Automatic Emapping method for IC (integrated circuit) carrier plate based on AI (artificial intelligence)

Abstract

The invention provides an automatic Emapping method of an IC carrier plate based on AI, and relates to the field of data processing. According to the method, the geometric shape of the IC carrier plate is stabilized through partition vacuum adsorption and temperature closed-loop adjustment, an upper view image and a lower view image are synchronously collected under the condition that the physical state is controlled, and an effective anchor point set is formed through confidence coefficient interlocking screening in combination with a detection frame model and a character recognition model; constructing a unified cross-view affine coordinate mapping parameter based on the effective anchor points, and establishing a one-to-one binding relationship between the coordinate marks and the grid units; gridding grading positioning is carried out under structural consistency constraint, high-precision detection point theoretical coordinates are obtained, finally, a motion platform is controlled to complete electrical detection, conduction impedance data is associated with space information, and a reliable Emapping result is generated. By implementing the technical scheme provided by the invention, the accuracy of automatic Emapping of the IC carrier plate can be improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and more specifically to an AI-based method for automatic emapping of IC carrier boards. Background Technology

[0002] In IC substrate electrical testing and emapping applications, the test results need to establish a stable, one-to-one mapping relationship with the specific physical location on the substrate to support defect localization, process traceability, and yield analysis. However, in actual automated emapping processes, IC substrates typically have structural characteristics such as thinness, large area, and multiple material composite layers, making them prone to warping, local undulations, or non-uniform thermal deformation during vacuum fixing, temperature changes, or long-term testing. This causes dynamic changes in the substrate's geometry during the testing cycle. Furthermore, existing automated testing systems often use top-view and bottom-view cameras to collaboratively complete marker recognition and test point positioning. However, imaging from different perspectives is difficult to maintain stable coordinate consistency under the influence of optical distortion, installation errors, and operational disturbances. Cross-perspective coordinate systems often rely on one-time calibration or ideal plane assumptions, making it difficult to adapt to real-time changes in the substrate's shape. Based on this, test points are usually obtained through step-by-step positioning or cumulative offset methods. The spatial positioning error accumulates continuously during multiple coordinate transformations and local corrections, ultimately causing a deviation between the actual contact position of the electrical test probe and the theoretical test position. This makes it impossible to reliably map electrical test results such as continuity impedance to the true physical location of the IC substrate.

[0003] The aforementioned problems, when combined, severely reduce the accuracy of automatic emapping, thus necessitating an AI-based method for automatic emapping of IC carrier boards. Summary of the Invention

[0004] This application provides an AI-based method for automatic emapping of IC carrier boards, which facilitates the improvement of the accuracy of automatic emapping of IC carrier boards.

[0005] The first aspect of this application provides an AI-based automatic emapping method for IC carrier boards. The method includes: performing partitioned vacuum adsorption fixation on the IC carrier board using a temperature-controlled vacuum adsorption platform, while simultaneously implementing closed-loop temperature regulation to ensure the IC carrier board maintains geometric stability and controlled thermal deformation during automatic emapping; under the partitioned vacuum adsorption fixation and closed-loop temperature regulation state, controlling an upward-viewing camera and a downward-viewing camera to simultaneously acquire image information of the IC carrier board, and based on the image information, locating marked areas using a detection box model and resolving coordinate marked characters using a character recognition model, while simultaneously combining detection confidence and recognition confidence interlocking judgment to form a set of effective anchor points; based on the spatial correspondence of each reference mark in the set of effective anchor points under the upward-viewing camera and the downward-viewing camera, constructing a unified coordinate system for the upward-viewing camera and the downward-viewing camera. An affine transformation matrix of the coordinate system is established, and this affine transformation matrix is ​​used as the basic coordinate mapping parameter for automatic embedding of the IC carrier board. Under the constraint of the affine transformation matrix, coordinate mark recognition and coordinate mapping processing are performed on the coordinate marks on the IC carrier board through the basic coordinate mapping parameter to establish a one-to-one binding relationship between the coordinate marks and the grid cells. Based on the one-to-one binding relationship between the coordinate marks and the grid cells, grid-based hierarchical positioning is performed under the constraints of the affine transformation matrix and structural consistency, so that the detection points can sequentially complete grid-level positioning and fine-tuning-level precision positioning to obtain the theoretical coordinates of the detection points. The motion platform is controlled to reach the target position according to the theoretical coordinates of the detection points, and electrical characteristic detection is performed on the IC carrier board. After collecting the conduction impedance data, it is associated and recorded with the corresponding coordinate marks, grid cells and theoretical coordinates of the detection points to generate the embedding result.

[0006] A second aspect of this application provides an AI-based automatic emapping device for IC carriers. The device includes an acquisition module and a processing module. The acquisition module is used to perform partitioned vacuum adsorption fixation on the IC carrier using a temperature-controlled vacuum adsorption platform, and simultaneously perform closed-loop temperature regulation to ensure that the IC carrier maintains geometric stability and controlled thermal deformation during the automatic emapping process. The processing module is used to control an upward-viewing camera and a downward-viewing camera to simultaneously acquire image information of the IC carrier under partitioned vacuum adsorption fixation and closed-loop temperature regulation. Based on the image information, it locates marked areas using a detection box model and parses coordinate marked characters using a character recognition model. Simultaneously, it combines the interlocking judgment of detection confidence and recognition confidence to form a set of effective anchor points. The processing module is also used to construct a unified upward-viewing camera coordinate system based on the spatial correspondence of each reference mark in the set of effective anchor points under the upward-viewing camera and the downward-viewing camera. The processing module is further configured to: establish a one-to-one binding relationship between coordinate markers and grid cells; perform grid-level hierarchical positioning based on the one-to-one binding relationship between coordinate markers and grid cells; and perform electrical characteristic detection on the IC carrier board, and associate and record the collected conduction impedance data with the corresponding coordinate markers, grid cells, and theoretical coordinates of the detection points to generate an Emaporizing result. The processing module is also configured to: perform coordinate marker recognition and coordinate mapping processing on coordinate markers on the IC carrier board under the constraints of the affine transformation matrix and the coordinate mapping basic parameters, under the constraints of the affine transformation matrix; perform grid-level positioning and fine-tuning positioning on coordinate markers on the IC carrier board under the constraints of the affine transformation matrix and the coordinate mapping basic parameters, under the constraints of the coordinate transformation matrix and the coordinate mapping basic parameters, under the constraints of the coordinate transformation matrix and the coordinate mapping basic parameters, under the constraints of the coordinate transformation matrix and the coordinate mapping basic parameters, under the constraints of the coordinate transformation matrix and the coordinate mapping basic parameters, under the constraints of the coordinate transformation matrix and the coordinate mapping basic parameters, under the constraints of the coordinate transformation matrix and the coordinate mapping basic parameters, under the constraints of the coordinate transformation matrix and the coordinate mapping basic parameters, under the constraints of the coordinate transformation matrix and the constraints of the structural consistency, under the constraints of the affine transformation matrix and the constraints of the structural consistency, under the constraints of the affine transformation matrix and the constraints of the structural consistency, under the constraints of the structural consistency, under the constraints of the coordinate transformation matrix and the constraints of the structural consistency, under ...

[0007] A third aspect of this application provides an electronic device including a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, and both the user interface and the network interface are used to communicate with other devices. The processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method described above.

[0008] A fourth aspect of this application provides a non-transitory computer-readable storage medium storing instructions that, when executed, perform the method described above.

[0009] In summary, one or more technical solutions provided in this application have at least the following technical effects or advantages: By unifying physical stability control, visual perception consistency constraints, and spatial positioning logic into a single technical chain, the automatic emapping of IC carrier boards is transformed from "empirical positioning and result recording" to "deterministic spatial mapping under controlled physical boundaries." On the one hand, through the partitioned vacuum adsorption fixation and closed-loop temperature regulation of the temperature-controlled vacuum adsorption platform, the geometric changes and thermal deformation of the IC carrier board during the entire detection process are constrained within a controllable range. This fundamentally reduces the impact of carrier board warpage, local undulations, and thermal drift on subsequent visual recognition and positioning accuracy, providing a stable physical basis for automatic emapping. On the other hand, through the synchronous acquisition of data from the top-view and bottom-view cameras and the collaborative recognition of the detection frame model and the character recognition model, and by introducing a confidence interlocking determination mechanism for detection confidence and recognition confidence, the anchor points entering the coordinate mapping process are ensured to simultaneously meet the dual constraints of spatial positioning reliability and character semantic reliability. This significantly reduces the risk of damage to the overall coordinate system caused by misidentification or misuse of markers.

[0010] Based on an effective set of anchor points, an affine transformation matrix is ​​constructed to unify the coordinate systems of the top-view and bottom-view cameras. This transforms the cross-view coordinate relationship from a one-time empirical calibration to a verifiable mapping relationship established based on real-time recognition results, thereby enhancing the system's adaptability to installation errors, viewing angle differences, and changes in operating conditions. Furthermore, by binding coordinate markers to grid cells one-to-one, the original continuous spatial problem is transformed into a gridded spatial problem with clear semantic boundaries. This ensures that subsequent positioning and detection revolve around stable spatial semantic cells, preventing unconstrained error propagation across the entire board. Combined with gridded hierarchical positioning executed using the affine transformation matrix and structural consistency constraints, the detection point positioning process converges gradually from coarse to fine, ensuring positioning efficiency while effectively suppressing the impact of local deformation, imaging noise, or cumulative offset on the final theoretical coordinates of the detection points.

