CN117969534B

CN117969534B - Optical lens detection method, device, equipment and storage medium

Info

Publication number: CN117969534B
Application number: CN202410374469.5A
Authority: CN
Inventors: 张盛峰; 杜肖剑; 陈永俊; 郭玲玲; 岑键年
Original assignee: Guangdong Kingding Optical Technology Co ltd
Current assignee: Guangdong Kingding Optical Technology Co ltd
Priority date: 2024-03-29
Filing date: 2024-03-29
Publication date: 2024-06-25
Anticipated expiration: 2044-03-29
Also published as: CN117969534A

Abstract

The application relates to the technical field of lens detection and discloses an optical lens detection method, an optical lens detection device, optical lens detection equipment and a storage medium. The method comprises the following steps: generating a transmission stripe in a lens detection system based on a detection light source and placing an optical lens to be detected in a stripe beam path; collecting an initial stripe image of an optical lens and performing color space transformation to obtain a target stripe image; performing optical lens defect identification through a target detection model to obtain first optical lens defect characteristics; performing dark field illumination and scattered light defect characteristic analysis on the optical lens to obtain second optical lens defect characteristics; performing Euclidean distance calculation on the defect characteristics of the first optical lens and the defect characteristics of the second optical lens to obtain target Euclidean distance data; model parameter multi-objective optimization is carried out through a non-dominant sorting genetic algorithm to obtain an optical lens detection model, and the detection accuracy of the optical lens is improved.

Description

Optical lens detection method, device, equipment and storage medium

Technical Field

The present application relates to the field of lens detection technologies, and in particular, to a method, an apparatus, a device, and a storage medium for detecting an optical lens.

Background

In the production and quality control process of optical lenses, it is important to accurately and efficiently detect defects of the optical lenses. The performance of an optical lens as a key component of a precision instrument directly affects the definition, precision and reliability of the whole optical system. Thus, any defect of the optical lens, whether surface scratches, internal bubbles, or irregular shapes, may cause a significant deterioration in the performance of the optical system. The traditional optical lens detection methods mainly depend on artificial vision or simple optical instruments, and have low efficiency and limitation on detection precision. Particularly for identification and classification of micro defects, manual detection often has difficulty in achieving satisfactory results.

With the rapid development of optical technology and image processing technology, an automatic and intelligent detection method gradually becomes a research hot spot. By using advanced image acquisition equipment and a deep learning algorithm, the accuracy and efficiency of optical lens defect detection can be greatly improved. However, existing automated detection methods still face some challenges. How to effectively extract the characteristic information of the optical lens from the complex background, and further accurately identify and classify the micro defects is a technical difficulty. In addition, given an optical lens, the types of defects are various, and the defect forms are different, which requires a high degree of flexibility and wide adaptability of the detection model.

Disclosure of Invention

The application provides an optical lens detection method, an optical lens detection device, optical lens detection equipment and a storage medium, which are used for improving the detection accuracy of an optical lens.

In a first aspect, the present application provides an optical lens detection method, including:

Generating transmission stripes in a lens detection system based on a preset detection light source, and placing an optical lens to be detected on a stripe beam path of the transmission stripes;

collecting an initial stripe image of the optical lens through a preset high-resolution camera, and performing color space transformation on the initial stripe image to obtain a target stripe image;

inputting the target stripe image into a preset target detection model to identify the defects of the optical lens, so as to obtain the defect characteristics of the first optical lens;

Dark field illumination and scattered light collection are carried out on the optical lens to obtain scattered spectrum distribution data, and optical lens defect characteristic analysis is carried out on the scattered spectrum distribution data to obtain second optical lens defect characteristics;

performing Euclidean distance calculation on the first optical lens defect characteristic and the second optical lens defect characteristic to obtain target Euclidean distance data;

and performing model parameter multi-objective optimization on the target detection model according to the target Euclidean distance data through a non-dominant ordering genetic algorithm to obtain an optical lens detection model.

In a second aspect, the present application provides an optical lens detection device including:

the detection module is used for generating transmission stripes in the lens detection system based on a preset detection light source and placing the optical lens to be detected on a stripe beam path of the transmission stripes;

the transformation module is used for acquiring an initial stripe image of the optical lens through a preset high-resolution camera and carrying out color space transformation on the initial stripe image to obtain a target stripe image;

the identification module is used for inputting the target stripe image into a preset target detection model to identify the defects of the optical lens, so as to obtain the defect characteristics of the first optical lens;

The analysis module is used for carrying out dark field illumination and scattered light collection on the optical lens to obtain scattered spectrum distribution data, and carrying out optical lens defect characteristic analysis on the scattered spectrum distribution data to obtain second optical lens defect characteristics;

The calculation module is used for carrying out Euclidean distance calculation on the first optical lens defect characteristic and the second optical lens defect characteristic to obtain target Euclidean distance data;

and the optimization module is used for carrying out model parameter multi-objective optimization on the target detection model according to the target Euclidean distance data through a non-dominant ordering genetic algorithm to obtain an optical lens detection model.

A third aspect of the present application provides an optical lens inspection apparatus comprising: a memory and at least one processor, the memory having instructions stored therein; the at least one processor invokes the instructions in the memory to cause the optical lens detection device to perform the optical lens detection method described above.

A fourth aspect of the present application provides a computer readable storage medium having instructions stored therein which, when run on a computer, cause the computer to perform the above-described optical lens detection method.

