Machine vision-based sewing machine table board processing quality online detection method

Abstract

The present application relates to sewing machine table board processing technical field, specifically disclose a sewing machine table board processing quality on-line detection method based on machine vision, the present application is through to the sewing machine table board carries out initial shooting and basic identification, determines target table board type, matches corresponding target table board model, carries out multiple posture adjustment control, after posture adjustment control, posture shooting is carried out, the corresponding posture adjustment of target table board model is carried out, based on multiple posture contrast image, carries out processing quality on-line comparison detection and processing, the present application is through to the sewing machine table board carries out initial shooting and type identification, matches target table board model and creates virtual rendering environment according to environmental monitoring data, obtains posture shooting image after multiple posture adjustment control, generates posture contrast image through posture identification and model corresponding adjustment, so as to realize automatic, comprehensive processing quality on-line detection, improve detection efficiency, avoid inconsistent and missed detection.

Description

Technical Field

[0001] This invention belongs to the field of sewing machine table processing technology, and particularly relates to an online detection method for sewing machine table processing quality based on machine vision. Background Technology

[0002] The processing of the sewing machine table is a crucial manufacturing step in the sewing machine assembly process, involving multiple precision procedures such as cutting, milling, drilling, and tapping. Its processing quality directly affects the assembly accuracy and operational stability of the sewing machine. Currently, the quality inspection of sewing machine tables still mainly relies on manual visual inspection or sampling inspection using simple measuring tools. This method has the following technical drawbacks in practical applications: 1. Low inspection efficiency: Manual inspection is difficult to keep up with the pace of high-speed automated production lines. The inspection process is time-consuming and tiring, becoming a bottleneck for production line efficiency. 2. High subjectivity of test results: Differences in the experience and judgment standards of different testers lead to poor consistency of test results, making it difficult to form a unified quality inspection standard; 3. Incomplete inspection coverage: Traditional methods usually only conduct spot checks on key locations of the platform, and cannot conduct full-view, multi-angle, and all-round inspections of the platform's structural dimensions, surface defects, installation features, etc., which can easily lead to missed inspections; 4. Lack of automated closed-loop mechanism: Existing methods are difficult to achieve automatic linkage between detection results and subsequent processing such as defect location and defect marking, making it difficult to support the need for closed-loop control of the detection process in the context of intelligent manufacturing.

[0003] As machine vision technology matures, some industries have attempted to introduce it into product quality inspection. However, existing technologies mostly focus on static imaging or image comparison under a single posture, which cannot adapt to the complex inspection needs of sewing machine tables with multiple postures, features, and workstations. They also lack the ability to dynamically adapt to the inspection environment, lighting conditions, and material properties, making it difficult to achieve stable and high-precision online quality inspection in real production lines. Summary of the Invention

[0004] The purpose of this invention is to provide an online inspection method for the processing quality of sewing machine table based on machine vision, which aims to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions: A machine vision-based online inspection method for the processing quality of sewing machine tabletops, the method specifically includes the following steps: The processing production line is monitored, and when the sewing machine table is transferred, the sewing machine table is initially photographed and basic identification is performed to determine the target table type. According to the target platform type, match the corresponding target platform model; perform environmental monitoring, obtain environmental monitoring data, create a virtual rendering environment based on the environmental monitoring data, and render and load the target platform model in the virtual rendering environment; The posture is adjusted and controlled during the movement of the sewing machine table; The sewing machine table is followed and the production line is paused after the posture adjustment control is performed. The posture image is captured. The posture of the captured image is recognized to determine the shooting posture, and the target platform model is adjusted accordingly in the virtual rendering environment to obtain a posture comparison image. Based on the posture comparison image, the posture captured image is subjected to online comparative detection and processing of processing quality.

[0006] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention achieves high-precision posture imaging of the sewing machine table under multiple posture adjustments by controlling the posture of the sewing machine table multiple times and combining it with a flow-following and production line pause mechanism. By aligning posture with the model posture, multiple posture comparison images are obtained, constructing a comprehensive detection system covering the table structure, hole positions, grooves, and surface texture, thus solving the core problems of "incomplete detection and easy omission" in traditional detection methods.

[0007] 2. After matching the target platform model, this invention dynamically creates a virtual rendering environment based on the lighting intensity, color temperature, background texture, and light source layout parameters of the actual production line, and parametrically adjusts the visual characteristics of the target platform model, such as material reflection and shadow projection. This high-fidelity virtual environment can generate posture comparison images that are highly consistent with the visual representation of the actual captured images, providing a standardized and reproducible reference benchmark for subsequent image comparison. This fundamentally eliminates the impact of environmental changes on the detection results and significantly improves the consistency and accuracy of the detection.

[0008] 3. In the processing quality inspection stage, this invention performs pixel-level alignment, differential processing, connected component analysis, and defect candidate block extraction based on posture comparison images and posture-captured images. Combined with a pre-set surface defect classification model, it achieves defect type identification and confidence level determination. This process integrates image processing and deep learning technologies, enabling high-precision, automated detection of dimensional deviations and surface defects, avoiding subjective errors associated with manual interpretation.

