Cement pole steel cage quality inspection device and inspection method based on machine vision

Abstract

The present invention discloses a quality inspection device and method for cement pole steel cage based on machine vision, and relates to the field of quality inspection. The following scheme is proposed, including a detection unit and a marking mechanism, the marking mechanism includes an electric push rod and a marking gun, the marking gun is fixed on the output end of the electric push rod, the electric push rod is used to extend and retract to push the marking gun to leave a mark on the steel cage, and the detection unit includes: a first acquisition module, used to collect weld spot images; a weld spot recognition module, used to input the weld spot image into a pre-built weld spot recognition model, and output the weld spot image category; a marking module, used to determine whether to generate a marking instruction based on the weld spot image category; a control module, based on the marking instruction, controls the extension and retraction of the electric push rod to mark the corresponding weld spot. By providing an electric push rod on the support plate and installing a marking gun at the output end of the electric push rod, the control module controls the extension and retraction of the electric push rod based on the marking instruction to mark the corresponding weld spot.

Description

Technical Field

[0001] The present invention relates to the field of quality inspection, and in particular to a quality inspection device and a inspection method for a cement pole steel cage based on machine vision. Background Art

[0002] In the process of making cement poles, steel cage is an important component. The quality of steel cage directly affects the strength and stability of cement poles. At present, the manufacturing process of steel cage is mainly divided into three steps: winding, welding and testing.

[0003] At present, both reinforcement winding and welding are produced by fully automatic or semi-automatic equipment. The full-automatic or semi-automatic equipment can grasp the uniformity of the auxiliary reinforcement spacing and is easy to detect. The quality inspection of the reinforcement cage weld point adopts manual visual inspection or machine vision inspection;

[0004] For example, a Chinese patent with authorization announcement number CN115984283B discloses an intelligent detection method for the welding quality of steel cages. First, a grayscale image of the steel cage welding is obtained, and the corresponding characteristic signal graph is obtained through a neural network, and the abnormal welding area is obtained through the brightness difference between the grayscale image and the pixel points at the normal welding position. The streamer vector field is obtained based on the gradient information of the pixel points in the abnormal area, and the convergence point and divergence point are obtained based on the streamer vector direction difference between the pixel points in the streamer vector field. Different convergence windows and divergence windows are adaptively obtained according to the positions of the convergence point and the divergence point, and the neural network is trained through different windows, thereby improving the training speed and recognition accuracy of the neural network. The trained neural network recognizes the steel cage welding image to be detected and determines the welding defects; it obtains the abnormal welding area through the brightness difference between the grayscale image and the pixel points at the normal welding position, and then determines the welding defects;

[0005] However, the fundamental purpose of steel cage quality inspection is to detect unqualified welds and repair them. The current manual visual inspection method can repair them on the spot or mark the unqualified welds and then repair them uniformly, but the efficiency is low, and manual visual inspection is labor-intensive and prone to visual fatigue, resulting in a decrease in detection accuracy. The current machine vision-based inspection method cannot quickly and accurately mark unqualified welds. Summary of the invention

[0006] In view of this, the purpose of the present invention is to propose a cement pole steel cage quality inspection device and inspection method based on machine vision, so as to realize real-time rapid and accurate marking of unqualified welds during the inspection process.

[0007] In order to achieve the above technical objectives, the present invention provides a cement pole steel cage quality detection device based on machine vision:

[0008] It includes a detection unit, a marking mechanism and a translation mechanism. The detection unit and the marking mechanism are arranged on the translation mechanism. The translation mechanism is used to drive the detection unit and the marking mechanism to translate. The marking mechanism includes an electric push rod and a marking gun. The marking gun is fixed on the output end of the electric push rod. The electric push rod is used to telescopically push the marking gun to leave a mark on the steel cage. The detection unit includes:

[0009] A first acquisition module is used to acquire welding point images;

[0010] A solder joint recognition module is used to input the solder joint image into a pre-built solder joint recognition model and output the solder joint image category;

[0011] A marking module, used for determining whether to generate a marking instruction based on the welding spot image category;

[0012] The control module controls the extension and retraction of the electric push rod to mark the corresponding welding points based on the marking instructions.

[0013] Preferably, the translation mechanism comprises a linear slide and a support plate, and the support plate is fixed on the output end of the linear slide.

