A three-dimensional detection device and method for surface defects of large-aperture curved optical elements

Abstract

The present invention relates to a device and method for three-dimensional detection of surface defects of large-aperture curved optical elements, belonging to the field of three-dimensional detection of surface defects of optical elements. The device includes a mobile platform, a 3D line spectrum confocal sensor, a calibration head, an industrial camera, a telecentric lens, a ring light source, a point light source, and a camera axis fine-tuning device, etc. The method includes the following steps: positioning and focusing the element to be tested; scanning the 3D line spectrum confocal sensor to obtain the defect position and three-dimensional information of the defect; switching to a two-dimensional defect detection module to obtain two-dimensional information of surface defects; and comprehensively analyzing the two-dimensional information and the three-dimensional information to obtain the three-dimensional detection results of each defect on the surface of the element to be tested. The present invention realizes rapid positioning and high-precision three-dimensional detection of surface defects of large-aperture curved optical elements by carrying two defect detection modules on a mobile platform.

Description

Technical Field

[0001] The present invention relates to the field of three-dimensional detection of surface defects of optical elements, and in particular to a device and method for three-dimensional detection of surface defects of large-aperture curved optical elements. Background Art

[0002] With the continuous development of advanced optical technology, large-aperture curved optical components are gradually used in astronomy, space optics and other fields. However, surface defects generated during the manufacturing and transportation of large-aperture curved optical components will reduce the beam quality and affect the normal operation of the system. Therefore, high-precision three-dimensional defect detection on their surfaces is of great significance.

[0003] At present, the three-dimensional detection of surface defects of curved optical components mostly adopts scanning probe surface profiler, white light interferometer, structured light detection method and three-dimensional point cloud measurement technology. Among them, the scanning probe surface profiler is a contact detection, which is easy to cause secondary damage to the component. When the white light interferometer is used for curved surface defect detection, its angle needs to be adjusted, and the operation is relatively cumbersome. Although the structured light three-dimensional detection has high sensitivity, the calculation is complicated. Non-contact three-dimensional point cloud measurement technology has the advantages of high precision and large detection range. The acquired point cloud data contains a lot of detailed information. While detecting the component morphology, the defect length, width and depth information can be obtained. Therefore, this method is gradually applied to the defect detection of optical components. Publication number: CN 110404816A, the application named a 3D curved glass defect detection device and method based on a robotic arm discloses that the curved glass is adsorbed and moved by a robotic arm, and two line scanning cameras are used to obtain the surface image of the component under different light sources to detect the defects. Publication number: CN117619759A, the invention name is a visual inspection system and method for industrial assembly objects, which discloses a method of grasping objects with a robotic arm and using multiple industrial cameras to collect multi-angle images to judge and identify component appearance defects. In the above three-dimensional defect detection of components, the robotic arm is often used to control the movement of components and the imaging system is fixed, but it is mostly used for defect detection of small-diameter curved optical components, and it is difficult to ensure that the optical components can be scanned in all directions. In order to detect large-diameter curved optical surface defects with high precision and obtain three-dimensional information of defects, further improvements are needed. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of existing three-dimensional defect detection technology, and to provide a three-dimensional detection device and method for surface defects of large-aperture curved optical components. The device uses a mobile platform to carry a 3D line spectrum confocal sensor and a camera lens module, and obtains the depth and defect location information of the defect based on the point cloud data obtained by the 3D line spectrum confocal sensor. According to the defect location information, the imaging module of the camera lens is switched to obtain the two-dimensional information of the defect. The two methods are combined to complete the three-dimensional high-precision detection of surface defects of curved optical components.

