Power distribution cabinet foot margin positioning and assembling method, system and device based on AI visual identification

Abstract

The invention discloses a power distribution cabinet foot margin positioning and assembling method, system and device based on AI visual identification, and relates to the technical field of electric power engineering, and the system comprises an image acquisition module, an analysis module and an execution module. According to the invention, the image acquisition module provides a rich basis by acquiring image data and constructing a model library, and the comparison unit calculates feature vector differences by adopting Manhattan distance, performs matching judgment by setting a threshold DA, and quickly matches corresponding foundation bolt models in the model library. An updating unit temporarily stores and marks an unknown model which fails to be matched, triggers an alarm to inform an engineer, judges whether misjudgment exists or not based on a visual tool, deletes the unknown model in time to avoid interference if the unknown model is misjudged, and stores the unknown model into a warehouse if the unknown model is a new model; the model base is continuously improved, and the overall performance and reliability of the system are improved.

Description

Technical Field

[0001] The present invention relates to the field of power engineering technology, and in particular to a method, system and device for positioning and assembling the anchors of a distribution cabinet based on AI visual recognition. Background Art

[0002] In the field of power engineering, distribution cabinets are key equipment for power distribution and control, and their installation quality directly impacts the stable operation of the power system. Anchor bolts, including foundation bolts, are crucial components connecting the distribution cabinet to the foundation. Their accurate positioning and assembly are crucial in the manufacture of distribution cabinets. Traditional methods for positioning and assembling anchor bolts rely primarily on manual labor, where workers rely on experience and simple tools to position and install them. With the continuous development of industrial automation technology, some simple mechanically assisted assembly methods have begun to emerge, such as the use of basic fixtures to assist in positioning.

[0003] In the existing technology, manual operation is affected by factors such as visual errors, operator proficiency and working status, and it is difficult to accurately control the position and posture of the anchor bolts. In actual production, the positioning deviation of the anchor bolts will cause the distribution cabinet to be unstable, and the slight deviation of the anchor bolts will cause the entire cabinet to tilt after accumulation, affecting its subsequent safety and stability. In addition, the distribution cabinet requires the use of a variety of anchor bolts of different specifications and models. Traditional methods are difficult to distinguish quickly and accurately, and assembly errors are prone to occur, further affecting product quality and production efficiency.

[0004] Therefore, a distribution cabinet anchor positioning and assembly method, system and device based on AI visual recognition are proposed to solve the above problems. Summary of the Invention

[0005] The main purpose of the present invention is to provide a method, system and device for positioning and assembling the anchors of a distribution cabinet based on AI visual recognition to solve the problems raised in the above background.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is: a distribution cabinet foot positioning and assembly system based on AI visual recognition, the system includes an image acquisition module, an analysis module and an execution module; The image acquisition module is used to capture images of anchor bolts and holes at multiple angles using an industrial camera mounted on a robotic arm, and transmit the captured image information to the analysis module via a Wi-Fi module after pre-processing. It also collects a large number of anchor bolt models and constructs a bolt model library. The analysis module is used to receive image information, extract features of the anchor bolts and the preset hole positions, identify the positions of the anchor bolts and the preset hole positions in the image, and compare the extracted feature vectors of the anchor bolts with corresponding anchor bolt models in the model library to quickly match the corresponding anchor bolt models. If the comparison fails, it is determined that the anchor bolt is a new model. The execution module is used to analyze the postures of the anchor bolts and the preset holes according to the extracted features and recognition results, and plan the motion path of the robot arm according to the postures of the anchor bolts and the positions and postures of the preset holes.

[0007] Preferably, the image acquisition module includes an acquisition unit, a pre-processing unit, a transmission unit and a construction unit; The acquisition unit is used to capture images of anchor bolts and holes at multiple angles using an industrial camera mounted on a robotic arm; The preprocessing module is used to preprocess the collected anchor bolt and hole position images, and the preprocessing includes grayscale conversion, filtering denoising and image enhancement; The transmission unit is used for the Wi-Fi module to transmit the pre-processed image information to the analysis module; The construction module is used for the data acquisition instrument to collect a large number of anchor bolt models, wherein the anchor bolt models include corresponding feature vectors and posture parameter information, and a model library is constructed based on MySQL.

