CN-122023347-A - Visual image layering detection system and method based on AI

CN122023347ACN 122023347 ACN122023347 ACN 122023347ACN-122023347-A

Abstract

The application discloses a visual image layering detection system and a visual image layering detection method based on AI, and relates to the technical field of image analysis, and the visual image layering detection system comprises the steps of acquiring a two-dimensional planar visual image of a sample, dividing a scanning area of the sample, counting to obtain an irregular subarea and a flat subarea of the sample, carrying out subarea scanning optimization analysis on the divided areas to obtain a high-resolution topography of the flat subarea of the sample and an optimized topography of the irregular subarea, carrying out rigidity phase analysis on the sample after subarea scanning optimization to obtain a rigidity distribution map and a phase distribution map of the sample, fusing the high-resolution topography of the flat subarea of the sample and the optimized topography of the irregular subarea to obtain a comprehensive topography of the sample, and further carrying out packaging quality detection according to the comprehensive topography map, the phase distribution map and the rigidity distribution map of the sample.

Inventors

YIN FANGTAO
WANG FAN
LU JINBO
ZHOU LONGSHENG
Xin Mingxin
LI YULONG
WANG GUANGYU
WU YANNI
SUN MINGXIA

Assignees

澳立奇科技股份有限公司

Dates

Publication Date: 20260512
Application Date: 20260131

Claims (10)