According to the technical scheme provided by the application, the micro defect characteristics of the surface of the optical lens can be effectively extracted by combining the initial stripe image acquired by the high-resolution camera and advanced image processing technology, such as color space transformation and frequency domain filtering. Further, the target detection model is utilized for deep learning analysis, so that high-precision identification and classification of defects can be realized, and the detection accuracy is greatly improved. Not only the traditional transmission fringe analysis technology is adopted, but also dark field illumination and scattering spectrum distribution data analysis are introduced, so that various defects from surface defects to internal defects and the like can be comprehensively detected. The multi-dimensional detection mode ensures the detection capability of the full coverage of the defects of the optical lens and reduces the risk of missing detection. Through automatic image acquisition and processing flow and efficient deep learning model analysis, the detection method can rapidly process a large amount of optical lens detection data. Compared with the traditional manual detection or simple automatic scheme, the speed and the production efficiency of detection are obviously improved, and the requirement of mass production is met. The non-dominant ordering genetic algorithm is used for multi-objective optimization of model parameters, so that the detection model can adapt to optical lens defect characteristics of different types and complexity, and the universality and the adaptability of the detection method are enhanced. This means that the same detection system can be applied to quality control of a variety of optical lenses, reducing the cost of equipment and technology replacement. The whole detection flow has high automation degree, and the defect analysis from the placement of the optical lens and the image acquisition can be realized through software control. The simple operation mode reduces the technical requirements for operators, so that the detection process is easier to manage and execute, and the detection accuracy of the optical lens is improved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings may be obtained based on these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic diagram of an embodiment of a method for detecting an optical lens in an embodiment of the application;

fig. 2 is a schematic diagram of an optical lens inspection apparatus according to an embodiment of the application.

Detailed Description

The embodiment of the application provides an optical lens detection method, an optical lens detection device, optical lens detection equipment and a storage medium. The terms "first," "second," "third," "fourth" and the like in the description and in the claims and in the above drawings, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments described herein may be implemented in other sequences than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed or inherent to such process, method, article, or apparatus.

For ease of understanding, a specific flow of an embodiment of the present application is described below with reference to fig. 1, and an embodiment of a method for detecting an optical lens in an embodiment of the present application includes:

step S101, generating transmission stripes in a lens detection system based on a preset detection light source, and placing an optical lens to be detected on a stripe beam path of the transmission stripes;

it is to be understood that the execution body of the present application may be an optical lens detection device, and may also be a terminal or a server, which is not limited herein. The embodiment of the application is described by taking a server as an execution main body as an example.

Specifically, wavelength screening is performed on light rays emitted by the detection light source, and a specific spectrum range meeting the stripe generation condition is selected. By controlling the wavelength, it is ensured that the resulting transmission fringes have the desired characteristics and sharpness. An initial light beam is generated from the screened spectral range. In order to accurately control the properties of the light beam and effectively generate the transmission fringes in the lens detection system, an optical modulator is used to phase modulate the initial light beam. The phase modulation can increase the stability of the transmitted fringes and make the generation of the fringes more controllable and flexible. The spatial light modulator applies a spatially varying phase pattern to the beam path, which imparts spatial encoding properties to the transmitted fringes. Spatially encoded fringes can provide more information for subsequent image processing and analysis and can enhance the ability of the system to detect microscopic defects on the lens surface. The beam broadening technology is adopted, the beam diameter is enlarged through the diffraction element, and the transmission stripes are ensured to cover the whole optical lens to be detected. The diffraction element can effectively expand the diameter of the light beam so that the transmission fringes uniformly cover the whole optical lens. The optical lens to be inspected is placed in the path of the transmitted fringe beam, ensuring that the fringe pattern is uniformly covered on the lens surface. In order to capture the fringe coverage status from multiple angles and further improve the accuracy and efficiency of the detection, the optical lens is angularly adjusted by the rotating stage, fringe variations on the lens surface are observed and analyzed from different viewing angles, and the quality of the optical lens is more comprehensively evaluated.

Step S102, acquiring an initial stripe image of an optical lens through a preset high-resolution camera, and performing color space transformation on the initial stripe image to obtain a target stripe image;

specifically, an initial fringe image of the optical lens is acquired by a preset high-resolution camera, and the high-resolution camera can capture tiny details and fringes on the surface of the optical lens. The color space of the initial fringe image is converted from the standard red-green-blue (RGB) color space into a spherical coordinate system, thereby more easily defining the color center in the spherical coordinate system and more accurately processing the image color information. In the spherical coordinate system, denoising operation is carried out on the initial stripe image, so that random noise in the image is effectively removed, and clearer stripe characteristics are reserved. The image is converted back to the sRGB color space, which is processed with its wide application and contrast characteristics. And (3) carrying out image merging by calculating the distance between the color of each pixel in the initial stripe image and the color center, so as to improve the definition and the discernability of stripes in the image. And denoising the combined images again to obtain clear and accurate denoising stripe images. And (3) carrying out frequency domain filtering on the de-noised fringe image, separating out high-frequency components in the image, and obtaining an enhanced fringe characteristic diagram. The frequency domain filtering can significantly highlight the streak feature making it more identifiable in subsequent edge detection. And identifying and marking stripe boundaries in the enhanced stripe feature map by an edge detection algorithm, wherein the accurate identification of the boundaries directly influences the subsequent segmentation and analysis effects. And carrying out lens region segmentation analysis on the boundary mark graph, distinguishing different regions on the lens, providing more detailed information for stripes in each region, and finally obtaining a target stripe image which can accurately reflect details on the optical lens.