[0008] Preferably, the analysis module includes an extraction unit, an identification unit, a comparison unit and an update unit; The extraction module is used to extract the features of the anchor bolts and the preset hole positions through CNN; The recognition unit is used to recognize the anchor bolts and preset hole positions in the image through YOLO.

[0009] Preferably, the comparison unit is used to compare the extracted anchor bolt features with the anchor model features in the model library by calculating the sum of the absolute differences of the two feature vectors on each coordinate axis and comparing them with a set threshold. The calculation formula is as follows: Among them, DM(x,y) represents the Manhattan distance between feature vectors x and y, which is used to measure the difference. x and y represent the extracted anchor bolt features and the anchor model features in the model library, respectively. n represents the dimension of the feature vector. i Represents the value of the i-th dimension of the feature vector x, y i Represents the value of the i-th dimension of the feature vector y; By setting the threshold to DA, when DM(x,y) is less than DA, the match is successful, and the corresponding model is the matched anchor bolt model. If there are multiple successful matches, the DM(x,y) closest to DA is selected. If DM(x,y) is greater than DA, the match fails and is marked as an unknown model.

[0010] Preferably, the update unit is used to receive unknown model information and temporarily store it in the model library and mark it as a model to be verified, while triggering an audible and visual alarm and a terminal alarm to notify engineers to intervene and make a judgment. The engineering design is used to use visualization tools to determine whether a misjudgment is caused by occlusion, deformation, damage and illumination. If it is determined to be a misjudgment, the model to be verified in the model library is deleted and the image information is re-collected. If it is determined to be a new model, the model to be verified is stored in the model library; the unknown model information includes the currently collected image and the extracted feature vector.

[0011] Preferably, the execution module includes a rotation unit, a translation change unit, an offset unit and an execution unit.

[0012] Preferably, the rotation unit constructs an initial rotation matrix based on the posture information of the corresponding anchor bolt in the model library, and the steps are as follows: S1: Construct the rotation angle θ around the x-axis x The rotation matrix R of the angle x , the formula is: S2: Construct the rotation angle θ around the y-axis y The rotation matrix R of the angle y , the formula is: S3: Construct the rotation angle θ around the z axis z The rotation matrix R of the angle z , the formula is: Among them, R x Represents the rotation of the anchor bolt around the x-axis in three-dimensional space x The rotation matrix of the angle, R y Represents the rotation of the anchor bolt around the y-axis in three-dimensional space y The rotation matrix of the angle, R z Represents the rotation of the anchor bolt around the z axis in three-dimensional space z The rotation matrix of the angle, θ x ,θ y and θ z Represents the rotation angle of the anchor bolt around the x, y and z axes respectively.

[0013] Preferably, the translation change unit is used to describe the position movement of the anchor bolt in three-dimensional space, and simultaneously synthesizes the rotation matrix and the translation matrix to describe the rotation and translation of the anchor bolt, and the steps are as follows: Step 1: Describe the position movement of the anchor bolt in three-dimensional space in the form of: Where T represents the translation matrix, tx Represents the translation of the anchor bolt along the x-axis in three-dimensional space, t y The translation of the anchor bolt along the y-axis in three-dimensional space, t z Represents the translation of the anchor bolt along the z-axis in three-dimensional space.