[0014] Preferably, the marking mechanism further comprises:

[0015] A connecting pile is fixed on the supporting plate, and the bottom of the electric push rod is hinged to the connecting pile;

[0016] The stepper motor is fixed on the connecting pile, and the output end is fixedly connected to the electric push rod. The stepper motor is used to drive the electric push rod to adjust the angle.

[0017] Preferably, the first acquisition module is fixed on the support plate, and the detection unit further comprises:

[0018] The second acquisition module is used to collect the vertical distance value between the steel cage and the detection end of the second acquisition module, and the detection end of the second acquisition module and the hinge point of the electric push rod and the connecting pile are located on the same horizontal line;

[0019] The marking adjustment module calculates the angle value based on the vertical distance value, and the control module controls the stepper motor to adjust the angle of the electric push rod based on the angle value.

[0020] Preferably, the angle value calculation expression is: Among them, θ is the angle value, h is the vertical distance value, and d is the horizontal distance value between the detection end of the second acquisition module and the hinge point between the electric push rod and the connecting pile.

[0021] Preferably, the mark adjustment module calculates the telescopic length value based on the vertical distance value, and the telescopic length value is used as the telescopic amount control parameter of the electric push rod; the calculation expression of the telescopic length value is: Where L is the telescopic length value, and Z is the shortest length value of the electric push rod plus the marking gun.

[0022] Preferably, the training method of the solder joint recognition model is:

[0023] Pre-collect multiple groups of historical solder joint images, detect each group of historical solder joint images to obtain feature data of each group of historical solder joint images, and generate corresponding actual annotations for each group of historical solder joint image feature data according to the category of the historical solder joint images when collected. When the category of the historical solder joint image is qualified, the actual annotation is 1, and when the category of the historical solder joint image is unqualified, the actual annotation is 0;

[0024] Each group of historical solder joint image feature data is used as the input of a machine learning model. The machine learning model takes the predicted annotation of each group of historical solder joint image feature data as output, takes the actual annotation as the prediction target, and takes minimizing the sum of the prediction accuracies of all historical solder joint image feature data as the training target; the machine learning model is trained until the sum of the prediction accuracies converges, then the training is stopped and the machine learning model obtained by the training is used as a historical solder joint image discrimination model, and the machine learning model is either a naive Bayes model or a support vector machine model.

[0025] A detection platform is fixed on the outer surface of the linear slide, and a roller group is evenly fixed on the bottom of the detection platform.

[0026] Preferably, a support assembly is fixed on the outer surface of the detection platform, and the support assembly is used to support the object to be detected, and the support assembly includes:

[0027] A first fixed pile, fixed on the detection platform;

[0028] The support frame has two sides hinged to the first fixed pile, and the roller group is fixed to the bottom of the support frame;

[0029] The support roller is hinged to the top of the first fixed pile.

[0030] Preferably, a flip driving assembly is provided on the side of the support assembly, and the flip driving assembly includes:

[0031] A second fixed pile, fixed on the detection platform;

[0032] An oil cylinder has one end hinged to the second fixed pile and the other end hinged to the support frame, and the oil cylinder is used to push the support frame to flip.

[0033] Preferably, two linear slides and two supporting plates are provided, the two linear slides are arranged side by side, and the detection unit and the marking mechanism are respectively fixed on the two supporting plates.

[0034] The quality detection method of cement pole steel cage based on machine vision is realized based on the above-mentioned quality detection device of cement pole steel cage based on machine vision, and the detection method comprises:

[0035] Pre-collect historical solder joint images, and train a solder joint recognition model that predicts the solder joint image category based on the historical solder joint images;

[0036] Collect solder joint images, input the solder joint images into the trained solder joint recognition model, and output the solder joint image categories;

[0037] It is determined whether to generate a marking instruction based on the welding spot image category. If a marking instruction is generated, the corresponding welding spot is marked.

[0038] Preferably, the method for determining whether to generate a marking instruction based on the solder joint image category is:

[0039] If the output solder joint image category is 1, no marking instruction is generated;

[0040] If the output solder joint image category is 0, a marking instruction is generated.

[0041] It can be seen from the above technical solutions that the present application has the following beneficial effects:

[0042] 1: By arranging an electric push rod on the support plate and installing a marking gun at the output end of the electric push rod, the control module controls the extension and retraction of the electric push rod based on the marking instruction to mark the corresponding welding point.