[0009] 4. Based on defect detection, this invention constructs a spatial mapping relationship between image coordinates and 3D model regions. Combined with multi-view cross-validation, it accurately maps the defect coordinates in the image to the 3D spatial coordinates of the target platform model, achieving precise defect positioning on the platform. This positioning data can directly drive subsequent defect marking and sorting processes, forming an automated closed loop from detection to processing, significantly improving the intelligence level and quality control capabilities of the production line. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention.

[0011] Figure 1 A flowchart of the method provided by an embodiment of the present invention is shown.

[0012] Figure 2 A flowchart of the flow monitoring and basic identification process in the method provided by an embodiment of the present invention is shown.

[0013] Figure 3 A flowchart illustrating the method for determining the target platform type provided in an embodiment of the present invention is shown.

[0014] Figure 4 A flowchart illustrating the matching of a target platform model in the method provided by an embodiment of the present invention is shown.

[0015] Figure 5 A flowchart illustrating multiple attitude adjustment controls in the method provided by an embodiment of the present invention is shown.

[0016] Figure 6 A flowchart of the corresponding attitude adjustment control in the method provided by the embodiments of the present invention is shown.

[0017] Figure 7 A flowchart illustrating the method for acquiring multiple pose images provided in an embodiment of the present invention is shown.

[0018] Figure 8 A flowchart of pose recognition and model adjustment in the method provided by an embodiment of the present invention is shown.

[0019] Figure 9 A flowchart of the online comparison detection and processing in the method provided by an embodiment of the present invention is shown.

[0020] Figure 10 The flowchart of the online comparison detection and judgment method provided in the embodiment of the present invention is shown. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0022] Understandably, in the current technology, the quality inspection of sewing machine table processing mainly relies on manual visual inspection or simple inspection. This is not only inefficient, but also prone to inconsistent results due to subjective differences in judgment among different inspectors. In addition, the inspection method usually only conducts spot checks on some key locations, resulting in incomplete quality assessment and easy to miss inspections.

[0023] To address the aforementioned issues, this invention monitors the production line flow. Upon entering the sewing machine table, the sewing machine table undergoes initial imaging and basic identification to determine the target table type. Based on the target table type, a corresponding target table model is matched and rendered. During the sewing machine table's flow, multiple posture adjustments are performed. The sewing machine table is followed during the flow; after the posture adjustments, the production line flow is paused, and posture images are captured. Posture recognition is performed on these images to determine multiple shooting postures, and the target table model is adjusted accordingly to obtain multiple posture comparison images. Based on these comparison images, online comparison detection and processing of the processing quality is performed on the multiple posture images. It can perform multiple posture adjustments and controls on the sewing machine table, match the target table model, and perform corresponding posture adjustments to acquire multiple posture comparison images. Then, it performs online comparison detection and processing of the processing quality of multiple posture images, thereby realizing automatic and comprehensive online detection of the processing quality of the sewing machine table, improving detection efficiency, and avoiding problems such as inconsistent detection results and missed detections.

[0024] Figure 1 A flowchart of the method provided by an embodiment of the present invention is shown.

[0025] Specifically, the online inspection method for sewing machine table processing quality based on machine vision includes the following steps: Step S101: Monitor the flow of the processing line. When the sewing machine table is transferred, take an initial picture and perform basic identification on the sewing machine table to determine the target table type.

[0026] In this embodiment of the invention, the sewing machine table processing line is monitored to obtain flow monitoring data. The flow monitoring data is then identified to determine whether a sewing machine table has entered the line. If a sewing machine table has entered the line, an initial image is taken of the entering sewing machine table. Then, basic feature recognition is performed on the initial image to extract the basic feature data of the entering sewing machine table. Feature matching is performed on the basic feature data in a preset feature database, and the feature matching results are recorded. Finally, based on the feature matching results, the target table type of the entering sewing machine table is determined from multiple preset processing table types.

[0027] Specifically, Figure 2 A flowchart of the flow monitoring and basic identification process in the method provided by an embodiment of the present invention is shown.

[0028] In a preferred embodiment of the present invention, the step of monitoring the flow of the processing line, and performing initial photography and basic identification on the sewing machine table when the machine enters the sewing machine table to determine the target table type, specifically includes the following steps: Step S1011: Monitor the flow of the processing production line and obtain flow monitoring data; Step S1012: Identify the flow monitoring data and determine whether a sewing machine table has flowed in; Step S1013: When a sewing machine table is being transported in, an initial photograph is taken of the sewing machine table to obtain an initial photographed image; Step S1014: Perform basic identification on the initial captured image to determine the target platform type.

[0029] Specifically, Figure 3 A flowchart illustrating the method for determining the target platform type provided in an embodiment of the present invention is shown.

[0030] In a preferred embodiment of the present invention, the step of performing basic identification on the initial captured image to determine the target platform type specifically includes the following steps: Step S10141: Perform basic feature recognition on the initial captured image and extract basic feature data; Step S10142: In the preset feature database, perform feature matching on the basic feature data and record the feature matching results; Step S10143: Based on the feature matching results, determine the target table type of the sewing machine table to be transferred in from multiple preset processing table types.