[0011] By controlling the motion platform to perform electrical characteristic testing based on the theoretical coordinates of the detection points, and linking the continuity impedance data with coordinate markers, grid cells, and the theoretical coordinates of the detection points, each electrical test result naturally possesses a clear, unique, and traceable spatial semantics. This solves the problem in existing automatic emapping systems where electrical test results are difficult to stably correspond to real physical locations. The overall solution not only improves the accuracy, consistency, and repeatability of emapping results but also significantly enhances the system's robustness and engineering usability under complex operating conditions. Attached Figure Description

[0012] Figure 1 A flowchart illustrating an AI-based automatic emapping method for IC carrier boards provided in this application embodiment; Figure 2A schematic diagram of a module for an AI-based automatic emapping device for IC carrier boards provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0013] Explanation of reference numerals in the attached figures: 21. Acquisition module; 22. Processing module; 31. Processor; 32. Communication bus; 33. User interface; 34. Network interface; 35. Memory. Detailed Implementation

[0014] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0015] In the description of the embodiments of this application, the words "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design that is described as "for example" or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design options. Rather, the use of the words "for example" or "for instance" is intended to present the relevant concepts in a specific manner.

[0016] In the description of the embodiments of this application, the term "multiple" means two or more. For example, multiple systems means two or more systems, and multiple screen terminals means two or more screen terminals. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0017] To address the aforementioned technical problems, this application provides an AI-based automatic emapping method for IC carrier boards, referring to... Figure 1 , Figure 1 This is a flowchart illustrating an AI-based automatic emapping method for IC carrier boards, provided in an embodiment of this application. The method is applied to a server and includes steps S110 to S160, as follows:

[0018] S110. The IC carrier is fixed by partitioned vacuum adsorption on a temperature-controlled vacuum adsorption platform, and the temperature is adjusted in a closed loop at the same time to keep the geometric shape of the IC carrier stable and the thermal deformation controlled during the automatic emapping process.

[0019] Specifically, the server, acting as the unified control hub of the automatic emapping system, establishes a two-way communication connection with the temperature-controlled vacuum adsorption platform and the motion platform to centrally complete command issuance, status acquisition, and coordinated scheduling control. The server can send zoned vacuum adsorption control commands and temperature setting commands to the temperature-controlled vacuum adsorption platform, and obtain vacuum pressure, vacuum flow, and temperature feedback in real time to ensure that the IC carrier board's fixed state and temperature control state meet stability requirements. At the same time, the server issues motion commands to the motion platform based on the theoretical coordinates of the detection points, and receives the motion platform's pose feedback and operating status in real time for performing position verification and precise positioning control. During the automatic emapping process, the server uniformly schedules the temperature-controlled vacuum adsorption platform and the motion platform based on the process status and feedback information, triggering motion and detection operations only when the physical conditions are stable, thereby achieving automatic control and coordinated linkage between the two.

[0020] A gridded adsorption surface is pre-processed on a temperature-controlled vacuum adsorption platform, ensuring that the gridded adsorption surface corresponds one-to-one with the outer dimensions, reference edges, and placement orientation of the IC carrier. Multiple grid units are formed within the gridded adsorption surface according to preset rows and columns. Each grid unit is connected to a vacuum pump, proportional valve, shut-off valve, pressure sensor, and flow sensor via an independent vacuum adsorption circuit, enabling each grid unit to independently perform suction, discharge, and measurement. Simultaneously, a positioning reference structure is set on the gridded adsorption surface, ensuring that the reference edge of the IC carrier fits snugly against the positioning reference structure after placement. This guarantees that the coordinate mapping parameters of the subsequent top-view and bottom-view cameras remain consistent with the spatial division of the grid units under the same mechanical reference for a long period. To avoid crosstalk caused by adsorption coupling between grid units, a one-way check valve and buffer volume are set in the vacuum adsorption circuit corresponding to each grid unit, preventing pressure fluctuations between adjacent grid units from propagating and spreading in a short time. This provides a controllable and independent execution object for subsequent dynamic balance adjustment.

[0021] After placing the IC carrier on the gridded adsorption surface, pre-suction is performed on the grid cells corresponding to the central region to form continuous adhesion and eliminate large-scale warping in the central region. Then, progressive suction is performed on the grid cells corresponding to the edge regions to gradually flatten the edge regions under the constraint of central adhesion and eliminate edge suspension. Finally, corner locking suction is performed on the grid cells corresponding to the corner regions to suppress corner torsion and local arching under the constraint of edge flattening. During the sequential opening of the vacuum adsorption circuit, the "instantaneous full opening" method is not used. Instead, a ramp-type opening control is used for each batch of grid cells to smoothly transition the vacuum pressure of the grid cells from the ambient pressure to the target vacuum pressure, thereby reducing the risk of stress concentration and permanent deformation caused by transient pressure difference pulling on the IC carrier. After the corner regions are suctioned, the vacuum adsorption circuits of the central region, edge region, and corner region are kept in a controlled holding state, so that the IC carrier enters the subsequent temperature closed-loop adjustment and stability determination stage under the same fixed boundary conditions.

[0022] In the fixed state of partitioned vacuum adsorption, vacuum pressure feedback and vacuum flow feedback are continuously collected for each grid cell. These two feedbacks are used together to determine whether the grid cell has local leakage, local suspension, or local over-adsorption. The vacuum flow feedback represents the suction compensation required to maintain the current vacuum pressure. If the vacuum flow feedback remains consistently high while the target vacuum pressure remains approximately constant, it indicates poor sealing or edge gaps in the grid cell. Therefore, it is necessary to adjust the vacuum adsorption intensity of this grid cell and its adjacent grid cells to achieve bonding redistribution. To suppress local warping or suspension caused by differences in adsorption intensity between adjacent grid cells, a dynamic balance adjustment strategy centered on "target vacuum pressure tracking and adjacent difference suppression" is constructed for each grid cell. This ensures that while the grid cell reaches the target vacuum pressure, the vacuum pressure difference with its adjacent grid cells is controlled within an allowable range, thereby stabilizing the bonding shape of the IC carrier in a continuous state without excessive local pulling. The dynamic balance adjustment can employ an adjacent difference penalty control target, where the control quantity of each grid cell is constrained by both its own error and adjacent differences. The control target can be expressed as follows:

[0023] in, This represents the dynamic balance adjustment target value of the i-th grid cell. The value changes in real time with the control process, and the smaller the value, the more it satisfies the "target tracking and adjacent consistency" requirement. This represents the vacuum pressure feedback of the i-th grid cell, which is collected in real time by the pressure sensor of that grid cell; The target vacuum pressure is preset by the process and can be adjusted according to the IC substrate type or temperature control status. This represents the adjacent difference penalty coefficient, used to balance the requirements of "rapidly reaching the target vacuum pressure" and "suppressing adjacent differences". The larger the value, the greater the emphasis on pressure consistency between adjacent grid cells; The set of adjacent grid cells of the i-th grid cell is determined according to the topological relationship of the adsorption surface of the gridded partition, including vertical, horizontal and vertical adsorption or further extended to diagonal adsorption. The principle of this formula is to incorporate the pressure error of a single grid cell and the pressure difference of adjacent grid cells into the same target, so that the control strategy can reduce its own error while automatically suppressing spatial discontinuities, thereby reducing local over-adsorption and local suspension and improving geometric stability.

[0024] While maintaining the fixed state of the partitioned vacuum adsorption, temperature closed-loop regulation is initiated. The temperature-controlled vacuum adsorption platform is divided into multiple temperature-controlled zones in a manner corresponding to the spatial distribution of grid cells. Each temperature-controlled zone has an independent heating or cooling execution channel and a temperature feedback acquisition point, thereby enabling spatial distribution regulation of the temperature in the IC carrier contact area. The temperature closed-loop regulation first drives the temperature-controlled vacuum adsorption platform to the target temperature window according to a preset temperature trajectory, and then enters a constant temperature hold state after reaching the target temperature window, so that temperature fluctuations are controlled and non-uniform thermal expansion of the IC carrier caused by temperature drift is avoided. When a temperature difference is detected between different temperature-controlled zones, instead of using a single average temperature as the sole control target, the temperature uniformity index is used as a constraint to make the temperature of each temperature-controlled zone tend to be consistent in the spatial dimension, thereby reducing non-uniform thermal deformation caused by local temperature differences. The temperature uniformity index can be characterized by the dispersion of the regional temperature, and can be expressed as the following formula:

[0025] in, The value indicates the temperature uniformity index; the smaller the value, the more consistent the temperature is across all temperature-controlled zones. M represents the number of temperature-controlled zones, which is determined by the temperature control zoning design of the temperature-controlled vacuum adsorption platform. This indicates the temperature feedback of the k-th temperature control zone, which is collected by the temperature sensor of that temperature control zone. This represents the average temperature feedback from all temperature-controlled zones, achieved through... The average is obtained; the principle of this formula is to quantify the degree of deviation of each temperature control zone from the average temperature in the form of standard deviation, thereby transforming "reducing temperature difference" into a calculable and thresholdable control constraint, so that the temperature closed-loop regulation not only tracks the target temperature window, but also actively suppresses the thermal deformation caused by spatial temperature difference.

