CPU disassembly and assembly positioning method and system based on real-time mainboard image

Abstract

The invention discloses a CPU disassembly and assembly positioning method and system based on a real-time mainboard image, and relates to the technical field of image localization, and the method comprises the steps: placing a CPU mainboard on disassembly and assembly equipment, carrying out the preliminary positioning, calibrating an angular point, and collecting a multi-view time sequence image through a camera; the depth of each visual angle time sequence image is calculated, image alignment is carried out to calculate the comprehensive depth, a neural radiation field is adopted to generate a three-dimensional point cloud according to the comprehensive depth, a dynamic region segmentation algorithm is used based on the three-dimensional point cloud to carry out point cloud segmentation and verification, and then CPU point cloud data is extracted; mapping the CPU point cloud data to a potential space to extract potential features, extracting local-global features through local-global comparison optimization, and detecting CPU integrity by using a deep learning algorithm; and deviation positioning detection is carried out according to the CPU point cloud data, and the dismounting position is optimized and adjusted. According to the invention, the positioning and dismounting precision of the CPU is greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of image positioning technology, in particular to a CPU disassembly and assembly positioning method and system based on real-time mainboard images. Background Art

[0002] With the increasing complexity of electronic products, especially computer hardware, precise component positioning and disassembly and assembly technologies play a vital role in industrial manufacturing and maintenance. In recent years, with the continuous development of automation equipment and robotics, automatic positioning technologies based on vision systems have gradually become mainstream. These technologies mainly use high-precision cameras and computer vision algorithms, combined with machine learning methods, to locate, identify, and detect the status of objects. In this process, three-dimensional vision technology has shown great application potential, especially in processing complex hardware components, especially the disassembly and maintenance of precision components such as CPUs and motherboards. The generation and processing of three-dimensional point cloud data is an important research direction in the current field of computer vision, and has made significant progress in precise positioning, object recognition, and damage detection.

[0003] However, existing 3D vision technologies mostly rely on 2D image recognition and template-based matching techniques. However, these methods often exhibit low accuracy when processing objects with high morphological complexity and from different perspectives, and are unable to cope with complex disassembly and assembly tasks. In particular, when performing CPU disassembly and positioning, they usually rely on manual intervention to adjust and optimize the disassembly path, which may cause equipment damage or reduced operational accuracy. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a CPU disassembly and assembly positioning method and system based on real-time motherboard images, which solves the problem that the existing technology often exhibits low accuracy when processing objects with high morphological complexity and at different viewing angles.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a CPU disassembly and positioning method based on real-time motherboard images, which comprises: Placing the CPU motherboard on the disassembly and assembly equipment for preliminary positioning and corner calibration, and collecting multi-view time-series images through a camera, wherein the disassembly and assembly equipment includes a disassembly and assembly platform and a robotic arm; Calculate the depth of each perspective time-series image and perform image alignment to calculate the comprehensive depth. Use neural radiation field to generate a 3D point cloud based on the comprehensive depth. Use dynamic region segmentation algorithm to segment the 3D point cloud and extract CPU point cloud data after verification. Map the CPU point cloud data to the latent space to extract potential features, extract local-global features through local-global comparison optimization, and use deep learning algorithms to detect CPU integrity; Deviation positioning detection is performed based on the CPU point cloud data, and the disassembly position is optimized and adjusted, and a disassembly record is generated for storage.

[0007] As a preferred solution of the CPU disassembly and assembly positioning method based on real-time motherboard images of the present invention, wherein: the depth calculation of each perspective time-series image and the image alignment calculation of the comprehensive depth refer to using OpenCV to calculate the intrinsic parameter matrix and extrinsic parameter matrix of the camera of each perspective according to the corner point positions in each frame of the time-series image of each perspective; Combine the intrinsic parameter matrix and the extrinsic parameter matrix to obtain the reprojection matrix P;

[0008] The time series image of each view is projected through the reprojection matrix, and the perspective projection is aligned according to the calibration corner point. The depth of the pixel points in the time series image of each view in each frame is calculated separately, and the average value of the corresponding pixel point depth of the time series image of all views is calculated as the comprehensive depth of each frame pixel point (x, y) .

