Method and apparatus for automatic centering control for metasurface lens

Abstract

The application discloses an automatic centering control method and device for a metasurface lens, and the method comprises the following steps: collecting an initial image from a visual camera and extracting a fusion feature map, performing edge analysis on the fusion feature map to obtain an edge attention weight map, and then performing pixel-by-pixel weighting on the fusion feature map to obtain an edge-enhanced feature map; performing edge prediction on the fusion feature map and combining the edge-enhanced feature map to obtain a center coordinate, and judging whether the center coordinate is qualified or not; if the center coordinate is qualified, a qualified prompt information is sent out; if the center coordinate is not qualified, an adjustment instruction is generated and sent to a three-jaw three-axis displacement table, and the steps of obtaining the center coordinate and centering are repeated again. Through image feature extraction and edge analysis, the edge features of the metasurface lens and the annular gasket can be quickly and accurately extracted, and the center deviation can be calculated, thereby providing accurate data support for subsequent centering operation, and the efficiency and accuracy of the centering control of the metasurface lens are greatly improved.

Description

Technical Field

[0001] This invention relates to the field of metasurface device control technology, and in particular to an automatic centering control method and apparatus for metasurface lenses. Background Technology

[0002] With the continuous development of optical technology and the increasing maturity of advanced nanofabrication technology, metasurface lenses, thanks to their subwavelength-scale artificial microstructure design, can achieve ultra-thin and integrated optical functions that are difficult to achieve with traditional optical components. They have been widely used in high-end imaging, lidar, quantum communication, and other fields. By integrating complex optical functions onto a small chip-level component, they can be more tightly integrated with other optoelectronic devices such as light sources and detectors, achieving more compact optical path designs, which has attracted much attention in scientific research and industry. However, to achieve accurate beam control by metasurface lenses, lens alignment is required. Currently, existing technologies rely on operators observing the lens edge against a reference mark using a microscope, requiring manual rotation of a fine-tuning frame for positioning and adjustment. This method is affected by visual fatigue and operator experience, resulting in large fluctuations in alignment accuracy. Furthermore, existing equipment is mostly designed for traditional spherical lenses, using a single optical path detection for lens positioning and adjustment, which cannot adapt to the asymmetric microstructure reflection characteristics of metasurface lenses. Therefore, there is an urgent need for a professional, high-precision, end-to-end automated alignment device in current production applications to achieve rapid and accurate positioning of metasurface lenses during assembly. Therefore, the existing methods for positioning and adjusting metasurface lenses in imaging devices suffer from low adjustment efficiency. Summary of the Invention

[0003] This invention provides an automatic centering control method and apparatus for metasurface lenses, aiming to solve the problem of low adjustment efficiency in existing methods for positioning and adjusting metasurface lenses in imaging devices.

[0004] In a first aspect, embodiments of the present invention provide an automatic centering control method for metasurface lenses, wherein the method is applied in a control terminal, the control terminal establishing communication connections with a vision camera and a three-pin three-axis displacement stage to achieve data information transmission, the three-pin structure on the three-pin three-axis displacement stage fixing the metasurface lens by vacuum adsorption, the vision camera, the microscopic magnification module, the hollow gasket adsorption stage, and the three-pin three-axis displacement stage being arranged sequentially along the same optical axis, and a linked annular supplementary light source being arranged around the periphery of the microscopic magnification module, the method comprising:

[0005] Upon receiving the start command, the corresponding fusion feature map is extracted from the initial image acquired by the visual camera according to the preset image feature extraction model;

[0006] The fused feature map is parsed according to a pre-set edge attention parsing model to obtain the corresponding edge attention weight map;

[0007] The fused feature map is weighted pixel-by-pixel based on the edge attention weight map to obtain the corresponding edge enhancement feature map;

[0008] The fused feature map is input into a preset edge prediction model to obtain the corresponding edge prediction information;

[0009] The edge prediction information and the edge enhancement feature map are fitted according to a preset fitting rule to obtain the corresponding center coordinates of the circle.

[0010] The coordinates of the center of the circle are judged to determine whether they meet the preset deviation judgment rules, so as to obtain the judgment result of whether they are qualified;

[0011] If the judgment result is unqualified, a corresponding adjustment command is generated based on the center coordinates and sent to the three-pin three-axis displacement stage to control the three-pin three-axis displacement stage to adjust the center position of the metasurface lens and return to execute the step of extracting the corresponding fusion feature map from the initial image acquired by the vision camera according to the preset image feature extraction model;

[0012] If the judgment result is qualified, a corresponding qualified prompt message will be issued.

[0013] Secondly, embodiments of the present invention also provide an automatic centering control device for metasurface lenses, wherein the device includes a control terminal, a vision camera, a three-pin three-axis displacement stage, a linked annular supplementary light source, a microscopic magnification module, and a hollow gasket adsorption stage. The three-pin structure on the three-pin three-axis displacement stage is fixedly fixed to the metasurface lens by vacuum adsorption. The three-pin three-axis displacement stage is fixedly mounted on a mechanical support platform. The control terminal establishes communication connections with the vision camera and the three-pin three-axis displacement stage to realize data information transmission. The vision camera, the microscopic magnification module, the hollow gasket adsorption stage, and the three-pin three-axis displacement stage are arranged sequentially along the same optical axis, and the linked annular supplementary light source is arranged around the periphery of the microscopic magnification module.

[0014] The control terminal is used to execute the automatic centering control method for metasurface lenses as described in the first aspect above.

[0015] Thirdly, embodiments of the present invention also provide a computer device, wherein the device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0016] Memory, used to store computer programs;

[0017] When the processor executes the program stored in the memory, it implements the steps of the automatic centering control method for metasurface lenses described in the first aspect above.

[0018] Fourthly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the automatic centering control method for metasurface lenses as described in the first aspect above.