[0043] 2: The telescopic marking method is used without affecting the detection unit, and the electric push rod can adjust the angle to keep it in the same straight line with the welding points of different heights, which can meet the marking of welding points of steel cages of various specifications. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0045] Figure 1 The overall structural diagram of the cement pole reinforcement cage quality detection device based on machine vision provided by the present invention;

[0046] Figure 2 A schematic diagram of a side cross-sectional structure of a cement pole reinforcement cage quality detection device based on machine vision provided by the present invention;

[0047] Figure 3 A detection simulation schematic diagram of a cement pole steel cage quality detection device based on machine vision provided by the present invention;

[0048] Figure 4A schematic diagram of the structure of a detection unit of a cement pole reinforcement cage quality detection device based on machine vision provided by the present invention;

[0049] Figure 5 A schematic diagram of the process of the cement pole reinforcement cage quality detection method based on machine vision provided by the present invention;

[0050] Figure 6 This is a schematic diagram of the top structure of the second embodiment provided by the present invention.

[0051] Description of the drawings: 1. Linear slide; 2. Support plate; 3. Electric push rod; 31. Connecting pile; 32. Marking gun; 33. Stepper motor; 4. Detection platform; 41. Roller group; 5. Support assembly; 51. First fixed pile; 52. Support frame; 53. Support roller; 6. Flip drive assembly; 61. Second fixed pile; 62. Cylinder; 7. First acquisition module; 8. Second acquisition module. DETAILED DESCRIPTION

[0052] The following description is merely exemplary in nature and is not intended to limit the present disclosure, application, and use. It should be understood that in all of these figures, the same or similar reference numerals indicate the same or similar parts and features. The various drawings only schematically represent the concepts and principles of the embodiments of the present disclosure, and do not necessarily show the specific dimensions and proportions of the various embodiments of the present disclosure. Specific parts in specific drawings may be exaggerated to illustrate the relevant details or structures of the embodiments of the present disclosure.

[0053] Embodiment 1

[0054] See also Figure 1 and Figure 4 As shown, a cement pole steel cage quality inspection device based on machine vision includes a detection unit, a marking mechanism and a translation mechanism. The detection unit and the marking mechanism are arranged on the translation mechanism. The translation mechanism is used to drive the detection unit and the marking mechanism to translate, thereby changing the detection point. The translation range of the translation mechanism depends on the length of the steel cage. The translation range of the translation mechanism is greater than or equal to the length of the steel cage, so as to ensure the complete detection of the steel cage.

[0055] Specifically, the marking mechanism includes an electric push rod 3, a marking gun 32 is fixed at the output end of the electric push rod 3, a connecting pile 31 is hinged at the bottom of the electric push rod 3, a stepper motor 33 is fixed on the connecting pile 31, the output end of the stepper motor 33 is fixedly connected to the electric push rod 3, the stepper motor 33 is used to drive the electric push rod 3 to adjust the angle, and the marking gun 32 is used to leave a mark on the steel cage. In this embodiment, the marking gun 32 adopts a structure that has the ability to quickly leave a mark on an object, such as a stamp or a spray gun, which is not specifically limited here; in this embodiment, the marking gun 32 is controlled to move to the welding point position for marking by telescoping the electric push rod 3. After marking, the electric push rod 3 can be reset with the marking gun 32 to avoid affecting the displacement of the steel cage during detection.

[0056] For details, see Figure 4 As shown, the detection unit includes a first acquisition module 7, a welding point recognition module, a marking module, a control module, a second acquisition module 8 and a marking adjustment module, and each module is connected through a wired and / or wireless network.

[0057] The first acquisition module 7 is used to acquire the welding point image. In this embodiment, the first acquisition module 7 uses a high-definition camera.