Step S103, inputting a target stripe image into a preset target detection model to identify defects of the optical lens, and obtaining defect characteristics of the first optical lens;

Specifically, feature encoding processing is performed on the target stripe image, deep features of the target stripe image are extracted through a convolutional neural network layer of the target detection model, and a feature encoding diagram is obtained, wherein the target detection model in the embodiment can adopt a STMask R-CNN model. The feature code map is input into a region suggestion network of the target detection model, candidate regions possibly associated with potential defects are generated according to feature codes of the images, the candidate regions are drawn into a plurality of defect candidate region maps, and regions possibly with the defects are located. And extracting multi-scale defect characteristics from each defect candidate region map through a multi-scale perception mechanism. And the multi-scale characteristic map is obtained by analyzing the performances of the candidate region on different scales, so that the recognition capability of the model on different sizes and morphological defects is effectively improved. Defects on optical lenses may exist in different sizes and shapes, and the characteristics of these defects can be more fully captured by multi-scale analysis. Semantic segmentation is carried out on the multi-scale feature map, and the specific shape and position of each defect are segmented through mask branches of the target detection model, so that a defect mask map is obtained. The mask branches can define the exact location of the defect and delineate the specific contours of the defect. Classifying each identified defect through the defect mask map, marking the specific type of the defect, and generating a defect classification map. The first optical lens defect feature is constructed by combining the defect mask map and the defect classification map.

Step S104, dark field illumination and scattered light collection are carried out on the optical lens to obtain scattered spectrum distribution data, and optical lens defect characteristic analysis is carried out on the scattered spectrum distribution data to obtain second optical lens defect characteristics;

specifically, the dark field illumination technology creates a special optical condition by applying uniformly distributed illumination, so that only light scattered from the surface defects of the lens can be captured, but light directly irradiated onto the lens is not directly collected, pure scattered light data scattered from the surface defects of the lens can be obtained, and the sensitivity and accuracy of defect detection are improved. And recording the spectral distribution of the pure scattered light data by a spectrometer to obtain scattered spectral distribution data. The spectral distribution of the scattered light can provide rich information about the defects on the lens surface, such as the size, shape, and possibly the nature of the defects. Fourier transforming the scatter spectrum distribution data converts the data in the time domain into the frequency domain such that the scatter spectrum features originally distributed in the temporal or spatial scale are converted into spectral features in the frequency domain. Fourier transforms can reveal potential patterns in the scatter spectrum data, making these patterns more readily captured by subsequent analysis and identification processes. And carrying out pattern recognition on the frequency domain scattering spectrum characteristic to distinguish scattering characteristics caused by different types of optical defects, and obtaining second optical lens defect characteristics. Specific frequency domain feature patterns are identified and these patterns are associated with specific types of optical defects.

Step S105, performing Euclidean distance calculation on the defect characteristics of the first optical lens and the defect characteristics of the second optical lens to obtain target Euclidean distance data;

Specifically, feature encoding is performed on the defect features of the first optical lens to obtain a first defect feature vector, and feature encoding is performed on the defect features of the second optical lens to obtain a second defect feature vector. The encoding process reduces complex defect features to a vector form that can be analyzed quantitatively. And respectively carrying out standardization processing on the first defect feature vector and the second defect feature vector to ensure that different feature vectors have comparability in size and scale, and obtaining a first target feature vector and a second target feature vector. Normalization eliminates deviations in the raw data due to different measurement scales. And carrying out Euclidean distance calculation on the first target feature vector and the second target feature vector, and quantitatively reflecting the similarity or the difference between the two groups of defect features. Euclidean distance calculation the distance between defect features is evaluated in multiple dimensions by measuring the straight line distance between two points in vector space, and the obtained target Euclidean distance data provides an intuitive and quantitative index for the similarity or difference of defect features.

And S106, performing model parameter multi-objective optimization on the target detection model according to the target Euclidean distance data through a non-dominant ordering genetic algorithm to obtain the optical lens detection model.

Specifically, a population of non-dominant ordered genetic algorithms is initialized, the population consisting of a plurality of individuals, each individual representing a set of parameter configurations of a target detection model. The initialization provides a diverse genetic library for genetic algorithms, ensuring that the algorithm can explore the best possible solution in a wide search space. And evaluating each individual in the population, and calculating the influence of each individual, namely the model parameter configuration, on the defect detection performance of the optical lens by taking the target Euclidean distance data as an evaluation standard. Through evaluation, each individual gets a performance score that reflects the test efficacy of the model under the parameter configuration. The population is ranked by a non-dominant ranking genetic algorithm, and a prioritization result for each individual is determined based on the dominant relationships between the individuals and the crowding distance. The non-dominant ranking genetic algorithm performs ranking by comparing the merits among individuals while considering the crowding distance to ensure the diversity of solutions, which ensures that the genetic algorithm can effectively evolve toward multi-objective optimization. And selecting individuals with high priority to perform crossover and mutation operation according to the priority sequencing result, and generating a new generation of parameter configuration by simulating a genetic and mutation mechanism in biological evolution. The crossover operation enables the genes of excellent individuals to be combined, and the mutation operation introduces new genetic mutation to increase the diversity of the population. And carrying out optimization solution on the model parameter configuration of the new generation, and continuously optimizing in an iterative mode until the optimal parameter configuration is found. The optimal parameter configuration is used for updating parameters of the target detection model, so that a model optimized for optical lens defect detection is obtained.