[0019] This invention provides an automatic centering control method for metasurface lenses. The method includes: acquiring an initial image from a vision camera and extracting a fused feature map; performing edge analysis on the fused feature map to obtain an edge attention weight map; then performing pixel-by-pixel weighting on the fused feature map to obtain an edge enhancement feature map; performing edge prediction on the fused feature map and fitting it with the edge enhancement feature map to obtain the center coordinates; determining whether the center coordinates are qualified; if qualified, issuing a qualified prompt message; if unqualified, generating an adjustment command and sending it to a three-axis displacement stage, and repeating the above steps of obtaining the center coordinates and centering. This method, through image feature extraction and edge analysis, can quickly and accurately extract the edge features of the metasurface lens and the annular washer and calculate the center deviation, providing accurate data support for subsequent centering operations, thereby significantly improving the efficiency and accuracy of centering control of the metasurface lens. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of an automatic centering control method for metasurface lenses provided in an embodiment of the present invention.

[0022] Figure 2 This is a schematic diagram illustrating an application scenario of the automatic centering control method for metasurface lenses provided in an embodiment of the present invention.

[0023] Figure 3 This is a schematic diagram illustrating the application effect of the automatic centering control method for metasurface lenses provided in this embodiment of the invention.

[0024] Figure 4 This is a schematic diagram illustrating another application effect of the automatic centering control method for metasurface lenses provided in this embodiment of the invention;

[0025] Figure 5This is a structural diagram of an automatic centering control device for metasurface lenses provided in an embodiment of the present invention;

[0026] Figure 6 The diagram shows the structure of the three-pin three-axis displacement stage in the automatic centering control device for metasurface lenses provided in this embodiment of the invention.

[0027] Figure 7 A partial structural diagram of an automatic centering control device for metasurface lenses provided in an embodiment of the present invention;

[0028] Figure 8 A schematic block diagram of a control terminal provided in an embodiment of the present invention;

[0029] Figure 9 This is a schematic block diagram of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0032] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0033] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0034] Please see Figure 1 As shown in the figure, an embodiment of this invention provides an automatic centering control method for metasurface lenses. This method is applied in a control terminal 10 and is executed by application software installed in the control terminal 10. Figure 2As shown, the control terminal 10 establishes communication connections with the vision camera 11 and the three-pin three-axis displacement stage 12 to realize the transmission of data information. The three-pin structure 121 on the three-pin three-axis displacement stage 12 fixes the metasurface lens by vacuum adsorption. The vision camera 11, the microscopic magnification module 13, the hollow gasket adsorption stage 14 and the three-pin three-axis displacement stage 12 are arranged sequentially along the same optical axis. The linked ring supplementary light source 15 is arranged around the periphery of the microscopic magnification module 13. The three-pin three-axis displacement stage 12 is used to fix the metasurface lens 122 and adjust its position. The linked annular supplementary light source 15 emits a beam to supplement the illumination of the metasurface lens and the three-pin three-axis displacement stage 12 below. The central area of ​​the hollow gasket adsorption stage 14 is hollowed out, and three adsorption holes are provided on the hollow gasket adsorption stage 14 for vacuum adsorption of the annular gasket to prevent slippage during the centering process. The vision camera 11 is used to acquire images containing the annular gasket and the metasurface lens, and the microscopic magnification module 13 is used to adjust the magnification during image acquisition. The metasurface lens includes a substrate and a metasurface structure disposed on the substrate. The metasurface structure is composed of several microstructures, which can be cylindrical, square, or cross-shaped structures. The control terminal 10 can be a terminal device with data processing and command transmission / reception functions, such as a laptop, desktop computer, tablet computer, or mobile phone. Figure 1 As shown, the method includes steps S110 to S180.

[0035] S110. Receive the start command and extract the corresponding fused feature map from the initial image acquired by the vision camera according to the preset image feature extraction model.

[0036] Upon receiving the user's start command, the process of centering control of the metasurface lens is initiated. First, an initial image acquired by a vision camera is obtained, and the corresponding fusion features are extracted from the acquired initial image based on an image feature extraction model.

[0037] In a specific embodiment, step S110 includes the following sub-steps: performing continuous convolution processing on the initial image according to multiple convolutional layers configured in the image feature extraction model, with each convolutional layer outputting a feature map; performing pooling compression processing on each feature map according to the pooling compression layer in the image feature extraction model to obtain a feature saliency score corresponding to each feature map; performing weight allocation according to the weight allocation function in the image feature extraction model and the feature saliency score to obtain dynamic weight coefficients corresponding to each feature map; upsampling each feature map according to a preset resolution parameter in the image feature extraction model to obtain an upsampled feature map corresponding to the resolution parameter; and weighted fusing the upsampled feature maps of each feature map according to the dynamic weight coefficients to obtain a corresponding fused feature map.

[0038] First, to optimize the detail-sensitive characteristics of submicron-level microstructures in metasurface lenses, an improved feature pyramid structure multi-scale feature fusion network was constructed as an image feature extraction model. By adding a high-resolution feature layer F2 to the classic YOLO v8 backbone network (CSPDarknet), which has a lower sampling rate of 1 / 8 compared to the traditional minimum feature layer, more detailed information of microstructure edges, such as the boundary contours of nanopillar arrays, can be preserved, thus solving the problem of feature loss at tiny edges in traditional algorithms.

[0039] The image feature extraction model is configured with multiple convolutional layers. These layers can perform continuous convolution processing on the initial image; that is, the feature map obtained from the convolution process of the previous convolutional layer is further processed by the next convolutional layer. Therefore, the feature map size obtained after convolution processing by each convolutional layer is different. For example, by performing convolution processing on the input image through various convolutional layers, four different scale feature maps—F2, F3, F4, and F5—are obtained through downsampling. Specifically, the F2 feature map is downsampled by 1 / 4, focusing on sub-micron level edges; the F3 feature map is downsampled by 1 / 8, preserving local details such as microstructure arrays; the F4 feature map is downsampled by 1 / 16, extracting medium-scale edges such as the boundary of a ring washer; and the F5 feature map is downsampled by 1 / 32, capturing global contours such as the overall shape of a lens.

