CN108760766B - An image stitching method for detecting micro-defects on the surface of large-diameter optical crystals - Google Patents

Abstract

An image stitching method for detecting micro-defects on the surface of large-diameter optical crystals relates to an image stitching method for detecting micro-defects. The present invention solves the problem that the image acquisition and defect identification links of large-diameter crystals take a long time in the current defect detection. The invention firstly scans the surface area of the large-diameter crystal element to be measured, and uses a detection microscope and a detection CCD to collect real-time images of the scanned area, and determines the size range and overlapping area size of a single picture; Collect image splicing and coordinate transformation of defect points, determine the position of each defect point in each image in the global coordinate system, and establish a defect database. The invention is suitable for image splicing for detection of micro-defects on the surface of optical crystals.

Description

An image stitching method for detecting micro-defects on the surface of large-diameter optical crystals

technical field

The invention belongs to the field of optical engineering, and particularly relates to an image stitching method for micro-defect detection.

Background technique

In order to alleviate the problems of shortage of fossil fuels and serious environmental pollution faced by mankind, countries around the world are developing laser-driven inertial confinement nuclear fusion devices in order to obtain controllable clean fusion energy. Nonlinear optical crystal materials represented by KDP are used to make photoelectric switches and frequency multipliers because of their unique optical properties, and they have become irreplaceable core components in current laser fusion projects. However, water-soluble growth and ultra-precision machining of large-diameter optical crystals are extremely difficult (the traditional preparation cycle for monolithic crystals is as long as two years). In addition, the mechanical processing and laser pretreatment of optical crystal elements will introduce micron-scale defect points on the surface of the optical crystal element. These defect points are very easy to induce laser damage in the environment of high-energy laser use, and expand rapidly in the subsequent high-energy laser targeting process. , resulting in the scrapping of the entire crystal element, and ultimately severely limiting the optical performance and service life of the large-diameter crystal element. At this stage, the problem of laser damage caused by micro-defects on the surface of optical crystals has become a technical bottleneck restricting the increase in the output energy of laser fusion devices. The development of control and removal technologies for micro-defects on the surface of large-diameter optical crystals is very useful for achieving the expected goal of laser fusion ignition. significant.

At present, the use of micromachining technology to precisely repair micro-defects on the surface of optical crystals is the most promising strategy to alleviate the laser damage of components caused by defects and prolong the service life of large-diameter optical crystal components. This strategy can achieve high precision and high quality. , The recycling of large-diameter expensive optical crystals, so as to ensure the stable operation of the high-energy load of the laser nuclear fusion device. Rapid and accurate detection of micro-defects on the surface of optical crystals is the key to their precise repair. First, the size and shape information of micro-defects on the surface of optical crystals directly determine the formulation of subsequent micro-mechanical repair strategies and the selection of process parameters. In addition, the detection accuracy of micro-defect points on the surface of the optical crystal will seriously affect the determination of the relative position of the tool and the defect points to be repaired during the repair process, which will ultimately affect the successful repair of the defect points. Most importantly, the target density of the laser fusion device requires that the replacement, inspection, repair and re-installation process of large-diameter crystal elements must be completed within 4 hours, that is, defect detection must have the characteristics of high efficiency. However, the micro-defects on the surface of large-diameter optical crystals are small in size, different in shape, and uneven in distribution. It usually takes several hours for image acquisition and defect identification of large-diameter crystals in defect detection. Moreover, through the efficient detection of micro-defects, only the defect information of a large number of components and local areas can be obtained, and the position coordinates of the defect points are only relative to the pixel coordinates of a single image. The position coordinates in the global coordinate system of the optical crystal surface are the most valuable information for repair. Therefore, it is urgent to develop an image stitching method for the detection of micro-defects on the surface of large-aperture optical crystals. Through efficient and high-precision stitching of batch and local defect images obtained by inspection, the defect position distribution within the surface range of full-aperture optical crystals can be obtained. , which provides important parameter information for the effective repair of large-diameter optical crystal components.

SUMMARY OF THE INVENTION

The present invention solves the problem that the image acquisition and defect identification links of large-diameter crystals take a long time in the current defect detection.