[0026] After the partitioned vacuum adsorption fixation and temperature closed-loop regulation reach a stable state, stability judgments are performed on the vacuum stability of each grid unit and the temperature stability of each temperature control zone. This ensures that the automatic emapping process is triggered only when the physical boundary conditions are sufficiently stable, thereby avoiding the establishment of unreliable coordinate mapping parameters during vacuum pressure fluctuations or temperature drift. The vacuum stability judgment focuses on the fluctuation degree of vacuum pressure feedback and the compensation strength of vacuum flow feedback, enabling the identification of both pressure fluctuations and hidden leaks. The temperature stability judgment focuses on the fluctuation degree of temperature feedback and the temperature uniformity index, enabling the identification of both overall temperature drift and regional temperature differences. Vacuum stability and temperature stability can be characterized as window fluctuation indices and compared with preset thresholds to form trigger conditions. An exemplary stability judgment index can be expressed as follows:

[0027]

[0028] in, Indicates the i-th grid cell in the time window The vacuum pressure fluctuation index is obtained by taking the difference between the maximum and minimum pressure values ​​within the time window; This represents the vacuum pressure feedback of the i-th grid cell at time t; The sliding time window for stability determination is set by the system and can be adjusted according to the process cycle. This indicates that the k-th temperature control zone is within the time window. The temperature fluctuation index within the time window is obtained by taking the difference between the maximum and minimum temperature values ​​within that time window; This represents the temperature feedback of the k-th temperature control zone at time t; the principle of this formula is to use the range within the window to characterize the short-time fluctuation amplitude, thereby transforming "stability" into a determinable condition; when all grid cells... All are below the preset vacuum stability threshold and all temperature control zones All are below the preset temperature stability threshold, and the temperature uniformity index When the uniformity is less than the preset threshold, the output shows that the vacuum stability and temperature stability simultaneously meet the preset conditions. At the moment the judgment result is established, the automatic emapping process is triggered, so that the subsequent synchronous acquisition, effective anchor point set construction and coordinate mapping basic parameter application are all based on a stable and reproducible carrier state.

[0029] S120. Under the conditions of partitioned vacuum adsorption fixation and temperature closed-loop regulation, the upper-view camera and the lower-view camera are controlled to synchronously acquire image information of the IC carrier board. Based on the image information, the marked area is located by the detection frame model and the coordinate marked character is analyzed by the character recognition model. At the same time, the set of effective anchor points is formed by combining the interlocking judgment of the detection confidence and the recognition confidence.

[0030] Specifically, under the condition that the partitioned vacuum adsorption fixation and temperature closed-loop regulation have passed the vacuum stability and temperature stability judgment respectively and remain stable, a unified synchronization trigger signal is simultaneously sent to the upward-looking camera and the downward-looking camera, so that the two cameras complete the exposure, readout and timestamp writing with the same time reference. This avoids the inconsistency in the two images caused by micro-deformation of the carrier plate, micro-vibration of the platform or micro-fluctuation of illumination due to time drift. Among them, the synchronization trigger signal refers to the unified trigger pulse or trigger command issued by the same hardware trigger source or the same real-time controller; the time reference refers to the consistent definition of "the same moment" by the two cameras; and the timestamp refers to the identification field that records the corresponding moment of the acquisition. The upward-looking camera refers to the camera that acquires a larger global image from above the carrier plate, and the downward-looking camera refers to the camera that acquires a local high-resolution image from below the carrier plate or through a specific optical path. The image information refers to the combination data containing the pixel matrix and the exposure parameters, gain parameters, light source parameters, timestamp and operating condition identifier bound to the matrix, so that subsequent identification and confidence interlock judgment can be traced back to the same physical boundary conditions.

[0031] Brightness equalization, noise suppression, and geometric distortion correction are performed on the synchronously acquired image information to ensure that the coordinate markers exhibit more consistent contrast, edge sharpness, and geometric shape under different regions, lighting conditions, and viewing angles. Brightness equalization refers to adjusting the brightness distribution of the image to be more uniform or more suitable for the recognition model. Common methods include histogram-based contrast-limited adaptive equalization. Noise suppression reduces the interference of random noise or texture noise on character edges and marker boundaries. Common methods include bilateral filtering or nonlocal mean filtering. Geometric distortion correction refers to inversely mapping radial and tangential distortions according to camera calibration parameters to make the geometric shape of the marked area closer to the real plane projection. For example, the pixel mapping relationship of distortion correction can be expressed as:

[0032] in, Represents the normalized coordinates of pixels in the distorted image, used to correspond to the original acquired image; Represents the normalized coordinates of the pixels after distortion correction, used to obtain the corrected image; This represents the radial distortion coefficient, obtained from offline camera calibration. This represents the tangential distortion coefficient, obtained from offline camera calibration. This represents the normalized radial distance to the optical axis. The principle of this set of formulas is to use a polynomial to approximate the radial stretching and tangential offset caused by the lens, and to achieve backsampling by establishing a mapping relationship from "distortion coordinates to distortion coordinates", thereby resampling the original image into a geometrically more consistent corrected image. After preprocessing, the processed image information is input into the detection box model to locate the corresponding marked regions with coordinates, and the detection confidence of each marked region is output. The detection box model refers to the model that aims to detect targets and outputs the position and category probability of the target bounding box. The marked region refers to the smallest enclosing area covering the coordinate marker characters and their necessary background boundaries. The detection confidence refers to the degree of confidence of the detection box model in that "the marked region is indeed a coordinate marker". At the same time, each marked region, its detection confidence, bounding box coordinates, and the camera identifier to which it belongs are written into the candidate anchor point set. The candidate anchor point set refers to the set of marked region records that have not yet passed the confidence interlocking judgment but are considered to be potentially valid anchor points.

[0033] For each marked region in the candidate anchor point set, the character recognition model is invoked to parse the coordinate marker characters within the marked region and output the recognition confidence score. This ensures that each marked region not only has a location result of "where" but also a semantic result of "what". The character recognition model refers to a model that recognizes character sequences in an image and outputs the character sequences and their probabilities. The coordinate marker characters refer to the combination of character indices and numeric indices used to represent grid coordinates. The character index refers to letters or equivalent symbols used to represent row or vertical serial numbers, and the numeric index refers to numeric strings used to represent column or horizontal serial numbers. The unified semantic definition refers to fixing the mapping rules from character indices to vertical serial numbers and from numeric indices to horizontal serial numbers throughout the automatic emapping process, and fixing the allowed character sets, character length ranges, and character validity verification rules, thereby avoiding the same character being interpreted as having different meanings at different stages. The recognition confidence score refers to the degree of confidence of the character recognition model in the correctness of the output coordinate marker characters. It can be obtained from character-level probability aggregation or from the overall sequence probability, and is bound and stored with the character validity verification result of the marked region, so that subsequent confidence interlocking judgments have a consistent input caliber.

[0034] The detection confidence score output by the detection box model and the recognition confidence score output by the character recognition model are jointly input into the confidence interlocking judgment logic to simultaneously constrain "correct positioning of the marked area" and "correct parsing of the coordinate marker character," and the corresponding record is written into the valid anchor point set only when both simultaneously meet the preset threshold conditions. The confidence interlocking judgment refers to a dual-condition gating strategy that requires both the detection confidence score and the recognition confidence score to reach the threshold before allowing passage. The preset threshold conditions refer to threshold pairs that are pre-set or adaptively updated online for different carrier types, lighting conditions, and camera configurations. For example, the interlocking judgment can be expressed as a logical condition:

[0035] in, A Boolean result indicating whether the coordinate marker passed the interlocking determination; This represents the detection confidence level, output by the detection box model, and its value is between 0 and 1. This represents the detection threshold, used to limit the minimum confidence requirement for "usable positioning results," and can be determined by historical statistics or online calibration. This represents the recognition confidence level, output by the character recognition model, and its value is between 0 and 1. The recognition threshold is used to define the minimum confidence requirement for "semantic usability of a character," and can be determined by sample difficulty stratification or operational condition stratification; symbol The expression represents a logical AND, requiring both conditions to be met simultaneously. The principle of this formula is to measure two independent error sources using probabilistic indicators and to use common gating to avoid erroneous anchor points caused by high-confidence misjudgments of a single model entering the subsequent coordinate mapping basic parameters and gridded hierarchical positioning link. The effective anchor point set refers to the set of anchor point records that have passed the interlocking judgment and can be used for subsequent affine transformation matrix construction or coordinate mapping processing. Its record fields include at least the marked area location, coordinate marker characters, detection confidence, recognition confidence, camera identifier, and timestamp.