[0014] Step 2: Combine the rotation matrix and the translation matrix to describe the rotation and translation of the anchor bolts simultaneously. The format is: Among them, H represents the homogeneous coordinate transformation matrix, R represents the combined rotation matrix (R = R x R y R z ), T represents the translation matrix; the offset unit is used to calculate the position offset and rotation angle offset of the anchor bolt position in the image and the standard position in the model library, the steps are as follows: (1) Calculate the position offset using the following formula: Δx=x i -x s ; Δy=y i -y s ; Among them, x i represents the actual horizontal coordinate value of the anchor bolt in the two-dimensional image, x s Represents the actual vertical coordinate value of the anchor bolt in the two-dimensional image, x s Represents the horizontal coordinate value of the corresponding position in the model library, y s Represents the ordinate value of the corresponding position in the model library, Δx represents the position offset of the x-axis, and Δy represents the position offset of the y-axis; (2) Calculate the rotation angle offset. The calculation formula is as follows: Among them, θ represents the rotation angle offset, k1 represents the slope of the straight line fitting the anchor bolt contour in the acquired image, and k2 represents the slope of the straight line fitting the anchor bolt contour in the model library.

[0015] The execution unit is used to calculate the position offset and perform posture adjustment calculation in combination with the initial homogeneous coordinate transformation matrix. If there is a position offset, the translation matrix is ​​adjusted by Δx and Δy to adjust t in T. x and t y If the rotation angle offset occurs, adjust the rotation matrix θ by θ x ,θ y and θ z , and then the actual posture of the anchor bolt in the current image is calculated through the homogeneous coordinate transformation matrix.

[0016] A distribution cabinet anchor positioning and assembly device based on AI visual recognition includes a robotic arm and a storage device, wherein a computer program is stored in the storage device. When the computer program is executed by the processor, the processor executes the steps of the system.

[0017] The distribution cabinet foot positioning and assembly method based on AI visual recognition includes the following steps: Step 1: Enter the image acquisition module, use the industrial camera to take images of the anchor bolts and holes from multiple angles, pre-process them, and transfer them to the analysis module. At the same time, a foundation bolt model library is constructed. Step 2: Enter the analysis module, use CNN to extract the features of the anchor bolts and preset holes, and use YOLO to identify the anchor bolts and preset hole positions in the image. The sum of the absolute differences between the anchor bolt feature vectors collected in the image and the feature vectors in the anchor bolt model on each coordinate axis is calculated, and the sum is compared with the set threshold to quickly match the corresponding anchor bolt model and update the anchor bolt model library. Step 3: Enter the execution module, build the initial rotation matrix and translation matrix based on the posture information of the corresponding anchor bolt in the model library, merge the rotation matrix and translation matrix into a homogeneous coordinate transformation matrix, and calculate the position offset and rotation angle offset of the anchor bolt. Based on the position offset and rotation angle offset, compensate the corresponding data in the homogeneous coordinate transformation matrix and calculate the actual posture of the anchor bolt in the current image.

[0018] The present invention has the following beneficial effects: 1. In the present invention, the image acquisition module provides a rich foundation by collecting image data and building a model library. The comparison unit uses Manhattan distance to calculate the difference in eigenvectors, and performs matching judgment by setting a threshold DA. It quickly matches the corresponding anchor bolt model in the model library, reducing assembly deviations caused by incorrect bolt model judgment or inaccurate posture analysis. The update unit temporarily stores and marks unknown models that fail to match, and triggers an alarm to notify engineers. It also determines whether it is a misjudgment based on a visualization tool. If it is a misjudgment, it will be deleted in time to avoid interference. If it is a new model, it will be stored in the library, and the model library will be continuously improved to improve the overall performance and reliability of the system.

[0019] 2. In the present invention, the rotation unit constructs an initial rotation matrix based on the posture information of the model library, and accurately describes the rotation posture of the anchor bolt from the three dimensions of x, y, and z axes, ensuring that the robotic arm can accurately grasp the initial orientation of the bolt. The translation change unit describes the position movement of the anchor bolt in three-dimensional space, and synthesizes the rotation matrix and the translation matrix into a homogeneous coordinate transformation matrix to achieve a comprehensive description of the bolt rotation and translation.