[0040] Further, based on the pooling compression layer in the image feature extraction model, each feature map is processed... F i Pooling compression processing is performed separately (subscript) i This represents the sequence number of the feature map. i =2,3,4,5), specifically, global average pooling is performed on each feature map to compress each feature map into a 1×1 vector, thereby obtaining the feature saliency score corresponding to each feature map, that is, each feature map can be given a feature saliency score; the specific process can be expressed by formula (1):

[0041] (1);

[0042] S(F i ) That is, the obtained first i The feature significance score corresponding to each feature map.

[0043] In the image feature extraction model, the weight allocation function and the feature saliency score are used to allocate weights. Then, the feature saliency score of each feature map can be weighted to obtain a dynamic weight coefficient. By constructing an attention-weighted fusion mechanism, a dynamic weight allocation strategy is adopted to replace the traditional equal weight or fixed weight fusion. The dynamic weight allocation to obtain the dynamic weight coefficient corresponding to each feature map can be expressed by formula (2):

[0044] (2);

[0045] w i That is, the first i Dynamic weight coefficients assigned to each feature map. exp Represented by the base of the natural logarithm e Exponentiation of bases; where i =2,3,4,5. By normalizing the scores using the above formula (2), the weight coefficients corresponding to each feature map can be obtained. w i By obtaining dynamic weight coefficients to dynamically adjust the feature contribution of each layer, the network can automatically focus on effective edge regions such as the edge of metasurface lenses and the edge of annular washers in complex backgrounds.

[0046] Each feature map is upsampled according to the preset resolution parameters in the image feature extraction model; for example, the preset resolution parameters are the feature maps. F 2 The corresponding resolution is represented by the feature map. F 2 Based on the resolution, the three feature maps F3, F4, and F5 are upsampled to the feature map. F 2 With a given resolution, the upsampled feature maps corresponding to each feature map can be obtained; where, the feature map F 2 The corresponding upsampled feature map is also known as the feature map. F 2 .

[0047] The upsampled feature maps of each feature map are weighted and fused according to the dynamic weight coefficients obtained in the above steps. That is, the upsampled feature maps are weighted and summed according to the dynamic weight coefficients. The specific calculation process can be expressed by formula (3):

[0048] (3);

[0049] F fusion The fused feature map obtained by weighted fusion. Upsample This is for upsampling processing.

[0050] In a specific embodiment, the control terminal also establishes a communication connection with the microscopic magnification module and the linked ring supplementary light source to realize the transmission of data information; before extracting the corresponding fused feature map from the initial image acquired by the visual camera according to the preset image feature extraction model, the method further includes: generating a corresponding supplementary light control command according to the supplementary light parameters set in the start command and sending it to the linked ring supplementary light source to control the linked ring supplementary light source to light up the light source; generating a corresponding magnification adjustment command according to the magnification set in the start command and sending it to the microscopic magnification module to control the microscopic magnification module to adjust the objective lens magnification; and sending a zero adjustment command to the three-pin three-axis displacement stage to control the three-pin three-axis displacement stage to adjust the metasurface lens to the starting position.

[0051] To improve the image quality acquired by the vision camera, a supplementary lighting control command can be generated based on the supplementary lighting parameters set in the startup command before image acquisition. This command is then sent to the linked ring supplementary lighting source to control its illumination. The supplementary lighting parameters include wavelength (selectable in three bands: 450nm / 520nm / 650nm), single-channel LED power (adjustable from 0-0.5W), ring-zone brightness (including independent settings for the brightness of the four quadrants), and illumination angle (adjustable from 0° to 60°).

[0052] Furthermore, a magnification adjustment command can be generated based on the magnification set in the startup command, and the magnification adjustment command can be sent to the microscopic magnification module to adjust the objective lens magnification, thereby achieving focusing and hardware calibration; the magnification can be set to 10×, 20× or 50×.

[0053] After completing the objective lens magnification adjustment, a zero-return adjustment command is sent to the three-axis displacement stage. The three-axis displacement stage is controlled by the zero-return adjustment command to adjust the metasurface lens to the starting position in order to perform the zero-return operation.

[0054] S120. Perform edge parsing on the fused feature map according to the preset edge attention parsing model to obtain the corresponding edge attention weight map.

[0055] Considering the asymmetric microstructure of metasurface lenses, such as anisotropic nanopores leading to uneven edge reflection, a microstructure edge enhancement processing step is designed to effectively distinguish edges from background noise. Specifically, the fused feature map is first subjected to edge parsing based on an edge attention parsing model, thereby obtaining an edge attention weight map corresponding to the fused feature map.

[0056] In a specific embodiment, step S120 includes the following sub-steps: capturing the edge contour of the fused feature map according to the two-directional gradient operator in the edge attention parsing model to obtain the corresponding edge response map; and performing dynamic edge weighting enhancement on the edge response map and the fused feature map according to the convolution kernel in the edge attention parsing model to obtain the corresponding edge attention weight map.

[0057] Firstly, edge contours are directly captured through gradient-guided edge response generation, which is more accurate than traditional convolution-based edge detection, especially for low-contrast microstructure edges (signal-to-noise ratio <15dB). The edge attention parsing model is equipped with a two-directional gradient operator, which directly captures edge contours to highlight low-contrast edges, such as the boundaries of dark areas in microstructures, and captures the corresponding edge response map. The specific process of edge contour capture can be represented by formula (4):

[0058] (4);

[0059] E This is the obtained edge response map, where ▽ x and ▽ y These represent the gradient operators in the horizontal and vertical directions, respectively.

[0060] Based on the obtained edge response map, the edge region features are further enhanced through a dynamic edge enhancement mechanism to obtain an edge attention weight map. M e By dynamically weighting and enhancing the edge response map and fused feature map using the convolutional kernel in the edge attention parsing model, the corresponding edge attention weight map can be obtained.

[0061] In a specific embodiment, the step of dynamically enhancing the edge response map and the fused feature map according to the convolution kernel in the edge attention parsing model to obtain the corresponding edge attention weight map includes: performing a convolution operation on the fused feature map according to the convolution kernel to obtain a first convolution operation result; compressing the first convolution operation result according to the activation function in the edge attention parsing model to obtain the corresponding global semantic weight map; performing a convolution operation on the edge response map according to the convolution kernel to obtain the corresponding second convolution operation result; and superimposing the second convolution operation result with the value corresponding to the pixel at the same position in the global semantic weight map and then compressing it through the activation function to obtain the corresponding edge attention weight map.