[0009] As a preferred solution of the CPU disassembly and positioning method based on real-time motherboard image of the present invention, wherein: the neural radiation field is used to generate a three-dimensional point cloud according to the comprehensive depth, and the point cloud is segmented and verified based on the three-dimensional point cloud using the dynamic region segmentation algorithm, and the CPU point cloud data is extracted after the comprehensive depth of each frame pixel is divided into the three-dimensional point cloud and the point cloud is segmented and verified. Generate a 3D point cloud as the z-axis coordinate, mark the time sequence t for each frame of the 3D point cloud, and calculate the depth difference of each pixel in consecutive frames And sum the pixel depth differences of all frames to obtain the temporal depth difference ; For each pixel in the continuous frames, calculate its depth variance in the continuous frame images ; Use the gradient calculation method to calculate the normal corresponding to the pixel point in each frame ; Based on pixel normals Calculate the normal difference for each pixel ; The temporal depth difference of each pixel , depth variance and normal differences Dynamic eigenvalues ​​are calculated using weighted summation using the dynamic region segmentation algorithm. ; According to the dynamic characteristic value Compare with the preset threshold to assign pixel labels. If the pixel value is greater than the preset threshold, the pixel is assigned a label of 1, otherwise it is assigned a label of 0. All pixels with the label of 1 are extracted to form a preliminary classification set; For each pixel in the preliminary classification set, a neighborhood radius r is set, and the pixels within the neighborhood radius are regarded as adjacent points. For each pixel, its dynamic object weight relative to the adjacent points is calculated. ; Synchronously calculate the normal difference between the current pixel and the adjacent points ; Finally, the weighted standard deviation is calculated based on the temporal depth difference between the pixel point and its adjacent points. ; Weighting Dynamic Objects , Normal Difference and weighted standard deviation Perform weighted summation and compare it with the set segmentation threshold. If the weighted summation result is greater than the set segmentation threshold, the point cloud point corresponding to the pixel is added to the dynamic area point cloud set, and all pixel points in the preliminary classification set are traversed to form a dynamic area point cloud set. The point cloud in the dynamic area point cloud set is used as the CPU point cloud.

[0010] As a preferred solution of the CPU disassembly and assembly positioning method based on real-time motherboard images of the present invention, wherein: mapping the CPU point cloud data to a latent space to extract potential features, extracting local-global features through local-global comparison optimization, and using a deep learning algorithm to detect CPU integrity refers to mapping the CPU point cloud data to a latent space to extract a potential feature representation z through a variational autoencoder; Extract global features based on the latent feature representation z through the first convolutional neural network , global style contrast learning is used to calculate the convolution channel mean and variance to form a global feature representation ; Construct a global contrast loss function based on global features and global feature representation ; At the same time, the potential space is divided into regions, and the local features are extracted according to the potential feature representation z through the second convolutional neural network And construct the local contrast loss function ; Comprehensive global contrast loss function and local contrast loss function Constructing the total loss function ; Based on the total loss function, the gradient descent method and the Adam optimizer are used to iteratively train the first convolutional neural network and the second convolutional neural network respectively. The training is stopped when the total loss function converges. The global features and local features are obtained by the trained first convolutional neural network and the second convolutional neural network respectively. A support vector machine model is constructed for classification detection. The model input is set as global features and local features, and the output is CPU integrity. The support vector machine model is trained using training data, and the global features and local features are input into the trained support vector machine model to obtain CPU integrity. If the CPU is complete, deviation positioning is continued. If it is incomplete, a defect warning is issued to alert the staff.