[0005] To achieve the above purpose, the technical scheme is as follows: a 3D detection device for surface defects of large-aperture curved optical elements, the device comprising a mobile platform, a 3D line spectrum confocal sensor, an industrial camera, a telecentric lens, a ring light source, a point light source, and a camera axis fine-tuning device; wherein the 3D line spectrum confocal sensor constitutes a 3D point cloud detection module; the industrial camera, the telecentric lens, the ring light source, the point light source, and the camera axis fine-tuning device constitute a 2D defect detection module; the 2D defect detection module is close to the front end of the 3D point cloud detection module; the 3D point cloud detection module and the 2D defect detection module are mounted on the mobile platform; the industrial camera is positioned at the The 3D line spectrum confocal sensor and the telecentric lens are located in the middle of the 3D line spectrum confocal sensor and the telecentric lens to control the 2D defect detection module to make fine adjustments up and down to achieve image focusing. The 3D point cloud detection module is used to obtain the point cloud data of the entire surface of the large-aperture curved optical element. The 2D defect detection module is used to obtain the 2D defect information of the large-aperture curved optical element.

[0006] The present invention also provides a three-dimensional detection method for surface defects of a large-aperture curved optical element, comprising the following steps:

[0007] Step S1: Positioning and focusing the component to be tested; including: placing the component to be tested on the sample placement platform, using the calibration head at the lower end of the 3D line spectrum confocal sensor to obtain the spatial coordinates of different positions of the component to be tested, and completing the positioning of the component to be tested; moving the 3D line spectrum confocal sensor downward to a position where the light spot is clear, and completing the focusing;

[0008] Step S2: Scanning the 3D line spectrum confocal sensor to obtain the defect position and the three-dimensional information of the defect; including: determining the scanning path of the 3D line spectrum confocal sensor and the moving speed of the mobile platform according to the curvature radius, caliber and spatial position of the component to be tested, scanning the component to be tested according to the predetermined path, saving the obtained point cloud data, and determining the defect position and three-dimensional information after analyzing and calculating the point cloud data;

[0009] Step S3: Switch to the two-dimensional defect detection module to obtain two-dimensional information of the defect; including: for a defect at a certain determined position, according to the offset between the two-dimensional defect detection module and the three-dimensional point cloud detection module, move the two-dimensional defect detection module to the defect, focus clearly and obtain a bright field or dark field image of the defect; if the defect is in multiple fields of view, customize the scanning path according to the defect position, complete the splicing of the bright field or dark field image of the entire defect and image processing, and then obtain the two-dimensional information of the defect;

[0010] Step S4: Comprehensively analyze the two-dimensional information and the three-dimensional information to obtain three-dimensional detection results of each defect on the surface of the component to be tested.

[0011] Compared with the prior art, the present invention has the following beneficial effects:

[0012] The present invention uses a 3D line spectrum confocal sensor module and a camera lens two-dimensional detection module to respectively obtain three-dimensional point cloud data and two-dimensional machine vision images, thereby achieving high-precision three-dimensional detection and quantitative evaluation of surface defects of curved optical elements.

[0013] The present invention can customize and plan the scanning path of the two-dimensional defect detection module according to the defect position obtained by the three-dimensional point cloud data, so as to realize fast and accurate detection.

[0014] The 3D line spectrum confocal sensor module and the two-dimensional detection module in the present invention can be installed on a variety of large mobile platforms, such as robotic arms, gantries, etc., through a rapid switching device, and can be used to detect large-aperture curved surface components with an aperture of 1-2m.

[0015] The 3D line spectrum confocal sensor module and the two-dimensional detection module in the present invention can be used separately for in-situ detection and off-line detection of surface defects of curved optical elements. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of a device in which the mobile platform proposed in the present invention is a mechanical arm.

[0017] Figure 2 It is a schematic diagram of a device in which the mobile platform proposed in the present invention is a gantry.

[0018] Figure 3 It is a flow chart of the method proposed by the present invention.

[0019] Figure 4 It is a point cloud image corresponding to a defect on the surface of the optical element in the embodiment.

[0020] Figure 5 It is a defect extracted after the point cloud image is processed in the embodiment.

[0021] Figure 6It is a dark field scattering image of the same defect obtained under a ring light source in the embodiment.

[0022] Figure 7 It is a complete scratch image stitched together from multiple dark-field scattering images.

[0023] Figure 8 is the minimum circumscribed rectangle corresponding to the scratch under dark field.