[0058] The solder joint recognition module is used to input the solder joint image into the pre-built solder joint recognition model and output the solder joint image category. The training method of the solder joint recognition model is:

[0059] Under the experimental environment, multiple groups of historical solder joint images are collected, and each group of historical solder joint images is detected to obtain the feature data of each group of historical solder joint images. The feature data of each group of historical solder joint images is used to generate corresponding actual annotations according to the category of the historical solder joint images when collected. When the category of the historical solder joint image is qualified, the actual annotation is 1, and when the category of the historical solder joint image is unqualified, the actual annotation is 0;

[0060] Each set of historical solder joint image feature data is used as the input of the machine learning model. The machine learning model uses the predicted annotation of each set of historical solder joint image feature data as the output, the actual annotation as the prediction target, and the minimization of the sum of the prediction accuracy of all historical solder joint image feature data as the training target; the calculation formula for the prediction accuracy is: k =(a k -w k ) 2 , where k is the number of the historical solder joint image feature data, z k is the prediction accuracy, a k is the predicted annotation value corresponding to the kth group of historical solder joint image feature data, w kis the actual annotation corresponding to the kth group of historical solder joint image feature data; the machine learning model is trained until the sum of the prediction accuracies reaches convergence. It should be noted that the convergence standard is determined by technical personnel in this field according to actual needs. The training is stopped and the trained machine learning model is used as a historical solder joint image discrimination model. The machine learning model is a naive Bayes model or a support vector machine model.

[0061] The marking module is used to determine whether to generate a marking instruction based on the solder joint image category. If the output solder joint image category is 1, the solder joint quality is qualified and no marking instruction is generated. If the output solder joint image category is 0, the solder joint quality is unqualified and a marking instruction is generated.

[0062] The control module controls the electric push rod 3 to extend and retract to mark the corresponding welding point based on the marking instruction. That is, when the marking module generates the marking instruction, the control signal is sent to the control module. The control module controls the electric push rod 3 to extend and retract once, so that the marking gun 32 contacts the welding point for marking.

[0063] The second acquisition module 8 is used to collect the vertical distance value between the steel cage and the detection end of the second acquisition module, and the detection end of the second acquisition module 8 and the hinge point of the electric push rod 3 and the connecting pile 31 are located on the same horizontal line. The purpose is that the detection end of the second acquisition module 8, the hinge point of the electric push rod 3 and the connecting pile 31 and the welding point connection line form a right triangle, which is convenient for calculating the angle between the electric push rod 3 and the connecting pile 31.

[0064] The marking adjustment module calculates the angle value based on the vertical distance value, and the control module controls the stepping motor 33 to adjust the angle of the electric push rod 3 based on the angle value.

[0065] Exemplarily, the detection unit and the marking mechanism are arranged below the steel cage, and the steel cage is driven to rotate by an external driving device. The weld point stops when it is placed directly above the first data acquisition and marking mechanism. The detection unit detects the weld point. If the quality of the weld point is unqualified, a marking instruction is generated, and the electric push rod 3 adjusts its angle to ensure that after the electric push rod 3 is extended, the marking gun 32 is just below the weld point, and the weld point is marked. This process is repeated to detect each weld point in the steel cage.

[0066] For details, see Figure 1 and Figure 3 As shown, the marking mechanism also includes: a linear slide 1, the outer surface of which is slidably connected to a support plate 2; a connecting pile 31 is fixed on the support plate 2;

[0067] The linear slide 1 is used to drive the support plate 2 to translate. The linear slide 1 of this embodiment can be of electromagnetic type or screw type. Taking the linear slide 1 with a screw structure as an example, the screw structure includes a reciprocating screw and a wire sleeve threadedly connected to the reciprocating screw. The reciprocating screw is driven to rotate by a servo motor. If the wire sleeve is limited, it can only translate along the axis of the reciprocating screw. The rotation of the reciprocating screw can push the wire sleeve to translate through the thread. The support plate 2 is installed on the wire sleeve and can translate along the axis of the screw. When the quality inspection of the steel cage is carried out, the linear slide 1 can be arranged parallel to the main reinforcement in the steel cage. When the steel cage rotates at a uniform speed, the linear slide 1 can translate at a uniform speed along the direction of the main reinforcement through the support plate 2 with the electric push rod 3, so that each welding point will pass above the marking gun 32, see Figure 3 shown.

[0068] For more details, see Figure 3 and Figure 4 As shown, the first acquisition module 7 is fixed on the support plate 2. As mentioned above, when the linear slide 1 moves with the support plate 2, each welding point also passes over the first acquisition module 7, so that the welding points are detected one by one. Figure 3 Marking U shown.