An image stitching method for detecting micro-defects on the surface of large-diameter optical crystals is realized based on a fast searching and micro-milling repair device for large-diameter KDP crystal surface micro-defects. There is a marble platform under the air flotation frame, a repair window is opened on the marble platform, and the position directly below the corresponding air flotation frame, and a microscope and a detection CCD (881) are arranged on the microscope moving platform bracket (121);

An image stitching method for detecting micro-defects on the surface of a large-diameter optical crystal, comprising the following steps:

Step 1. Scan the surface area of the large-diameter crystal element to be measured, and use a detection microscope and a detection CCD to collect real-time images of the scanned area to obtain a batch of local and single scanned images of the large-diameter crystal element;

The detection CCD is a charge-coupled device image sensor;

During the process of collecting images, the overlapping dimensions of the field of view of the detection CCD in the moving directions of the X and Y axes are Δm and Δn respectively;

Step 2. Determine the size range and overlapping area size of a single image:

Move the large-diameter crystal element so that the edge of the crystal element is located in the repair window to ensure that there are no defects on the crystal surface observed in the detection CCD;

Then the micro-milling tool starts to rotate, and the three-axis linkage platform of the tool is lifted until the front end profile of the tool can be observed in the detection CCD, and the tool feeding is stopped; at this time, the tool is located at a certain position P ₁ in the detection CCD, and the image I ₁ is collected; Next, the moving tool moves in the X direction, the moving distance is x, reaches the position P ₂ , and collects the image I ₂ ; continues to move x to reach P ₃ , and collects the image I ₃ ; then continuously moves twice in the Y direction, and the moving distance is x, Collect images I ₄ and I ₅ respectively;

The images of five positions are processed, and the image feature matching algorithm is used to extract the feature points to calculate the pixel distance of the tool movement in the image.

Using the feature points concentrated at the position of the tool tip, the two images are matched in turn according to the moving sequence of the feature points; the pixel distance of the feature points moving in the X and Y directions is calculated respectively, and the average value of the feature points in the X and Y directions is calculated. Offset, that is, the average pixel moving distance ΔP of the feature point;

This results in the actual magnification K of the image:

K=(α/x)·ΔP

where α is the pixel size;

Calculate the size range of a single image through the actual magnification K of the image, W and H are the width and height of each image respectively; and determine the overlapping size of each captured image;

Step 3. Based on the coordinate system translation transformation method, the stitching of the collected images and the coordinate transformation of the defect points are realized:

Each image in the single image acquisition is named in the manner of "Xm _{x -Yny} ", where m _x and _ny represent the number of scanning steps traveled in the X and Y directions respectively, that is, represent the real-time location of the captured image. Position, the images with the same m _x or _ny number have the same X or Y coordinate position; assuming there are n images in the Y direction and m scanned images in the X direction, in the global composed of m × n scanned images In the coordinate system, it is assumed that the upper left corner is still the origin; I _i,j represents the images numbered i, j in the X and Y directions respectively, i, j=0, 1, 2..., then I _{i, j} The position of the coordinate origin O _i,j in the global coordinate system is O' _i,j , the coordinate value of each image in the global coordinate system (i·(W-Δm),j·(H-Δn)); O _i,j represents the coordinate origin of the image coordinate system corresponding to each picture;

Then, the position of each defect point in each image in the global coordinate system is determined, and a defect database is established.

Further, the image stitching method for detecting micro-defects on the surface of a large-diameter optical crystal further comprises the following steps:

Step 4. Use the coordinate system translation transformation method for image stitching:

First, according to the number and arrangement of images, a blank "canvas" is established. The size of the canvas is determined by the size of the images and the overlapping parts;

Then, through the position information of the defect points in the defect database, coordinate transformation is carried out on the positions of all defect points in each image, and at the corresponding position on the "canvas", according to the size of the defect point, draw a fitting ellipse indicating its shape and size; Complete image stitching for micro-defect detection containing only defect information.

Further, before determining the size range of a single image and the size of the overlapping area, a laser interferometer needs to be used to measure the positioning errors of the X and Y axes, and to perform following error compensation.

Further, the x is 500 μm.

The present invention has the following beneficial effects:

(1) By comparing the efficiency and accuracy based on the image feature matching algorithm and the present invention, the present invention takes no more than 1 minute to complete the entire process of image stitching for a 90mm×90mm collected image.