[0024] Fig. 9 is a bright field scattering image of the same defect obtained under a point light source in the embodiment.

[0025] Fig.10 It is a complete scratch image stitched together from multiple bright field scattering images.

[0026] Fig.11 is the minimum circumscribed rectangle corresponding to the scratch under bright field.

[0027] Description of Figure Numbers:

[0028] 1. Mobile platform; 2. 3D line spectrum confocal sensor; 3. Camera axis fine-tuning device; 4. Industrial camera; 5. Telecentric lens; 6. Point light source; 7. Sensor probe; 8. Ring light source; 9. Calibration head; 10. Component to be tested; 11. Sample placement platform. DETAILED DESCRIPTION

[0029] The present invention is described in detail below with reference to the accompanying drawings and embodiments. It should be noted that the protection scope of the present invention is not limited to the scope described in the embodiments.

[0030] The present invention proposes a three-dimensional detection device for surface defects of large-aperture curved optical elements, such as Figure 1 , Figure 2 The device includes a mobile platform 1, a 3D line spectrum confocal sensor 2, a camera axis fine-tuning device 3, an industrial camera 4, a high-resolution telecentric lens 5, a point light source 6, a sensor probe 7, a ring light source 8, a calibration head 9, a component to be measured 10, and a sample placement platform 11. Figure 1 In the example, the mobile platform 1 adopts a robotic arm; Figure 2In the figure, the mobile platform 1 adopts a gantry. The 3D line spectrum confocal sensor 2 is the main component of the three-dimensional point cloud detection module, and the industrial camera 4, the telecentric lens 5, the ring light source 8, the point light source 6, and the camera axis fine-tuning device 3 are components of the two-dimensional defect detection module; wherein the two-dimensional defect detection module is close to the front end of the three-dimensional point cloud detection module. The sensor probe 7 is located directly below the 3D line spectrum confocal sensor 2. The sensor probe 7 is used to emit line light and receive line light reflected by the component to be tested 10; the calibration head 9 is installed below the 3D line spectrum confocal sensor 2 and is located between the sensor probe 7 and the two-dimensional defect detection module. The position coordinates of three points on the boundary of the component to be tested 10 are determined by the calibration head 9, and then the spatial position of the component to be tested 10 is determined; the industrial camera 4 is located at the top of the two-dimensional defect detection module to realize the image acquisition function; the telecentric lens 5 is installed at the lower end of the industrial camera 4 to achieve clear imaging; the annular light source 8 is installed at the lower end of the telecentric lens 5 to achieve dark field lighting; the point light source 6 is installed at the opening on the side of the telecentric lens to achieve bright field lighting; the camera axis fine-tuning device 3 is located between the 3D line spectrum confocal sensor 2 and the telecentric lens 5, and can control the two-dimensional defect detection module to perform fine-tuning up and down to achieve image focus.

[0031] The mobile platform 1 drives the 3D line spectrum confocal sensor 2 to scan the large-aperture curved optical component along a predetermined path to obtain point cloud data of the entire surface of the component.

[0032] The two-dimensional defect detection module can choose to be illuminated by a point light source 6 or an annular light source 8, that is, a set of equipment can realize two imaging modes, bright field and dark field. The point light source 6 is refracted by the lens opening to realize bright field illumination, and obtain a machine vision image of a dark defect with a bright background; the annular light source 8 is located below the lens to realize dark field illumination, and obtain a machine vision image of a bright defect with a dark background. The two images are combined and analyzed to obtain more accurate two-dimensional information of the defect.

[0033] The two-dimensional defect detection module can obtain a clear defect image by combining coarse adjustment and fine adjustment. Coarse adjustment is to control the movement of the mobile platform 1 until the defect appears in the picture. Fine adjustment is to control the movement of the module through the camera axis fine adjustment device 3 to make the defect image clear.