In the embodiment of the application, the micro defect characteristics of the surface of the optical lens can be effectively extracted by combining the initial stripe image acquired by the high-resolution camera and advanced image processing technology, such as color space transformation and frequency domain filtering. Further, the target detection model is utilized for deep learning analysis, so that high-precision identification and classification of defects can be realized, and the detection accuracy is greatly improved. Not only the traditional transmission fringe analysis technology is adopted, but also dark field illumination and scattering spectrum distribution data analysis are introduced, so that various defects from surface defects to internal defects and the like can be comprehensively detected. The multi-dimensional detection mode ensures the detection capability of the full coverage of the defects of the optical lens and reduces the risk of missing detection. Through automatic image acquisition and processing flow and efficient deep learning model analysis, the detection method can rapidly process a large amount of optical lens detection data. Compared with the traditional manual detection or simple automatic scheme, the speed and the production efficiency of detection are obviously improved, and the requirement of mass production is met. The non-dominant ordering genetic algorithm is used for multi-objective optimization of model parameters, so that the detection model can adapt to optical lens defect characteristics of different types and complexity, and the universality and the adaptability of the detection method are enhanced. This means that the same detection system can be applied to quality control of a variety of optical lenses, reducing the cost of equipment and technology replacement. The whole detection flow has high automation degree, and the defect analysis from the placement of the optical lens and the image acquisition can be realized through software control. The simple operation mode reduces the technical requirements for operators, so that the detection process is easier to manage and execute, and the detection accuracy of the optical lens is improved.

In a specific embodiment, the process of executing step S101 may specifically include the following steps:

(1) Wavelength screening is carried out on the light rays emitted by the detection light source, and a spectrum range meeting the stripe generation condition is selected;

(2) Generating an initial light beam according to the spectrum range, carrying out phase modulation on the initial light beam through an optical modulator, and generating transmission stripes in a lens detection system;

(3) Applying a spatially varying phase pattern on the beam path by the spatial light modulator to provide the transmission fringes with spatially encoded properties;

(4) The beam widening technology is adopted, the beam diameter is enlarged through a diffraction element, and the transmission stripes are ensured to cover the whole optical lens to be detected;

(5) The optical lens to be detected is placed in the beam path of the transmitted stripes to ensure that the stripe pattern uniformly covers the lens surface, and the stripe covering state is captured at a plurality of angles by adjusting the optical lens through the rotating table.

Specifically, wavelength screening is performed on light rays emitted by the detection light source, and a spectrum range meeting the stripe generation condition is selected. The propagation and scattering properties of light of different wavelengths in the optical material may be different, and appropriate wavelength selection may enhance the sharpness and contrast of the transmitted fringes, thereby making it easier to identify defects on the lens. For example, for a particular optical lens material, it may be found that using a medium wavelength range light source, such as green light, maximizes the contrast of scattered light to direct light while ensuring fringe sharpness. An initial beam of light is generated based on the selected spectral range. The light emitted by the laser or other light source is subjected to a certain optical treatment, for example by means of a filter, to ensure that the wavelength of the light beam corresponds to a predetermined spectral range. The initial beam is phase modulated using an optical modulator, such as a Liquid Crystal Modulator (LCM) or an acousto-optic modulator (AOM), to change the propagation phase of the optical wave, thereby generating a transmission fringe having a specific characteristic in the optical detection system. The phase modulation enables the phase distribution of the transmitted light wave front to change according to a preset mode, and the change directly affects the formation and the characteristics of the transmission stripes, so that an accurate control means is provided for the generation of the transmission stripes on the surface of the lens. In order to provide the transmission fringes with spatially encoded properties, the detected information content is further increased, by applying a spatially varying phase pattern on the beam path by means of a Spatial Light Modulator (SLM). The SLM can locally adjust the phase of the light beam at the microscopic level, by designing a specific phase pattern, the transmitted fringes can not only carry information about the light beam itself, but also encode spatial characteristics about the object to be detected, and such fringes can significantly improve the ability to identify fine defects during interaction with the lens surface. In order to ensure that the transmission fringes can uniformly cover the entire optical lens to be inspected, beam broadening techniques are employed by introducing a diffraction element, such as a diffraction grating or fresnel lens, in the beam path. The diffraction element is used for adjusting the propagation direction and shape of the light beam according to the diffraction principle, so that the diameter of the light beam is enlarged, the size of the transmission stripe is expanded enough to cover the whole lens surface, and the comprehensiveness and uniformity of detection are ensured. The optical lens to be detected is placed in the path of the transmitted fringe beam, ensuring that the fringe pattern covers the lens surface uniformly. By using the rotating table to make position and angle adjustments to the optical lens, the streak coverage state is captured from multiple directions, which helps to comprehensively evaluate the optical characteristics of the lens surface, and also improves the ability to identify defects on the lens surface.

In a specific embodiment, the process of executing step S102 may specifically include the following steps:

(1) Collecting an initial stripe image of an optical lens through a preset high-resolution camera, converting the color space of the initial stripe image from standard red, green and blue to a spherical coordinate system, and defining a color center in the spherical coordinate system;

(2) Performing denoising operation on the initial stripe image in the spherical coordinate system, and converting the denoised image back to the sRGB color space;

(3) Image combination is carried out according to the distance between the color of each pixel in the initial stripe image and the color center, and denoising treatment is carried out on the combined image, so that a denoised stripe image is obtained;

(4) Performing frequency domain filtering on the de-noised fringe image, and separating high-frequency components in the de-noised fringe image to obtain an enhanced fringe feature diagram;

(5) Identifying and marking the stripe boundary on the optical lens in the enhanced stripe characteristic map through an edge detection algorithm to obtain a boundary marking map;

(6) And carrying out lens region segmentation analysis on the boundary mark graph to obtain a target stripe image.