[0031] Specifically, in an actual production line environment, to accurately identify the type of sewing machine table entering the inspection station, this invention provides the following specific implementation methods: Photoelectric sensors or vision triggers are installed above or on the side of the conveyor belt in the production line to monitor in real time whether a worktable enters the predetermined detection area. When a sensor signal is triggered, the control system records the trigger moment. Activate the high-definition industrial camera, pre-installed at a specific location on the production line, and use a ring light source or coaxial light source to capture initial images of the worktable from above or to the side, under preset fixed exposure time and gain parameters. To ensure image quality, the camera lens plane should be parallel to the intended main feature plane of the worktable (such as the upper surface of the worktable). The initial captured image undergoes a basic recognition process: first, an edge detection algorithm is used to extract the outer contour of the platform; then, a contour matching algorithm (such as shape context-based matching) is used to compare it with standard contour templates for various types of platforms stored in a pre-set feature database; the feature database pre-stores contour data, main hole distribution topology, and geometric parameters of specific markers (such as positioning holes and grooves) for different platform models; if the matching degree between the image contour and a certain model template exceeds a preset threshold (e.g., 85%), then that model is identified as the target platform type. If the matching fails, the system issues an alarm, prompting manual intervention or transferring the platform to the inspection area.

[0032] Furthermore, the online inspection method for sewing machine table processing quality based on machine vision also includes the following steps: Step S102: Match the corresponding target platform model according to the target platform type; perform environmental monitoring, obtain environmental monitoring data, create a virtual rendering environment based on the environmental monitoring data, and render and load the target platform model in the virtual rendering environment.

[0033] In this embodiment of the invention, according to the type of the target platform, the target platform model corresponding to the sewing machine platform that is transferred into the preset model database is matched. At the same time, the processing line of the sewing machine platform is monitored for environmental factors such as brightness and color temperature to obtain environmental monitoring data. Then, based on the environmental monitoring data, a corresponding virtual rendering environment is created in the preset 3D software, the target platform model is imported, and the target platform model is rendered and loaded accordingly.

[0034] Specifically, Figure 4 A flowchart illustrating the matching of a target platform model in the method provided by an embodiment of the present invention is shown.

[0035] In a preferred embodiment of the present invention, the steps include: matching a corresponding target platform model according to the target platform type; performing environmental monitoring to obtain environmental monitoring data; creating a virtual rendering environment based on the environmental monitoring data; and rendering and loading the target platform model in the virtual rendering environment. Step S1021: According to the target platform type, match the corresponding target platform model from the preset model database; Step S1022: Conduct environmental monitoring and obtain environmental monitoring data; Step S1023: Create a corresponding virtual rendering environment based on the environmental monitoring data; Step S1024: In the virtual rendering environment, the target platform model is rendered and loaded.

[0036] Specifically, based on the environmental monitoring data, a corresponding virtual rendering environment is created, and the specific steps are as follows: Illumination intensity and color temperature data that affect visual imaging are obtained from environmental monitoring data; using the illumination intensity and color temperature data, a basic rendering scene template is obtained by matching in a preset virtual rendering scene template library. Texture features are extracted from the background area in the initial captured image to obtain background texture data; the background texture data is used as texture mapping parameters to dynamically replace the background texture in the basic rendering scene template to generate a customized rendering scene. In the customized rendering scene, light intensity data, color temperature data, and preset production line light source layout parameters are used to construct a virtual lighting model with directional light sources, and the number and direction of the light sources are matched with the actual light source layout of the processing production line. The virtual lighting model is used to perform lighting rendering on the customized rendering scene to obtain a customized rendering scene with completed lighting rendering. Extract the material attribute data of the target platform model from its preset attributes; based on the material attribute data, in the customized rendering scene after completing the lighting rendering, parametrically adjust the reflection characteristics and shadow projection characteristics of the target platform model surface to obtain a high-fidelity virtual detection environment consistent with the visual performance of the sewing machine platform during posture shooting. In a high-fidelity virtual inspection environment, the target platform model is loaded, and the virtual lighting model and background texture data in the high-fidelity virtual inspection environment are used to render and load the target platform model to obtain the corresponding virtual rendering environment.

[0037] Specifically, to achieve a high degree of consistency between the real-world images and the virtual model images, the construction process of the virtual rendering environment is as follows: In the environmental monitoring process, multiple integrated color temperature and illuminance sensors are deployed around the production line's testing station, located at key positions corresponding to the posture of the worktable to be tested. Real-time data collected includes: ambient light intensity (lux), correlated color temperature (Kelvin), and the directional angle of the main light source. When creating the virtual rendering environment, a 3D rendering engine (Unity) is used. Based on the collected light intensity and color temperature data, a main directional light is created in the engine to simulate the main lighting of the production line, and several point light sources or spotlights are created to simulate auxiliary lighting. The intensity, color (obtained through color temperature conversion), and 3D spatial position of the light sources must match the measured data. For the background texture, a background area excluding the platform and containing only the conveyor belt and surrounding fixed structures is cropped from the initial captured image. Image processing algorithms are used to extract the texture map of this area and map it onto the corresponding 3D model surface in the virtual scene, making the virtual background visually consistent with the real background. Finally, after loading the target platform model, the corresponding material shaders are set in the rendering engine according to the model's material properties (such as roughness and metallicity parameters of metal, wood, and plastic). By adjusting parameters such as material reflectivity and specular intensity, the reflection and shadow performance of the virtual platform under virtual lighting is made to match the visual characteristics of the real platform under production line lighting, thereby completing the construction of a high-fidelity virtual rendering environment.