[0011] As a preferred solution of the CPU disassembly and assembly positioning method based on real-time motherboard images described in the present invention, the method includes placing the CPU motherboard on the disassembly and assembly equipment for preliminary positioning and calibrating corner points, and collecting multi-view time-series images through a camera, which means placing the CPU motherboard on the disassembly and assembly platform and manually performing preliminary positioning, and obtaining multi-view time-series images by installing a multi-view camera on a robotic arm during the positioning process.

[0012] As a preferred solution of the CPU disassembly and assembly positioning method based on real-time motherboard images described in the present invention, the deviation positioning detection and optimization adjustment of the disassembly position based on the CPU point cloud data refers to calculating the distance deviation value based on the robot arm position coordinates and the CPU point cloud coordinates after performing the CPU integrity detection. If the distance deviation value is greater than the set threshold, the positioning deviation between the robot arm and the CPU is judged, and the shortest path planning algorithm is used to plan the robot arm movement path based on the robot arm position coordinates and the CPU point cloud coordinates, and the robot arm is controlled to move to optimize the disassembly position.

[0013] As a preferred solution of the CPU disassembly and assembly positioning method based on real-time motherboard image described in the present invention, wherein: the generating of disassembly records for storage refers to generating and storing disassembly records of the CPU point cloud data and the robot arm movement path after the CPU is disassembled.

[0014] In a second aspect, the present invention provides a CPU disassembly and assembly positioning system based on real-time motherboard images, comprising: The data collection module is used to place the CPU motherboard on the disassembly and assembly equipment for preliminary positioning and corner calibration, and collect multi-view time-series images through the camera; The point cloud extraction module is used to calculate the depth of each perspective time-series image and perform image alignment to calculate the comprehensive depth. It uses the neural radiation field to generate a 3D point cloud based on the comprehensive depth. Based on the 3D point cloud, it uses the dynamic region segmentation algorithm to perform point cloud segmentation and extract CPU point cloud data after verification. The CPU detection module is used to map the CPU point cloud data to the latent space to extract potential features, extract local-global features through local-global comparison optimization, and use deep learning algorithms to detect CPU integrity; The storage module is adjusted to detect deviation positioning based on CPU point cloud data, optimize and adjust the disassembly position, and generate disassembly records for storage.

[0015] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the CPU disassembly and positioning method based on real-time motherboard images as described in the first aspect of the present invention is implemented.

[0016] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the CPU disassembly and positioning method based on real-time motherboard image as described in the first aspect of the present invention.

[0017] The beneficial effects of the present invention are as follows: the present invention uses a neural radiation field to generate a three-dimensional point cloud, and combines it with a local-global contrast optimization algorithm to perform depth calculation and alignment of images from multiple perspectives, thereby greatly improving the CPU positioning and disassembly accuracy. It also extracts accurate CPU point cloud data based on a dynamic region segmentation algorithm, and can achieve automated and high-precision CPU positioning and integrity detection during the disassembly process. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 This is a flow chart of the CPU disassembly and assembly positioning method based on real-time motherboard images in Example 1.

[0020] Figure 2 This is a structural diagram of the CPU disassembly and positioning system based on real-time motherboard images in Example 1. DETAILED DESCRIPTION

[0021] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0022] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0023] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0024] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a CPU disassembly and positioning method based on real-time motherboard images, comprising the following steps: S1. Place the CPU motherboard on the disassembly and assembly equipment for preliminary positioning and corner calibration, and collect multi-view time-series images through a camera. The disassembly and assembly equipment includes a disassembly and assembly platform and a robotic arm; Specifically, the CPU mainboard is placed on the disassembly and assembly equipment for preliminary positioning and corner calibration, and multi-view time-series images are collected through the camera. This means that the CPU mainboard is placed on the disassembly and assembly platform and manually preliminarily positioned. During the positioning process, a multi-view camera is installed on the robotic arm to obtain multi-view time-series images.