Specifically, an initial fringe image of the optical lens is acquired by a preset high-resolution camera. The color space of the initial stripe image is converted from the standard Red Green Blue (RGB) color space to a spherical coordinate system so that the color information of the image can be processed and analyzed in a more intuitive space, and the definition of the color center provides a reference point for subsequent image processing and analysis work. In the spherical coordinate system, denoising operation is performed on the initial fringe image. Random noise in the image is effectively removed using image processing techniques such as wavelet transform or median filtering without compromising the structure of the fringes themselves. The denoised image is converted back to the sRGB color space, preserving the sharpness of the denoised image. Image merging is performed according to the distance between the color of each pixel and the color center in the initial stripe image, the distance between the color value of each point in the image and the color center is calculated, and the image is recombined based on the distance information so as to highlight the characteristics of the stripe. And denoising the combined images again to ensure that the obtained denoising stripe image is as clear and accurate as possible while retaining key characteristics. To further enhance the fringe features, the de-noised fringe image is frequency domain filtered. By applying fourier transform or the like to transform the image from the spatial domain to the frequency domain, the high frequency components in the image, i.e. those defining the edges and details of the fringes, are separated, resulting in an enhanced fringe signature. The stripe boundaries on the optical lens in the enhanced stripe feature map are identified and marked by an edge detection algorithm, such as the Canny edge detection algorithm. Finally, the target stripe image is obtained by carrying out lens region segmentation analysis on the boundary mark image. Through image analysis technology, including image segmentation algorithm and machine learning technology, each region of the optical lens is ensured to be accurately separated from the boundary mark graph, and finally a clearly defined target stripe image containing relevant optical lens quality information is obtained. The target fringe image can reveal the location and nature of the defect on the lens and reflect the severity of the defect.

In a specific embodiment, the process of executing step S103 may specifically include the following steps:

(1) Performing feature coding processing on the target stripe image, and extracting deep features of the target stripe image through a convolutional neural network layer of a target detection model to obtain a feature coding diagram;

(2) Inputting the feature code image into a region suggestion network of a target detection model, and generating candidate regions related to potential defects through the region suggestion network to obtain a plurality of defect candidate region images;

(3) Extracting multi-scale defect characteristics of each defect candidate region map through a multi-scale perception mechanism to obtain a multi-scale characteristic map;

(4) Carrying out semantic segmentation on the multi-scale feature map, and segmenting the shape and the position of each defect through mask branches of the target detection model to obtain a defect mask map;

(5) Classifying each identified defect through a defect mask map, marking the defect type, and obtaining a defect classification map;

(6) And constructing the defect characteristics of the first optical lens according to the defect mask diagram and the defect classification diagram.

Specifically, feature encoding processing is performed on the target stripe image, and stripe features in the image are converted into a form which can be understood and processed by a computer. And deep feature extraction is carried out on the coded target stripe image through a convolutional neural network layer of the target detection model. With the multi-layer structure of the convolutional neural network, deep and abstract features in the image can be captured, which contain information about the surface state of the optical lens. For example, the convolution layer can identify these detail features when processing a lens image with fine scratches. The feature code map is input into a region suggestion network (RPN) of the target detection model that is capable of generating candidate regions associated with potential defects. The RPN can predict which regions are most likely to contain defects by analyzing the feature information in the feature code map, thereby generating a plurality of defect candidate region maps. In order to improve the accuracy and the fineness of detection, multi-scale defect feature extraction is carried out on each defect candidate region map through a multi-scale sensing mechanism, so that a multi-scale feature map is obtained. The multi-scale processing enables the model to observe and analyze defects on different scales, enhancing the recognition capability of the model for defects of various sizes and shapes. And performing semantic segmentation on the multi-scale feature map, and dividing the specific shape and position of each defect by a mask branch of the target detection model to generate a defect mask map. The feature regions in the multi-scale feature map are mapped back into the original image space, identifying the boundaries of each defect. Classifying each identified defect based on the defect mask map, and marking the defect type to obtain a defect classification map. Learning understanding of the shape, size, and pattern of defects using models classifies defects into different categories such as scratches, stains, bubbles, and the like.

In a specific embodiment, the process of executing step S104 may specifically include the following steps:

(1) Based on dark field illumination technology, uniformly distributed illumination is applied to the optical lens, so that only light scattered by defects on the surface of the lens is ensured to be captured, and pure scattered light data is obtained;

(2) Recording the spectrum distribution of the pure scattered light data by a spectrometer to obtain scattered spectrum distribution data;

(3) Performing Fourier transform on the scattered spectrum distribution data, and converting the scattered spectrum distribution data on a time domain into spectrum characteristics on a frequency domain to obtain the frequency domain scattered spectrum characteristics;

(4) And carrying out pattern recognition on the frequency domain scattering spectrum characteristics to distinguish scattering characteristics caused by different types of optical defects, and obtaining second optical lens defect characteristics.

In particular, dark field illumination techniques enable light from the sample itself to be captured by the imaging system rather than directly illuminating the light. This technique ensures that only light scattered by the lens surface defects is captured by applying uniformly distributed illumination, resulting in pure scattered light data. Dark field illumination is achieved by placing the light source beside the sample rather than directly aiming the sample so that direct light does not directly enter the lens of the detection device. Only light scattered from the surface of the sample, such as scratches, particles, or other irregularities, will be captured by the lens. For example, in the case of inspecting an optical lens, if there are fine scratches on the surface, the scratches scatter light radiated sideways, and the scattered light is captured by the inspection apparatus, and a smooth area without defects does not scatter light into the inspection apparatus, thereby achieving high-contrast defect imaging. And recording the spectral distribution of the pure scattered light data by a spectrometer to obtain scattered spectral distribution data. The spectrometer is able to accurately measure the light intensities of different wavelengths, providing detailed information about how the light interacts with the optical lens surface. This information reflects the spectral characteristics of different types of defects, as different surface defects, such as scratches and pits, scatter light in different ways, resulting in differences in spectral distribution. Fourier transforming the scatter spectral distribution data converts the data from the time domain to the frequency domain. Fourier transforms analyze scattered light data from another angle, i.e., by spectral features in the frequency domain. This conversion process reveals periodicity and patterns in the data, which helps identify scattering features caused by different defect types. For example, one particular type of defect may produce unique peaks or patterns in the frequency domain, while these features may be less noticeable in the time domain data. And finally, carrying out pattern recognition on the frequency domain scattering spectrum characteristics, and distinguishing scattering characteristics caused by different types of optical defects, so as to obtain second optical lens defect characteristics. Patterns in the frequency domain scatter spectral features are automatically identified and classified by data analysis and machine learning techniques. The specific frequency domain features of different defect types are identified by training a machine learning model, and scattering spectrum data is automatically converted into specific defect identification results.