[0038] Furthermore, the online inspection method for sewing machine table processing quality based on machine vision also includes the following steps: Step S103: During the rotation of the sewing machine table, multiple posture adjustment controls are performed.

[0039] In this embodiment of the invention, multiple posture adjustment positions on the processing production line of the sewing machine table are determined, and the dynamic flow of the sewing machine table is monitored in real time to obtain the real-time flow position of the sewing machine table during the flow process. Based on the multiple posture adjustment positions, the real-time flow position is matched and analyzed to determine whether posture adjustment is required. According to the target table type, the adjustment process parameters for multiple posture adjustments are obtained. If it is determined that posture adjustment is required, the target adjustment parameter corresponding to the currently matched posture adjustment position is selected from the multiple adjustment process parameters. Then, according to the target adjustment parameter, a corresponding posture adjustment signal is generated, and the corresponding posture adjustment control is performed according to the posture adjustment signal.

[0040] Specifically, Figure 5 A flowchart illustrating multiple attitude adjustment controls in the method provided by an embodiment of the present invention is shown.

[0041] In a preferred embodiment of the present invention, the multiple posture adjustment controls during the rotation of the sewing machine table specifically include the following steps: Step S1031: Determine multiple attitude adjustment positions; Step S1032: Real-time acquisition of the real-time position of the sewing machine table during the transfer process; Step S1033: Based on multiple posture adjustment positions, perform matching analysis on the real-time flow position to determine whether posture adjustment is required; Step S1034: When attitude adjustment is required, perform the corresponding attitude adjustment control.

[0042] Specifically, Figure 6 A flowchart of the corresponding attitude adjustment control in the method provided by the embodiments of the present invention is shown.

[0043] In a preferred embodiment of the present invention, the step of performing corresponding attitude adjustment control when attitude adjustment is required specifically includes the following steps: Step S10341: Obtain the adjustment process parameters for multiple attitude adjustments based on the target platform type; Step S10342: When attitude adjustment is required, select the target adjustment parameter from multiple adjustment process parameters; Step S10343: Generate a corresponding attitude adjustment signal according to the target adjustment parameters; Step S10344: Perform corresponding attitude adjustment control according to the attitude adjustment signal.

[0044] It should be noted that, based on the target plate type, the attitude adjustment sequence of that type of plate is obtained from the process parameter database. The attitude adjustment sequence includes several attitude adjustment tasks, each of which includes: adjusting position coordinates, target attitude parameters, and adjusting actuator parameters. The system acquires the real-time position of the sewing machine table during its rotation and reads the conveyor line position data via an encoder. When the difference between the real-time position and a certain posture adjustment position is less than a set threshold, it determines that posture adjustment is required. An attitude adjustment signal is generated based on the target adjustment parameters. The attitude adjustment signal includes: control command type, action parameters, and action sequence. The control command type includes clamping, rotation, tilting, and releasing. The action parameters include rotation angle, speed, and acceleration. The action sequence includes action order and interval time. The attitude adjustment actuator is a multi-degree-of-freedom mechanical gripper that performs corresponding actions according to the attitude adjustment signal to adjust the sewing machine table to the target attitude.

[0045] It should be noted that the specific implementation method of attitude adjustment control is as follows: The attitude adjustment position is a series of spatial points preset according to the target table type. For example, the first adjustment position is used to rotate the table 90 degrees around the vertical axis to detect side features; the second adjustment position is used to tilt the table 30 degrees around the horizontal axis to detect the bottom milling quality. The production line position is obtained in real time by a rotary encoder installed on the conveyor belt drive shaft; the control system converts the encoder pulse count into the cumulative distance the platform moves along the conveyor belt, thereby accurately knowing the real-time position of the platform. When the absolute value of the deviation between the real-time position and any preset attitude adjustment position is less than the preset positioning accuracy (e.g., ±2 mm), the control system sends an attitude adjustment command to the multi-axis robot or dedicated tilting mechanism. The command includes target attitude parameters: for example, target rotation angle, angular velocity, and angular acceleration. The actuator (such as a pneumatic or servo-driven gripper) first clamps onto the process boss or dedicated positioning point on the edge of the platform, and then performs a rotation or tilting motion according to the command parameters. After the motion is completed, the gripper releases and resets, awaiting the next adjustment command.

[0046] Furthermore, the online inspection method for sewing machine table processing quality based on machine vision also includes the following steps: Step S104: Follow the flow of the sewing machine table. After the posture adjustment control, perform production line flow pause control and take posture pictures to obtain multiple posture pictures.

[0047] In this embodiment of the invention, the sewing machine table is followed according to the real-time flow position. After the posture adjustment control, a production line pause command is generated. Then, the production line flow pause control is performed according to the production line pause command. During multiple posture adjustment controls, corresponding posture images are captured to obtain multiple posture images of the sewing machine table.

[0048] Specifically, Figure 7 A flowchart illustrating the method for acquiring multiple pose images provided in an embodiment of the present invention is shown.