[0036] For coordinate markers that fail the confidence interlocking judgment, a secondary recognition process is triggered to improve the recognition pass rate under conditions of blurriness, occlusion, reflection, or low contrast, while preventing the direct discarding of low-quality samples, which could lead to insufficient anchor points or uneven anchor point distribution. The secondary recognition process refers to improving separability by performing feature enhancement on the marked region and its neighboring regions without changing the two core models: the detection box model and the character recognition model. Joint recognition is then performed again, and the confidence interlocking judgment is repeated. Feature enhancement involves edge enhancement, contrast enhancement, de-reflection processing, or adaptive sharpening of local regions to make the grayscale difference between character strokes and the background more significant, the boundaries more continuous, and the breaks fewer. The neighboring regions refer to the areas extending around the marked region. The text scope is used to preserve complete character information even when the marked area is too small or the character strokes overflow. Joint recognition refers to first relocating the marked area on the feature-enhanced image by the detection box model to correct the bounding box, and then re-parse the coordinate marked characters in the corrected marked area by the character recognition model to avoid the coupling between "recognition failure due to positioning deviation" and "recognition failure in turn cannot correct positioning". The secondary interlocking judgment uses the same threshold caliber or adjusts the threshold caliber according to the abnormal working condition strategy, and only adds the coordinate marks that pass the secondary interlocking judgment to the effective anchor point set. At the same time, the secondary processing identifier and processing parameters are written into the record field so that the quality stratification of the effective anchor point set and the robustness of the affine transformation matrix can be weighted and controlled in the future.

[0037] S130. Based on the spatial correspondence of each reference mark in the effective anchor point set under the top-view camera and the bottom-view camera, construct an affine transformation matrix that unifies the coordinate system of the top-view camera and the coordinate system of the bottom-view camera, and determine the affine transformation matrix as the coordinate mapping basis parameter for the automatic emapping of the IC carrier board.

[0038] Specifically, after the set of valid anchor points has passed consistency verification, the set of valid anchor points is confirmed as the benchmark anchor point set, and the partitioned vacuum adsorption fixation and temperature closed-loop regulation are maintained in a stable state, so that the benchmark markers are not affected by the geometric drift of the carrier plate during acquisition and extraction. The marker region corresponding to each benchmark marker is located in the images from the upward-looking camera and the downward-looking camera, and the center coordinates of the marker region are extracted as the pixel coordinate representation of the benchmark marker, or geometric fitting is performed on the boundary of the marker region to obtain the sub-pixel center coordinates as the fine pixel coordinate representation of the benchmark marker. Here, the center coordinates of the marker region refer to the center point coordinates calculated from the bounding box or contour of the marker region, and the sub-pixel coordinates are used as the fine pixel coordinate representation of the benchmark marker. The center coordinates refer to the non-integer pixel coordinates obtained through methods such as boundary contour fitting, centroid estimation, or edge sub-pixel interpolation. The coordinates of each reference mark in the image of the upward-looking camera are designated as point pairs in the coordinate set of the upward-looking camera, and the coordinates of the same reference mark in the image of the downward-looking camera are designated as corresponding point pairs in the coordinate set of the downward-looking camera. The two-way pairing is performed according to the reference mark identifier, thereby forming a spatial correspondence between the coordinate sets of the upward-looking camera and the coordinate sets of the downward-looking camera. The spatial correspondence refers to the pairing relationship of the same reference mark in different camera coordinate systems. The coordinate systems of the upward-looking camera and the downward-looking camera refer to the two-dimensional pixel coordinate systems established with the imaging planes of the upward-looking camera and the downward-looking camera as references, respectively.

[0039] A geometric consistency check is performed based on the spatial correspondence between the top-view camera coordinate set and the bottom-view camera coordinate set to ensure the consistency of the relative topology of each reference marker. The geometric consistency check verifies the relative distance ordering, relative orientation relationship, and convexity and directionality of the multi-point configuration between reference markers. Topological consistency means that the relative connection relationships of reference markers in the two coordinate systems do not intersect, flip, or are incorrectly rearranged, thus avoiding incorrect pairings from entering the fitting process. After the geometric consistency check passes, an affine transformation matrix is ​​constructed using a robust fitting method. Robust fitting refers to a fitting strategy that can still obtain stable mapping parameters even with a small number of abnormal pairing points or high positioning noise. It can use residuals as a basis to screen interior points and reduce the impact of outliers on the fitting. The affine transformation matrix refers to a two-dimensional linear transformation and translation combination used to characterize translation, rotation, and scale / shear relationships. Mapping points in the top-view camera coordinate system to points in the bottom-view camera coordinate system can be expressed as follows:

[0040] in, This represents the reference marker coordinates in the upward-looking camera coordinate system, and the values ​​are taken from the upward-looking camera coordinate set; The linear transformation parameters are used to jointly characterize rotation, scaling, and shearing relationships, and their values ​​are determined by the fitting process. The translation parameter, representing the offset of the origins of the two coordinate systems, is determined by the fitting process. The principle behind this formula is to use an invertible two-dimensional linear transformation to describe the rotation, scaling, and non-orthogonal shearing between the coordinate axes, and to superimpose a translation term to describe the origin offset, thereby achieving a stable mapping between the imaging coordinates of the two cameras under the condition that the plane assumption holds. Robust fitting can be achieved by minimizing the sum of squared residuals of the inlier set, and the inlier set is selected using a residual threshold. An exemplary objective function can be expressed as follows:

[0041] Where A represents by The constructed two-dimensional linear transformation matrix; Indicates the translation parameter; This represents the coordinates of the i-th reference marker in the upward-looking camera coordinate system; This represents the coordinates of the i-th reference marker in the downward-viewing camera coordinate system; This represents the set of interior points, which is determined by both geometric consistency checks and residual selection. The Euclidean norm of a two-dimensional vector is used to measure the magnitude of the mapping error. The principle of this formula is to make the coordinates after affine mapping as close as possible to the corresponding observed coordinates, and to suppress the pull of outliers on the parameters by accumulating the error only on the set of interior points.

[0042] After the affine transformation matrix is ​​constructed, a mapping consistency verification is performed on it. Mapping consistency verification involves quantitatively evaluating the mapping effect using known reference marker pairing relationships and determining whether the evaluation result meets preset accuracy conditions. Specifically, each reference marker coordinate in the upward-looking camera coordinate set is mapped to the downward-looking camera coordinate system using the affine transformation matrix. The mapping deviation between the mapping result and the corresponding coordinates in the downward-looking camera coordinate set is calculated, and a threshold determination is made for the statistical quantity of the mapping deviation. The mapping deviation refers to the distance error between the mapped coordinates and their actual corresponding coordinates. The preset accuracy conditions refer to the constraints on the allowable range of the mapping deviation, and their values ​​are determined by the process positioning accuracy requirements. An exemplary mapping deviation can be expressed as follows:

[0043] in, This represents the mapping deviation of the i-th reference mark; This represents the true coordinates of the i-th reference marker in the downward-viewing camera coordinate system, and its value comes from the downward-viewing camera coordinate set; Indicates will The predicted coordinates obtained through affine transformation matrix mapping are output by the mapping operation. When the mapping deviation of all reference marks meets the preset accuracy conditions, or when the maximum value, mean, quantile, and other statistics meet the preset accuracy conditions, the affine transformation matrix is ​​solidified as the coordinate mapping basic parameters for automatic IC carrier emapping. This is then bound and stored with the current partitioned vacuum adsorption fixed state identifier, temperature closed-loop adjustment state identifier, and synchronous trigger acquisition identifier. This ensures that subsequent coordinate mark recognition and coordinate mapping processing can ensure consistency with the current physical conditions when calling these coordinate mapping basic parameters. When the mapping deviation does not meet the preset accuracy conditions, the affine transformation matrix is ​​not solidified. Instead, the process backtracks to the reference mark extraction stage of the reference anchor point set to obtain more reliable reference mark coordinates or removes abnormal pairing points and refits until the mapping consistency verification is passed before solidifying it as the coordinate mapping basic parameters.

[0044] S140. Under the constraint of the affine transformation matrix, coordinate mark recognition and coordinate mapping processing are performed on the coordinate marks on the IC carrier through the coordinate mapping basic parameters to establish a one-to-one binding relationship between the coordinate marks and the mesh cells.

[0045] Specifically, the spatial position of the marked area corresponding to the coordinate marker in the upward-viewing camera coordinate system is determined as the center coordinates or sub-pixel center coordinates of the marked area in the upward-viewing camera coordinate system. This coordinate is used as the mapping source point of the coordinate mapping basic parameters, thereby ensuring that the generation of the predicted focus position is consistent with the coordinate definition when constructing the preceding affine transformation matrix. Subsequently, the coordinate mapping basic parameters are called to perform cross-coordinate system mapping on the mapping source point to obtain the predicted focus position in the downward-viewing camera coordinate system. The predicted focus position is then converted into the center of the focus window of the downward-viewing camera and the target position of the motion platform. This allows the downward-viewing camera to directly enter the field of view highly related to the coordinate marker without global search during local fine imaging. The predicted focus position refers to the target pixel coordinates obtained after mapping from the upward-viewing camera coordinate system to the downward-viewing camera coordinate system through the coordinate mapping basic parameters. It is used to indicate the local imaging center position that the downward-viewing camera should be aligned with. Local fine imaging refers to the imaging method of acquiring local image information with higher effective resolution within a smaller field of view. The character edge sharpness is improved by reducing the imaging window, adjusting the lens magnification, and optimizing the exposure and ring light source parameters.