[0025] S2. Calculate the depth of each perspective time-series image and perform image alignment to calculate the comprehensive depth. Generate a 3D point cloud based on the comprehensive depth using the neural radiation field. Use the dynamic region segmentation algorithm based on the 3D point cloud to segment the point cloud and extract the CPU point cloud data after verification. Specifically, calculating the depth of each viewpoint sequential image and performing image alignment to calculate the comprehensive depth means using OpenCV to calculate the intrinsic parameter matrix and extrinsic parameter matrix of the viewpoint camera based on the corner point positions in each frame of the sequential image of each viewpoint. The extrinsic parameter matrix includes the rotation matrix R and the displacement vector T. The intrinsic parameter matrix K is usually: ; in and are the horizontal and vertical focal lengths, and is the corner point position; The intrinsic parameter matrix and the extrinsic parameter matrix are combined to obtain the reprojection matrix P, which is expressed as: ; The time series image of each view is projected through the reprojection matrix, and the perspective projection is aligned according to the calibration corner point. The depth of the pixel points in the time series image of each view in each frame is calculated separately, and the average value of the corresponding pixel point depth of the time series image of all views is calculated as the comprehensive depth of each frame pixel point (x, y) .

[0026] Through precise calculation of the internal and external parameter matrices and generation of the reprojection matrix, images captured from multiple perspectives can be synthesized in a unified coordinate system. The depth information of each perspective image is precisely integrated in this way, avoiding errors caused by perspective differences. This process ensures the high accuracy of the final depth map, providing reliable data support for subsequent 3D reconstruction and object detection. The aligned images can be comprehensively calculated based on the depth information of each perspective, ensuring that the depth value of each pixel is highly consistent with its position in the actual 3D space. This process enhances the accuracy of the depth map, allowing multi-view data to provide more accurate physical space information when performing further 3D reconstruction or detection. By calculating the pixel depth of each frame and combining the depth data from the time-series images of all perspectives, the comprehensive depth of each pixel is obtained. This method can effectively eliminate the errors caused by a single perspective and improve the accuracy and reliability of the image depth information. The final comprehensive depth calculation can provide more realistic and accurate object depth data, which is of great significance for subsequent object recognition, positioning, and disassembly processes.

[0027] Furthermore, the neural radiation field is used to generate a three-dimensional point cloud based on the comprehensive depth. The dynamic region segmentation algorithm is used based on the three-dimensional point cloud to segment the point cloud and extract the CPU point cloud data after verification. The comprehensive depth of each pixel in each frame is calculated. Generate a 3D point cloud as the z-axis coordinate, mark the time sequence t for each frame of the 3D point cloud, and calculate the depth difference of each pixel in consecutive frames And sum the pixel depth differences of all frames to obtain the temporal depth difference : ; in is the pixel depth value of the t-1 frame, is the total number of frames; For each pixel in the continuous frames, calculate its depth variance in the continuous frame images ; Use the gradient calculation method to calculate the normal corresponding to the pixel point in each frame : ; Based on pixel normals Calculate the normal difference for each pixel : ; The temporal depth difference of each pixel , depth variance and normal differences Dynamic eigenvalues ​​are calculated using weighted summation using the dynamic region segmentation algorithm. : ; in is the weight, which is used to balance the impact of normal difference on the segmentation result; According to the dynamic characteristic value Compare with the preset threshold to assign pixel labels. If the pixel value is greater than the preset threshold, the pixel is assigned a label of 1, otherwise it is assigned a label of 0. All pixels with the label of 1 are extracted to form a preliminary classification set; For each pixel in the preliminary classification set, a neighborhood radius r is set, and the pixels within the neighborhood radius are regarded as adjacent points. For each pixel, its dynamic object weight relative to the adjacent points is calculated. : ; Where M is the total number of adjacent points, is the time depth difference between the i-th adjacent points; Synchronously calculate the normal difference between the current pixel and the adjacent points : ; in is the normal difference between the i-th adjacent points; Finally, the weighted standard deviation is calculated based on the temporal depth difference between the pixel point and its adjacent points. : ; Weighting Dynamic Objects , Normal Difference and weighted standard deviation Perform weighted summation and compare it with the set segmentation threshold. If the weighted summation result is greater than the set segmentation threshold, the point cloud point corresponding to the pixel is added to the dynamic area point cloud set, and all pixel points in the preliminary classification set are traversed to form a dynamic area point cloud set. The point cloud in the dynamic area point cloud set is used as the CPU point cloud.