[0034] The present invention proposes a method for three-dimensional detection of surface defects of large-aperture curved optical elements, which uses the above-mentioned three-dimensional detection device for surface defects of large-aperture curved optical elements. The implementation process is as follows: Figure 3 As shown. Specifically including the following steps:

[0035] Step S1, positioning and focusing of the component to be tested; specifically, the component to be tested 10 is placed on the sample placement platform 11, the mobile platform 1 is equipped with a 3D line spectrum confocal sensor 2, and is positioned using a calibration head 9. The 3D line spectrum confocal sensor 2 is moved down to a position where the light spot is clear to complete the focusing.

[0036] Step S2, the 3D line spectrum confocal sensor 2 scans to obtain the defect position and obtains the three-dimensional information of the defect; specifically: the scanning path of the 3D line spectrum confocal sensor 2 is determined according to the position, caliber and curvature radius of the component to be tested. The point cloud data obtained after scanning is processed to determine the position of each defect and the depth information of the defect. Among them, the point cloud data processing method steps are:

[0037] Step S2.1 uses point cloud data to fit a plane; including: using the RANSAC algorithm to fit a plane, the plane fitting formula and equation coefficient solution formula are as follows:

[0038] (1)

[0039] Among them, A, B, C, and D are the coefficients of the plane equation. are the coordinates of any three points in the point cloud data, is the coordinate of the point in the point cloud data.

[0040] Step S2.2 determines the defect location; including: calculating the distance from each point to the fitting plane. If the distance is greater than the set distance threshold T1, the point is considered to be a defect point, and then the location of each point on the defect is determined;

[0041] (2)

[0042] Among them, i is the i-th point in the point cloud dataset, is the coordinate of the point, is the distance from the point to the fitting plane.

[0043] Step S2.3 point cloud clustering; including: using Euclidean clustering to distinguish different defects. Calculating the Euclidean distance between a defect point and its k neighboring points. If the Euclidean distance is less than the Euclidean distance threshold T2, the neighboring point is placed in the clustering of the defect;

[0044] (3)

[0045] Among them, q is the Euclidean distance between the defect point and a neighboring point, is the position coordinate of the defect point, is the position coordinate of a nearby point.

[0046] Step S2.4 obtains three-dimensional defect information; including: for the same cluster, determining the difference between the maximum and minimum coordinates of the defect points in the cluster in each direction to determine the depth, length, width and other information of the defect.

[0047] Step S3, switch to the two-dimensional defect detection module to obtain two-dimensional information of surface defects; specifically: according to the location of the surface defect of the component, switch to the two-dimensional defect detection module composed of the industrial camera 4 and the high-resolution telecentric lens 5, select the point light source 6 or the ring light source 8 for illumination, and use the camera axis fine-tuning device 3 to focus, and obtain a bright field or dark field defect image. The SIFT stitching algorithm is used to stitch the multiple bright field images or dark field images obtained, and the stitched bright field or dark field defect images are processed; specifically, the following steps are included:

[0048] Step S3.1 constructs a Gaussian scale space; including: convolving the image with Gaussian kernels of different scale coefficients:

[0049] (4)

[0050] in, is a Gaussian scale space function of a certain scale, is a Gaussian function, σ is a Gaussian scaling coefficient, The defect image to be stitched is Grayscale value at coordinates.

[0051] Step S3.2 constructs a differential Gaussian scale space; including: subtracting adjacent Gaussian functions and then convolving them:

[0052] (5)

[0053] in, is a Gaussian function with variable scale, is the Gaussian difference scale space function, and k is the scaling factor between adjacent layers.

[0054] Step S3.3 determines the main direction of the feature point; including: if a sampling point in the Gaussian difference pyramid is an extreme point in its 3×3×3 neighborhood, then the extreme point is the feature point. Calculate the gradient amplitude and gradient direction within the 3σ neighborhood centered on the feature point. Use the histogram to count the gradient distribution of pixels in the neighborhood, and the direction with the largest gradient amplitude is the main direction of the feature point;

[0055] (6)

[0056] in, For The gradient amplitude of the feature point at the coordinate, For The gradient direction of the feature point at coordinates. For The Gaussian scale-space value of the point at coordinates, For The Gaussian scale-space value of the point at coordinates, For The Gaussian scale-space value of the point at coordinates, For The Gaussian scale-space value of the point at coordinates.