[0046] After the downward-looking camera completes local fine imaging based on the predicted focus position, the obtained image information is used as the sole basis for confirming the valid coordinate markers. This ensures that the confirmation of valid coordinate markers is based on the image condition with the highest character recognizability, rather than on the global image condition of the upward-looking camera. The detection box model is called again in the image information to locate the boundary of the marked region and output the detection confidence score. Simultaneously, the character recognition model is called to parse the coordinate marker characters within the marked region and output the recognition confidence score. The detection confidence score and the recognition confidence score are input into the confidence score interlocking judgment logic. Only when both the detection confidence score and the recognition confidence score simultaneously meet the preset threshold condition and the coordinate marker character passes the character validity check is the coordinate marker... Coordinate markers are confirmed as valid coordinate markers. Valid coordinate markers refer to coordinate markers that can be reliably located and reliably parsed. Character validity verification refers to verifying whether the characters of the coordinate marker belong to the preset allowed character set, whether the character index belongs to the preset character index set, and whether the numeric index belongs to the preset numeric index range. When the confidence interlock judgment fails, feature enhancement and secondary recognition processing are triggered to improve the contrast, edge continuity, and stroke integrity of the marker area before re-judging. This improves the pass rate of valid coordinate marker confirmation without reducing the threshold and avoids low-confidence coordinate markers from entering the subsequent grid cell identification stage, which would cause spatial semantic errors.

[0047] After valid coordinate markers are confirmed, character and numeric indices are extracted from the coordinate marker characters, maintaining a unified semantic definition for both. This ensures that character indices are used only to determine vertical indices, and numeric indices are used only to determine horizontal indices, guaranteeing consistency in subsequent grid cell identifier determination rules. Preset semantic rules are implemented using a fixed mapping table and fixed numerical parsing rules. Character indices are mapped to vertical indices via table lookup, and numeric indices are mapped to horizontal indices via string parsing. Furthermore, grid cell identifiers are generated based on the vertical and horizontal indices, making each grid cell identifier a unique identifier for the coordinate marker. The vertical indices refer to the grid cell's position in the vertical arrangement, the horizontal indices refer to the grid cell's position in the horizontal arrangement, and the grid cell identifier is a unique identifier field determined by both the vertical and horizontal indices. For example, the calculation of the grid cell identifier can be represented as follows:

[0048] Where G represents the integer encoding form of the grid cell identifier, used to uniquely identify the grid cell in the binding record; The horizontal grid number represents the total number of grid cells, determined by the number of columns of the gridded partition adsorption surface; r represents the vertical index, obtained from the character index through a preset mapping table, with a value range of 1 to the total number of vertical grid cells; c represents the horizontal index, obtained from the numeric index through string parsing, with a value range of 1 to... The principle of this formula is to convert the two-dimensional row and column index into a one-dimensional unique code, so that subsequent storage, retrieval and association of records can complete the grid cell location with a single field, while the vertical and horizontal serial numbers can still be recovered through inverse operations. After obtaining the grid cell identifier, the coordinate marker identifier, grid cell identifier, character index, numeric index, vertical serial number, horizontal serial number, marked area position in the upward camera coordinate system, predicted focus position in the downward camera coordinate system, marked area position in local fine imaging, detection confidence and recognition confidence are written into the binding record, thereby forming a one-to-one binding relationship between coordinate markers and grid cells, and providing stable spatial semantic constraints and traceable data basis for subsequent gridded hierarchical positioning.

[0049] S150. Based on the one-to-one binding relationship between coordinate markers and grid cells, gridded hierarchical positioning is performed under the constraints of affine transformation matrix and structural consistency, so that the detection points can complete grid-level positioning and fine-tuning-level positioning in sequence to obtain the theoretical coordinates of the detection points.

[0050] Specifically, after writing the one-to-one binding relationship between coordinate markers and grid cells into the binding record, the binding record is loaded as a whole into the positioning task context. This positioning task context simultaneously carries the coordinate marker identifier, grid cell identifier, the position of the coordinate marker in the upward and downward camera coordinate systems, the detection confidence and recognition confidence, and the version identifier of the affine transformation matrix bound to it. This ensures that the subsequent positioning process can always be traced back to the same coordinate mapping basic parameters and the same physical conditions. Under the constraints of the affine transformation matrix, the detection points to be detected are classified according to their respective grid cell identifiers, and the grid cell is defined as the smallest positioning unit. This allows grid-level positioning to converge the spatial search range with the grid cell as the boundary, rather than blindly scanning the entire board area. Here, the positioning task context refers to the runtime data structure used to organize all the data and states required for positioning. The grid cell identifier refers to the grid cell number uniquely determined by the vertical and horizontal serial numbers. Grid-level positioning refers to the coarse positioning process at the grid cell scale, used to quickly guide the motion platform to the target area neighborhood and establish a spatial starting point for subsequent fine-tuning positioning.

[0051] In grid-level positioning, the coordinate marker positions associated with the grid cell are preferentially selected as grid-level positioning anchor points. An affine transformation matrix is ​​used to map these anchor points from the top-view camera coordinate system to the bottom-view camera coordinate system, thereby generating the predicted neighborhood position of the motion platform and controlling the motion platform to move into the predicted neighborhood. This ensures that the motion platform's positioning process is established around the verified coordinate markers rather than directly approaching uncertain detection points. Here, the predicted neighborhood refers to the spatial neighborhood centered on the mapped coordinate marker position and covering the range of the grid cell. The grid-level positioning anchor point refers to the reference coordinate point used to guide the motion platform into the target grid cell. Local verification refers to the bottom-view camera re-images and re-identifies the coordinate markers after the motion platform reaches the predicted neighborhood to confirm the consistency of the spatial position of the grid cell under the current partitioned vacuum adsorption fixation and temperature closed-loop regulation conditions.

[0052] After grid-level localization is completed and local verification confirms the consistency of the spatial positions of the grid cells, fine-tuning-level precision localization is initiated to converge the spatial position of the detection point from the grid cell scale to a fine scale suitable for electrical characteristic detection. Under the constraint of the affine transformation matrix, the downward-looking camera acquires local image information of the detection point's neighborhood. Local feature sets related to the detection point are extracted from the local image information, and the motion platform pose is iteratively corrected in the visual loop to gradually converge the spatial position of the detection point. Fine-tuning-level precision localization refers to the precision localization process achieved through repeated imaging and pose fine-tuning within a small range. Local image information refers to the high-resolution image acquired by the downward-looking camera in a local fine imaging mode. The local feature set refers to the feature set used to stably describe the geometric structure of the detection point's neighborhood. The local feature set may include pad boundaries, via edges, trace corners, and the boundary contours or center features of the detection point target. The visual loop refers to the feedback iterative process of "imaging-feature extraction-deviation calculation-pose correction-re-imaging". For example, the difference between the reference center and the current center of the local feature set can be used to drive pose correction, as shown below:

[0053] in, The k-th iteration represents the pose correction applied to the motion platform, which may include planar displacement components and necessary attitude fine-tuning components; K represents the pose correction gain matrix, which is used to convert pixel or local coordinate deviations into corrections for the motion platform. Its value is determined by the calibration ratio, motion platform resolution, and convergence speed requirements. The target reference center is defined by the set of local features and is determined by the expected detection point center or the best alignment center obtained by template matching. This represents the current feature center extracted from the local image information in the k-th iteration. The principle of this formula is to transform the "deviation of the detection point in the image" into the "pose correction that the motion platform needs to compensate for", and to control the correction magnitude of each iteration through the gain matrix, so that the visual loop can achieve stable convergence within a finite number of steps and avoid overshoot oscillation.

[0054] During the fine-tuning stage of precise positioning, structural consistency constraints are continuously applied to ensure that the convergence direction and range of the detection points do not deviate from the reasonable geometric relationships between grid cells, thereby suppressing the propagation of positioning errors caused by local deformation or imaging disturbances. Structural consistency constraints refer to a set of rules that constrain the positioning results within the current grid cell using the relative geometric relationships between adjacent grid cells. Relative geometric relationships refer to the spatial consistency of relative displacement, relative direction, and relative scale that the coordinate marker positions of adjacent grid cells should satisfy. Positioning error propagation refers to the phenomenon where a local error in a grid cell spreads to adjacent grid cells through cumulative correction, error mapping, or erroneous feature association, causing systematic deviations. In implementation, the coordinate marker positions of the current grid cell and its adjacent grid cells are used as reference objects for structural consistency constraints, calculated in each iteration. Before the candidate correction direction is constrained by projection or boundary clipping, the candidate correction amount will not cause the detection point to cross the reasonable boundary of the current grid cell in space, or cause abnormal jumps in the relative displacement relationship of adjacent grid cells. This restricts the degree of freedom of fine-tuning and precise positioning within the range of "mesh topology and physical morphology interpretability", making the final convergence result more stable and more reproducible.