In a specific embodiment, the process of executing step S105 may specifically include the following steps:

(1) Performing feature coding on the defect features of the first optical lens to obtain a first defect feature vector, and performing feature coding on the defect features of the second optical lens to obtain a second defect feature vector;

(2) Performing feature vector set normalization processing on the first defect feature vector and the second defect feature vector to obtain a first target feature vector and a second target feature vector;

(3) And carrying out Euclidean distance calculation on the first target feature vector and the second target feature vector to obtain target Euclidean distance data.

Specifically, feature encoding is performed on the defect features of the first optical lens to obtain a first defect feature vector, feature encoding is performed on the defect features of the second optical lens to obtain a second defect feature vector, and the defect features are extracted and quantized. For example, if two different types of defects, such as scratches and dust, are found in the inspection process, the shape, size, depth, and other features of the defects are extracted through the feature encoding process, and the information is converted into a series of values to form a defect feature vector. In order to enable these feature vectors to be compared and analyzed under the same standard, they are normalized. The normalization process ensures comparability between different feature vectors. Different defects may have great differences in terms of original size, strength, etc., and by adjusting all feature vectors to the same scale, the influence of these variables is eliminated, resulting in a first target feature vector and a second target feature vector. And carrying out Euclidean distance calculation on the normalized target feature vector to obtain target Euclidean distance data. Euclidean distance is a method for measuring the distance between two points, and can be used for calculating the similarity or difference between two defect feature vectors. The difference between the two different defect features is quantified by calculating the Euclidean distance between the first target feature vector and the second target feature vector. The smaller the distance value, the more similar the two defect features are; conversely, the larger the distance value, the more pronounced the difference between defect features.

In a specific embodiment, the process of executing step S106 may specifically include the following steps:

(1) Initializing a population of non-dominant ordered genetic algorithm, each individual representing a parameter configuration of a set of target detection models;

(2) Evaluating each individual, taking the target Euclidean distance data as an evaluation standard, and calculating the influence of each individual on the defect detection performance of the optical lens to obtain the performance score of each individual;

(3) Sorting the population by a non-dominant sorting genetic algorithm, and determining a priority sorting result of each individual based on the dominant relationship and the crowding distance among the individuals;

(4) According to the priority ordering result, selecting individuals with high priority to perform crossing and mutation operations, and generating a new generation of parameter configuration;

(5) And carrying out optimization solution on the model parameter configuration of the new generation to obtain the optimal parameter configuration, and carrying out model parameter updating on the target detection model through the optimal parameter configuration to obtain the optical lens detection model.

Specifically, a population of non-dominant ordered genetic algorithms is initialized, wherein each individual represents a specific set of parameter configurations of the target detection model. For example, considering STMask R-CNN for identifying and classifying defects on an optical lens, the parametric configuration of the model may include the number of filters of the convolution layer, the learning rate, the regularization strength, etc. The initialization process may randomly generate a series of such parameter configurations, each of which is intended to capture and process defect features in the image data in a different manner. Each individual was evaluated to determine their performance in the optical lens defect detection task. And taking the target Euclidean distance data as an evaluation standard, and measuring the influence of each individual, namely parameter configuration, on the model detection performance. The evaluation process quantifies the effect of each parameter configuration by calculating the Euclidean distance between the predicted result of the model on the images of known defects and the actual defect by means of a validation set consisting of the images of the optical lens of these defects. Each individual obtains a score based on their performance that reflects the efficiency and accuracy of the parameter configuration over the defect detection task. Individuals in the population are ranked by a non-dominant ranking genetic algorithm. The dominant relationship and crowding distance between individuals were examined. In a multi-objective optimization problem, one solution may perform well on one objective and not on another objective, and the non-dominant ranking genetic algorithm ranks by determining which solutions are not dominant by the other solutions on all objectives. At the same time, the calculation of the crowding distance ensures that the solutions selected to enter the next generation are evenly distributed in the solution space, thereby ensuring the diversity of the solutions. According to the individual priority ordering result, the algorithm selects a part of individuals with the best performance to perform crossing and mutation operation, simulates the genetic mutation process in biological evolution, and generates a new generation of parameter configuration. The interleaving operation allows two superior solutions to "swap" some of their parameters in the hope of producing more superior offspring. Mutation operations introduce new genetic diversity by randomly changing the values of certain parameters. And carrying out optimization solution on the new generation of model parameter configuration to determine the final and optimal parameter configuration. This typically involves multiple rounds of crossover, mutation and selection processes until a point is found where performance is no longer significantly improved. This optimal parameter configuration is then used to update the target inspection model, resulting in a model that is highly optimized for the specific optical lens defect inspection task.