[0028] The use of NeRF enables the present invention to accurately generate three-dimensional point clouds in complex dynamic environments, especially in conditions with different lighting, occlusions, and viewing angles. This plays an important role in subsequent dynamic object detection and CPU point cloud extraction. Through high-quality point cloud generation, the system can obtain more accurate object shapes during the disassembly and assembly process, laying the foundation for further positioning, classification, and analysis. Through the calculation of time-depth difference, the present invention can accurately capture the depth changes of objects at different time points, thereby distinguishing dynamic objects from static backgrounds. This depth difference calculation based on time changes effectively improves the system's adaptability in real-time dynamic environments, and can dynamically update depth map data, thereby ensuring CPU Accurate identification and tracking of dynamic objects during the disassembly and assembly process, and the calculation of normal differences can effectively identify changes in the object surface in continuous frames, especially for objects with complex shapes and irregular movements. This method can more accurately capture the characteristic changes on the object surface and distinguish dynamic objects from the background. Combined with the dynamic area segmentation algorithm, the system can accurately extract dynamic areas to ensure high-quality presentation of dynamic objects in three-dimensional reconstruction. The dynamic area segmentation algorithm performs weighted calculations on the temporal depth difference, normal difference and depth variance of each pixel, making the algorithm highly adaptable and robust in dynamic environments. Through this method, the system can efficiently segment the point cloud of dynamic objects and accurately extract target point cloud data during the CPU disassembly and assembly process, thereby improving the system's disassembly and assembly accuracy and efficiency.

[0029] S3, mapping the CPU point cloud data to the latent space to extract potential features, extracting local-global features through local-global comparison optimization, and using deep learning algorithms to detect CPU integrity; Specifically, the CPU point cloud data is mapped to the latent space to extract potential features, local-global features are extracted through local-global comparison optimization, and the CPU integrity is detected using a deep learning algorithm. The CPU point cloud data is mapped to the latent space through a variational autoencoder to extract the latent feature representation z: ; in is a variational autoencoder, are the variational autoencoder parameters, It is CPU point cloud data; Extract global features based on the latent feature representation z through the first convolutional neural network , global style contrast learning is used to calculate the convolution channel mean and variance to form a global feature representation : ; in and are the convolution channel mean and variance, respectively, obtained through the first convolutional neural network; Construct a global contrast loss function based on global features and global feature representation : ; in Represent z as the latent feature and the global feature The cosine similarity of is the loss parameter; At the same time, the potential space is divided into regions, and the local features are extracted according to the potential feature representation z through the second convolutional neural network And construct the local contrast loss function : ; in Local features for the i-th and j-th latent space partition regions The cosine similarity of Comprehensive global contrast loss function and local contrast loss function Constructing the total loss function : ; Where λ is a hyperparameter used to balance global and local feature learning; Based on the total loss function, the gradient descent method and the Adam optimizer are used to iteratively train the first convolutional neural network and the second convolutional neural network respectively. The training is stopped when the total loss function converges. The global features and local features are obtained by the trained first convolutional neural network and the second convolutional neural network respectively. A support vector machine model is constructed for classification detection. The model input is set as global features and local features, and the output is CPU integrity. The support vector machine model is trained using training data, and the global features and local features are input into the trained support vector machine model to obtain CPU integrity. If the CPU is complete, deviation positioning is continued. If it is incomplete, a defect warning is issued to alert the staff.