[0069] Specifically, the angle value calculation expression is Wherein, θ is the angle value, arctan inverse tangent function, h is the vertical distance value, and d is the horizontal distance value between the detection end of the second acquisition module 8 and the hinge point between the electric push rod 3 and the connecting pile 31, which is measured by a person skilled in the art; for example, if h is 50 mm and d is 200 mm, That is, the electric push rod 3 needs to be adjusted to an angle of about 14.04 degrees so that the telescopic path of the electric push rod 3 is spatially aligned with the welding point.

[0070] More specifically, the marking adjustment module calculates the telescopic length value based on the vertical distance value, and the telescopic length value is used as the telescopic amount control parameter of the electric push rod 3; the calculation expression of the telescopic length value is: Wherein L is the telescopic length value, Z is the shortest length value of the electric push rod 3 plus the marking gun 32, which is measured by a technician in this field on the electric push rod 3 plus the marking gun 32. If Z sets a virtual position, that is, adds a virtual position value, the distance between the marking gun 32 and the welding point after the final mark is extended can be controlled. If the marking gun 32 adopts a contact marking structure, such as a seal, it is best if the distance between the final mark and the welding point after the mark is extended is less than or equal to 0, to ensure that the marking gun 32 has sufficient pressure applied to the welding point to ensure the marking effect; if it is a non-contact marking structure, such as a paint spraying structure, the marking gun 32 may not directly contact the welding point to avoid the marking gun 32 colliding with the welding point and causing damage to the marking gun 32; the purpose is to control the telescopic length of the marking gun 32 to avoid the marking gun 32 being too close or too far from the welding point after the final extension, thereby affecting the marking effect.

[0071] Embodiment 2

[0072] See also Figure 5 As shown, based on the above embodiments, the improvement of the second embodiment is that two linear slides 1 and two support plates 2 are provided, the two linear slides 1 are arranged side by side, the detection unit and the marking mechanism are respectively fixed on the two support plates 2, and the marking gun 32, the detection end of the first acquisition module 7 and the detection end of the second acquisition module 8 are located on the same line, parallel to the main reinforcement in the steel cage, the two linear slides 1 have the same translation speed, and the linear slide 1 carrying the detection unit works first, that is, the time difference between the detection unit and the marking mechanism passing the same point is fixed, that is, after the detection unit passes the point X detection, the marking mechanism will pass the point X after N time intervals, N is a value greater than 0, so that the marking work can be performed after N time intervals.

[0073] The purpose of this embodiment is to respectively drive the detection unit and the marking mechanism to translate by two linear slides 1, and after the detection unit detects unqualified welds, the marking mechanism will mark them after an interval of N time periods. In this way, even if the detection unit needs to take images for analysis and detection, if the computing power of the hardware equipment equipped with the analysis system is insufficient and the detection time is long, the N time intervals can be used to take images for analysis and detection, thereby reducing the pause time of the detection time. After the first data detection module in the detection unit takes the weld image, the linear slide 1 can continue to translate and prepare for the detection of the next detection point, and the weld identification module can perform analysis work in the process from the current detection point to the next detection point. In this way, the marking time and the detection time do not need to be superimposed, thereby improving the overall detection efficiency.

[0074] Embodiment 3

[0075] Based on the above embodiments, see Figure 1 and Figure 2 As shown, a detection platform 4 is fixed on the outer surface of the linear slide 1, and a roller group 41 is evenly fixed on the bottom of the detection platform 4 to facilitate the movement of the detection platform 4. A support assembly 5 is fixed on the outer surface of the detection platform 4, and the support assembly 5 is used to support the object to be detected. The support assembly 5 includes: a first fixed pile 51, fixed on the detection platform 4; a support frame 52, both sides of which are hinged to the first fixed pile 51, and the roller group 41 is fixed to the bottom of the support frame 52; a support roller 53, which is hinged to the top of the first fixed pile 51; when testing, the steel cage can be placed on the support roller 53, and the support roller 53 can be connected to the output end of the driving motor (not shown), so that the rotation of the support roller 53 can drive the steel cage to rotate and adjust the detection position.

[0076] A flipping drive assembly 6 is provided on the side of the support assembly 5, and the flipping drive assembly 6 includes: a second fixed pile 61, fixed on the detection platform 4; a cylinder 62, one end of which is hinged to the second fixed pile 61, and the other end is hinged to the support frame 52, and the cylinder 62 is used to push the support frame 52 to flip; the support assembly 5 is connected by a movable structure, and the flipping of the support assembly 5 is controlled by the flipping drive assembly 6. The purpose is that after the inspection of the steel cage, the single-sided support assembly 5 can be flipped to lower the single-sided height of the detection platform 4, and the steel cage can roll away from the detection platform 4 from high to low, which is convenient for the steel cage to go up and down the detection platform 4.