[0020] 3. In the present invention, the offset unit accurately finds the difference between the anchor bolt in the image and the standard position by quantitatively calculating the position offset and the rotation angle offset. The execution unit performs posture adjustment calculation based on the offset combined with the initial homogeneous coordinate transformation matrix, and adjusts the translation matrix and the rotation matrix accordingly according to the position and rotation angle offset, thereby accurately calculating the actual posture of the bolt, ensuring that the robotic arm can operate accurately according to the actual state of the bolt, greatly improving the accuracy and automation of the anchor bolt positioning and assembly, and effectively ensuring the quality and efficiency of the distribution cabinet assembly. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flow chart of the power distribution cabinet anchor positioning and assembly method based on AI visual recognition of the present invention; Figure 2 This is a flow chart of the distribution cabinet anchor positioning and assembly system based on AI visual recognition of the present invention. DETAILED DESCRIPTION

[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0023] Implementation See also Figure 1 and Figure 2 , the present invention provides a technical solution: a distribution cabinet foot positioning and assembly system based on AI visual recognition, the system includes an image acquisition module, an analysis module and an execution module; The image acquisition module is used to capture images of anchor bolts and holes from multiple angles using an industrial camera mounted on a robotic arm. The captured image information is pre-processed and transmitted to the analysis module via the Wi-Fi module. A large number of anchor bolt models are collected and a bolt model library is constructed. The analysis module is used to receive image information, extract the features of the anchor bolts and the preset hole positions, identify the positions of the anchor bolts and the preset hole positions in the image, and compare the extracted feature vectors of the anchor bolts with the corresponding anchor bolt models in the model library to quickly match the corresponding anchor bolt models. If the comparison fails, the anchor bolt is determined to be a new model. The execution module is used to analyze the posture of the anchor bolts and the preset holes according to the extracted features and recognition results, and plan the motion path of the robot arm according to the posture of the anchor bolts and the position and posture of the preset holes.

[0024] The image acquisition module includes an acquisition unit, a pre-processing unit, a transmission unit and a construction unit; The acquisition unit is used to capture images of anchor bolts and holes at multiple angles using an industrial camera mounted on the robotic arm; The preprocessing module is used to preprocess the collected anchor bolt and hole images, including grayscale conversion, filtering and denoising, and image enhancement; The transmission unit is used by the Wi-Fi module to transmit the pre-processed image information to the analysis module; The construction module is used for the data acquisition instrument to collect a large number of anchor bolt models, which include corresponding feature vectors and posture parameter information, and build a model library based on MySQL.

[0025] In this embodiment, the acquisition unit captures images of anchor bolts and holes from multiple angles, thereby obtaining all-round visual information and avoiding feature omissions due to a single viewing angle, providing a rich data basis for subsequent accurate identification and analysis. After preprocessing, the preprocessed image information is quickly transmitted to the analysis module via the Wi-Fi module to ensure the timeliness of data flow. The construction unit collects a large number of anchor bolt models and builds a model library based on MySQL to provide rich samples for comparison and identification, thereby enhancing the adaptability and versatility of the system.

[0026] Implementation 2: Please refer to Figure 1 and Figure 2 ,The present invention provides a technical solution: based on the implementation of the first embodiment, the analysis module includes an extraction unit, an identification unit, a comparison unit and an update unit; The extraction module is used to extract the features of anchor bolts and preset hole positions through CNN; The recognition unit is used to identify the anchor bolts and preset hole positions in the image through YOLO.

[0027] The comparison unit is used to compare the extracted anchor bolt features with the anchor model features in the model library by calculating the sum of the absolute differences between the two feature vectors on each coordinate axis and comparing them with the set threshold. The calculation formula is as follows: Among them, DM(x,y) represents the Manhattan distance between feature vectors x and y, which is used to measure the difference. x and y represent the extracted anchor bolt features and the anchor model features in the model library, respectively. n represents the dimension of the feature vector. i Represents the value of the i-th dimension of the feature vector x, y i Represents the value of the i-th dimension of the feature vector y; By setting the threshold to DA, when DM(x,y) is less than DA, the match is successful, and the corresponding model is the matched anchor bolt model. If there are multiple successful matches, the DM(x,y) closest to DA is selected. If DM(x,y) is greater than DA, the match fails and is marked as an unknown model.