[0049] In a preferred embodiment of the present invention, the following steps are specifically included: Following the shift and follow of the sewing machine table, after posture adjustment control, the production line shift is paused, and posture images are captured to obtain multiple posture images. Step S1041: Follow the real-time flow position of the sewing machine table. Step S1042: After attitude adjustment control, a production line pause command is generated; Step S1043: Perform production line flow suspension control according to the production line suspension instruction; In step S1044, during each of the multiple attitude adjustment controls, corresponding attitude shots are taken to acquire multiple attitude shot images.

[0050] It should be noted that after each attitude adjustment control is completed, the system immediately executes the following shooting procedure: Send a "pause" command to the drive motor controller of the production line conveyor belt and set a short, stable waiting time (e.g., 0.5 seconds) to eliminate the slight vibrations of the platform and conveyor belt caused by inertia and ensure that the platform is stationary. The system triggers a vision imaging subsystem. This subsystem includes multiple industrial cameras with pre-calibrated internal and external parameters, which are fixedly installed at different angles (such as directly above or to the side) of the inspection station. The control system automatically selects the camera with the best viewing angle based on the current preset posture of the platform to capture an image that clearly shows the features to be inspected. After the image is captured, the system automatically records the corresponding posture label (e.g., "frontal horizontal posture", "left tilt 30 degrees posture"), camera parameters, and timestamp, and immediately sends a "resume flow" command to the production line controller, so that the platform continues to move to the next adjustment position.

[0051] Furthermore, the online inspection method for sewing machine table processing quality based on machine vision also includes the following steps: Step S105: Perform posture recognition on the multiple posture images to determine multiple shooting postures, and adjust the corresponding posture of the target platform model in the virtual rendering environment to obtain multiple posture comparison images.

[0052] In this embodiment of the invention, by performing posture recognition on multiple posture-captured images, the shooting posture of the sewing machine table at different posture adjustment positions is determined. Then, according to the multiple shooting postures, the target table model is adjusted accordingly. At the same time, the shooting angle analysis is performed on multiple posture-captured images to determine multiple corresponding angle parameters. Then, according to the multiple angle parameters, the target table model after posture adjustment is cropped accordingly to obtain multiple posture comparison images.

[0053] Specifically, Figure 8 A flowchart of pose recognition and model adjustment in the method provided by an embodiment of the present invention is shown.

[0054] In a preferred embodiment of the present invention, the step of performing pose recognition on multiple pose-captured images to determine multiple shooting poses, and then adjusting the target platform model in the virtual rendering environment to obtain multiple pose comparison images specifically includes the following steps: Step S1051: Perform pose recognition on the multiple pose images to determine multiple shooting poses; Step S1052: Adjust the posture of the target platform model according to the multiple shooting postures. Step S1053: Obtain the viewing angle parameters corresponding to the multiple posture-captured images; Step S1054: According to the multiple viewpoint parameters, the target platform model after attitude adjustment is cropped accordingly to obtain multiple attitude comparison images.

[0055] Specifically, according to the aforementioned viewpoint parameters, the target platform model after attitude adjustment is cropped accordingly to obtain an attitude comparison image. The specific steps are as follows: The position and orientation parameters of the virtual camera in three-dimensional space are determined based on the viewpoint parameters, and the virtual lighting model parameters corresponding to the pose-captured images are extracted from the virtual rendering environment. The pose of the virtual camera is set in the virtual rendering environment according to the position and orientation parameters, and the imaging parameters of the virtual camera are matched and adjusted using the virtual lighting model parameters to obtain the virtual camera after the settings are completed. The virtual camera, once set up, is used to capture images of the target platform model after its posture has been adjusted, in order to obtain an initial cropped image. Extract image resolution and image format parameters from the pose-captured image; based on the image resolution and image format parameters, perform resolution and format conversion on the initial cropped image to obtain an intermediate cropped image with the same image size and image encoding as the pose-captured image; use the intermediate cropped image as the pose reference image.

[0056] Furthermore, this step generates a pose comparison image that is highly consistent with the real-world image in terms of visual performance, size, and encoding by dynamically matching the viewpoint, lighting, and imaging parameters, significantly improving the accuracy and reliability of online comparison detection.

[0057] It should be noted that the specific steps for obtaining the pose comparison image are as follows: Pose recognition is performed on the captured images. A matching algorithm based on feature points (such as scale-invariant feature transform) is used to match the salient feature points of the platform in the image (such as corner points and hole centers) with the corresponding 3D feature points on the target platform model. By solving the perspective n-point problem, the spatial pose of the camera relative to the platform (i.e., the shooting pose) at the time of shooting is accurately calculated. In the constructed virtual rendering environment, the virtual camera is moved to the spatial pose calculated in step S1051 and the same internal parameters (focal length, principal point coordinates, distortion parameters) are set. In the virtual environment, based on the actual posture adjustment parameters executed in step S103, the loaded target platform model is transformed in three-dimensional space so that its posture is completely consistent with the posture of the real platform at the time of shooting. The target platform model, after attitude adjustment, is rendered using a virtual camera to generate a virtual image. To ensure comparability between the virtual and real-world images, post-processing is required: appropriate amounts of Gaussian or salt-and-pepper noise are added to the virtual image based on the noise model of the real-world camera; simultaneously, the grayscale histogram of the virtual image is adjusted to match the overall brightness and contrast of the real-world image. The virtual image processed in this way becomes the final attitude comparison image.