[0062] The first convolution operation result is obtained by performing convolution operation on the fused feature map according to the convolution kernel. The first convolution operation result is then compressed according to the activation function to obtain the global semantic weight map. The second convolution operation result is obtained by performing convolution operation on the edge response map through the convolution kernel to solve the problem of fracture and noise at the edge of the metasurface. The convolution kernel can be expressed by formula (5):

[0063] (5);

[0064] The result of the second convolution operation is obtained, along with two values ​​corresponding to the same pixel in the global semantic weight map. These two values ​​are then superimposed, and the superimposed values ​​corresponding to each pixel are compressed to [0,1] using an activation function, thus obtaining the edge attention weight map. By weighting and enhancing the edge region features, the edge blurring problem caused by asymmetric reflection of the metasurface is solved. The specific process of obtaining the edge attention weight map can be expressed by formula (6):

[0065] (6);

[0066] Where σ() is equivalent to Sigmoid (), Conv () means performing convolution operations using a convolution kernel; M e This is the resulting edge attention weight map.

[0067] S130. The fused feature map is weighted pixel by pixel according to the edge attention weight map to obtain the corresponding edge enhancement feature map.

[0068] Using edge attention weight map M e For fused feature maps F fusion Pixel-by-pixel weighting is performed to enhance edge features. The specific processing procedure is shown in formula (7):

[0069] (7);

[0070] in, F e This represents the constructed edge enhancement feature map. This indicates pixel-by-pixel multiplication. F fusion Represents the fused feature map. M e This represents the edge attention weight map.

[0071] S140. Input the fused feature map into a preset edge prediction model to obtain the corresponding edge prediction information.

[0072] The obtained fused feature map is input into a pre-set edge prediction model to obtain edge prediction information. The edge prediction model is a neural network model built based on artificial intelligence, comprising an input layer, an intermediate layer, and an output layer. The input layer is used to input the fused feature map, the intermediate layer performs correlation analysis on the feature information input from the input layer, and transmits the output results to the output layer. The value output by the output layer is the edge prediction information, which includes the coordinate values ​​of each edge point in the predicted bounding box.

[0073] To improve the accuracy of the edge prediction model, training data containing edge box markers needs to be input into the edge prediction model for iterative training before using it. The training process is to optimize and adjust the model parameters in the edge prediction model. To adapt to the characteristic that the edge fitting quality of metasurface lenses determines the centering accuracy, this application constructs an edge loss enhancement intersection over union (ELEIoU) function during model training. This loss function can be specifically expressed by formula (8):

[0074] (8);

[0075] in, L ELEIoU This represents the edge loss enhancement loss function. AR This represents the aspect ratio loss. EL This represents the edge fitting residual loss (Edge Loss). Aspect Ratio Loss. AR and edge fitting residual loss EL The formulas (9) and (10) are used to express the following respectively:

[0076] (9);

[0077] (10);

[0078] β =0.2, which represents the weighting coefficient for aspect ratio loss; γ =0.3, representing the weighting coefficient of the marginal fitting residual loss; b Indicates the predicted bounding box. b gt Represents the true bounding box (the edge box label in the training data). Used to measure the overlap between the predicted bounding box and the ground truth bounding box; ρTo predict the center distance between the bounding box and the ground truth bounding box, c The minimum diagonal length of the bounding box rectangle, with a penalty for center offset; w / h To predict the aspect ratio of the bounding box, w gt / h gt This represents the actual aspect ratio of the frame. d k Indicates the first k Distance from each edge point to the center of the fitted circle N The number of edge points. This represents the average distance from N edge points to the center of the fitted circle. During iterative training of the model, the edge fitting residual loss is introduced, incorporating the residuals from the edge points to the fitted circle into the loss constraint. This makes the entire intelligent model more focused on edge accuracy optimization during training, and the multi-dimensional constraints improve the localization stability loss value. L ELEIoU (Edge loss enhancement loss function), which optimizes and adjusts the model parameters in the edge prediction model based on gradient descent and backpropagation.

[0079] S150. Fit the edge prediction information and the edge enhancement feature map according to the preset fitting rules to obtain the corresponding center coordinates.

[0080] Traditional object detection algorithms only output bounding box coordinates, with center localization relying on simple geometric fitting. To meet the micrometer-level accuracy requirements of metasurface lenses and mitigate optical distortion interference, this example constructs a complete high-precision sub-pixel-level center localization algorithm chain. In the specific implementation, edge prediction information and edge enhancement feature maps can be fitted according to fitting rules to obtain the corresponding center coordinates.

[0081] In a specific embodiment, step S150 includes the following sub-steps: filtering the edge prediction information and the edge points corresponding to the edge enhancement feature map according to the random sampling consensus algorithm in the fitting rule to obtain a corresponding set of effective edge points; assigning weights to each effective edge point in the set of effective edge points according to the allocation function in the fitting rule to obtain the weight value corresponding to each effective edge point; performing weighted least squares fitting on the set of effective edge points according to the objective function in the fitting rule and the weight value corresponding to each effective edge point to obtain a corresponding edge fitting curve; and extracting the center coordinates of the lens and the center coordinates of the washer from each edge fitting curve as the corresponding center coordinates.

[0082] First, the edge prediction information and the edge enhancement feature map are filtered according to the random sampling consensus algorithm in the fitting rules to obtain the corresponding effective edge point set. Then, the RANSAC (Random Sample Consensus) algorithm is introduced to filter the robust edge lines in the edge prediction information and edge enhancement feature map, eliminating false detection points caused by microstructure reflection noise, thus obtaining the effective edge point set containing effective edge points. It is ensured that the probability of the filtered effective edge points being located within the true bounding box (the edge box label in the training data) is ≥95%. The resulting effective edge points are then combined into the effective edge point set {( x k ,y k )} N k=1 .