(2) The image coordinate system conversion and splicing method proposed by the present invention can quickly realize the splicing and defect coordinate conversion of batch local images through coordinate system translation transformation according to the size of the overlapping part between adjacent collected images, and convert each collected image into The defect point information is extracted and summarized in the global coordinate system of the entire crystal surface;

(3) The present invention calibrates the magnification of the defect detection microscope, determines the size of the field of view of the microscope and the detection CCD and the size of the overlapping portion of a single captured picture, and provides necessary parameter information for image splicing;

Description of drawings

Figure 1 is a top view of the inspection microscope and inspection CCD of the large-diameter KDP crystal surface micro-defect fast search and micro-milling repair device;

Fig. 2 is a schematic diagram of the overlapping area between the scanned images of the large-diameter optical crystal;

Fig. 3 is the defect scanning image mosaic map based on image feature matching;

4 is a schematic diagram of a magnification calibration process;

5 is a schematic diagram of feature point matching;

Fig. 6 is a schematic diagram of coordinate transformation splicing effect;

Fig. 7 is the data information sectional view of traversing the database to read each image;

FIG. 8 is a schematic diagram of a mosaic image finally formed by using the present invention in an embodiment.

Detailed ways

Specific implementation one:

An image stitching method for detecting micro-defects on the surface of large-diameter optical crystals is realized based on the rapid search and micro-milling repairing device for micro-defects on the surface of large-diameter KDP crystals (application number: 201310744691.1). The search and micro-milling repair device is shown in Figure 1. Based on the rapid search and micro-milling repair device for micro-defects on the surface of large-diameter KDP crystals, there is a marble platform under the air flotation frame. Repair window, microscope mobile platform bracket 121 is provided with microscope and detection CCD881;

An image stitching method for detecting micro-defects on the surface of a large-diameter optical crystal, comprising the following steps:

Step 1. Based on the "continuous motion acquisition" raster scanning scheme of large-diameter crystal elements, scan the surface area of the large-diameter crystal element to be measured, and use the detection microscope and detection CCD to collect real-time images of the scanned area to obtain batches of large-diameter crystal elements Partial and single scanned images of ;

The detection CCD (Charge Coupled Device) is a charge coupled device image sensor;

During the process of collecting images, the overlapping dimensions of the field of view of the detection CCD in the moving directions of the X and Y axes are Δm and Δn respectively;

The "continuous motion acquisition" raster scanning means that the optical crystal moves continuously along the raster scanning path, and at the same time, the upper detection CCD collects an image at a certain time interval (that is, one scanning step for each movement of the crystal); During the process, there is a certain margin (represented by Δm and Δn respectively) between the scanning step in the X and Y directions and the visual field distance in the corresponding direction of the image. The relative positional relationship can provide the necessary common overlapping area for splicing between images, as shown in Figure 2;

Step 2. Determine the size range and overlapping area size of a single image:

Move the large-diameter crystal element so that the edge of the crystal element is located in the repair window to ensure that there are no obvious defects on the crystal surface observed in the detection CCD;

Then the micro-milling tool starts to rotate, and the three-axis linkage platform of the tool is lifted until the contour of the front end of the tool can be clearly observed in the detection CCD, the feed of the tool is stopped, and the tool cannot contact the lower surface of the upper crystal; The tool is located at a certain position P ₁ , and the image I ₁ is collected; then, the tool is moved by the motor in the X direction, the moving distance is x, and the position P ₂ is reached, and the image I ₂ is collected; continue to move x to reach P ₃ , and the image I ₃ is collected; Then continuously move twice in the Y direction, the moving distance is x, and collect images I ₄ and I ₅ respectively, as shown in FIG. 4 ;

Due to the high motion accuracy of the tool moving motor, the distance can be used as the standard length to effectively control the error; although the imaging unit of the detection CCD is theoretically a standard square, there may also be a certain error, through the simultaneous calibration of the X and Y directions , which can suppress the imaging distortion of the detection CCD;

The images of five positions are processed, and the image feature matching algorithm is used to extract the feature points to calculate the pixel distance of the tool movement in the image, which can avoid the error when manually selecting the reference point, and the accuracy is higher;

Since there are no defect points with obvious features on the crystal surface, the feature points concentrated at the position of the tool tip are used to match two images in sequence according to the moving sequence of the feature points. The effect is shown in Figure 5; Matching operation, calculate the pixel distance of the feature points moving in the X and Y directions respectively, and calculate the average offset of the feature points in the X and Y directions, that is, the average pixel moving distance ΔP of the feature points;