[0028] The update unit is used to receive unknown model information and temporarily store it in the model library and mark it as a model to be verified. At the same time, it triggers sound and light alarms and terminal alarms to notify engineers to intervene and make judgments. The engineering design is used to use visualization tools to determine whether misjudgment is caused by occlusion, deformation, damage and illumination. If it is determined to be a misjudgment, the model to be verified in the model library will be deleted and the image information will be re-collected. If it is determined to be a new model, the model to be verified will be stored in the model library; the unknown model information includes the current collected image and the extracted feature vector.

[0029] Specifically, the threshold DA is developed by analyzing a large amount of historical assembly data, statistically comparing the feature vectors of successfully assembled anchor bolts, and conducting a large number of experiments to test the impact of different DA values ​​on matching effects in different scenarios, observing indicators such as matching success rate and false positive rate, and selecting the DA value that optimizes system performance based on the experimental results.

[0030] In this embodiment, the extraction unit uses CNN technology to deeply explore the features of anchor bolts and preset holes, effectively capture complex details, and the YOLO algorithm quickly determines the positions of anchor bolts and preset holes in the image, providing accurate spatial information for robotic arm operation and ensuring the accuracy of the assembly process. The comparison unit uses Manhattan distance to calculate the difference in feature vectors, and performs matching judgment by setting a threshold DA to quickly match the corresponding anchor bolt model, reducing assembly deviations caused by incorrect bolt model judgment or inaccurate posture analysis. The update unit temporarily stores and marks unknown models that fail to match, and triggers an alarm to notify engineers. It also determines whether it is a misjudgment based on visualization tools. If it is a misjudgment, it will be deleted in time to avoid interference. If it is a new model, it will be stored in the warehouse, and the model library will be continuously improved to improve the overall performance and reliability of the system.

[0031] Implementation 3: Please refer to Figure 1 and Figure 2 The present invention provides a technical solution: based on implementation one, the execution module includes a rotation unit, a translation change unit, an offset unit and an execution unit.

[0032] The rotation unit constructs the initial rotation matrix based on the posture information of the corresponding anchor bolt in the model library. The steps are as follows: S1: Construct the rotation angle λ around the x-axis x The rotation matrix R of the angle x , the formula is: S2: Construct the rotation angle λ around the y-axis y The rotation matrix R of the angle y , the formula is: S3: Construct the rotation angle θ around the z axis zThe rotation matrix R of the angle z , the formula is: Among them, R x Represents the rotation of the anchor bolt around the x-axis in three-dimensional space x The rotation matrix of the angle, R y Represents the rotation of the anchor bolt around the y-axis in three-dimensional space y The rotation matrix of the angle, R z Represents the rotation of the anchor bolt around the z axis in three-dimensional space z The rotation matrix of the angle, θ x ,θ y and θ z Represents the rotation angle of the anchor bolt around the x, y and z axes respectively.

[0033] The translation change unit is used to describe the position movement of the anchor bolt in three-dimensional space. The rotation matrix and the translation matrix are synthesized to describe the rotation and translation of the anchor bolt. The steps are as follows: Step 1: Describe the position movement of the anchor bolt in three-dimensional space in the form of: Where T represents the translation matrix, t x Represents the translation of the anchor bolt along the x-axis in three-dimensional space, t y The translation of the anchor bolt along the y-axis in three-dimensional space, t z Represents the translation of the anchor bolt along the z-axis in three-dimensional space.