[0058] Furthermore, the online inspection method for sewing machine table processing quality based on machine vision also includes the following steps: Step S106: Based on the multiple posture comparison images, perform online comparison detection and processing of the processing quality of the multiple posture images.

[0059] In this embodiment of the invention, multiple posture comparison images are used as standards to perform online comparative detection of processing quality on multiple posture images. Specifically, this includes online comparative detection of structural dimensions on multiple posture images, recording a first detection result, and online comparative detection of surface defects on multiple posture images, recording a second detection result. Then, by combining the first and second detection results, the processing quality of the incoming sewing machine table is analyzed to determine whether there are any processing quality defects. If no processing quality defects are determined, the sewing machine table is transferred to the next process. If processing quality defects are determined, quality defect data is acquired, and based on the quality defect data, the quality defect of the sewing machine table is located to acquire defect location data. Then, according to the defect location data, the sewing machine table is marked with defects, and the marked sewing machine table is transferred to the recycling area.

[0060] Understandably, online comparative inspection of structural dimensions includes online comparative inspection of hole diameter, online comparative inspection of hole position offset, online comparative inspection of groove structure dimensions, and online comparative inspection of peripheral dimensions; online comparative inspection of surface defects includes comparison of surface texture features, location and classification of defect features, and assessment of defect severity.

[0061] Specifically, Figure 9 A flowchart of the online comparison detection and processing in the method provided by an embodiment of the present invention is shown.

[0062] In a preferred embodiment of the present invention, the online comparison detection and processing of the processing quality of the multiple posture-captured images based on the multiple posture comparison images specifically includes the following steps: Step S1061: Based on the multiple posture comparison images, perform online comparison detection of the processing quality of the multiple posture images to determine whether there are any processing quality defects.

[0063] Specifically, Figure 10 The flowchart of the online comparison detection and judgment method provided in the embodiment of the present invention is shown.

[0064] In a preferred embodiment of the present invention, the step of performing online comparative detection of processing quality on multiple posture-captured images based on multiple posture comparison images to determine whether there are processing quality defects specifically includes the following steps: Step S10611: Based on the multiple posture comparison images, perform online comparison detection of structural dimensions on the multiple posture images and record the first detection result; Step S10612: Based on multiple posture comparison images, perform online comparative detection of surface defects on multiple posture captured images, and record the second detection result; Step S10613: Combine the first detection result and the second detection result to perform processing quality analysis and determine whether there are processing quality defects.

[0065] Specifically, for the structural dimension detection results and the surface defect detection results, detection result vectors are constructed separately. Each element in the structural dimension detection result vector represents whether a certain type of dimensional deviation exceeds the standard; each element in the surface defect detection result vector represents whether a certain type of surface defect exists. A weight vector is set, and the weight values ​​are preset according to the degree of influence of each inspection item on the processing quality; the weight value of the aperture accuracy is higher than the weight value of the outer dimension, and the weight value of the critical surface defect is higher than the weight value of the non-critical surface defect. The comprehensive quality score is calculated as follows: the structural dimension detection result vector is multiplied by the corresponding weight vector to obtain the structural dimension weighted score; the surface defect detection result vector is multiplied by the corresponding weight vector to obtain the surface defect weighted score; the structural dimension weighted score and the surface defect weighted score are added together to obtain the total weighted score; the total weighted score is divided by the sum of all weight values ​​to obtain the comprehensive quality score on a percentage basis.

[0066] It should be noted that when the overall quality score is less than the set threshold, it is determined that there is a processing quality defect; otherwise, it is determined to be qualified. When defects are found, the defect information from each inspection item is summarized to form quality defect data, including defect type, defect location, defect size, and defect severity.

[0067] Furthermore, this step uses vector weighting and a percentage-based scoring system to quantify and integrate structural dimensions and surface defects according to their degree of impact, thereby achieving a unified and objective evaluation of processing quality, effectively avoiding subjective bias and supporting automatic defect identification.

[0068] It should be noted that...

[0069] Furthermore, the online comparison detection and processing of the processing quality of the multiple posture-captured images based on multiple posture comparison images also includes the following steps: Step S1062: If a processing quality defect is determined to exist, quality defect data is obtained.

[0070] Step S1063: Based on the quality defect data, locate the quality defect and obtain defect location data.

[0071] Step S1064: According to the defect location data, perform defect marking processing on the sewing machine table.