[0077] Embodiment 4.

[0078] See also Figure 5 As shown, based on the above embodiment, the quality detection method of cement pole steel cage based on machine vision is implemented by the above-mentioned cement pole steel cage quality detection device based on machine vision, and the detection method includes:

[0079] Pre-collect historical solder joint images, and train a solder joint recognition model that predicts the solder joint image category based on the historical solder joint images;

[0080] Collect solder joint images, input the solder joint images into the trained solder joint recognition model, and output the solder joint image categories;

[0081] It is determined whether to generate a marking instruction based on the welding spot image category. If a marking instruction is generated, the corresponding welding spot is marked.

[0082] Specifically, the training method of the solder joint recognition model is:

[0083] Under the experimental environment, multiple groups of historical solder joint images are collected, and each group of historical solder joint images is detected to obtain the feature data of each group of historical solder joint images. The feature data of each group of historical solder joint images is used to generate corresponding actual annotations according to the category of the historical solder joint images when collected. When the category of the historical solder joint image is qualified, the actual annotation is 1, and when the category of the historical solder joint image is unqualified, the actual annotation is 0;

[0084] Each set of historical solder joint image feature data is used as the input of the machine learning model. The machine learning model uses the predicted annotation of each set of historical solder joint image feature data as the output, the actual annotation as the prediction target, and the minimization of the sum of the prediction accuracy of all historical solder joint image feature data as the training target; the calculation formula for the prediction accuracy is: k =(a k -w k ) 2 , where k is the number of the historical solder joint image feature data, z k is the prediction accuracy, a k is the predicted annotation value corresponding to the kth group of historical solder joint image feature data, w kis the actual annotation corresponding to the kth group of historical solder joint image feature data; the machine learning model is trained until the sum of the prediction accuracies reaches convergence. It should be noted that the convergence standard is determined by technical personnel in this field according to actual needs. The training is stopped and the trained machine learning model is used as a historical solder joint image discrimination model. The machine learning model is a naive Bayes model or a support vector machine model.

[0085] Specifically, the method for determining whether to generate a marking instruction based on the solder joint image category is as follows:

[0086] If the output solder joint image category is 1, no marking instruction is generated;

[0087] If the output solder joint image category is 0, a marking instruction is generated.

[0088] The exemplary implementation scheme of the present disclosure is described in detail above with reference to the preferred embodiments. However, it can be understood by those skilled in the art that, without departing from the concept of the present disclosure, various modifications and variations can be made to the above-mentioned specific embodiments, and various technical features and structures proposed in the present disclosure can be combined in various ways without exceeding the protection scope of the present disclosure, which is determined by the attached claims.

Claims

1. The cement pole steel cage quality detection device based on machine vision is characterized by: The invention comprises a detection unit, a marking mechanism and a translation mechanism, wherein the detection unit and the marking mechanism are arranged on the translation mechanism, the translation mechanism is used to drive the detection unit and the marking mechanism to translate, the marking mechanism comprises an electric push rod (3) and a marking gun (32), the marking gun (32) is fixed on the output end of the electric push rod (3), the electric push rod (3) is used to telescopically push the marking gun (32) to leave a mark on the steel cage, the translation mechanism comprises a linear slide (1) and a support plate (2), the support plate (2) is fixed on the output end of the linear slide (1), and the detection unit comprises: A first acquisition module (7) for acquiring images of welding spots, wherein the first acquisition module (7) is fixed on the support plate (2); A solder joint recognition module is used to input the solder joint image into a pre-built solder joint recognition model and output the solder joint image category; A marking module, used for determining whether to generate a marking instruction based on the welding spot image category; A control module controls the electric push rod (3) to extend and retract to mark the corresponding welding point based on the marking instruction; A second acquisition module (8) is used to acquire a vertical distance value between the steel cage and a detection end of the second acquisition module, and the detection end of the second acquisition module (8) is located on the same horizontal line as a hinge point between the electric push rod (3) and the connecting pile (31); A marking adjustment module calculates an angle value based on the vertical distance value, and a control module controls a stepping motor (33) to adjust the angle of the electric push rod (3) based on the angle value; The angle value calculation expression is: ,in, is the angle value, is the vertical distance value, is the horizontal distance value between the detection end of the second acquisition module (8) and the hinge point between the electric push rod (3) and the connecting pile (31); The mark adjustment module calculates the telescopic length value based on the vertical distance value, and the telescopic length value is used as a telescopic amount control parameter of the electric push rod (3); the calculation expression of the telescopic length value is: ,in is the telescopic length value, and Z is the shortest length value of the electric push rod (3) and the marking gun (32).