The time for each image matching operation is about 13s. Since the magnification calibration process only needs to be performed once before the optical crystal scan, the efficiency is acceptable; from this, the actual magnification K of the image is obtained:

K=(α/x)·ΔP=(3.45/500)·ΔP

where α is the pixel size;

Calculate the size range of a single image by the actual magnification K of the image, W and H are the width and height of each image respectively; and determine the overlapping size of each captured image; taking the magnification of 2.25X as an example, this When the image width W=3.766mm, Δm=0.266mm (corresponding to 196pixels), height H=3.156mm, Δn=0.156mm (corresponding to 102pixels);

Step 3. Based on the coordinate system translation transformation method, the stitching of the collected images and the coordinate transformation of the defect points are realized:

The stitching principle based on the coordinate system translation transformation method is shown in Figure 6, in which each image in the single image acquisition is named in the way of "Xm _{x -Yny} ", where m _x and _ny represent the X and Y directions respectively. The number of past scanning steps, that is, the real-time position of the captured image, the images with the same m _x or _ny number have the same X or Y coordinate position; assuming that there are n images in the Y direction, and there are n images in the X direction m scanned images, in the global coordinate system composed of m×n scanned images, it is assumed that the upper left corner is still the origin; I _i,j represent the images numbered i, j in the X and Y directions, i, j j=0,1,2..., then the position of the coordinate origin O _i,j of I _i ,j in the global coordinate system is O' _i,j , the coordinate value of each image in the global coordinate system (i ·(W-Δm),j·(H-Δn)), as shown in Figure 6; O _i,j represents the coordinate origin of the image coordinate system corresponding to each picture;

Then, the position of each defect point in each image in the global coordinate system is determined, and a defect database is established.

Specific implementation two:

The image stitching method for detecting micro-defects on the surface of large-diameter optical crystals described in this embodiment is characterized in that it further comprises the following steps:

Step 4. Use the coordinate system translation transformation method for image stitching:

The crystal scanning image is essentially a carrier of information, most of the information contained in the image can be ignored, and only the parameter information of the defect points obtained through detection is the real important information; therefore, it can be seen from Figure 6 that, The essence of image stitching through coordinate system transformation is to extract and summarize the information of limited defect points in each image, while ignoring the complete image that contains a lot of useless information;

First, according to the number and arrangement of images, a blank "canvas" is established. The size of the canvas is determined by the size of the images and the overlapping parts;

Then, through the position information of the defect points in the defect database, coordinate transformation is carried out on the positions of all defect points in each image, and at the corresponding position on the "canvas", according to the size of the defect point, draw a fitting ellipse indicating its shape and size; In this way, the information of all defect points in the entire scanning range can be intuitively read from the "canvas"; image stitching for micro-defect detection containing only defect information is completed.

Other steps and parameters are the same as in the first embodiment

Specific implementation three:

In this embodiment, before determining the size range of a single image and the size of the overlapping area, a laser interferometer needs to be used to measure the positioning errors of the X and Y axes, and to perform following error compensation.

In order to ensure the applicability and accuracy of the present invention, it is necessary to improve the motion characteristics of the X-axis and Y-axis linear units of the crystal scanning motion, measure the positioning error and compensate:

The premise of transforming and splicing based on the image coordinate system is to ensure that the optical crystal scanning motion position has a high accuracy and can determine the size of the overlapping part between the images collected each time;

The servo motor and linear motor of the optical crystal moving axis are installed with a grating ruler to realize the position signal feedback. By adjusting the PID parameters of the motor closed-loop control system, the motion steady-state characteristics and dynamic characteristics of the X and Y axes can be improved;

Using a laser interferometer to measure the positioning errors of the X and Y axes, and perform following error compensation, the positioning errors of the two axes can be controlled within 3μm (300mm stroke); at this time, the position accuracy of the optical crystal scanning motion can meet the requirements.