[0030] By using a variational autoencoder to map point cloud data to a latent space, the present invention can extract a more compact and representative feature representation from high-dimensional data. This feature representation helps reduce redundant information, making subsequent global and local feature extraction more efficient. It also helps improve the generalization ability of the model and reduce overfitting. By using the first and second convolutional neural networks to extract global features and local features respectively, the present invention can simultaneously capture the overall shape and local damage features of the CPU. This joint extraction of global and local features makes the model more comprehensive when detecting CPU integrity and can accurately identify possible local damage or defects in the CPU. Global style contrast learning improves the modeling ability of the overall features of the object by calculating the mean and variance of the convolution channel, while the local contrast loss enables the model to focus on the detailed areas of the CPU. This method ensures the effective learning of different features and can improve the accuracy of the model in detecting CPU integrity, especially when processing CPUs with complex shapes, and can effectively identify their global and local defects. By using a support vector machine to classify the extracted global features and local features, the present invention can accurately detect CPU The introduction of support vector machine not only improves the classification accuracy, but also enhances the robustness of the model, which can effectively deal with different CPU models and damage conditions, thereby providing reliable decision support for automated disassembly and assembly.

[0031] S4. Perform deviation positioning detection and optimize the disassembly position based on the CPU point cloud data, and generate a disassembly record for storage; Specifically, performing deviation positioning detection and optimizing and adjusting the disassembly position based on the CPU point cloud data means calculating the distance deviation value based on the robot arm position coordinates and the CPU point cloud coordinates after performing the CPU integrity detection. If the distance deviation value is greater than the set threshold, the positioning deviation between the robot arm and the CPU is judged. The shortest path planning algorithm is used according to the robot arm position coordinates and the CPU point cloud coordinates to plan the robot arm movement path and control the robot arm to move to optimize the disassembly position.

[0032] By utilizing the precise matching of the CPU's three-dimensional point cloud data and the robotic arm's position and comparing the deviation between the two, the deviation between the robotic arm and the target CPU can be dynamically detected and corrected to ensure the accuracy of the position during the disassembly process. The shortest path planning algorithm automatically calculates the optimal movement path of the robotic arm, avoiding unnecessary and redundant movement of the robotic arm during the disassembly process and improving the overall disassembly efficiency. Especially when the robot's working space is limited or the disassembly environment is complex, the algorithm can effectively avoid collisions or path intersections between the robotic arm and other objects, further improving the safety and accuracy of the disassembly process.

[0033] Furthermore, generating a disassembly record for storage refers to generating a disassembly record for storage using the CPU point cloud data and the robot arm movement path after the CPU is disassembled.

[0034] This embodiment also provides a CPU disassembly and assembly positioning system based on real-time motherboard images, including: The data collection module is used to place the CPU motherboard on the disassembly and assembly equipment for preliminary positioning and corner calibration, and collect multi-view time-series images through the camera; The point cloud extraction module is used to calculate the depth of each perspective time-series image and perform image alignment to calculate the comprehensive depth. It uses the neural radiation field to generate a 3D point cloud based on the comprehensive depth. Based on the 3D point cloud, it uses the dynamic region segmentation algorithm to perform point cloud segmentation and extract CPU point cloud data after verification. The CPU detection module is used to map the CPU point cloud data to the latent space to extract potential features, extract local-global features through local-global comparison optimization, and use deep learning algorithms to detect CPU integrity; The storage module is adjusted to detect deviation positioning based on CPU point cloud data, optimize and adjust the disassembly position, and generate disassembly records for storage.

[0035] This embodiment also provides a computer device, which is suitable for the CPU disassembly and positioning method based on real-time motherboard images, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the CPU disassembly and positioning method based on real-time motherboard images proposed in the above embodiment.

[0036] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0037] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the CPU disassembly and positioning method based on a real-time motherboard image as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0038] In summary, the present invention uses a neural radiation field to generate a three-dimensional point cloud, and combines it with a local-global contrast optimization algorithm to perform depth calculation and alignment of images from multiple perspectives, thereby greatly improving the accuracy of CPU positioning and disassembly. It also extracts accurate CPU point cloud data based on a dynamic region segmentation algorithm, and can achieve automated and high-precision CPU positioning and integrity detection during the disassembly and assembly process.