The method for detecting an optical lens in the embodiment of the present application is described above, and the optical lens detecting device in the embodiment of the present application is described below, referring to fig. 2, where an embodiment of the optical lens detecting device in the embodiment of the present application includes:

The detection module 201 is configured to generate a transmission stripe in the lens detection system based on a preset detection light source, and place an optical lens to be detected on a stripe beam path of the transmission stripe;

The transformation module 202 is configured to collect an initial stripe image of the optical lens through a preset high-resolution camera, and perform color space transformation on the initial stripe image to obtain a target stripe image;

The recognition module 203 is configured to input a target stripe image into a preset target detection model to perform optical lens defect recognition, so as to obtain a first optical lens defect feature;

the analysis module 204 is used for carrying out dark field illumination and scattered light collection on the optical lens to obtain scattered spectrum distribution data, and carrying out optical lens defect characteristic analysis on the scattered spectrum distribution data to obtain second optical lens defect characteristics;

the calculating module 205 is configured to perform euclidean distance calculation on the first optical lens defect feature and the second optical lens defect feature to obtain target euclidean distance data;

and the optimization module 206 is configured to perform model parameter multi-objective optimization on the target detection model according to the target euclidean distance data by using a non-dominant ordering genetic algorithm, so as to obtain an optical lens detection model.

Through the cooperation of the components, the micro defect characteristics of the surface of the optical lens can be effectively extracted by combining an initial stripe image acquired by a high-resolution camera and advanced image processing technologies such as color space transformation and frequency domain filtering. Further, the target detection model is utilized for deep learning analysis, so that high-precision identification and classification of defects can be realized, and the detection accuracy is greatly improved. Not only the traditional transmission fringe analysis technology is adopted, but also dark field illumination and scattering spectrum distribution data analysis are introduced, so that various defects from surface defects to internal defects and the like can be comprehensively detected. The multi-dimensional detection mode ensures the detection capability of the full coverage of the defects of the optical lens and reduces the risk of missing detection. Through automatic image acquisition and processing flow and efficient deep learning model analysis, the detection method can rapidly process a large amount of optical lens detection data. Compared with the traditional manual detection or simple automatic scheme, the speed and the production efficiency of detection are obviously improved, and the requirement of mass production is met. The non-dominant ordering genetic algorithm is used for multi-objective optimization of model parameters, so that the detection model can adapt to optical lens defect characteristics of different types and complexity, and the universality and the adaptability of the detection method are enhanced. This means that the same detection system can be applied to quality control of a variety of optical lenses, reducing the cost of equipment and technology replacement. The whole detection flow has high automation degree, and the defect analysis from the placement of the optical lens and the image acquisition can be realized through software control. The simple operation mode reduces the technical requirements for operators, so that the detection process is easier to manage and execute, and the detection accuracy of the optical lens is improved.

The present application also provides an optical lens inspection apparatus including a memory and a processor, the memory storing computer readable instructions which, when executed by the processor, cause the processor to perform the steps of the optical lens inspection method in the above embodiments.

The present application also provides a computer readable storage medium, which may be a non-volatile computer readable storage medium, or may be a volatile computer readable storage medium, having stored therein instructions that, when executed on a computer, cause the computer to perform the steps of the optical lens detection method.

It will be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, systems and units may refer to the corresponding processes in the foregoing method embodiments, which are not repeated herein.

The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be embodied essentially or in part or all of the technical solution or in part in the form of a software product stored in a storage medium, including instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a read-only memory (ROM), a random access memory (random acceS memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

The above embodiments are only for illustrating the technical solution of the present application, and not for limiting the same; although the application has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An optical lens detection method, characterized in that the optical lens detection method comprises:

Inputting the target stripe image into a preset target detection model to identify the defects of the optical lens, so as to obtain the defect characteristics of the first optical lens; the method specifically comprises the following steps: performing feature coding processing on the target stripe image, and extracting deep features of the target stripe image through a convolutional neural network layer of a target detection model to obtain a feature coding diagram; inputting the feature code map to a region suggestion network of the target detection model, and generating candidate regions related to potential defects through the region suggestion network to obtain a plurality of defect candidate region maps; extracting multi-scale defect characteristics of each defect candidate region map through a multi-scale perception mechanism to obtain a multi-scale characteristic map; performing semantic segmentation on the multi-scale feature map, and segmenting the shape and the position of each defect through mask branches of the target detection model to obtain a defect mask map; classifying each identified defect through the defect mask map, marking the defect type, and obtaining a defect classification map; constructing a first optical lens defect characteristic according to the defect mask map and the defect classification map;

Dark field illumination and scattered light collection are carried out on the optical lens to obtain scattered spectrum distribution data, and optical lens defect characteristic analysis is carried out on the scattered spectrum distribution data to obtain second optical lens defect characteristics; the method specifically comprises the following steps: based on dark field illumination technology, uniformly distributed illumination is applied to the optical lens, so that only light scattered by defects on the surface of the lens is ensured to be captured, and pure scattered light data are obtained; recording the spectrum distribution of the pure scattered light data by a spectrometer to obtain scattered spectrum distribution data; performing Fourier transform on the scattered spectrum distribution data, and converting the scattered spectrum distribution data in a time domain into spectrum characteristics in a frequency domain to obtain the frequency domain scattered spectrum characteristics; performing mode identification on the frequency domain scattering spectrum characteristics to distinguish scattering characteristics caused by different types of optical defects, so as to obtain second optical lens defect characteristics;

2. The method according to claim 1, wherein the generating a transmission stripe in a lens inspection system based on a preset inspection light source and placing an optical lens to be inspected in a stripe beam path of the transmission stripe comprises:

wavelength screening is carried out on the light rays emitted by the detection light source, and a spectrum range meeting the stripe generation condition is selected;

generating an initial light beam according to the spectrum range, carrying out phase modulation on the initial light beam through an optical modulator, and generating transmission stripes in a lens detection system;

Applying a spatially varying phase pattern on the beam path by a spatial light modulator, the transmission fringes having spatially encoded properties;

adopting a beam broadening technology, and expanding the beam diameter through a diffraction element to ensure that the transmission stripes cover the whole optical lens to be detected;

The optical lens to be detected is placed in the beam path of the transmitted stripes to ensure that the stripe pattern uniformly covers the lens surface, and the stripe covering state is captured at a plurality of angles by adjusting the optical lens through the rotating table.