[0034] Step 2: Combine the rotation matrix and the translation matrix to describe the rotation and translation of the anchor bolts simultaneously. The format is: Among them, H represents the homogeneous coordinate transformation matrix, R represents the combined rotation matrix (R = R x R y R z ), T represents the translation matrix; The offset unit is used to calculate the position offset and rotation angle offset between the anchor bolt position in the image and the standard position in the model library. The steps are as follows: (1) Calculate the position offset using the following formula: Δx=x i -x s ; Δy=y i -y s ; Among them, x i represents the actual horizontal coordinate value of the anchor bolt in the two-dimensional image, x sRepresents the actual vertical coordinate value of the anchor bolt in the two-dimensional image, x s Represents the horizontal coordinate value of the corresponding position in the model library, y s Represents the ordinate value of the corresponding position in the model library, Δx represents the position offset of the x-axis, and Δy represents the position offset of the y-axis; (2) Calculate the rotation angle offset. The calculation formula is as follows: Among them, θ represents the rotation angle offset, k1 represents the slope of the straight line fitting the anchor bolt contour in the acquired image, and k2 represents the slope of the straight line fitting the anchor bolt contour in the model library.

[0035] The execution unit is used to calculate the position offset and perform attitude adjustment calculations based on the initial homogeneous coordinate transformation matrix. If there is a position offset, the translation matrix is ​​adjusted by Δx and Δy to adjust t in T. x and t y If the rotation angle offset occurs, adjust the rotation matrix θ by θ x ,θ y and θ z , and then the actual posture of the anchor bolt in the current image is calculated through the homogeneous coordinate transformation matrix.

[0036] Specifically, by calculating the actual posture of the anchor bolt in the current image and using algorithm A, combined with the actual posture of the bolt and assembly requirements, a collision-free, efficient and precise motion path for the robotic arm is planned, allowing the robotic arm to approach and operate the bolt with an appropriate posture.

[0037] In this embodiment, the rotation unit constructs an initial rotation matrix based on the posture information of the model library, and accurately describes the rotation posture of the anchor bolt from the three dimensions of the x, y, and z axes to ensure that the robotic arm can accurately grasp the initial orientation of the bolt. The translation change unit describes the position movement of the anchor bolt in three-dimensional space, and combines the rotation matrix and the translation matrix into a homogeneous coordinate transformation matrix to achieve a comprehensive description of the bolt rotation and translation. The offset unit accurately finds the difference between the anchor bolt in the image and the standard position by quantitatively calculating the position offset and the rotation angle offset. The execution unit performs posture adjustment calculation based on the offset combined with the initial homogeneous coordinate transformation matrix, and adjusts the translation matrix and the rotation matrix accordingly based on the position and rotation angle offset, thereby accurately calculating the actual posture of the bolt, ensuring that the robotic arm can accurately operate according to the actual state of the bolt, greatly improving the accuracy and automation of the anchor bolt positioning and assembly, and effectively ensuring the quality and efficiency of the distribution cabinet assembly.

[0038] Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations may be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. The power distribution cabinet anchor positioning and assembly system based on AI visual recognition is characterized by: The system includes an image acquisition module, an analysis module and an execution module; The image acquisition module is used to capture images of anchor bolts and holes at multiple angles using an industrial camera mounted on a robotic arm, and transmit the captured image information to the analysis module via a Wi-Fi module after pre-processing. It also collects a large number of anchor bolt models and constructs a bolt model library. The analysis module is used to receive image information, extract features of the anchor bolts and the preset hole positions, identify the positions of the anchor bolts and the preset hole positions in the image, and compare the extracted feature vectors of the anchor bolts with corresponding anchor bolt models in the model library to quickly match the corresponding anchor bolt models. If the comparison fails, it is determined that the anchor bolt is a new model. The execution module is used to analyze the postures of the anchor bolts and the preset holes according to the extracted features and recognition results, and plan the motion path of the robot arm according to the postures of the anchor bolts and the positions and postures of the preset holes.