1. The AI-based visual image layering detection method is characterized by comprising the following steps: Acquiring a two-dimensional plane visual image of the sample by using camera equipment, dividing a scanning area of the sample based on the two-dimensional plane visual image of the sample, and counting to obtain an irregular sub-area and a flat sub-area of the sample; Scanning and optimizing analysis is carried out on the irregular subarea and the flat subarea of the segmented sample, and a high-resolution topography map of the flat subarea and an optimized topography map of the irregular subarea of the sample are obtained; carrying out rigidity phase analysis on the sample after the partition scanning optimization to obtain a rigidity distribution diagram and a phase distribution diagram of the sample; and fusing the high-resolution topography map of the flat subarea and the optimized topography map of the irregular subarea of the sample to obtain a comprehensive topography map of the sample, and further detecting the packaging quality according to the comprehensive topography map, the phase distribution map and the rigidity distribution map of the sample.
2. The AI-based visual image layering detection method as set forth in claim 1, wherein said sample-based two-dimensional planar visual image segments a scanned area of a sample, and the specific segmentation process is: Carrying out image preprocessing on a two-dimensional plane visual image of a sample, taking each pixel point in the two-dimensional plane visual image of the sample after the image preprocessing as a data sample, constructing a feature vector for each pixel point, wherein the feature vector of each pixel point comprises a color intensity value and a local texture feature value, clustering the feature vectors of all the pixel points by adopting a clustering algorithm, marking a region with simple texture and uniform color as a flat sub-region according to a clustering result, marking a region with complex texture or abrupt color change as an irregular sub-region, and carrying out statistics to obtain the irregular sub-region and the flat sub-region of the sample.
3. The AI-based visual image layering detection method of claim 1, wherein said scanning optimization analysis is performed on irregular and flat subareas of the segmented sample, and comprises the following specific analysis processes: Scanning the flat subarea of the sample by using the vibration mode of the AFM to obtain a high-resolution topography map of the flat subarea of the sample, scanning the irregular subarea of the sample by using the contact mode of the AFM to obtain a basic topography map of the irregular subarea of the sample, and optimizing the basic topography map of the irregular subarea of the sample to obtain an optimized topography map of the irregular subarea of the sample.
4. The AI-based visual image layering detection method of claim 3, wherein said obtaining an optimized morphology map of irregularly segmented regions of said sample comprises the following steps: Scanning an irregular subarea of a sample by using a contact mode of an AFM, synchronously collecting instantaneous acceleration and instantaneous speed of each pixel point probe in the vertical direction of the irregular subarea of the sample in the scanning process, recording a difference value between the instantaneous acceleration of each pixel point probe in the vertical direction and corresponding reference instantaneous acceleration as an instantaneous acceleration deviation value of each pixel point probe in the vertical direction, comparing an absolute value of the instantaneous acceleration deviation value of each pixel point probe in the vertical direction with a set instantaneous acceleration deviation threshold value, and recording the pixel point as a pixel point to be optimized if the absolute value of the instantaneous acceleration deviation value of the pixel point probe in the vertical direction is higher than the set instantaneous acceleration deviation threshold value, or else, not marking the pixel point as the pixel point to be optimized; And counting to obtain each pixel point to be optimized, and optimizing each pixel point to be optimized by combining the basic topography map of the irregular subarea of the sample, the instantaneous acceleration deviation value, the instantaneous speed and the set natural resonance frequency of each pixel point to be optimized in the vertical direction of the probe to obtain the optimized topography map of the irregular subarea of the sample.
5. The AI-based visual image layering detection method of claim 4, wherein said optimizing each pixel to be optimized comprises the following steps: The method comprises the steps of preprocessing data of instantaneous acceleration deviation values, instantaneous speeds and set natural resonance frequencies of pixel point probes to be optimized in the vertical direction, and inputting the data into a trained compensation optimization model to obtain optimized compensation quantity of each pixel point to be optimized; Reading an original height value of each pixel point to be optimized from a basic topography map of an irregular sub-area of a sample, if the instantaneous speed of the pixel point to be optimized in the vertical direction is greater than zero, indicating that the pixel point to be optimized probe is moving upwards, carrying out negative optimization on the pixel point to be optimized, subtracting a corresponding optimization compensation amount from the original height value of the pixel point to be optimized to obtain a target height value of the pixel point to be optimized, if the instantaneous speed of the pixel point to be optimized in the vertical direction is less than zero, indicating that the pixel point to be optimized is moving downwards, carrying out positive optimization on the pixel point to be optimized, adding the corresponding optimization compensation amount to the original height value of the pixel point to be optimized to obtain a target height value of the pixel point to be optimized, and replacing original values in the basic topography map with the target height values of all marked pixel points to be optimized to obtain an optimized topography map of the irregular sub-area of the sample.
6. The AI-based visual image layering detection method of claim 1, wherein the stiffness phase analysis is performed on the sample after the zonal scanning optimization, and the specific analysis process is as follows: Arranging a plurality of monitoring points in each sub-region of a sample by adopting a characteristic guide point distribution method, synchronously recording the space coordinates of each monitoring point, controlling a probe to vertically execute a force-distance nano indentation micro experiment on each monitoring point to obtain a contact stiffness value of each monitoring point, and obtaining a stiffness distribution diagram of the sample based on the space coordinates and the contact stiffness values of all the monitoring points, wherein each sub-region of the sample comprises an irregular sub-region and a flat sub-region of the sample; scanning the sample by using the vibration mode of the AFM to obtain a phase distribution diagram of the sample.
7. The AI-based visual image layering detection method of claim 1, wherein said packaging quality detection is based on a comprehensive topography, a phase distribution and a stiffness distribution of a sample, specifically a three-level layering determination: the method comprises the steps of a first layer, performing primary screening on rigidity threshold values of all sub-areas of a sample based on a rigidity judgment threshold value and a comprehensive morphology graph of the sample, and marking a high rigidity candidate area, a to-be-positioned area and a suspension area, wherein the rigidity judgment threshold value comprises a rigidity judgment first threshold value and a rigidity judgment second threshold value; The second layer is used for judging phase signals of the high-rigidity candidate area and the area to be determined, and finally judging a bonding area, a part of bonding area or a suspension area; And the third layer is used for transmitting the packaging quality layering judging result comprising a bonding region, a part of bonding region, a suspension region and a high-rigidity abnormal region to an AI visual image detection platform for visual output and display.
8. The AI-based visual image layering detection method of claim 7, wherein the stiffness threshold preliminary screening is performed on each divided area of the sample based on the stiffness determination threshold and the comprehensive morphology map of the sample, and comprises the following specific steps: And (3) carrying out region detection on the comprehensive morphology graph of the sample, marking the subarea as a high-rigidity candidate region if the average value of the subarea contact rigidity is larger than the second threshold value for rigidity judgment, comparing the average value of the subarea contact rigidity with the first threshold value for rigidity judgment if the average value of the subarea contact rigidity is smaller than or equal to the second threshold value for rigidity judgment, marking the subarea as a pending region if the average value of the subarea contact rigidity is larger than or equal to the first threshold value for rigidity judgment, and marking the subarea as a suspended region if the average value of the subarea contact rigidity is smaller than the first threshold value for rigidity judgment.
9. The AI-based visual image layering detection method of claim 7, wherein said phase signal determination for said high stiffness candidate region and said region to be localized comprises the following steps: Judging the phase signal of the high-rigidity candidate region, if the phase signal of the high-rigidity candidate region is normal, finally judging the lamination region, if the phase signal of the high-rigidity candidate region is abnormal, marking the high-rigidity candidate region as a high-rigidity abnormal region, judging the phase signal of the region to be localized, if the phase signal of the region to be localized is normal, finally judging the lamination region as a part, and if the phase signal of the region to be localized is abnormal, marking the region to be localized as a suspension region.
10. An AI-based visual image layering detection system, comprising: the image acquisition and segmentation module is used for acquiring a two-dimensional plane visual image of the sample by using the image pickup equipment, further segmenting a scanning area of the sample based on the two-dimensional plane visual image of the sample, and counting to obtain an irregular sub-area and a flat sub-area of the sample; The partition scanning module is used for scanning and optimizing analysis on the irregular subareas and the flat subareas of the segmented sample to obtain a high-resolution topography map of the flat subareas and an optimized topography map of the irregular subareas of the sample; the rigidity phase diagram analysis module is used for carrying out rigidity phase analysis on the sample after the partition scanning optimization to obtain a rigidity distribution diagram and a phase distribution diagram of the sample; The image fusion detection module is used for fusing the high-resolution topography map of the flat subarea of the sample and the optimized topography map of the irregular subarea to obtain the comprehensive topography map of the sample, and further carrying out packaging quality detection according to the comprehensive topography map, the phase distribution map and the rigidity distribution map of the sample.