[0057] Step S3.4 generates a feature point descriptor; including: taking the feature point as the center, calculating the gradient information in 8 directions in a 4×4 area to obtain a 128-dimensional feature point descriptor.

[0058] Step S3.5: feature point matching and image stitching; including: calculating the distance between all feature point descriptors on different images, and the feature point pairs whose distance is less than the distance threshold T3 are matched point pairs. The image transformation matrix is ​​calculated according to the spatial position of each matching point pair to complete the image stitching.

[0059] Step S3.6 filters the spliced ​​image, including: using Gaussian filtering to smooth and denoise the spliced ​​image, and the Gaussian filter function is as follows:

[0060] (7)

[0061] Among them, G(x,y) is the Gaussian filter function, is the location coordinate.

[0062] Step S3.6 background unevenness correction; including: to avoid the influence of uneven illumination, top hat transformation is performed on the filtered dark field image:

[0063] (8)

[0064] Among them, I1 is the dark field stitching image after correction of illumination, f1 is the dark field stitching image before correction, b is the structural element, Represents the opening operation in morphological processing.

[0065] Perform bottom hat transformation on the filtered bright field stitching image:

[0066] (9)

[0067] Among them, I2 is the bright field stitching image after correction of illumination, f2 is the bright field stitching image before correction, b is the structural element, and • represents the closing operation in morphological processing.

[0068] Step S3.7: binarization of the image;

[0069] The maximum inter-class variance method is used to determine the binarization threshold T4 of the dark field or bright field stitching image after the correction illumination. The grayscale value of the pixel point in the image whose grayscale value is greater than the binarization threshold T4 is set to 1, and the grayscale value of the pixel point whose grayscale value is less than the binarization threshold T4 is set to 0.

[0070] Step S3.8: obtaining defect two-dimensional information;

[0071] The dark field stitching image and the bright field stitching image are processed using eight connected domains respectively. The pixels in the same connected domain constitute a defect. The defect is selected using the minimum bounding rectangle. The length of the minimum bounding rectangle is the length of the defect, and the width of the minimum bounding rectangle is the width of the defect.

[0072] Step S4, comprehensive analysis, obtain the defect three-dimensional detection results; specifically: the depth obtained from the point cloud data processing results is used as the defect depth. Since the image processing methods used in bright field and dark field are slightly different, there is a certain deviation, so the average of the length and width of the two is selected as the length and width of the scratch. Specific embodiments

[0073] First, two defect inspection modules are fixed on a moving platform, and a component to be tested 10 is placed on a sample placement platform 11 .

[0074] At the beginning of the test, the scanning path is planned according to the position, diameter and curvature radius of the component 10 to be tested, and the scanning is completed to obtain point cloud data. Figure 4 After plane fitting and clustering of the point cloud image, defect information can be extracted, as shown in Figure 5 As shown, the defect length is 8.228mm, the width is 0.148mm, and the maximum depth is 10.900um.

[0075] According to the position of the defect on the surface of the component, a two-dimensional defect detection module composed of an industrial camera 4 and a high-magnification telecentric lens 5 is used for detection. After splicing, filtering, background correction, binarization and connected domain determination of the acquired scattering image, the two-dimensional information of the defect is obtained. The dark field defect image obtained when using a ring light source is obtained, such as Figure 6 As shown; obtain the spliced ​​dark field scattering image, such as Figure 7 As shown; Get the minimum circumscribed rectangle image corresponding to the scratch under dark field, such as Figure 8 As shown; Get the bright field defect image obtained when using a point light source, such as Fig. 9 As shown; obtain the spliced ​​bright field scattering image, such as Fig.10 As shown; Get the minimum circumscribed rectangle image corresponding to the scratch under bright field, such as Fig.11 As shown. For this embodiment, the length of the scratch obtained in the dark field is 8.280mm, and the width is 0.197mm, and the length of the scratch obtained in the bright field is 8.295mm, and the width is 0.197mm. Due to the different processing methods adopted in the bright field and the dark field, there is a certain deviation, so the average value of the two is selected as the length and width of the scratch, that is, the length of the scratch is 8.288mm, and the width is 0.197mm.