2. The power distribution cabinet anchor positioning and assembly system based on AI visual recognition according to claim 1 is characterized in that: The image acquisition module includes an acquisition unit, a pre-processing unit, a transmission unit and a construction unit; The acquisition unit is used to capture images of anchor bolts and holes at multiple angles using an industrial camera mounted on a robotic arm; The preprocessing module is used to preprocess the collected anchor bolt and hole position images, and the preprocessing includes grayscale conversion, filtering denoising and image enhancement; The transmission unit is used for the Wi-Fi module to transmit the pre-processed image information to the analysis module; The construction module is used for the data acquisition instrument to collect a large number of anchor bolt models, wherein the anchor bolt models include corresponding feature vectors and posture parameter information, and a model library is constructed based on MySQL.

3. The power distribution cabinet anchor positioning and assembly system based on AI visual recognition according to claim 1 is characterized in that: The analysis module includes an extraction unit, an identification unit, a comparison unit and an update unit; The extraction module is used to extract the features of the anchor bolts and the preset hole positions through CNN; The recognition unit is used to recognize the anchor bolts and preset hole positions in the image through YOLO.

4. The power distribution cabinet anchor positioning and assembly system based on AI visual recognition according to claim 3 is characterized in that: The comparison unit is used to compare the extracted anchor bolt features with the anchor model features in the model library by calculating the sum of the absolute differences of the two feature vectors on each coordinate axis and comparing them with the set threshold. The calculation formula is as follows: Among them, DM(x,y) represents the Manhattan distance between feature vectors x and y, which is used to measure the difference. x and y represent the extracted anchor bolt features and the anchor model features in the model library, respectively. n represents the dimension of the feature vector. i Represents the value of the i-th dimension of the feature vector x, y i Represents the value of the i-th dimension of the feature vector y; By setting the threshold to DA, when DM(x,y) is less than DA, the match is successful, and the corresponding model is the matched anchor bolt model. If there are multiple successful matches, the DM(x,y) closest to DA is selected. If DM(x,y) is greater than DA, the match fails and is marked as an unknown model.

5. The power distribution cabinet anchor positioning and assembly system based on AI visual recognition according to claim 4 is characterized in that: The update unit is used to receive unknown model information and temporarily store it in the model library and mark it as a model to be verified. At the same time, it triggers an audible and visual alarm and a terminal alarm to notify engineers to intervene and make a judgment. The engineering design is used to use visualization tools to determine whether a misjudgment is caused by occlusion, deformation, damage and illumination. If it is determined to be a misjudgment, the model to be verified in the model library is deleted and the image information is re-collected. If it is determined to be a new model, the model to be verified is stored in the model library; the unknown model information includes the currently collected image and the extracted feature vector.

6. The power distribution cabinet anchor positioning and assembly system based on AI visual recognition according to claim 1 is characterized in that: The execution module includes a rotation unit, a translation change unit, an offset unit and an execution unit.

7. The power distribution cabinet anchor positioning and assembly system based on AI visual recognition according to claim 6 is characterized in that: The rotation unit constructs an initial rotation matrix based on the posture information of the corresponding anchor bolt in the model library, and the steps are as follows: S1: Construct the rotation angle θ around the x-axis x The rotation matrix R of the angle x , the formula is: S2: Construct the rotation angle θ around the y-axis y The rotation matrix R of the angle y , the formula is: S3: Construct the rotation angle θ around the z axis z The rotation matrix R of the angle z , the formula is: Among them, R x Represents the rotation of the anchor bolt around the x-axis in three-dimensional space x The rotation matrix of the angle, R y Represents the rotation of the anchor bolt around the y-axis in three-dimensional space y The rotation matrix of the angle, R z Represents the rotation of the anchor bolt around the z axis in three-dimensional space z The rotation matrix of the angle, θ x ,θ y and θ z Represents the rotation angle of the anchor bolt around the x, y and z axes respectively.

