspective of information theory.
6. The method according to claim 1, wherein in step S3, the sharpness quantization calculation uses at least one operator of the Sobel gradient amplitude method and the laplace energy function method, the sharpness is represented by the image focusing degree, and the sharpness score is highest when the defect is focused on the physical layer where the defect is located.
7. The method of claim 1, wherein the invasiveness evaluation determination rule of the defect to the critical layer in the step S4 is that if the definition corresponding to the level where the main body of the defect is located is a peak value, the surface fixed point images of the adjacent critical layers show a sub-high definition score and form a sub-peak value, the defect is determined to have caused the invasiveness damage of the indentation and the bulge on the surfaces of the adjacent critical layers, and the invasiveness depth and the damage degree of the defect to the critical layers are quantified through the ratio of the main peak value to the sub-peak value and the Z-axis position distance corresponding to the main peak value and the sub-peak value.
8. The method of any of claims 1-7, wherein the multilayer transparency comprises any of an ultra thin glass cover plate, an OLED display panel, an LCD liquid crystal module, a touch screen sensor, a multilayer optical film, a composite transparency, and the defect comprises at least one of bubbles, impurity particles, scratches.
9. The defect positioning system for realizing the method according to any one of claims 1-7, comprising an imaging module, a motion control module, a light source module, a calculation and control module and a data storage and communication module, wherein the imaging module comprises an industrial camera and a matched high-magnification objective lens, is used for carrying out global scanning imaging and high-resolution layered imaging of different depth-of-field positions on a multilayer transparent body, the motion control module comprises a Z-axis precise lifting mechanism and a driving unit thereof, is used for driving the camera or a sample table to move along the Z-axis direction by a set stepping amount or realizing fixed-point focusing at a critical layer depth, the light source module is used for providing illumination for the detection of the multilayer transparent body, the calculation and control module comprises an industrial computer or an embedded processing unit, is used for coordinating the time sequence of camera acquisition, Z-axis motion control and light source triggering, and is also used for executing algorithms of image fusion, ROI extraction, similarity calculation, definition calculation, defect positioning and invasiveness evaluation, and the data storage and communication module is used for storing image data, calculation data and detection results in the detection process.
10. The defect positioning system according to claim 9, wherein the calculation and control module is internally provided with an image preprocessing module, a similarity calculation module, a definition analysis module and an invasiveness evaluation module, wherein the image preprocessing module is used for realizing noise reduction and feature enhancement of the accumulated gray level information graph, the similarity calculation module is used for pre-storing calculation algorithms of normalized cross-correlation, structural similarity indexes and mutual information, the definition analysis module is used for pre-storing calculation operators of a Sobel gradient amplitude method and a Laplace energy function method, and the invasiveness evaluation module is used for completing quantitative determination of defect invasiveness according to a definition primary peak value and a double-sided detection result.

Description

Multi-depth-of-field image fusion multi-layer transparent body defect positioning invasiveness assessment method and defect positioning system Technical Field The invention relates to the technical field of machine vision industrial detection, in particular to a multi-depth-of-field image fusion multi-layer transparent body defect positioning invasiveness assessment method and a defect positioning system, which are mainly applied to high-precision defect detection of industrial ultrathin glass, multi-layer display panels and composite transparent materials, and are particularly suitable for detection and performance influence assessment of internal layering defects of transparent bodies. Background In the high-end manufacturing fields of ultrathin glass, vehicle-mounted display screens, mobile phone cover plates and the like, products are formed by laminating a plurality of layers of transparent materials such as a glass substrate, an optical adhesive OCA layer, a functional film layer, a protective film PF layer and the like, and tiny defects such as bubbles, impurity particles and scratches in the products can seriously influence the optical performance and the structural strength of the products, so that the high-precision detection of the internal defects of the multilayer transparent bodies is a key link in the production process. In the prior art, a multi-layer focusing mode is adopted for defect detection of a multi-layer transparent body, namely, the focal length of a camera is manually or semi-automatically adjusted, a series of images are shot at different depths, then an operator browses the images frame by frame, and the most clear imaging frame of the defect is judged subjectively by means of human eyes, so that the layer on which the defect is positioned is estimated. However, the method has obvious technical bottleneck and cannot meet the requirements of industrial mass and high-beat automatic production: the whole defect positioning process is highly dependent on experience and working state of operators, the manual frame-by-frame analysis mode is low in efficiency and poor in repeatability of detection results, and detection requirements of an automatic production line are difficult to adapt; The prior method has no effective anti-interference algorithm, black speckle noise generated when the PF layer of the protective film is imaged due to the material characteristics, surface dust or slight scratch of a non-critical layer, which is as clear as a real defect on a specific focal plane, is easy to be misjudged as a core defect, causes over-killing of products and reduces the production yield; The defect invasiveness cannot be estimated, the detection dimension is single, the prior art can only roughly judge the approximate position of the defect, can not accurately estimate whether the defect invades or damages a core functional layer of a product, and can not quantify the damage degree of the defect, so that an effective basis cannot be provided for judging the risk level of the defect, and the high-quality detection requirement of a high-end transparent material is difficult to meet. In summary, there is a need in the art for an intelligent detection method capable of realizing full-automatic and high-precision positioning of defect positions of a multilayer transparent body and evaluating invasiveness of defects to a critical layer in an energy-based manner, so as to solve various defects in the prior art. Disclosure of Invention The invention aims to overcome the defects that the existing multilayer transparent defect detection technology depends on manpower, has poor anti-interference performance and cannot evaluate defect invasiveness, provides a multilayer transparent defect positioning invasiveness evaluation method for multi-depth image fusion, and simultaneously provides a defect positioning system for realizing the method. The method has the specific purposes of realizing full automation and micron-sized accurate positioning of the Z-axis position of the internal defect of the multilayer transparent material, completely discarding manual intervention, improving detection efficiency and consistency, establishing a robust image processing flow, effectively eliminating interference such as imaging noise, surface dust, slight scratch and the like of non-critical layers such as a protective film PF layer, accurately locking real defects, reducing misjudgment rate, accurately evaluating whether the defects such as particles cause destructive influence on a core layer by combining uniform step scanning with fixed-point shooting of the critical layers and detecting and verifying the front surface and the back surface, quantifying invasion degree, providing scientific quantification basis for judging the risk level of the defects, constructing a hardware system adapting to the detection method, realizing cooperative work of imaging, motion control and data processing, and providing stable