[0083] Further, the effective edge points in the effective edge point set are weighted according to the allocation function to obtain the weight value corresponding to each effective edge point. Then, each effective edge point is assigned a weight value. The allocation function can be expressed by formula (11):

[0084] (11);

[0085] in, w k The weight values ​​are dynamically assigned based on the distance from the effective edge points to the initial circle. By assigning weight values, points closer to the true edge contribute more, thereby reducing the fitting error. d k Let α be the distance from the kth valid edge point to the initial fitted circle, where α = 0.5.

[0086] Further, based on the objective function in the fitting rules and the weight values ​​corresponding to each effective edge point, weighted least squares fitting is performed on the effective edge points in the effective edge point set to obtain the corresponding edge fitting curve. The objective function can be expressed by formula (12):

[0087] (12);

[0088] The Newton-Raphson algorithm is used to iteratively update a, b, and r until convergence. The resulting three parameters a, b, and r can be used to construct the corresponding edge fitting curve. Since this embodiment requires determining the coordinates of the lens center and the washer center, it is necessary to fit the effective pixels corresponding to the lens to construct the lens edge fitting curve, and to fit the effective pixels corresponding to the annular washer to construct the annular washer edge fitting curve. The above-mentioned use of weighted least squares to fit the center and radius can effectively handle noise and different confidence levels in the data points.

[0089] Based on the obtained edge fitting curves, the lens center coordinates of the lens edge fitting curve and the washer center coordinates of the annular washer edge fitting curve are extracted as the corresponding center coordinates. Furthermore, considering the influence of optical distortion, this example integrates optical distortion correction, using the camera intrinsic distortion correction matrix... D =[ k 1 , k 2 , p 1 , p 2 , k 3 By integrating the center-of-circle calculation process, systematic errors caused by microscopic optical path distortion are eliminated, ultimately achieving sub-pixel-level positioning. A schematic diagram of the visual inspection results in this implementation is shown below. Figure 4 As shown, it includes the identified annular washer and metasurface lens, as well as the coordinates of two center points; where the red point is the center coordinate of the metasurface lens and the green point is the center coordinate of the annular washer.

[0090] S160. Determine whether the center coordinates of the circle meet the preset deviation judgment rules to obtain a judgment result of whether it is qualified.

[0091] The system then checks whether the center coordinates meet the pre-set deviation judgment rules. If they do, a qualified result is obtained; otherwise, an unqualified result is obtained.

[0092] In a specific embodiment, step S160 includes the following sub-steps: calculating the geometric distance between the center coordinates of the lens and the center coordinates of the washer in the center coordinates; determining whether the geometric distance is within the distance range set in the deviation judgment rule, so as to obtain a judgment result of whether it is qualified.

[0093] The geometric distance between the lens center coordinates and the washer center coordinates can be calculated using the aforementioned coordinate system. Specifically, the lateral difference between the two center coordinates and the horizontal axis, and the lateral difference between the two center coordinates and the vertical axis can be calculated. The square root of the sum of the squares of the lateral and vertical differences gives the geometric distance, which reflects the difference between the two center coordinates. It is then determined whether this geometric distance falls within the distance range specified in the deviation judgment rules. If it does, the judgment is considered satisfactory; otherwise, it is considered unsatisfactory.

[0094] S170. If the judgment result is unqualified, generate a corresponding adjustment command based on the center coordinates and send it to the three-pin three-axis displacement stage to control the three-pin three-axis displacement stage to adjust the center position of the metasurface lens and return to execute the step of extracting the corresponding fusion feature map from the initial image acquired by the vision camera according to the preset image feature extraction model.

[0095] If the judgment result is unqualified, it indicates that the difference between the current lens center coordinates and the washer center coordinates is too large. At this time, an adjustment command can be generated according to the deviation angle and deviation distance corresponding to the center coordinates. The adjustment command can control the three-axis displacement stage of the three-pin to drive the metasurface lens to move, so as to adjust the center position of the metasurface lens. After the adjustment is completed, the process can return to step S110 and re-acquire the initial image for analysis, repeating steps S110-S160.

[0096] S180. If the judgment result is qualified, issue the corresponding qualified prompt message.

[0097] If the judgment result is qualified, a qualified prompt message can be issued at this time. Afterwards, the vacuum adsorption is released, and the robotic arm removes the centered metasurface lens for subsequent processing.

[0098] The automatic centering control method for metasurface lenses disclosed in the above embodiments includes: acquiring an initial image from a vision camera and extracting a fused feature map; performing edge analysis on the fused feature map to obtain an edge attention weight map; then performing pixel-by-pixel weighting on the fused feature map to obtain an edge enhancement feature map; performing edge prediction on the fused feature map and fitting it with the edge enhancement feature map to obtain the center coordinates; determining whether the center coordinates are qualified; if qualified, issuing a qualified prompt message; if unqualified, generating an adjustment command and sending it to a three-axis displacement stage, and repeating the above steps of obtaining the center coordinates and centering. This method, through image feature extraction and edge analysis, can quickly and accurately extract the edge features of the metasurface lens and the annular washer and calculate the center deviation, providing accurate data support for subsequent centering operations, thereby significantly improving the efficiency and accuracy of centering control of the metasurface lens.

[0099] This invention also provides an automatic centering control device for metasurface lenses, such as... Figure 2 and Figure 5As shown, the device includes a control terminal 10, a vision camera 11, a three-pin three-axis displacement stage 12, a linked ring supplementary light source 15, a microscopic magnification module 13, and a hollow gasket adsorption stage 14. The three-pin structure 121 on the three-pin three-axis displacement stage 12 uses vacuum adsorption to assist in fixing the metasurface lens 122. The three-pin three-axis displacement stage 12 is fixedly mounted on a mechanical support stage 16. The control terminal 10 establishes communication connections with the vision camera 11 and the three-pin three-axis displacement stage 12 to achieve data transmission. The vision camera 11, the microscopic magnification module 13, the hollow gasket adsorption stage 14, and the three-pin three-axis displacement stage 12 are arranged sequentially along the same optical axis, and the linked ring supplementary light source 15 is arranged around the periphery of the microscopic magnification module. The control terminal 10 is used to execute any embodiment of the aforementioned automatic centering control method for metasurface lenses. Specifically, please refer to... Figure 2 , Figure 2 This is a schematic diagram illustrating an application scenario of the automatic centering control method for metasurface lenses provided in this embodiment of the invention.