[0055] When the fine-tuning stage precision positioning meets the stable convergence condition, the theoretical coordinates of the detection point are generated. The theoretical coordinates of the detection point are then associated with the corresponding grid cell identifier, coordinate marker identifier, and affine transformation matrix version identifier and written into the detection point record. This ensures that subsequent electrical characteristic detection can be performed based on the same spatial semantics and the same mapping foundation. Stable convergence refers to the fact that the center deviation of the local feature set is less than a preset threshold and the deviation change trend is no longer significant in several consecutive iterations. The theoretical coordinates of the detection point refer to the coordinate results representing the spatial position of the detection point under the constraints of the current physical working condition and the current coordinate mapping foundation parameters. The detection point record refers to the structured record used to carry "theoretical coordinates of the detection point - grid cell identifier - coordinate marker identifier - working condition identifier - quality index". To ensure that the theoretical coordinates of the detection point can be used for motion platform control and result traceability, the theoretical coordinates of the detection point retain both the pixel coordinate expression in the downward camera coordinate system and the position expression in the motion platform coordinate system. The two expressions are bound together through the version identifier of the same affine transformation matrix, thereby ensuring a stable, unique, and traceable correspondence between the subsequent conduction impedance data and the theoretical coordinates of the detection point.

[0056] S160. Based on the theoretical coordinates of the detection point, control the motion platform to reach the target position, perform electrical characteristic detection on the IC carrier board, and after collecting the conduction impedance data, associate and record it with the corresponding coordinate marks, grid cells and theoretical coordinates of the detection point to generate Emappping results.

[0057] Specifically, under the premise of continuous stability in the partitioned vacuum adsorption fixation state and the temperature closed-loop regulation state, the theoretical coordinates of the detection point are written into the target pose register area of ​​the motion platform. The motion platform then generates a time-parameterized trajectory from the current position to the target position neighborhood under the condition of satisfying the preset safety path constraints. The motion platform first moves along a safe height layer away from the IC substrate surface, then enters the target position neighborhood and performs deceleration control to reach the target position. After reaching the target position, the real-time pose feedback and the theoretical coordinates of the detection point are read for consistency verification. To simultaneously constrain position error, attitude error, and the dynamic controllability of the deceleration phase, the consistency verification is extended to a comprehensive pose deviation index that includes position deviation, attitude deviation, and velocity deviation. This comprehensive pose deviation index is compared with the preset pose conditions, thereby jointly defining the "accessible electrical characteristic detection" state in both spatial and dynamic aspects. The comprehensive pose deviation index can be expressed as:

[0058] in, This represents the overall pose deviation index. The smaller the value, the closer the motion platform is to the target pose and the more stable its dynamic state is. Represents the position deviation vector. Provided by real-time pose feedback, Given from the theoretical coordinates of the detection point; Represents the linear velocity deviation vector. For real-time linear velocity, To determine the desired linear velocity, either the zero vector or a preset micro-motion velocity is taken at the instant of positioning. Represents the angular velocity deviation vector. For real-time angular velocity, The desired angular velocity is taken as the zero vector at the instant of arrival. These are weight matrices for position, linear velocity, and angular velocity, used to reflect the sensitivity to deviations of different directions or magnitudes. The values ​​are set by the resolution of the motion platform, the probe contact tolerance, and the process requirements. For real-time attitude quaternions, The target pose quaternion is obtained by dot product of the two. Characterizes the degree of pose alignment; The attitude term weights are used to adjust the impact of attitude deviations on the overall index. This formula incorporates position, velocity, angular velocity, and attitude consistency into the same index, ensuring that the motion platform not only "reaches the target position" but also "is sufficiently statically stable and attitude-aligned after reaching the target position," thereby reducing measurement disturbances caused by micro-motions or attitude deviations during probe pressing. If the value is not greater than the threshold corresponding to the preset pose condition, the electrical characteristic detection is initiated; otherwise, the value remains within the neighborhood of the target position and fine-tuning continues while repeating the consistency check.

[0059] After the motion platform is stably positioned, the electrical testing unit applies preset electrical test conditions to the IC carrier board at the target location and collects conduction impedance data. Simultaneously, it monitors the motion platform's pose stability and the zoned vacuum adsorption fixation state within the sampling window to avoid interference from mechanical or fixation abnormalities in the electrical characteristic test results. To improve the robustness of the conduction impedance data to contact resistance fluctuations, fixture parasitic impedance, and noise, the preset electrical test conditions employ multi-frequency excitation and synchronous sampling. This allows the conduction impedance data to be characterized by complex impedances at multiple frequency points, and the final conduction impedance estimate is obtained through weighted robust fitting. This transforms the problem of single-point measurement being susceptible to interference into a robust estimation problem under multi-frequency consistency constraints. The multi-frequency complex impedance can be expressed as:

[0060] in, The complex impedance at the k-th frequency point is represented by v[n], which represents the voltage sampling sequence at the n-th sampling moment within the sampling window, acquired by the electrical testing unit; i[n] represents the corresponding current sampling sequence, acquired by the electrical testing unit; N represents the number of sampling points, determined by the sampling duration and sampling rate. Here, j is the complex exponential kernel function, and j is the imaginary unit. The frequency point corresponding to index k is set by the excitation frequency set; this formula obtains the complex impedance of each frequency point by extracting the frequency domain components of the voltage and current sequences and taking their ratios, thereby transforming the time domain noise and phase information into a frequency domain representation that can be used for consistency constraints; after obtaining the multi-frequency complex impedance, the robust estimation of the conduction impedance data can be expressed in the form of a fitting with Huber loss and weights as follows:

[0061] in, This represents the final estimated value of the on-resistance data, which can be in complex or amplitude form to meet the process definition; K represents the number of frequency points involved in the estimation; This represents the weight of the k-th frequency point, and its value can be set according to the signal-to-noise ratio, contact stability, or parasitic sensitivity of that frequency point. This represents the Huber loss function. The piecewise threshold of the loss function is used to apply an approximate quadratic penalty in small residual regions and to reduce the impact of outliers in large residual regions. This represents the difference between the candidate impedance and the frequency-point complex impedance. By robustly aggregating errors at multiple frequencies, this formula ensures that occasional contact jitter or parasitic anomalies at individual frequencies do not dominate the final conduction impedance data, thereby improving the repeatability of electrical characteristic testing under complex operating conditions.

[0062] During the on-resistance data acquisition process, the pose stability of the motion platform and the zoned vacuum adsorption fixation state are continuously monitored. The stability judgment is expanded from "single range" to "statistical covariance and confidence bounds," enabling the stability judgment to simultaneously reflect the impact of fluctuation amplitude, fluctuation correlation, and abnormal peaks. The pose stability of the motion platform can be described by the weighted covariance of the position deviation vector within the sampling window, and further given as a stability index in the form of confidence bounds, thereby statistically limiting the spatial jitter during the sampling period. An exemplary stability index can be expressed as:

[0063] in, This represents the pose stability index within the sampling window; the smaller the value, the more stable the pose. Represents the position deviation vector In the sampling window The covariance matrix within the window is obtained by sampling multiple times within the window. Calculated; This represents the stability weighting matrix, used to reflect the differences in the impact of jitter in different directions on the measurement; The first term represents the largest eigenvalue of the matrix, used to characterize the amplitude in the most unfavorable direction of the weighted jitter; the second term is the peak-sensitive term, where... For peak weight, It is the Euclidean norm. Used to capture the maximum transient deviation within the window. Used to provide typical deviation levels. To prevent the use of tiny constants with zero denominators, this formula characterizes the sustained vibration intensity using the maximum eigenvalue of the covariance and the abnormal spikes using the ratio of the maximum value to the median, ensuring that the stability assessment neither ignores sustained micro-vibrations nor occasional impacts; when When the preset stability threshold is exceeded, sampling is stopped and the process returns to the fine-tuning and consistency verification stage. At the same time, the state vector composed of vacuum pressure feedback and vacuum flow feedback of each grid unit can be used to monitor the covariance of the fixed state of vacuum adsorption in the partition, so as to avoid false anomalies caused by changes in probe contact state due to local leakage or local suspension.

[0064] After the on-resistance data acquisition is completed, the on-resistance data is bound to the theoretical coordinates of the detection points, and the affine transformation matrix identifier, the partitioned vacuum adsorption fixation state identifier, and the temperature closed-loop regulation state identifier are written into the detection record. This forms an Emapplication result containing spatial location, physical conditions, and electrical response information. After completing the Emapplication result record for a single detection point, the detection completion status is updated according to the grid cell identifier, and subsequent detection points within the same grid cell are executed until the electrical characteristics of all detection points belonging to the IC carrier are detected and the entire board Emapplication result is output. To ensure that the entire board Emapplication result has both "spatial visualization consistency" and "quality statistical availability," robust aggregation with spatial adjacency constraints can be performed on the on-resistance data within the same grid cell during the aggregation stage, and grid cell-level statistical fields can be output. This allows the spatial clustering characteristics of outliers to be explicitly expressed. The grid cell-level robust aggregation can be represented as follows:

[0065] in, This represents the on-resistance value of the grid cell g, used for the grid cell-level output of the whole-board Emapplication results; This represents the set of detection points belonging to grid cell g, determined by the grid cell identifier; This represents the estimated on-resistance value corresponding to the detection point p, which is derived from the multi-frequency robust estimation result. This represents the weight of the detection point, and its value can be set based on the pose stability index of the detection point, the quality of the confidence interlock result, or the sampling noise level. This represents the Huber loss function. This is the loss threshold for the grid cell, used to adjust the suppression of outliers; This represents the adjacency consistency weight, used to control the strength of spatial smoothing constraints; This represents the set of grid cells adjacent to grid cell g; The value represents the adjacency weight of the grid cells, and can be set according to the length of the shared edge, the division of the process area, or the similarity of the thermal field. This formula suppresses outliers within the grid cells by the first term and suppresses unreasonable abrupt changes in the adjacency relationship of the grid cells by the second term. This makes the whole board Emapplication results retain the local anomalies caused by real defects and reduce the spatial noise caused by occasional measurement disturbances, thereby improving the usability of the whole board Emapplication results in defect location and process traceability.