8. The power distribution cabinet anchor positioning and assembly system based on AI visual recognition according to claim 6 is characterized in that: The translation change unit is used to describe the position movement of the anchor bolt in three-dimensional space. The rotation matrix and the translation matrix are synthesized to describe the rotation and translation of the anchor bolt. The steps are as follows: Step 1: Describe the position movement of the anchor bolt in three-dimensional space in the form of: Where T represents the translation matrix, t x Represents the translation of the anchor bolt along the x-axis in three-dimensional space, t y The translation of the anchor bolt along the y-axis in three-dimensional space, t z Represents the translation of the anchor bolt along the z-axis in three-dimensional space; Step 2: Combine the rotation matrix and the translation matrix to describe the rotation and translation of the anchor bolts simultaneously. The format is: Among them, H represents the homogeneous coordinate transformation matrix, R represents the combined rotation matrix (R = R x R y R z ), T represents the translation matrix; The offset unit is used to calculate the position offset and rotation angle offset between the anchor bolt position in the image and the standard position in the model library, and the steps are as follows: (1) Calculate the position offset using the following formula: Δx=x i -x s ;Δy=y i -y s ; Among them, x i represents the actual horizontal coordinate value of the anchor bolt in the two-dimensional image, x s Represents the actual vertical coordinate value of the anchor bolt in the two-dimensional image, x s Represents the horizontal coordinate value of the corresponding position in the model library, y s Represents the ordinate value of the corresponding position in the model library, Δx represents the position offset of the x-axis, and Δy represents the position offset of the y-axis; (2) Calculate the rotation angle offset. The calculation formula is as follows: Where θ represents the rotation angle offset, k1 represents the slope of the straight line fitting the anchor bolt contour in the acquired image, and k2 represents the slope of the straight line fitting the anchor bolt contour in the model library; The execution unit is used to calculate the position offset and perform posture adjustment calculation in combination with the initial homogeneous coordinate transformation matrix. If there is a position offset, the translation matrix is ​​adjusted by Δx and Δy to adjust t in T. x and t y If the rotation angle offset occurs, adjust the rotation matrix θ by θ x ,θ y and θ z , and then the actual posture of the anchor bolt in the current image is calculated through the homogeneous coordinate transformation matrix.

9. A power distribution cabinet anchor positioning and assembly device based on AI visual recognition, referring to the power distribution cabinet anchor positioning and assembly system based on AI visual recognition according to any one of claims 1-8, characterized in that: The system comprises a robotic arm and a storage, wherein a computer program is stored in the storage, and when the computer program is executed by the processor, the processor executes the steps of the system according to any one of claims 1 to 8.

10. A method for positioning and assembling the anchors of a power distribution cabinet based on AI visual recognition, referring to the system for positioning and assembling the anchors of a power distribution cabinet based on AI visual recognition according to any one of claims 1 to 9, characterized in that: The following steps are included: Step 1: Enter the image acquisition module, use the industrial camera to take images of the anchor bolts and holes from multiple angles, pre-process them, and transfer them to the analysis module. At the same time, a foundation bolt model library is constructed. Step 2: Enter the analysis module, use CNN to extract the features of the anchor bolts and preset holes, and use YOLO to identify the anchor bolts and preset hole positions in the image. The sum of the absolute differences between the anchor bolt feature vectors collected in the image and the feature vectors in the anchor bolt model on each coordinate axis is calculated, and the sum is compared with the set threshold to quickly match the corresponding anchor bolt model and update the anchor bolt model library. Step 3: Enter the execution module, build the initial rotation matrix and translation matrix based on the posture information of the corresponding anchor bolt in the model library, merge the rotation matrix and translation matrix into a homogeneous coordinate transformation matrix, and calculate the position offset and rotation angle offset of the anchor bolt. Based on the position offset and rotation angle offset, compensate the corresponding data in the homogeneous coordinate transformation matrix and calculate the actual posture of the anchor bolt in the current image.

Images