3. The method for detecting an optical lens according to claim 1, wherein the capturing an initial fringe image of the optical lens by a preset high-resolution camera and performing color space transformation on the initial fringe image to obtain a target fringe image comprises:

collecting an initial stripe image of the optical lens by a preset high-resolution camera, converting the color space of the initial stripe image from standard red, green and blue to a spherical coordinate system, and defining a color center in the spherical coordinate system;

Performing denoising operation on the initial stripe image in the spherical coordinate system, and converting the denoised image back to an sRGB color space;

image combination is carried out according to the distance between the color of each pixel in the initial stripe image and the color center, and denoising processing is carried out on the combined image, so that a denoising stripe image is obtained;

Performing frequency domain filtering on the de-noised fringe image, and separating out high-frequency components in the de-noised fringe image to obtain an enhanced fringe feature diagram;

identifying and marking the fringe boundary on the optical lens in the enhanced fringe feature map through an edge detection algorithm to obtain a boundary mark map;

and carrying out lens region segmentation analysis on the boundary mark graph to obtain a target stripe image.

4. The method according to claim 1, wherein performing euclidean distance calculation on the first optical lens defect feature and the second optical lens defect feature to obtain target euclidean distance data comprises:

Performing feature coding on the defect features of the first optical lens to obtain a first defect feature vector, and performing feature coding on the defect features of the second optical lens to obtain a second defect feature vector;

Performing feature vector set normalization processing on the first defect feature vector and the second defect feature vector to obtain a first target feature vector and a second target feature vector;

And carrying out Euclidean distance calculation on the first target feature vector and the second target feature vector to obtain target Euclidean distance data.

5. The method according to claim 1, wherein the performing model parameter multi-objective optimization on the target detection model according to the target euclidean distance data by using a non-dominant ordering genetic algorithm to obtain an optical lens detection model comprises:

Initializing a population of non-dominant ordered genetic algorithms, each individual representing a set of parameter configurations of the target detection model;

Evaluating each individual, taking the target Euclidean distance data as an evaluation standard, and calculating the influence of each individual on the defect detection performance of the optical lens to obtain the performance score of each individual;

sorting the population by the non-dominant sorting genetic algorithm, and determining a priority sorting result of each individual based on the dominant relationship and the crowding distance among the individuals;

according to the priority ordering result, selecting individuals with high priority to perform crossing and mutation operations, and generating a new generation of parameter configuration;

and carrying out optimization solution on the new generation of model parameter configuration to obtain optimal parameter configuration, and carrying out model parameter updating on the target detection model through the optimal parameter configuration to obtain an optical lens detection model.

6. An optical lens inspection device, comprising:

The identification module is used for inputting the target stripe image into a preset target detection model to identify the defects of the optical lens, so as to obtain the defect characteristics of the first optical lens; the method specifically comprises the following steps: performing feature coding processing on the target stripe image, and extracting deep features of the target stripe image through a convolutional neural network layer of a target detection model to obtain a feature coding diagram; inputting the feature code map to a region suggestion network of the target detection model, and generating candidate regions related to potential defects through the region suggestion network to obtain a plurality of defect candidate region maps; extracting multi-scale defect characteristics of each defect candidate region map through a multi-scale perception mechanism to obtain a multi-scale characteristic map; performing semantic segmentation on the multi-scale feature map, and segmenting the shape and the position of each defect through mask branches of the target detection model to obtain a defect mask map; classifying each identified defect through the defect mask map, marking the defect type, and obtaining a defect classification map; constructing a first optical lens defect characteristic according to the defect mask map and the defect classification map;

The analysis module is used for carrying out dark field illumination and scattered light collection on the optical lens to obtain scattered spectrum distribution data, and carrying out optical lens defect characteristic analysis on the scattered spectrum distribution data to obtain second optical lens defect characteristics; the method specifically comprises the following steps: based on dark field illumination technology, uniformly distributed illumination is applied to the optical lens, so that only light scattered by defects on the surface of the lens is ensured to be captured, and pure scattered light data are obtained; recording the spectrum distribution of the pure scattered light data by a spectrometer to obtain scattered spectrum distribution data; performing Fourier transform on the scattered spectrum distribution data, and converting the scattered spectrum distribution data in a time domain into spectrum characteristics in a frequency domain to obtain the frequency domain scattered spectrum characteristics; performing mode identification on the frequency domain scattering spectrum characteristics to distinguish scattering characteristics caused by different types of optical defects, so as to obtain second optical lens defect characteristics;

7. An optical lens inspection apparatus, characterized by comprising: a memory and at least one processor, the memory having instructions stored therein;

the at least one processor invokes the instructions in the memory to cause the optical lens detection device to perform the optical lens detection method of any one of claims 1-5.

8. A computer readable storage medium having instructions stored thereon, which when executed by a processor, implement the optical lens detection method of any of claims 1-5.