[0076] Finally, the acquired two-dimensional defect information and three-dimensional defect information of the sample are comprehensively analyzed. In the process of two-dimensional data processing, the width of the defect is obtained by using the minimum circumscribed rectangle, and the width may be enlarged. Therefore, the width and depth of the defect can be obtained by processing the three-dimensional point cloud data, and the length can be obtained by processing the two-dimensional image. That is, the length of the scratch defect in the embodiment is 8.288mm, the width is 0.148mm, and the maximum depth is 10.900um. Thus, a three-dimensional detection device and method for surface defects of large-diameter curved optical elements is realized.

[0077] The above embodiments are only for illustrating the technical idea of ​​the present invention, and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the present invention.

Claims

1. A three-dimensional detection method for surface defects of large-aperture curved optical elements, the method is implemented based on a three-dimensional detection device for surface defects of large-aperture curved optical elements, the device comprising a mobile platform, a 3D line spectrum confocal sensor, an industrial camera, a telecentric lens, a ring light source, a point light source, and a camera axis fine-tuning device; wherein: The 3D line spectrum confocal sensor constitutes a three-dimensional point cloud detection module; the industrial camera, telecentric lens, annular light source, point light source, and camera axis fine-tuning device constitute a two-dimensional defect detection module; the two-dimensional defect detection module is close to the front end of the three-dimensional point cloud detection module; the three-dimensional point cloud detection module and the two-dimensional defect detection module are mounted on a mobile platform; the method comprises the following steps: Step S1: Positioning and focusing the component to be tested; including: placing the component to be tested on the sample placement platform, using the calibration head at the lower end of the 3D line spectrum confocal sensor to obtain the spatial coordinates of different positions of the component to be tested, and completing the positioning of the component to be tested; moving the 3D line spectrum confocal sensor downward to a position where the light spot is clear, and completing the focusing; Step S2: Scanning the 3D line spectrum confocal sensor to obtain the defect position and the three-dimensional information of the defect; including: determining the scanning path of the 3D line spectrum confocal sensor and the moving speed of the mobile platform according to the curvature radius, caliber and spatial position of the component to be tested, scanning the component to be tested according to the predetermined path, saving the obtained point cloud data, and determining the defect position and three-dimensional information after analyzing and calculating the point cloud data; Step S3: Switch to the two-dimensional defect detection module to obtain two-dimensional information of the defect; including: for a defect at a certain determined position, according to the offset between the two-dimensional defect detection module and the three-dimensional point cloud detection module, move the two-dimensional defect detection module to the defect, focus clearly and obtain a bright field or dark field image of the defect; if the defect is in multiple fields of view, customize the scanning path according to the defect position, complete the splicing of the bright field or dark field image of the entire defect and image processing, and then obtain the two-dimensional information of the defect; Step S4: Comprehensively analyze the two-dimensional information and the three-dimensional information to obtain three-dimensional detection results of each defect on the surface of the component to be tested.

2. A three-dimensional detection method for surface defects of large-aperture curved optical elements according to claim 1, characterized in that: In step S2, the specific steps of point cloud data analysis and calculation are as follows: Step S2.1: Fitting a plane using point cloud data; including: Perform plane fitting, the plane fitting formula and the plane equation coefficient solution formula are as follows: （1） Among them, A, B, C, and D are the coefficients of the plane equation. are the coordinates of any three points in the point cloud data; is the coordinate of the point in the point cloud data; Step S2.2: Determine the defect position; including: calculating the distance from each point to the fitting plane; if the distance is greater than the set distance threshold T1, the point is considered to be a point on the defect, and then the position of each point on the defect is determined; （2） Among them, i is the i-th point in the point cloud dataset, is the coordinate of the point, is the distance from the point to the fitting plane; Step S2.3: point cloud clustering; including: using Euclidean clustering to distinguish different defects; specifically: calculating the Euclidean distance between a point on the defect and its k neighboring points; if the Euclidean distance is less than the Euclidean distance threshold T2, the neighboring point is placed in the clustering of the defect; （3） Among them, q is the Euclidean distance between the point on the defect and a neighboring point, is the position coordinate of the point on the defect, is the position coordinate of a nearby point; Step S2.4: Obtain three-dimensional information of points on the defect; including: for the same cluster, determine the difference between the maximum and minimum coordinates of the points on the defect in each direction within the cluster to determine the depth, length and width information of the defect.