Description

Visual image layering detection system and method based on AI Technical Field The invention relates to the technical field of image analysis, in particular to an AI-based visual image layering detection system and method. Background In the technical field of visual image detection, along with continuous advancement of industrial automation and intellectualization, accurate and efficient image detection becomes one of core technologies in manufacturing, quality control and safety monitoring, and especially in a complex production environment, the quality and detail of an image are often affected by various factors, so that the traditional image detection method is difficult to meet the high-precision requirement. The prior art, such as the Chinese patent application publication No. CN116843632B, discloses an automatic nanoparticle detection method based on deep learning and AFM images. The method comprises the steps of collecting an AFM original image of the nano particles, processing the AFM original image to obtain a normalized height image, establishing a deep learning model and training the model, collecting an AFM original image of the nano particles to be detected, processing the AFM original image, inputting the detected normalized height image into the trained model, outputting the region, agglomeration information and confidence information of the nano particles to be displayed in a detection visualization height image, obtaining pixel-size factors, size information, size distribution information and agglomeration proportion of the nano particles according to the original detection image and the detection height image, and finally realizing automatic detection of the nano particles. In combination with the above scheme, in the technical field of image detection, detection and analysis are generally performed on a specific target in an image, however, in the detection process by using an AFM, the problem that abnormal oscillation occurs to a probe due to the fact that the interaction force between the tip of the AFM and the surface of a material dynamically changes along with the appearance of a substrate exists, the appearance of the substrate is such as a deep and narrow groove, a steep edge and the like, for example, when the probe spans the upper edge of the groove, the tip is captured by the adsorption force of the surface of the groove after being temporarily separated from contact, so that an abnormal accelerated dropping is generated, a scanning system erroneously interprets the instability process of the tip as that the probe is tracking a real geometric feature, so that a dropping signal is erroneously recorded as a part of the appearance of the side wall of the groove, the problem that the depth of the groove is exaggerated, the trailing of the edge profile is distorted and the like occurs in the final image, and the acquired appearance image is distorted is caused, so that the accurate measurement of the areas such as the depth of the surface groove and the edge profile is affected, and the false diagnosis of the subsequent package quality detection is caused. Disclosure of Invention Aiming at the defects of the prior art, the invention provides an AI-based visual image layering detection system and an AI-based visual image layering detection method, which can effectively solve the problems related to the background art. In order to achieve the above object, the first aspect of the present invention is achieved by the following technical solutions, including: and acquiring a two-dimensional plane visual image of the sample by using the image pickup equipment, dividing a scanning area of the sample based on the two-dimensional plane visual image of the sample, and counting to obtain an irregular sub-area and a flat sub-area of the sample. Scanning and optimizing analysis is carried out on the irregular subarea and the flat subarea of the segmented sample, and a high-resolution topography map of the flat subarea and an optimized topography map of the irregular subarea of the sample are obtained. And (3) carrying out rigidity phase analysis on the sample after the partition scanning optimization to obtain a rigidity distribution diagram and a phase distribution diagram of the sample. And fusing the high-resolution topography map of the flat subarea and the optimized topography map of the irregular subarea of the sample to obtain a comprehensive topography map of the sample, and further detecting the packaging quality according to the comprehensive topography map, the phase distribution map and the rigidity distribution map of the sample. The method comprises the steps of carrying out image preprocessing on a two-dimensional plane visual image of a sample, taking each pixel point in the two-dimensional plane visual image of the sample after image preprocessing as a data sample, constructing a characteristic vector for each pixel point, wherein the characteristic vector of each pixel point com