2. The cement pole steel cage quality detection device based on machine vision according to claim 1 is characterized in that: Labelling agencies also include: A connecting pile (31) is fixed on the supporting plate (2), and the bottom of the electric push rod (3) is hinged to the connecting pile (31); The stepper motor (33) is fixed on the connection pile (31), and the output end is fixedly connected to the electric push rod (3). The stepper motor (33) is used to drive the electric push rod (3) to adjust the angle.

3. The cement pole steel cage quality detection device based on machine vision according to claim 1 is characterized in that: The training method of the solder joint recognition model is: Pre-collect multiple groups of historical solder joint images, detect each group of historical solder joint images to obtain feature data of each group of historical solder joint images, and generate corresponding actual annotations for each group of historical solder joint image feature data according to the category of the historical solder joint images when collected. When the category of the historical solder joint image is qualified, the actual annotation is 1, and when the category of the historical solder joint image is unqualified, the actual annotation is 0; Each set of historical solder joint image feature data is used as input of a machine learning model, wherein the machine learning model uses the predicted annotation of each set of historical solder joint image feature data as output, uses the actual annotation as a prediction target, and uses minimizing the sum of the prediction accuracies of all historical solder joint image feature data as a training target; The machine learning model is trained until the sum of the prediction accuracies reaches convergence, the training is stopped and the trained machine learning model is used as a historical solder joint image discrimination model, and the machine learning model is one of a naive Bayes model or a support vector machine model.

4. The cement pole steel cage quality detection device based on machine vision according to claim 1 is characterized in that: A detection platform (4) is fixed on the outer surface of the linear slide (1), and a roller group (41) is evenly fixed on the bottom of the detection platform (4).

5. The cement pole steel cage quality detection device based on machine vision according to claim 4 is characterized in that: A support assembly (5) is fixed on the outer surface of the detection platform (4), and the support assembly (5) is used to support the object to be detected. The support assembly (5) comprises: A first fixing pile (51) fixed on the detection platform (4); A support frame (52) having both sides hinged to the first fixed pile (51), and the roller group (41) is fixed to the bottom of the support frame (52); The supporting roller (53) is hinged to the top of the first fixed pile (51).

6. The cement pole steel cage quality detection device based on machine vision according to claim 5 is characterized in that: A flip drive assembly (6) is provided on the side of the support assembly (5), and the flip drive assembly (6) comprises: A second fixing pile (61) is fixed on the detection platform (4); An oil cylinder (62) has one end hinged to the second fixed pile (61) and the other end hinged to the support frame (52), and the oil cylinder (62) is used to push the support frame (52) to flip.

7. The cement pole steel cage quality detection device based on machine vision according to claim 1 is characterized in that: Two linear slides (1) and two support plates (2) are provided, the two linear slides (1) are arranged side by side, and the detection unit and the marking mechanism are respectively fixed on the two support plates (2).

8. A method for detecting quality of cement pole reinforcement cage based on machine vision, which is implemented based on a device for detecting quality of cement pole reinforcement cage based on machine vision according to any one of claims 1 to 7, characterized in that: The detection method comprises: Pre-collect historical solder joint images, and train a solder joint recognition model that predicts the solder joint image category based on the historical solder joint images; Collect solder joint images, input the solder joint images into the trained solder joint recognition model, and output the solder joint image categories; It is determined whether to generate a marking instruction based on the welding spot image category. If a marking instruction is generated, the corresponding welding spot is marked.

9. The method for quality inspection of cement pole reinforcement cage based on machine vision according to claim 8 is characterized in that: The method for determining whether to generate a marking instruction based on the solder joint image category is: If the output solder joint image category is 1, no marking instruction is generated; If the output solder joint image category is 0, a marking instruction is generated.

Images