3. The three-dimensional detection method for surface defects of large-aperture curved optical elements according to claim 1, characterized in that: In step S3, the specific steps of bright field or dark field image stitching and image processing of the defect are as follows: Step S3.1: Constructing a Gaussian scale space; including: convolving the image with Gaussian kernels of different scale coefficients: （4） in, is a Gaussian scale space function of a certain scale, is a Gaussian function, σ The defect image to be stitched is Gray value at coordinates; Step S3.2: constructing a differential Gaussian scale space; including: subtracting adjacent Gaussian functions and then convolving them: （5） in, is a Gaussian function with variable scale, is the Gaussian difference scale space function, k is the scaling factor between adjacent layers; Step S3.3: determining the main direction of the feature point; including: if a sampling point in the Gaussian difference pyramid is an extreme point in its neighborhood, then the extreme point is the feature point, calculating the gradient amplitude and gradient direction within the 3σ neighborhood range centered on the feature point, using a histogram to count the gradient distribution of pixels in the neighborhood, and the direction with the largest gradient amplitude is the main direction of the feature point; （6） in, For The gradient amplitude of the feature point at the coordinate, For The gradient direction of the feature point at the coordinate, For The Gaussian scale-space value of the point at coordinates, For The Gaussian scale-space value of the point at coordinates, For The Gaussian scale-space value of the point at coordinates, For Gaussian scale-space value of the point at coordinates; Step S3.4: generating a feature point descriptor; including: taking the feature point as the center, calculating the gradient information in multiple directions in multiple regions to obtain a multi-dimensional feature point descriptor; Step S3.5: feature point matching and image stitching; including: calculating the distances between all feature point descriptors on different images, wherein the feature point pairs whose distances are less than the distance threshold T3 are matching point pairs, calculating the image transformation matrix according to the spatial positions of the matching point pairs, and completing the image stitching to obtain a dark field stitching image or a bright field stitching image; Step S3.6: filtering the dark field stitching image or the bright field stitching image; including: using a Gaussian filter to smooth and remove noise from the image, the Gaussian filter function is as follows: （7） Among them, G(x,y) is the Gaussian filter function, (x,y) is the position coordinate; Step S3.7: Background unevenness correction; including: performing top-hat transformation on the filtered dark field stitching image: （8） Among them, I1 is the dark field stitching image after correction of illumination, f1 is the dark field stitching image before correction, b is the structural element, Represents the opening operation in morphological processing; Perform bottom hat transformation on the filtered bright field stitching image: （9） Among them, I2 is the bright field stitching image after correction of illumination, f2 is the bright field stitching image before correction, b is the structural element, • represents the closing operation in morphological processing; Step S3.8: image binarization; including: using the maximum inter-class variance method to determine the binarization threshold T4 of the dark field stitching image or the bright field stitching image after the correction illumination, respectively, setting the gray value of the pixel point whose gray value is greater than the binarization threshold T4 in the image to 1, and setting the gray value of the pixel point whose gray value is less than the threshold binarization threshold T4 to 0; Step S3.9: Obtain two-dimensional defect information; including: using eight connected domains to process the binarized dark field stitching image and the bright field stitching image respectively, the pixels in the same connected domain constitute a defect, and the defect is selected using a minimum bounding rectangle, the length of the minimum bounding rectangle is the length of the defect, and the width of the minimum bounding rectangle is the width of the defect.

4. The three-dimensional detection method for surface defects of large-aperture curved optical elements according to claim 1, characterized in that: The component to be tested is a large-aperture curved optical component.

Images