CN-121830720-B - Automatic detection and classification system for surface defects of optical element

CN121830720BCN 121830720 BCN121830720 BCN 121830720BCN-121830720-B

Abstract

The invention relates to the technical field of precise optical detection and discloses an automatic detection and classification system for surface defects of an optical element, which comprises a hemispherical programmable LED lattice light source array, an industrial area array camera, an electric control liquid crystal polarization modulator and a central processing unit, wherein the processor drives the light source array to scan to obtain a time sequence image sequence, a maximum translation characteristic vector and a time domain gray scale variance of a region of interest are extracted, a photometric displacement gradient product is calculated and generated, the system shunts the region of interest to a reconstruction verification or polarization diagnosis queue according to the index, a track reconstruction module confirms surface adhesion defects by verifying diffuse reflection continuous translation attributes, and a polarization module confirms structural damage defects by calculating local polarization contrast. The invention utilizes photometric displacement gradient product and polarization modulation technique to realize accurate classification of surface foreign matters and structural damage, and reduces the over-killing rate of the detection system.

Inventors

ZENG ZHICHAO
ZHANG LAN
XU YAGUO
LI WEI
GONG YIDONG
YANG MEIJIAN

Assignees

上海芬创信息科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260313

Claims (8)

1. The automatic detection and classification system for the surface defects of the optical element is characterized by comprising a hemispherical programmable LED lattice light source array, an industrial area array camera comprising a camera lens, an electric control liquid crystal polarization modulator and a central processing unit, wherein the electric control liquid crystal polarization modulator is assembled at the light emitting end of the hemispherical programmable LED lattice light source array and in front of the camera lens, and the central processing unit comprises: The initialization and image acquisition module is used for driving the hemispherical programmable LED lattice light source array to scan and triggering the industrial area array camera to acquire an initial time sequence image sequence; the positioning and feature extraction module is used for performing maximum projection and segmentation on the sequence to extract a region of interest, and extracting a maximum translation feature vector and a time domain gray scale variance of the region of interest; The parameter calculation and scheduling module is used for calculating a photometric displacement gradient product and pushing the corresponding region of interest into a reconstruction verification queue or a polarization diagnosis queue respectively; the track reconstruction module is used for verifying diffuse reflection continuous translation attribute of the region of interest in the reconstruction verification queue so as to confirm the surface adhesion defect region; The polarization modulation and structure analysis module is used for calculating the local polarization contrast of the region of interest in the polarization diagnosis queue through optical rotation modulation so as to diagnose the structural damage defect region; the self-adaptive classification and output module is used for extracting classification labels and mapping the classification labels to a three-dimensional absolute coordinate system to generate a classification detection topology report; the parameter calculation and scheduling module performs multiplicative joint weighting on the modular length of the maximum translation characteristic vector and the time domain gray scale variance, and calculates and generates the photometric displacement gradient product by combining a base noise intensity mean value corrected by thermal drift along with a working condition environment; And the track reconstruction module maps the two-dimensional pixel translation characteristic to a three-dimensional physical reflection tangent plane through the industrial area array camera reference matrix to generate a predicted three-dimensional physical coordinate array, and compares the coincidence degree of the gray scale centroid displacement track of the secondary image acquisition and the predicted three-dimensional physical coordinate array.
2. The automatic detection and classification system for surface defects of optical elements according to claim 1, wherein the industrial area array camera is erected above a reference plane of the optical elements to be detected in an orthogonal overlooking posture, the hemispherical programmable LED lattice light source arrays are distributed in an equal space azimuth angle, and the central processing unit is connected with a light source array controller of the hemispherical programmable LED lattice light source arrays, trigger pins of the industrial area array camera and a polarization modulator driving circuit of the electric control liquid crystal polarization modulator through buses.
3. The system according to claim 1, wherein the initialization and image acquisition module is configured to plan the lighting timing by analyzing a large-step archimedes spiral space topology mathematical model, and to set a hardware anti-shake delay margin greater than the light emitting diode charging climbing time between the sending of the address lighting signal and the sending of the hard trigger pulse signal.
4. The system for automatically detecting and classifying surface defects of optical elements according to claim 1, wherein the positioning and feature extraction module is integrated with a frequency domain adaptive positioning switching mechanism, wherein when the area of the connected domain of the region of interest reaches a preset threshold, a gray centroid method is adopted to extract the maximum translation feature vector, and when the area of the connected domain of the region of interest does not reach the preset threshold, a phase correlation algorithm based on a frequency domain cross power spectrum is adopted to extract the maximum translation feature vector.
5. The system of claim 1, wherein the localization and feature extraction module locks the temporal variation of the extremum pixel intensities within the region of interest by traversing the initial temporal image sequence and generates the temporal gray scale variance from a central peak intensity dataset calculation at different light source operating points.
6. The automated optical element surface defect detection and classification system of claim 1, wherein the parameter calculation and scheduling module performs a two-wire state machine decision to push the reconstruction validation queue when the photometric displacement gradient product falls within a first class classification model threshold and push the polarization diagnostic queue when the maximum translational feature vector approaches a quantization noise level and a time-domain gray scale variance matches a second class classification model threshold.
7. An automatic detection and classification system for optical element surface defects according to claim 1, wherein the adaptive classification and output module is built with a residual convolutional neural network, the residual convolutional neural network takes a local high-quality gray image matrix of the region of interest as primary input data, and takes the local polarization contrast as an auxiliary spatial scalar to splice with an image feature vector of the local high-quality gray image matrix to output the classification label.
8. The system of claim 1, wherein the central processor further comprises a dynamic reference update and adaptive compensation module for performing full-extinction background acquisition at idle timing to update a base background noise intensity mean value and for performing electro-optic compensation based on an adaptive compensation scalar for planar reference calibration target residual luminous flux to calculate the hemispherical programmable LED array light source array drive current.