[0039] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A CPU disassembly and positioning method based on real-time motherboard images, characterized by: include, Placing the CPU motherboard on the disassembly and assembly equipment for preliminary positioning and corner calibration, and collecting multi-view time-series images through a camera, wherein the disassembly and assembly equipment includes a disassembly and assembly platform and a robotic arm; Calculate the depth of each perspective time-series image and perform image alignment to calculate the comprehensive depth. Use neural radiation field to generate a 3D point cloud based on the comprehensive depth. Use dynamic region segmentation algorithm to segment the 3D point cloud and extract CPU point cloud data after verification. Map the CPU point cloud data to the latent space to extract potential features, extract local-global features through local-global comparison optimization, and use deep learning algorithms to detect CPU integrity; Deviation positioning detection is performed based on the CPU point cloud data, and the disassembly position is optimized and adjusted, and a disassembly record is generated for storage.

2. The CPU disassembly and positioning method based on real-time motherboard image according to claim 1, characterized in that: Calculating the depth of each viewpoint sequential image and performing image alignment to calculate the comprehensive depth refers to using OpenCV to calculate the intrinsic parameter matrix and extrinsic parameter matrix of the viewpoint camera according to the corner point positions in each frame of the sequential image of each viewpoint; Combine the intrinsic parameter matrix and the extrinsic parameter matrix to obtain the reprojection matrix P; The time series image of each view is projected through the reprojection matrix, and the perspective projection is aligned according to the calibration corner point. The depth of the pixel points in the time series image of each view in each frame is calculated separately, and the average value of the corresponding pixel point depth of the time series image of all views is calculated as the comprehensive depth of each frame pixel point (x, y) .

3. The CPU disassembly and positioning method based on real-time motherboard image according to claim 2, characterized in that: The neural radiation field is used to generate a three-dimensional point cloud according to the comprehensive depth, and the point cloud is segmented and verified based on the three-dimensional point cloud using a dynamic region segmentation algorithm. The CPU point cloud data is extracted after the comprehensive depth of each pixel in the frame is converted into the point cloud data. Generate a 3D point cloud as the z-axis coordinate, mark the time sequence t for each frame of the 3D point cloud, and calculate the depth difference of each pixel in consecutive frames And sum the pixel depth differences of all frames to obtain the temporal depth difference ; For each pixel in the continuous frames, calculate its depth variance in the continuous frame images ; Use the gradient calculation method to calculate the normal corresponding to the pixel point in each frame ; Based on pixel normals Calculate the normal difference for each pixel ; The temporal depth difference of each pixel , depth variance and normal differences Dynamic eigenvalues ​​are calculated using weighted summation using the dynamic region segmentation algorithm. ; According to the dynamic eigenvalue Compare with the preset threshold to assign pixel labels. If the pixel value is greater than the preset threshold, the pixel is assigned a label of 1, otherwise it is assigned a label of 0. All pixels with the label of 1 are extracted to form a preliminary classification set; For each pixel in the preliminary classification set, a neighborhood radius r is set, and the pixels within the neighborhood radius are regarded as adjacent points. For each pixel, its dynamic object weight relative to the adjacent points is calculated. ; Synchronously calculate the normal difference between the current pixel and the adjacent points ; Finally, the weighted standard deviation is calculated based on the temporal depth difference between the pixel point and its adjacent points. ; Weighting Dynamic Objects , Normal Difference and weighted standard deviation Perform weighted summation and compare it with the set segmentation threshold. If the weighted summation result is greater than the set segmentation threshold, the point cloud point corresponding to the pixel is added to the dynamic area point cloud set, and all pixel points in the preliminary classification set are traversed to form a dynamic area point cloud set. The point cloud in the dynamic area point cloud set is used as the CPU point cloud.