[0066] This application also provides an AI-based automatic emapping device for IC carrier boards, see reference. Figure 2 , Figure 2This is a schematic diagram of a module for an AI-based automatic IC carrier embedding device provided in an embodiment of this application. The device is a server, which includes an acquisition module 21 and a processing module 22. The acquisition module 21 is used to perform partitioned vacuum adsorption fixation on the IC carrier through a temperature-controlled vacuum adsorption platform, and simultaneously perform closed-loop temperature regulation to ensure that the IC carrier maintains geometric stability and controlled thermal deformation during the automatic embedding process. The processing module 22 is used to control the upper and lower cameras to simultaneously acquire image information of the IC carrier under the partitioned vacuum adsorption fixation and temperature closed-loop regulation state, and to locate the marked area through a detection box model and parse the coordinate marked characters through a character recognition model based on the image information. At the same time, it combines the interlocking judgment of detection confidence and recognition confidence to form a set of effective anchor points. The processing module 22 is also used to construct a unified system based on the spatial correspondence of each reference mark in the set of effective anchor points under the upper and lower cameras. The processing module 22 is used to establish a coordinate mapping basis for automatic emapping of the IC carrier board by performing coordinate mark recognition and coordinate mapping processing on the coordinate marks on the IC carrier board under the constraint of the affine transformation matrix and through the coordinate mapping basis parameters, and establishing a one-to-one binding relationship between the coordinate marks and the grid cells. The processing module 22 is also used to perform gridded hierarchical positioning under the constraints of the affine transformation matrix and structural consistency based on the one-to-one binding relationship between the coordinate marks and the grid cells, so that the detection points can complete the grid-level positioning and fine-tuning level positioning in sequence to obtain the theoretical coordinates of the detection points. The processing module 22 is also used to control the motion platform to reach the target position according to the theoretical coordinates of the detection points, and perform electrical characteristic detection on the IC carrier board, and after collecting the conduction impedance data, associate and record it with the corresponding coordinate marks, grid cells and theoretical coordinates of the detection points to generate emapping results.

[0067] It should be noted that the above embodiments of the apparatus are only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0068] This application also provides an electronic device, with reference to... Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: at least one processor 31, at least one network interface 34, a user interface 33, a memory 35, and at least one communication bus 32.

[0069] The communication bus 32 is used to enable communication between these components.

[0070] The user interface 33 may include a display screen and a camera. Optionally, the user interface 33 may also include a standard wired interface and a wireless interface.

[0071] The network interface 34 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0072] The processor 31 may include one or more processing cores. The processor 31 connects to various parts of the server via various interfaces and lines, executing instructions, programs, code sets, or instruction sets stored in the memory 35, and calling data stored in the memory 35 to perform various server functions and process data. Optionally, the processor 31 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 31 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content to be displayed on the screen; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 31 and may be implemented as a separate chip.

[0073] The memory 35 may include random access memory (RAM) or read-only memory. Optionally, the memory 35 may include a non-transitory computer-readable storage medium. The memory 35 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 35 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 35 may also be at least one storage device located remotely from the aforementioned processor 31. Figure 3 As shown, the memory 35, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application program for an AI-based automatic emapping method for IC carrier boards.

[0074] exist Figure 3 In the electronic device shown, the user interface 33 is mainly used to provide an input interface for the user and to obtain the user input data; while the processor 31 can be used to call an application stored in the memory 35 for an AI-based IC carrier automatic emapping method. When executed by one or more processors, the electronic device performs one or more methods as described in the above embodiments.

[0075] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0076] This application also provides a non-transitory computer-readable storage medium storing instructions. When executed by one or more processors, these instructions cause an electronic device to perform one or more of the methods described in the above embodiments.

[0077] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0078] In the several embodiments provided in this application, it should be understood that the disclosed apparatus 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. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual couplings or direct couplings or communication connections may be through some service interfaces; indirect couplings or communication connections between apparatuses or units may be electrical or other forms.

[0079] 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 this embodiment according to actual needs.

[0080] Furthermore, the functional units in the various embodiments of this application 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.

[0081] 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 device (CMD). Based on this understanding, the technical solution of this application, 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 memory 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 of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0082] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of other embodiments of this disclosure upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A method for automatic emapping of IC carrier boards based on AI, characterized in that, The method includes: The IC carrier is fixed by partitioned vacuum adsorption on a temperature-controlled vacuum adsorption platform, and the temperature is adjusted in a closed loop simultaneously to ensure that the IC carrier maintains geometric stability and thermal deformation is controlled during the automatic emapping process. Under the conditions of partitioned vacuum adsorption fixation and temperature closed-loop regulation, the upper-view camera and the lower-view camera are controlled to simultaneously acquire image information of the IC carrier board. Based on the image information, the marked area is located by the detection box model and the coordinate marked character is analyzed by the character recognition model. At the same time, the confidence of detection and recognition are interlocked to form an effective anchor point set. Based on the spatial correspondence of each reference mark in the set of effective anchor points under the top-view camera and the bottom-view camera, an affine transformation matrix is ​​constructed to unify the coordinate system of the top-view camera and the coordinate system of the bottom-view camera, and the affine transformation matrix is ​​determined as the coordinate mapping basic parameter for automatic emapping of IC carrier board. Under the constraint of the affine transformation matrix, coordinate mark recognition and coordinate mapping processing are performed on the coordinate marks on the IC carrier through the coordinate mapping basic parameters to establish a one-to-one binding relationship between coordinate marks and mesh cells; Based on the one-to-one binding relationship between coordinate markers and grid cells, gridded hierarchical positioning is performed under the constraints of the affine transformation matrix and structural consistency, so that the detection points can sequentially complete grid-level positioning and fine-tuning-level precise positioning to obtain the theoretical coordinates of the detection points. The motion platform is controlled to reach the target position based on the theoretical coordinates of the detection point, and the electrical characteristics of the IC carrier board are tested. After collecting the conduction impedance data, it is associated with the corresponding coordinate markers, grid cells and theoretical coordinates of the detection point to generate an Emaplidding result.

2. The AI-based automatic emapping method for IC carrier boards according to claim 1, characterized in that, The process involves performing zoned vacuum adsorption fixation of the IC carrier on a temperature-controlled vacuum adsorption platform, with simultaneous closed-loop temperature regulation, to ensure that the IC carrier maintains geometric stability and controlled thermal deformation during automatic emmapping. Specifically, this includes: The temperature-controlled vacuum adsorption platform is pre-constructed with a gridded adsorption surface that matches the size and shape of the IC carrier. The gridded adsorption surface is divided into multiple independently controlled grid units, and each grid unit is connected to an independent vacuum adsorption circuit, so as to achieve coordinated control of overall adsorption and local adsorption on the same temperature-controlled vacuum adsorption platform. After the IC carrier is placed on the gridded adsorption surface, the vacuum adsorption circuit of the corresponding grid unit is activated in sequence according to the spatial order of the central area, edge area and corner area, so that the IC carrier is gradually attached and formed into a partitioned vacuum adsorption fixation state. In the partitioned vacuum adsorption fixation state, the vacuum pressure feedback and vacuum flow feedback of each grid unit are obtained, and the vacuum adsorption intensity of each grid unit is dynamically balanced and adjusted based on the vacuum pressure feedback and the vacuum flow feedback to suppress local warping or local suspension caused by the adsorption intensity difference between adjacent grid units. The temperature closed-loop adjustment of the temperature-controlled vacuum adsorption platform is started simultaneously, and the temperature-controlled vacuum adsorption platform is divided into temperature-controlled areas corresponding to the spatial distribution of the grid units. Temperature feedback acquisition and differentiated temperature adjustment are performed on each temperature-controlled area to make the temperature-controlled vacuum adsorption platform form a temperature field with controlled temperature uniformity in the spatial dimension and controlled temperature fluctuation in the time dimension, so as to reduce the non-uniform thermal deformation of IC carrier board caused by local temperature difference. After the partitioned vacuum adsorption fixation and temperature closed-loop regulation reach a stable state, the vacuum stability of each grid unit and the temperature stability of each temperature control area are respectively judged. When the vacuum stability and the temperature stability simultaneously meet the preset conditions, the automatic Emapplication process is executed.