Description

Automatic detection and classification system for surface defects of optical element Technical Field The invention relates to the technical field of precision optical detection, in particular to an automatic detection and classification system for surface defects of an optical element. Background In the process of manufacturing and quality detection of precision optical elements, accurate identification and classification of surface defects is a key link for determining the qualification rate of products. The existing automatic optical detection technology mostly adopts a dark field scattering imaging or bright field transmission imaging scheme, and defects are judged by acquiring brightness characteristics of defect areas. However, the conventional detection scheme based on the light intensity information has a significant limitation in distinguishing the surface-attached foreign matter from the damage of the element substrate structure. As the dust, fiber and other attachments in the environment and scratches and pits of the elements are all high-brightness scattering points in a single-intensity image, the detection algorithm is difficult to effectively distinguish the scattering points only by morphological characteristics, so that a large number of false alarm phenomena occur in industrial production, and the over-killing rate of the system is improved. In addition, complex environmental factors in the industrial field also present challenges for detection stability. Because the light source has physical attenuation in long-time operation, and the fluctuation of the ambient temperature easily causes the photoelectric device to generate thermal drift, the background noise and the contrast of the imaging system are nonlinear. The instability makes the preset detection threshold difficult to keep universal under different working conditions, frequent manual calibration is often needed, maintenance cost is increased, and continuity of automatic detection is reduced. Therefore, development of a classification detection system capable of suppressing environmental interference, having a physical scale quantification determination capability, and accurately distinguishing an attached matter from a structural damage has become an urgent need for improving the detection efficiency of an optical element. Disclosure of Invention Aiming at the defects of the prior art, the invention provides an automatic detection and classification system for surface defects of an optical element, and solves the problems that the false alarm rate is high and the long-term stability of the detection system is affected by industrial environment interference due to the fact that the surface adhering foreign matters cannot be accurately distinguished from the damage of a substrate structure in the prior detection technology. The invention aims at realizing the above purposes by adopting the following technical scheme that the automatic detection and classification system of the surface defects of the optical element comprises a hemispherical programmable LED lattice light source array, an industrial area array camera comprising a camera lens, an electric control liquid crystal polarization modulator arranged at the light emitting end of the hemispherical programmable LED lattice light source array and in front of the camera lens and a central processing unit, wherein the central processing unit comprises a central processing unit and comprises: The initialization and image acquisition module is used for driving the hemispherical programmable LED lattice light source array to scan and triggering the industrial area array camera to acquire an initial time sequence image sequence; the positioning and feature extraction module is used for performing maximum projection and segmentation on the sequence to extract a region of interest, and extracting a maximum translation feature vector and a time domain gray scale variance of the region of interest; The parameter calculation and scheduling module is used for calculating a photometric displacement gradient product and pushing the corresponding region of interest into a reconstruction verification queue or a polarization diagnosis queue respectively; the track reconstruction module is used for verifying diffuse reflection continuous translation attribute of the region of interest in the reconstruction verification queue so as to confirm the surface adhesion defect region; The polarization modulation and structure analysis module is used for calculating the local polarization contrast of the region of interest in the polarization diagnosis queue through optical rotation modulation so as to diagnose the structural damage defect region; and the self-adaptive classification and output module is used for extracting classification labels and mapping the classification labels to a three-dimensional absolute coordinate system to generate a classification detection topology report. The industrial area