[0072] Specifically, based on the aforementioned quality defect data, quality defects are located and defect location data is obtained. The specific steps are as follows: Extract defect type labels and defect location coordinates from quality defect data; use defect type labels to determine the preset detection area on the sewing machine table corresponding to the defect type labels, and obtain the image coordinates of the defect in the posture-captured image based on the defect location coordinates; Obtain the shooting posture and viewpoint parameters corresponding to the shooting posture and viewpoint parameters in the shooting posture and viewpoint image; based on the shooting posture and viewpoint parameters, retrieve the viewpoint image of the target platform model that matches the shooting posture and viewpoint parameters from the virtual rendering environment; extract the 3D model region corresponding to the preset detection area from the viewpoint image of the target platform model. Establish a spatial mapping relationship between image coordinates and the 3D model region; use the spatial mapping relationship to convert the image coordinates into 3D spatial coordinates on the target platform model to obtain the preliminary positioning coordinates of the defect on the target platform model; Acquire adjacent posture images corresponding to the posture image; perform cross-validation on the adjacent posture images to obtain the cross-validation results; based on the cross-validation results, determine the image coordinates of the defect in the adjacent posture images, and correct the preliminary positioning coordinates to obtain the precise three-dimensional spatial coordinates of the defect on the target platform model. The precise three-dimensional spatial coordinates are associated with and stored with the defect type label, and the associated and stored precise three-dimensional spatial coordinates and defect type label are used as defect location data.

[0073] Furthermore, defect type labels and defect location coordinates are extracted from the quality defect data, where the defect location coordinates are the pixel coordinates of the defect in the posture-captured image. Obtain the shooting posture parameters and viewpoint parameters corresponding to the posture-captured image; based on the shooting posture parameters, retrieve the viewpoint image of the target platform model in that posture from the virtual rendering environment; Based on the defect type label, the preset detection area corresponding to the defect type is found from the preset detection area mapping table, and the 3D model area corresponding to the preset detection area is extracted from the target platform model perspective image.

[0074] It should be noted that a spatial mapping relationship between image coordinates and the 3D model region is established. The specific method is as follows: obtain the intrinsic and extrinsic parameter matrices of the virtual camera; construct the projection relationship from the camera coordinate system to the image coordinate system; for 3D points within the preset detection area, fit the spatial plane equation of the region using the least squares method; solve the system of image coordinates and spatial plane equations to obtain the preliminary 3D spatial coordinates of the defect on the target platform model.

[0075] Furthermore, acquire at least one posture image adjacent to the current posture image, where adjacent is defined as being adjacent in position in the sewing machine table flow sequence or adjacent in time in the posture adjustment sequence. Repeat the above steps for images taken in adjacent poses to obtain the image coordinates of the defect in the adjacent images, and calculate the preliminary three-dimensional spatial coordinates of the image coordinates in the adjacent viewpoints; Calculate the spatial distance between the initial positioning coordinates and the positioning coordinates under adjacent viewpoints; when the spatial distance is less than a set threshold, take the average of the two coordinates as the corrected accurate three-dimensional spatial coordinates; when the spatial distance is greater than the threshold, perform a weighted average based on the confidence scores under each viewpoint, and give greater weight to the ones with higher confidence scores.

[0076] Furthermore, the corrected precise three-dimensional spatial coordinates are associated with the defect type label and stored to form defect location data, which is used for subsequent defect marking processing.

[0077] Furthermore, this step establishes a spatial mapping between image coordinates and the 3D model, and uses cross-verification of adjacent viewpoints to correct the positioning, thereby achieving precise 3D spatial positioning of the defect on the sewing machine table model, providing a reliable basis for subsequent marking processing.

[0078] Specifically, based on the posture comparison image, online comparative detection of surface defects is performed on the posture-captured image, and the second detection result is recorded. The specific steps are as follows: Obtain the shooting posture corresponding to the shooting posture image, and extract the posture reference image corresponding to the shooting posture from the posture reference image as the reference reference image; use the shooting posture to determine the target detection area exposed by the sewing machine table in the posture shooting image; In the reference reference image, the corresponding reference region image is obtained based on the target detection region; in the pose-captured image, the test region image is obtained by using the target detection region; the test region image and the reference region image are aligned pixel-level to obtain the aligned test region image and the aligned reference region image. The aligned test region image and the aligned reference region image are differentially processed to generate a difference image; connected component analysis is performed on the difference image to extract the difference region; based on the position of the difference region in the pose image, defect candidate image blocks are extracted from the pose image. The candidate image block of the defect is input into the preset surface defect classification model. The preset surface defect classification model is used to extract texture features and identify defect types of the candidate image block of the defect to obtain the defect type label and defect confidence score corresponding to the candidate image block of the defect. The defect confidence score is used to determine whether the candidate image block of the defect belongs to a real surface defect and obtain the defect judgment result. Based on the defect judgment results, when a candidate image block is determined to be a real surface defect, the defect type label, defect location coordinates, and defect size data corresponding to the candidate image block are recorded. The defect type label, defect location coordinates, and defect size data are then summarized to obtain the second detection result.

[0079] Furthermore, in this step, the shooting posture parameters corresponding to the shooting posture image are obtained. The shooting posture parameters include the rotation angle and tilt angle of the platform in three-dimensional space. Based on the shooting posture parameters and combined with the pre-calibrated detectable surface area in the target platform model, the detection area exposed by the sewing machine platform in the image under the current shooting posture is calculated. The specific calculation method is as follows: extract the normal vector of each detection surface from the target platform model; calculate the angle between the shooting view direction and the normal vector of each detection surface; mark the detection surface with the angle less than the set threshold as the exposed surface in the current posture; and take the projection area of ​​the exposed surface on the image plane as the target detection area.