[0100] In a more specific embodiment, the three-pin structure consists of three tungsten carbide pins arranged in an equilateral triangle, as shown in the specific structure below. Figure 6 and Figure 7 As shown. The three-pin triaxial displacement stage includes a three-pin structure, wherein the three-pin structure consists of three tungsten carbide pins arranged in an equilateral triangle. The three-pin structure is a modular design, and the pin spacing can be adjusted according to actual conditions. In this example, the pin spacing is set to 2mm, the top spherical radius R=0.005mm, and the surface roughness Ra≤0.02μm. The metasurface lens is fixed by vacuum adsorption, and the adsorption force is adjustable from 0.5-2N. The detachable three-pin structure is shown below. Figure 6 As shown, the specific structure of the ejector head 123 is as follows: Figure 7 As shown.

[0101] Specifically, the mechanical support platform is made of granite with a flatness of ≤0.01mm / m and a load-bearing capacity of over 100kg. A three-point horizontal adjustment mechanism allows for precise adjustment of the horizontality to ±0.02mm / m, effectively suppressing external vibration interference, with a vibration attenuation rate >90%@10-200Hz. The three-axis displacement stage consists of X-axis, Y-axis, and Z-axis displacement mechanisms connected in series. The X, Y, and Z axes utilize crossed roller guides and piezoelectric ceramic drives, with a stroke ≥20mm, positioning accuracy ≤±0.1μm, and repeatability ≤±0.05μm. The three-ejector structure is modularly connected to the three-axis displacement stage, allowing for easy installation and replacement of ejector pins of different specifications.

[0102] The hollow gasket adsorption platform has a hollow design in the middle and three adsorption holes for vacuum adsorption of the annular gasket to prevent movement during the alignment process.

[0103] The aforementioned microscopic magnification module employs an infinity-corrected optical system. The objective lens magnification can be switched via an electric turntable to 0.5×, 0.6×, 0.7×, 0.8×, ..., 20×, with a numerical aperture NA of 0.25 / 0.45 / 0.85. Combined with a 2× eyepiece, it achieves continuous magnification from 0.5× to 20×, meeting the imaging requirements of metasurface lenses of different sizes.

[0104] The linked ring-shaped supplementary lighting source consists of a ring-shaped lamp holder, 16 independently controlled LED beads with selectable wavelengths of 450nm / 520nm / 650nm, and a driving circuit. The lamp holder has an inner diameter of 30mm, an outer diameter of 50mm, and a coaxiality of ≤0.1mm with the microscopic optical path. Each LED has a power of 0.5W, with brightness continuously adjustable from 0-100%. It achieves color temperature adjustment from 1000-10000K through PWM dimming, supports independent control of four quadrants of ring-shaped zoned lighting, and can dynamically optimize the lighting angle from 0° to 60° based on the orientation of the metasurface microstructure.

[0105] The vision camera is a 20-megapixel high-definition area array CCD (charge-coupled device) industrial camera with a pixel size of 3.2μm×3.2μm, a frame rate of ≥30fps@full resolution, equipped with a global shutter to eliminate motion blur, a dynamic range of ≥70dB, and supports dual interface data transmission of USB 3.0 (Universal Serial Bus 3.0) and GigE (Gigabit Ethernet).

[0106] The control terminal applies the aforementioned automatic centering control method, which is a visual detection algorithm for sub-pixel-level edge extraction and center fitting of metasurface lenses. It processes images captured by a vision camera after dynamic illumination optimization and magnification to identify and detect the precise center position of the matching washer and the center position of the metasurface lens structure. The acquired initial image is shown below. Figure 3 As shown.

[0107] like Figure 8As shown, the control terminal 10 is specifically configured with the following units: a fusion feature map acquisition unit 110, used to receive a start command and extract the corresponding fusion feature map from the initial image acquired by the visual camera according to a preset image feature extraction model; an attention weight map acquisition unit 120, used to perform edge parsing on the fusion feature map according to a preset edge attention parsing model to obtain the corresponding edge attention weight map; an edge enhancement feature map acquisition unit 130, used to perform pixel-by-pixel weighting on the fusion feature map according to the edge attention weight map to obtain the corresponding edge enhancement feature map; an edge prediction information acquisition unit 140, used to input the fusion feature map into a preset edge prediction model to obtain the corresponding edge prediction information; and a center coordinate acquisition unit 150, used to obtain the center coordinates according to a preset edge prediction model. The fitting rules are used to fit the edge prediction information and the edge enhancement feature map to obtain the corresponding center coordinates; the judgment result acquisition unit 160 is used to judge whether the center coordinates meet the preset deviation judgment rules to obtain a judgment result of whether it is qualified; the adjustment instruction sending unit 170 is used to generate a corresponding adjustment instruction based on the center coordinates and send it to the three-pin three-axis displacement stage if the judgment result is unqualified, so as to control the three-pin three-axis displacement stage to adjust the center position of the metasurface lens and return to execute the step of extracting the corresponding fusion feature map from the initial image acquired by the vision camera according to the preset image feature extraction model; the prompt information issuing unit 180 is used to issue a corresponding qualified prompt information if the judgment result is qualified.

[0108] The automatic centering control device for metasurface lenses provided in this embodiment of the invention applies the aforementioned automatic centering control method for metasurface lenses. It acquires an initial image from a vision camera and extracts a fused feature map. Edge analysis is performed on the fused feature map to obtain an edge attention weight map, and then pixel-by-pixel weighting is applied to the fused feature map to obtain an edge enhancement feature map. Edge prediction is performed on the fused feature map, and the center coordinates are obtained by fitting the edge enhancement feature map. The center coordinates are then judged to be qualified; if qualified, a qualified prompt message is issued; if unqualified, an adjustment command is generated and sent to the three-axis displacement stage, and the above steps of obtaining the center coordinates and centering are repeated. This method, through image feature extraction and edge analysis, can quickly and accurately extract the edge features of the metasurface lens and the annular washer and calculate the center deviation, providing accurate data support for subsequent centering operations, thereby significantly improving the efficiency and accuracy of centering control of the metasurface lens.