3. The AI-based automatic emapping method for IC carrier boards according to claim 1, characterized in that, Under the conditions of partitioned vacuum adsorption fixation and closed-loop temperature regulation, the upper-view camera and lower-view camera are controlled to simultaneously acquire image information of the IC carrier board. Based on the image information, the marked area is located using a detection frame model, and the coordinates of the marked characters are analyzed using a character recognition model. Simultaneously, a set of effective anchor points is formed by combining the interlocking judgment of detection confidence and recognition confidence. Specifically, this includes: After the partitioned vacuum adsorption fixation and temperature closed-loop regulation are in a stable state, the upper-view camera and the lower-view camera are controlled by a unified synchronous trigger signal to acquire image information of the IC carrier board at the same time reference. After performing brightness equalization, noise suppression and geometric distortion correction on the image information, the processed image information is input into the detection box model to locate the coordinates of the corresponding marked regions, and a detection confidence score is output for each marked region. At the same time, the marked regions are written into the candidate anchor point set. For each marked region in the candidate anchor point set, the character recognition model is invoked to parse the coordinate marked characters within the marked region and output the recognition confidence score, while maintaining a unified semantic definition between character index and numeric index during the character parsing process; The detection confidence and the recognition confidence are interlocked with the input confidence logic. When the detection confidence and the recognition confidence simultaneously meet the preset threshold condition, the corresponding coordinate marker is written into the set of valid anchor points. For coordinate markers that fail the confidence interlocking determination, a secondary recognition process is triggered. After performing feature enhancement on the marker area and its neighboring areas, the detection box model and the character recognition model are called for joint recognition, and the confidence interlocking determination is performed a second time.

4. The AI-based automatic emapping method for IC carrier boards according to claim 1, characterized in that, Based on the spatial correspondence of each reference mark in the effective anchor point set under the upward-view camera and the downward-view camera, an affine transformation matrix is ​​constructed to unify the coordinate systems of the upward-view camera and the downward-view camera. This affine transformation matrix is ​​then determined as the fundamental coordinate mapping parameter for automatic emapping of the IC carrier board. Specifically, this includes: After the set of effective anchor points passes the consistency check, the set of effective anchor points is confirmed as the set of reference anchor points. Under the stable state of partitioned vacuum adsorption fixation and temperature closed-loop regulation, the center coordinates of the marked area corresponding to each reference mark or the sub-pixel center coordinates obtained by fitting the boundary of the marked area are extracted from the image of the top-view camera and the image of the bottom-view camera respectively, forming a one-to-one corresponding set of coordinates of the top-view camera and the bottom-view camera. A geometric consistency check is performed based on the spatial correspondence between the coordinate sets of the top-view camera and the coordinate sets of the bottom-view camera to ensure that the relative topological structure of each reference mark is consistent. After the geometric consistency check is passed, an affine transformation matrix for characterizing translation, rotation and scale and shearing relationships is constructed by a robust fitting method. After the affine transformation matrix is ​​constructed, a mapping consistency verification is performed on the affine transformation matrix. When the reference mark mapping deviation obtained based on the affine transformation matrix meets the preset accuracy condition, the affine transformation matrix is ​​solidified as the coordinate mapping basic parameters for automatic emapping of the IC carrier board.

5. The AI-based automatic emapping method for IC carrier boards according to claim 1, characterized in that, Under the constraint of the affine transformation matrix, the coordinate marker recognition and mapping processing on the coordinate markers on the IC carrier is performed through the coordinate mapping basic parameters to establish a one-to-one binding relationship between the coordinate markers and the mesh cells. Specifically, this includes: The spatial position of the marked area corresponding to the coordinate marker in the upward camera coordinate system is mapped to the downward camera coordinate system through the coordinate mapping basic parameters to generate a predicted focus position, and the downward camera is controlled to perform local fine imaging based on the predicted focus position. In the image information obtained from the local fine imaging, valid coordinate markers are confirmed; Based on the character index and numeric index of the valid coordinate marker, the corresponding vertical and horizontal serial numbers are determined according to preset semantic rules, and the grid cell identifier to which the coordinate marker belongs is determined according to the vertical and horizontal serial numbers, thereby establishing a one-to-one binding relationship between the coordinate marker and the grid cell.

6. The AI-based automatic emapping method for IC carrier boards according to claim 1, characterized in that, Based on the one-to-one binding relationship between coordinate markers and grid cells, grid-based hierarchical positioning is performed under the constraints of the affine transformation matrix and structural consistency, so that the detection points sequentially complete grid-level positioning and fine-tuning-level precise positioning to obtain the theoretical coordinates of the detection points. Specifically, this includes: The one-to-one binding relationship is loaded into the localization task context, and the detection points are classified according to their grid cell identifiers under the constraints of the affine transformation matrix. Grid-level localization is performed with the grid cell as the smallest localization unit. In grid-level positioning, based on the coordinate marker positions associated with grid cells, a predicted neighborhood of the motion platform is generated through the affine transformation matrix, and the motion platform is controlled to move to the predicted neighborhood. At the same time, a downward-looking camera is used to locally verify the coordinate markers in the predicted neighborhood to confirm the consistency of the spatial position of the grid cells under the current physical conditions. After the grid-level localization is completed, under the constraints of the affine transformation matrix and structural consistency, fine-tuning level localization is performed on the detection point. The fine-tuning level localization acquires local image information of the neighborhood of the detection point through the downward-looking camera and extracts the local feature set related to the detection point. Based on the local feature set, the pose of the motion platform is iteratively corrected in the visual closed loop so that the spatial position of the detection point gradually converges. In the fine-tuning stage of precise positioning, the relative geometric relationship between adjacent grid cells is used as a structural consistency constraint to limit the convergence direction and convergence range of the detection points, so as to suppress the propagation of positioning errors caused by local deformation or imaging disturbances. After the fine-tuning level positioning reaches stable convergence, the theoretical coordinates of the detection point are generated.

7. The AI-based automatic emapping method for IC carrier boards according to claim 1, characterized in that, The process involves controlling the motion platform to reach the target position based on the theoretical coordinates of the detection points, performing electrical characteristic testing on the IC carrier board, and, after acquiring the continuity impedance data, associating and recording it with the corresponding coordinate markers, grid cells, and theoretical coordinates of the detection points to generate an Emaplilation result. Specifically, this includes: Under the premise that the partitioned vacuum adsorption fixation state and the temperature closed-loop regulation state remain stable, the theoretical coordinates of the detection point are loaded as the target pose input of the motion platform, and the motion platform is controlled to move along the preset safe path to the neighborhood of the target position and then decelerate to the target position. At the same time, the consistency is checked by reading the real-time pose feedback of the motion platform and the theoretical coordinates of the detection point. When the pose deviation meets the preset pose conditions, the electrical characteristic detection is entered. After the motion platform is stably positioned, the electrical testing unit applies preset electrical test conditions to the IC carrier board at the target location and collects the corresponding conduction impedance data. During the conduction impedance data collection process, the motion platform's pose stability and partitioned vacuum adsorption fixation status are monitored to avoid mechanical or fixation abnormalities from interfering with the electrical characteristic test results. After the on-resistance data is acquired, the on-resistance data is bound to the theoretical coordinates of the detection point. At the same time, the corresponding affine transformation matrix identifier, partitioned vacuum adsorption fixation state identifier, and temperature closed-loop regulation state identifier are written into the detection record, thereby forming an Emapping result containing spatial location, physical conditions, and electrical response information. After recording the Emapplication result for a single test point, the test completion status is updated based on the grid cell identifier, and subsequent test points within the same grid cell are executed until the electrical characteristic test of the test points belonging to the IC carrier board is completed and the overall board Emapplication result is output.

8. An AI-based automatic emapping device for IC carrier boards, characterized in that, The apparatus is used to perform the AI-based automatic emapping method for IC carrier boards as described in any one of claims 1 to 7, the apparatus comprising an acquisition module and a processing module, wherein... The acquisition module is used to perform partitioned vacuum adsorption fixation on the IC carrier board through a temperature-controlled vacuum adsorption platform, and simultaneously perform closed-loop temperature regulation so that the IC carrier board maintains geometric stability and thermal deformation is controlled during the automatic emapping process. The processing module is used to control the upper and lower cameras to simultaneously acquire image information of the IC carrier board under the partitioned vacuum adsorption fixation and temperature closed-loop regulation state, and to locate the marked area through the detection box model and parse the coordinate marked character through the character recognition model based on the image information. At the same time, it combines the interlocking judgment of the detection confidence and the recognition confidence to form an effective anchor point set. The processing module is also used to construct an affine transformation matrix that unifies the coordinate system of the top-view camera and the coordinate system of the bottom-view camera based on the spatial correspondence of each reference mark in the set of effective anchor points under the top-view camera and the bottom-view camera, and to determine the affine transformation matrix as the coordinate mapping basic parameter for automatic emapping of the IC carrier board. The processing module is also used to perform coordinate mark recognition and coordinate mapping processing on the coordinate marks on the IC carrier board under the constraints of the affine transformation matrix, and to establish a one-to-one binding relationship between the coordinate marks and the mesh cells. The processing module is also used to perform gridded hierarchical positioning based on the one-to-one binding relationship between coordinate markers and grid cells, under the constraints of the affine transformation matrix and structural consistency, so that the detection point can complete grid-level positioning and fine-tuning-level positioning in sequence to obtain the theoretical coordinates of the detection point; The processing module is also used to control the motion platform to reach the target position based on the theoretical coordinates of the detection point, perform electrical characteristic detection on the IC carrier board, and after collecting the conduction impedance data, associate and record it with the corresponding coordinate markers, grid cells and theoretical coordinates of the detection point to generate Emaplid results.

9. An electronic device, characterized in that, The electronic device includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions. The user interface and the network interface are both used to communicate with other devices. The processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores instructions that, when executed, perform the method as described in any one of claims 1 to 7.

Images