Other steps and parameters are the same as in the first or second embodiment

Example

The image stitching directly through the image feature matching algorithm and the present invention will be compared and explained:

Install the optical crystal element to be tested on the large-diameter KDP crystal surface micro-defect fast search and micro-milling repair device (application number: 201310744691.1);

Based on the "continuous motion acquisition" raster scanning scheme of large-aperture crystal elements, the surface area of the large-aperture crystal element to be measured is scanned, and a detection microscope and a detection CCD are used to collect real-time images of the scanned area, so as to obtain local and single scanned image;

The detection CCD (Charge Coupled Device) is a charge coupled device image sensor;

Then adopt the stitching method based on the image feature matching algorithm to perform image stitching and the present invention to perform image stitching:

(1), adopt the stitching method based on image feature matching algorithm:

The image feature matching algorithm stitching method is a comprehensive and complex implementation process, which needs to call a large number of image algorithms to support the implementation, mainly including image feature searching and matching, lens calibration, image shape transformation, light compensation and fusion, etc. In the defect detection system shown in Figure 1, due to the horizontal movement of the optical crystal, the detection CCD and the surface of the optical crystal are always in a vertical state, and the brightness of the light source is constant, so there is no deformation and distortion between each collected image, and the process of image stitching can be simplified. There are two steps for feature point searching, matching and image fusion. The specific implementation steps are as follows:

(11) Finding and matching feature points of locally collected images. Feature points (corner points) refer to points in the image that are stable and can clearly reflect the changing characteristics of the image. Described mathematically, feature points have obvious derivatives in both orthogonal directions. Here, the feature point extraction and matching method commonly used in the field of computer vision - SIFT algorithm is used. The SIFT algorithm (Speed UpRobust Features) is stable in the extraction of feature points under geometric deformation and illumination changes, and is suitable for the extraction of local objects. At the same time, it also has the characteristics of high-speed extraction.

The Hessian matrix is the core of the SURF algorithm. Assuming that there is a function f(x,y), the corresponding Hessian matrix consists of the function and its partial derivatives:

The discriminant of the Hessian matrix is its eigenvalue, and whether the point is an extreme point is determined by positive and negative. For an image, replace f(x,y) with pixel values I(x,y) at different positions, select the second-order standard Gaussian function as the filter, and calculate the second-order partial derivative through the convolution between specific kernels, so that Get the Hessian of an image of scale σ:

In the formula, u=(x,y) ^T , L(u,σ)=G(σ)*I(u), and the inter-kernel function is the second-order partial derivative of the Gaussian function:

或

or

The feature points can be discriminated by calculating the value of the discriminant of H(u,σ). In order to balance the error between the exact value and the approximate value, the following discriminant is used:

det(H _approx )=D _xx D _yy -(0.9D _xy ) ² (4)

After finding the feature points, the SURF algorithm constructs a pyramid-shaped image space: a multi-layer image is obtained by changing the size of the filter kernel, and then the feature points are precisely located at each layer in the image space, and finally the Haar of each feature point needs to be calculated. The wavelet response, used to determine the main directions of the feature points.

(1) Fusion of locally collected images. Image fusion is to superimpose the two images according to the overlapping area between the two images calculated from the matching result in (1). In the crystal scan image, the overlapping area of the image can be approximated as a rectangle due to the stability of illumination and geometric conditions. Taking the left and right overlapping images as an example, the fused image can be divided into three parts: the left and right exclusive areas and the public area. The left and right exclusive areas retain the corresponding pixel values of the original images, and the common area is superimposed according to the overlap ratio. The corresponding pixel values of the left and right images.

The process of image feature point extraction, matching and image fusion is realized through the cross-platform open source computer vision image processing library OpenCV (Open Source Computer Vision Library). Here, three optical crystal scan pictures are selected, first stitched in pairs, and then stitched again, the result is shown in the figure 3 shown.

In the source image, two obvious defect points can be found. After the stitching is completed, the actual overlapping area can be calculated according to the stitched image pixels, so the coordinate system from a single image to multiple images can be determined. conversion relationship. In this cycle, the image stitching of the entire optical crystal surface can be realized.