[0109] This invention provides an automatic centering control device for metasurface lenses. The device uses an intelligent linkage control module to control a three-pin three-axis displacement stage module, a hollow washer adsorption stage module, a high-resolution microscopic magnification optical path module, and a linkage ring supplementary light source module. Through an image acquisition module and a vision algorithm module, the device enables rapid and accurate alignment of the reference component and the target element during the assembly of the metasurface lens. Compared with existing technologies, the advantages of this invention are as follows: 1. It achieves end-to-end integration and automation of the system, including the collaborative work of hardware components and the smooth integration of software algorithms, ensuring the efficient operation of the entire automatic alignment process and realizing efficient, fast, and accurate wafer-level index detection of metasurface lenses; 2. It designs and optimizes the linkage ring supplementary light source module, controls the generation of ring surface light, ensures the stability of light source properties, provides multi-band, zoned, and angle adjustable illumination, and provides consistent and reliable light source output in conjunction with the edge enhancement algorithm; 3. It designs a three-pin three-axis displacement stage suitable for metasurface lenses. The three specially made tungsten carbide pins are distributed in an equilateral triangle with a spherical top. With the assistance of vacuum adsorption, the metasurface lens can be quickly and accurately guided to move; 4. It proposes a sub-pixel level edge extraction and circle center fitting visual detection algorithm for metasurface lenses, combined with the nanometer-level motion accuracy of the piezoelectric driven displacement stage, to meet the high-precision assembly requirements of metasurface lenses.

[0110] The above-described automatic centering control method for metasurface lenses can be implemented as a computer program, which can be used in applications such as... Figure 9 It runs on the computer device shown.

[0111] Please see Figure 9 , Figure 9 This is a schematic block diagram of a computer device provided in an embodiment of the present invention. The computer device can be a control terminal for executing an automatic centering control method for metasurface lenses to perform automatic centering control on the metasurface lenses.

[0112] See Figure 9 The computer device 500 includes a processor 502, a memory, and a communication interface 505 connected via a communication bus 501. The memory may include a storage medium 503 and internal memory 504.

[0113] The storage medium 503 may store an operating system 5031 and a computer program 5032. When the computer program 5032 is executed, it causes the processor 502 to execute an automatic centering control method for metasurface lenses. The storage medium 503 may be a volatile storage medium or a non-volatile storage medium.

[0114] The processor 502 provides computing and control capabilities to support the operation of the entire computer device 500.

[0115] The internal memory 504 provides an environment for the operation of the computer program 5032 in the storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute an automatic centering control method for metasurface lenses.

[0116] This communication interface 505 is used for network communication, such as providing data transmission. Those skilled in the art will understand that... Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device 500 to which the present invention is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0117] The processor 502 is used to run the computer program 5032 stored in the memory to implement the corresponding functions in the above-described automatic centering control method for metasurface lenses.

[0118] Those skilled in the art will understand that Figure 9 The embodiments of the computer device shown do not constitute a limitation on the specific configuration of the computer device. In other embodiments, the computer device may include more or fewer components than illustrated, or combine certain components, or have different component arrangements. For example, in some embodiments, the computer device may include only memory and a processor. In such embodiments, the structure and function of the memory and processor are different from those shown. Figure 9 The embodiments shown are consistent and will not be repeated here.

[0119] It should be understood that, in this embodiment of the invention, the processor 502 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0120] In another embodiment of the invention, a computer-readable storage medium is provided. This computer-readable storage medium may be volatile or non-volatile. The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps included in the above-described automatic centering control method for metasurface lenses.

[0121] Those skilled in the art will readily understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention.

[0122] In the embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Units with the same function may be grouped into one unit. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, or it may be an electrical, mechanical, or other form of connection.

[0123] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of the present invention, depending on actual needs.

[0124] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0125] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a computer-readable storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned computer-readable storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), magnetic disks, or optical disks.

[0126] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An automatic centering control method for metasurface lenses, characterized in that, The method is applied in a control terminal, which establishes communication connections with a vision camera and a three-pin three-axis displacement stage to transmit data. The three-pin structure on the three-pin three-axis displacement stage uses vacuum adsorption to fix the metasurface lens. The vision camera, microscopic magnification module, hollow gasket adsorption stage, and the three-pin three-axis displacement stage are arranged sequentially along the same optical axis. A linked annular supplementary light source is arranged around the periphery of the microscopic magnification module. The method includes: Upon receiving the start command, the corresponding fusion feature map is extracted from the initial image acquired by the visual camera according to the preset image feature extraction model; The fused feature map is parsed according to a pre-set edge attention parsing model to obtain the corresponding edge attention weight map; The fused feature map is weighted pixel-by-pixel based on the edge attention weight map to obtain the corresponding edge enhancement feature map; The fused feature map is input into a preset edge prediction model to obtain the corresponding edge prediction information; The edge prediction information and the edge enhancement feature map are fitted according to a preset fitting rule to obtain the corresponding center coordinates of the circle. The coordinates of the center of the circle are judged to determine whether they meet the preset deviation judgment rules, so as to obtain the judgment result of whether they are qualified; If the judgment result is unqualified, a corresponding adjustment command is generated based on the center coordinates and sent to the three-pin three-axis displacement stage to control the three-pin three-axis displacement stage to adjust the center position of the metasurface lens and return to execute the step of extracting the corresponding fusion feature map from the initial image acquired by the vision camera according to the preset image feature extraction model; If the judgment result is qualified, a corresponding qualified prompt message will be issued.

2. The automatic centering control method for metasurface lenses according to claim 1, characterized in that, The step of extracting the corresponding fused feature map from the initial image acquired by the visual camera according to the preset image feature extraction model includes: The initial image is subjected to continuous convolution processing based on multiple convolutional layers configured in the image feature extraction model, and each convolutional layer outputs a feature map respectively. The pooling compression layer in the image feature extraction model is used to perform pooling compression processing on each feature map to obtain the feature saliency score corresponding to each feature map. Weights are assigned according to the weight allocation function in the image feature extraction model and the feature saliency score to obtain the dynamic weight coefficients corresponding to each feature map; Each feature map is upsampled according to the preset resolution parameters in the image feature extraction model to obtain an upsampled feature map corresponding to the resolution parameters; The upsampled feature maps of each feature map are weighted and fused according to the dynamic weight coefficients to obtain the corresponding fused feature map.