4. The CPU disassembly and positioning method based on real-time motherboard image according to claim 3, characterized in that: Mapping the CPU point cloud data to a latent space to extract potential features, extracting local-global features through local-global comparison optimization, and detecting CPU integrity using a deep learning algorithm refers to mapping the CPU point cloud data to a latent space to extract a potential feature representation z through a variational autoencoder; Extract global features based on the latent feature representation z through the first convolutional neural network , global style contrast learning is used to calculate the convolution channel mean and variance to form a global feature representation ; Construct a global contrast loss function based on global features and global feature representation ; At the same time, the potential space is divided into regions, and the local features are extracted according to the potential feature representation z through the second convolutional neural network And construct the local contrast loss function ; Comprehensive global contrast loss function and local contrast loss function Constructing the total loss function ; Based on the total loss function, the gradient descent method and the Adam optimizer are used to iteratively train the first convolutional neural network and the second convolutional neural network respectively. The training is stopped when the total loss function converges. The global features and local features are obtained by the trained first convolutional neural network and the second convolutional neural network respectively. A support vector machine model is constructed for classification detection. The model input is set as global features and local features, and the output is CPU integrity. The support vector machine model is trained using training data, and the global features and local features are input into the trained support vector machine model to obtain CPU integrity. If the CPU is complete, deviation positioning is continued. If it is incomplete, a defect warning is issued to alert the staff.

5. The CPU disassembly and positioning method based on real-time motherboard image according to claim 4, characterized in that: Placing the CPU mainboard on the disassembly and assembly equipment for preliminary positioning and corner point calibration, and collecting multi-perspective time-series images through the camera means placing the CPU mainboard on the disassembly and assembly platform and manually performing preliminary positioning, and obtaining multi-perspective time-series images by installing a multi-perspective camera on the robotic arm during the positioning process.

6. The CPU disassembly and positioning method based on real-time motherboard image according to claim 5, characterized in that: The deviation positioning detection and optimization adjustment of the disassembly position based on the CPU point cloud data refers to calculating the distance deviation value based on the robot arm position coordinates and the CPU point cloud coordinates after the CPU integrity detection. If the distance deviation value is greater than the set threshold, the positioning deviation between the robot arm and the CPU is judged, and the shortest path planning algorithm is used according to the robot arm position coordinates and the CPU point cloud coordinates to plan the robot arm movement path and control the robot arm to move and optimize the disassembly position.

7. The CPU disassembly and positioning method based on real-time motherboard image according to claim 6, characterized in that: The generating and storing of the disassembly record refers to generating and storing the disassembly record of the CPU point cloud data and the movement path of the robot arm after the CPU is disassembled.

8. A CPU disassembly and positioning system based on real-time motherboard images, based on the CPU disassembly and positioning method based on real-time motherboard images according to any one of claims 1 to 7, characterized in that: include, The data collection module is used to place the CPU motherboard on the disassembly and assembly equipment for preliminary positioning and corner calibration, and collect multi-view time-series images through the camera; The point cloud extraction module is used to calculate the depth of each perspective time-series image and perform image alignment to calculate the comprehensive depth. It uses the neural radiation field to generate a 3D point cloud based on the comprehensive depth. Based on the 3D point cloud, it uses the dynamic region segmentation algorithm to perform point cloud segmentation and extract CPU point cloud data after verification. The CPU detection module is used to map CPU point cloud data to the latent space to extract potential features, extract local-global features through local-global comparison optimization, and use deep learning algorithms to detect CPU integrity; The storage module is adjusted to detect deviation positioning based on CPU point cloud data, optimize and adjust the disassembly position, and generate disassembly records for storage.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the CPU disassembly and assembly positioning method based on real-time motherboard image are implemented as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the CPU disassembly and assembly positioning method based on real-time motherboard images according to any one of claims 1 to 7 are implemented.

Images