(2) Realize the stitching of the collected images based on the coordinate system translation transformation method:

(21) Determine the scanning range according to the number (quantity) of the collected images, that is, the scale range of the crystal coordinate system. In order to quickly verify the stitching function of locally collected images on the surface of the optical crystal and ensure the relative positional relationship between each image is accurate, in this example, the actual scanning size of the optical crystal is reduced to 90mm×90mm, and the overlap between adjacent images is appropriately increased. area size. To this end, the magnification of the microscope was reduced from 2.25X to 1.5X, and the scanning step was reduced to Δx=Δy=3.0mm. Under this condition, 31×31=961 images need to be collected, occupying about 5GB of hard disk space. The width and height of the image is 2456×2058 pixels. At this time, the size of the overlapping part of the captured image is Δm=1152 and Δn=754. From this, the coordinate range is calculated as X _scale = 30×2456-29×1152=40272, Y _scale =30×2058- 29×754=39874. In theory, although the scanning ranges in the X and Y directions are equal, in the outermost circle of the actual scanning range, the image field of view is larger than Δx and Δy, resulting in the coordinate range of the image being larger than the scanning range.

(22) Read the database information written in the automatic defect detection, and perform coordinate transformation on the position of each image defect point according to the image number sequence. In the automatic defect detection program, a defect point information table named in the order of the images is established in the database for 961 images, and the details of each defect point in the image are recorded. The correspondence relation between the data table number N and the image name number "Xm _{x -Yny} " is N=31m _x + _ny . According to the coordinate conversion image stitching method, it is assumed that there is a defect point in the image "XpYq.bmp" at (x _i , x _j ) (pixel), if the coordinate origin is set at the scanning starting point, the transformed global coordinate is (x _i +p·(W-Δm),x _j +q·(H-Δn)); actually set the origin as the center of the crystal, then the coordinates are (x _i +p·(W-Δm)-19565,x _j + q·(H-Δn)-19565). In order to record the information of all defect points in the global coordinate system, a new data table "tbAllDefectsInfo" is created, and the fields in the table are set to the same name as the image; each time the coordinate transformation of a defect point is completed, a corresponding row is written into the data table. field value. The above process is repeated until the entire database is traversed, and the data information of each image is read, as shown in Figure 7.

(23) Create a new background image and mark the positions of all defect points. In actual operation, because the image coordinate range of the full scan range is too large, when creating a new image, because the pixel size exceeds the size of the allocated memory, the new image cannot be created (the actual maximum image that can be newly created does not exceed 15000 × 15000 pixels). Here, the coordinate range is reduced proportionally, and if the current coordinate range is reduced to 1/5, the image size is 8054×7974 pixels. At this time, the coordinates of the defect point are also reduced to 1/5, and the defect area is reduced to 1/25. When the area is less than 1, the defect point is represented by 1 pixel. The final stitched image is shown in Figure 8.

In Figure 8, the white background is used as the crystal surface, the black dots represent the defect points on the crystal surface, and the diameter of the dots represents the defect area. A total of 394 defect points were actually detected. Generally speaking, the distribution of defect points on the crystal has a certain randomness. Since the image is saved in jpg format, the size of the image stitched by the actual 90mm×90mm scanning range is only 1.02MB; in terms of efficiency, the entire process from reading and writing the database to the final image stitching takes less than 1 minute, which is very efficient. Since the process of coordinate conversion and splicing is completed by digital operation, there is basically no error. This verifies the technological feasibility of image stitching using the coordinate transformation method.

Considering the efficiency and accuracy of image stitching, the feasibility comparison of two methods based on image feature matching algorithm stitching and image coordinate system conversion stitching:

For the stitching method based on the image feature matching algorithm, it can be accurate to 1 pixel, and to a certain extent, the error caused by the non-parallelism between the width and height of the CCD and the direction of the crystal motion axis can be eliminated. However, this image stitching method has serious efficiency problems for image stitching of large-diameter optical surfaces. In the case of no image pixel compression, an average stitching takes about 10s. If the surface of the large-diameter optical crystal of 410mm×410mm is scanned in full range, there are 118×360/3=14160 images in theory. This splicing method is not feasible in terms of efficiency. In addition, the image stitching algorithm operates on the complete image, and the image is operated in the memory during the operation of the program, and the size of each bmp format image is 4.82MB. As the stitched images become larger and larger, subsequent image stitching of large-sized optical surfaces cannot be achieved.

Compared with the stitching method based on the image feature matching algorithm, the image coordinate system conversion stitching method is more efficient and intuitive, and it is more feasible for image stitching for the detection of micro-defects on the surface of large-diameter optical crystals. Even though the image stitching method will cause certain errors due to the limitation of the scanning speed, in practice, the stitching error can be reduced by increasing the acquisition frequency of the crystal motion position signal in the numerical control program, so as to realize the surface defect detection of large-area optical crystals It is a high-precision stitching of images.