3. The automatic centering control method for metasurface lenses according to claim 1, characterized in that, The step of performing edge parsing on the fused feature map according to a preset edge attention parsing model to obtain the corresponding edge attention weight map includes: The edge contour is captured by the two-directional gradient operator in the edge attention parsing model to obtain the corresponding edge response map; Dynamic edge weighting enhancement is performed on the edge response map and the fused feature map based on the convolution kernel in the edge attention parsing model to obtain the corresponding edge attention weight map.

4. The automatic centering control method for metasurface lenses according to claim 3, characterized in that, The step of dynamically enhancing the edge response map and the fused feature map based on the convolutional kernels in the edge attention parsing model to obtain the corresponding edge attention weight map includes: The fused feature map is convolved using the convolution kernel to obtain the first convolution result. The first convolution operation result is compressed according to the activation function in the edge attention parsing model to obtain the corresponding global semantic weight map; The edge response map is convolved according to the convolution kernel to obtain the corresponding second convolution result; The result of the second convolution operation is superimposed with the value of the pixel at the same position in the global semantic weight map, and then compressed through the activation function to obtain the corresponding edge attention weight map.

5. The automatic centering control method for metasurface lenses according to claim 1, characterized in that, The step of fitting the edge prediction information and the edge enhancement feature map according to a preset fitting rule to obtain the corresponding circle center coordinates includes: The edge prediction information and the edge points corresponding to the edge enhancement feature map are filtered according to the random sampling consensus algorithm in the fitting rules to obtain the corresponding set of valid edge points; According to the allocation function in the fitting rule, the weights of each valid edge point in the set of valid edge points are allocated to obtain the weight value corresponding to each valid edge point. Based on the objective function in the fitting rule and the weight values ​​corresponding to each effective edge point, a weighted least squares fitting is performed on the set of effective edge points to obtain the corresponding edge fitting curve. The center coordinates of the lens and the center coordinates of the washer are extracted from the edge fitting curves respectively as the corresponding center coordinates.

6. The automatic centering control method for metasurface lenses according to claim 5, characterized in that, The step of judging whether the coordinates of the center of the circle meet the preset deviation judgment rules to obtain a judgment result of whether it is qualified includes: Calculate the geometric distance between the center coordinates of the lens and the center coordinates of the washer in the aforementioned center coordinate system; Determine whether the geometric distance is within the distance range set in the deviation judgment rule to obtain a judgment result of whether it is qualified.

7. The automatic centering control method for metasurface lenses according to claim 6, characterized in that, The control terminal establishes a communication connection with the microscopic magnification module and the linked ring light source to realize the transmission of data information; before extracting the corresponding fused feature map from the initial image acquired by the visual camera according to the preset image feature extraction model, the method further includes: The corresponding supplementary lighting control command is generated according to the supplementary lighting parameters set in the start command and sent to the linked ring supplementary lighting source to control the linked ring supplementary lighting source to light up the light source. The corresponding magnification adjustment command is generated according to the magnification set in the start command and sent to the micro-magnification module to control the micro-magnification module to adjust the objective lens magnification. A zero-adjustment command is sent to the three-pin three-axis displacement stage to control the three-pin three-axis displacement stage to adjust the metasurface lens to the starting position.

8. An automatic centering control device for metasurface lenses, characterized in that, The device includes a control terminal, a vision camera, a three-pin three-axis displacement stage, a linked ring supplementary light source, a microscopic magnification module, and a hollow gasket adsorption stage. The three-pin structure on the three-pin three-axis displacement stage is fixed to the metasurface lens by vacuum adsorption. The three-pin three-axis displacement stage is fixedly mounted on a mechanical support platform. The control terminal establishes communication connections with the vision camera and the three-pin three-axis displacement stage to realize data information transmission. The vision camera, the microscopic magnification module, the hollow gasket adsorption stage, and the three-pin three-axis displacement stage are arranged sequentially along the same optical axis, and the linked ring supplementary light source is arranged around the periphery of the microscopic magnification module. The control terminal is used to execute the automatic centering control method for metasurface lenses as described in any one of claims 1-7.

9. The automatic centering control device for metasurface lenses according to claim 8, characterized in that, The device also includes a unit configured in the control terminal: The fusion feature map acquisition unit is used to receive a start command and extract the corresponding fusion feature map from the initial image acquired by the visual camera according to a preset image feature extraction model. The attention weight map acquisition unit is used to perform edge parsing on the fused feature map according to the preset edge attention parsing model to obtain the corresponding edge attention weight map; The edge enhancement feature map acquisition unit is used to perform pixel-by-pixel weighting on the fused feature map according to the edge attention weight map to obtain the corresponding edge enhancement feature map; An edge prediction information acquisition unit is used to input the fused feature map into a preset edge prediction model to obtain the corresponding edge prediction information; The center coordinate acquisition unit is used to fit the edge prediction information and the edge enhancement feature map according to the preset fitting rules to obtain the corresponding center coordinates. The judgment result acquisition unit is used to judge whether the center coordinates of the circle meet the preset deviation judgment rules, so as to obtain the judgment result of whether it is qualified; An adjustment instruction sending unit is used to generate a corresponding adjustment instruction based on the center coordinates and send it to the three-pin three-axis displacement stage if the judgment result is unqualified, so as to control the three-pin three-axis displacement stage to adjust the center position of the metasurface lens and return to execute the extraction of the corresponding fusion feature map from the initial image acquired by the vision camera according to the preset image feature extraction model. The prompting unit is used to issue a corresponding qualified prompting message if the judgment result is qualified.

10. The automatic centering control device for metasurface lenses according to claim 8 or 9, characterized in that, The three-pin structure consists of three tungsten carbide pins arranged in an equilateral triangle.

Images