Claims (3)

1. An image stitching method for the detection of micro-defects on the surface of large-diameter optical crystals is realized based on the rapid search and micro-milling repairing device for micro-defects on the surface of large-diameter KDP crystals. A marble platform is arranged under the air-floating frame of the repairing device, a repairing window is opened on the marble platform at a position directly below the corresponding air-floating frame, and a microscope and a detection CCD (881) are arranged on the microscope moving platform bracket (121); It is characterized in that, comprises the following steps: Step 1. Scan the surface area of the large-diameter crystal element to be measured, and use a detection microscope and a detection CCD to collect real-time images of the scanned area to obtain a batch of local and single scanned images of the large-diameter crystal element; The detection CCD is a charge-coupled device image sensor; During the process of collecting images, the overlapping dimensions of the field of view of the detection CCD in the moving directions of the X and Y axes are Δm and Δn respectively; Step 2. Determine the size range and overlapping area size of a single image: Move the large-diameter crystal element so that the edge of the crystal element is located in the repair window to ensure that there are no defects on the crystal surface observed in the detection CCD; Then the micro-milling tool starts to rotate, and the three-axis linkage platform of the tool is lifted until the front end profile of the tool can be observed in the detection CCD, and the tool feeding is stopped; at this time, the tool is located at a certain position P ₁ in the detection CCD, and the image I ₁ is collected; Next, the moving tool moves in the X direction, the moving distance is x, reaches the position P ₂ , and collects the image I ₂ ; continues to move x to reach P ₃ , and collects the image I ₃ ; then continuously moves twice in the Y direction, and the moving distance is x, Collect images I ₄ and I ₅ respectively; The images of five positions are processed, and the image feature matching algorithm is used to extract the feature points to calculate the pixel distance of the tool movement in the image. Using the feature points concentrated at the position of the tool tip, the two images are matched in turn according to the moving sequence of the feature points; the pixel distance of the feature points moving in the X and Y directions is calculated respectively, and the average value of the feature points in the X and Y directions is calculated. Offset, that is, the average pixel moving distance ΔP of the feature point; This results in the actual magnification K of the image: K=(α/x)·ΔP where α is the pixel size; Calculate the size range of a single image through the actual magnification K of the image, W and H are the width and height of each image respectively; and determine the overlapping size of each captured image; Step 3. Based on the coordinate system translation transformation method, the stitching of the collected images and the coordinate transformation of the defect points are realized: Each image in the single image acquisition is named in the manner of "Xm _{x -Yny} ", where m _x and _ny represent the number of scanning steps traveled in the X and Y directions respectively, that is, represent the real-time location of the captured image. Position, the images with the same m _x or _ny number have the same X or Y coordinate position; assuming there are n images in the Y direction and m scanned images in the X direction, in the global composed of m × n scanned images In the coordinate system, it is assumed that the upper left corner is still the origin; I _i,j represents the images numbered i, j in the X and Y directions respectively, i, j=0, 1, 2..., then I _{i, j} The position of the coordinate origin O _i,j in the global coordinate system is O' _i,j , the coordinate value of each image in the global coordinate system (i·(W-Δm),j·(H-Δn)); O _i,j represents the coordinate origin of the image coordinate system corresponding to each picture; Then determine the position of each defect point in the global coordinate system in each image, and establish a defect database; Step 4. Use the coordinate system translation transformation method to practice image stitching: First, according to the number and arrangement of images, a blank "canvas" is established. The size of the canvas is determined by the size of the images and the overlapping parts; Then, through the position information of the defect points in the defect database, coordinate transformation is carried out on the positions of all defect points in each image, and at the corresponding position on the "canvas", according to the size of the defect point, draw a fitting ellipse indicating its shape and size; Complete image stitching for micro-defect detection containing only defect information. 2. the image splicing method that a kind of large-diameter optical crystal surface micro-defect detection according to claim 1 is used, it is characterized in that, need to utilize laser interferometer to measure X, Y before determining the size range of single picture and overlapping area size The positioning error of the axis, and the following error compensation is performed. 3 . The image stitching method for detecting micro-defects on the surface of large-diameter optical crystals according to claim 2 , wherein the moving distance x described in step 2 is 500 μm. 4 .

Images