[0080] Furthermore, in this step, a reference region image is cropped from the reference reference image based on the target detection region; and a test region image of the same spatial range is cropped from the pose-captured image. Pixel-level alignment employs a feature-point-based registration method: feature points are extracted from both the reference region image and the image to be tested; feature points are matched using a feature matcher; matching point pairs are selected using a filtering algorithm, and the perspective transformation matrix is ​​calculated; the image to be tested is transformed according to the perspective transformation matrix to align it with the reference region image at the pixel level.

[0081] It should be noted that after converting the aligned test area image and the reference area image into grayscale images, the difference in grayscale values ​​is calculated pixel by pixel to generate a difference image; specifically, the absolute value of the difference in grayscale values ​​at the same pixel position between the test area image and the reference area image is calculated. The difference image is binarized to set an adaptive threshold. This threshold is obtained by calculating the mean and standard deviation of the gray values ​​of the difference image and adding the mean and standard deviation by a preset multiple. Pixels with gray values ​​greater than this threshold are marked as difference pixels. Connectivity analysis is performed on the binarized difference image. The eight-neighbor connectivity rule is used to extract connected regions with an area greater than a set threshold as difference regions.

[0082] It should be noted that, based on the position of the difference region in the posture-captured image, the bounding rectangle of the difference region is expanded, and the expansion margin is a preset ratio of the width and height of the difference region to form a defect candidate region; the defect candidate region is then extracted from the posture-captured image to obtain a defect candidate image block.

[0083] Furthermore, the candidate image patches for defects are input into a preset surface defect classification model. The surface defect classification model is a classification network based on a lightweight convolutional neural network, and the network structure includes: an input layer for receiving image patches of fixed size; a convolutional layer for extracting texture features layer by layer; a global average pooling layer; and a fully connected layer for outputting the probability distribution of defect types.

[0084] The network output layer uses a classification activation function to output defect type labels and their confidence scores. Defect type labels include scratches, dents, burrs, oil stains, uneven coatings, etc.

[0085] When the confidence score is greater than a set threshold, it is determined to be a real surface defect, and the defect type label, defect location coordinates, and defect size data are recorded. The defect location coordinates are the coordinates of the center point of the defect candidate region in the original pose image, and the defect size data are the width and height of the bounding rectangle of the defect candidate region.

[0086] Furthermore, this step achieves accurate extraction and automatic identification of surface defects through pixel-level alignment, differentiation, and deep learning classification of the pose comparison image and the actual image, effectively reducing false detections and completely recording defect information.

[0087] It should be noted that the specific implementation of online structural dimension comparison and detection is as follows: Using the pose comparison image as a reference, structural dimensions are detected in the pose-captured images; specific steps include: Edge extraction: A sub-pixel edge extraction algorithm is used to extract key edges of the platform in the pose capture image and pose comparison image, such as outer contour edges, hole edges, and slot edges. Feature localization: The least squares method is used to fit the circular edge to obtain the coordinates of the hole center and the hole diameter; the straight edge is fitted to obtain the edge position, and then the length and width of the platform, as well as the distance between holes and between holes and edges are calculated; Size Comparison: All dimensional parameters (in pixels) measured in the pose-captured images are converted to actual physical dimensions (in millimeters) according to camera calibration coefficients, and then compared item by item with the standard dimensions of corresponding features in the pose comparison image. The deviation value is calculated. Result determination: Set the tolerance range for each dimension (for example, the hole diameter tolerance is ±0.05 mm, and the hole spacing tolerance is ±0.1 mm); if the deviation value exceeds the corresponding tolerance range, record the dimensional defect, including the defect type (such as "hole diameter is too large"), the measured value, the standard value, and the deviation amount, and summarize them as the first test result.

[0088] The detailed process for online comparative detection of surface defects is as follows: After obtaining the difference image and extracting the defect candidate image blocks, the following further processing is performed: The surface defect classification model uses a pre-trained lightweight convolutional neural network; the training dataset of this network contains a large number of labeled image patches of surface defects on the platform (such as scratches, pits, burrs, oil stains) and control image patches without defects. After inputting candidate image patches for defects into the model, the model outputs a probability vector, for example, [scratches: 90%, dents: 5%, burrs: 3%, normal: 2%]. A classification confidence threshold of 75% is set. When the maximum probability value exceeds this threshold, the classification result is accepted, and the candidate image patch is labeled as a true defect of the corresponding type. For each image patch identified as a genuine defect, the system records the following information as part of the second detection result: defect type label (e.g., "scratch"), defect location (i.e., the pixel coordinates of the center point of the image patch in the original pose image), defect size (the width and height of the image patch, representing the physical extent of the defect), and defect severity (which can be quantified by the probability value output by the model or by calculating the average pixel grayscale difference in the defect area, such as "minor", "moderate", or "severe"). It should be understood that although the steps in the flowcharts of the various embodiments of the present invention are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the various embodiments may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0089] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0090] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0091] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

[0092] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A machine vision-based online inspection method for the processing quality of sewing machine tabletops, characterized in that, The method specifically includes the following steps: The processing production line is monitored, and when the sewing machine table is transferred, the sewing machine table is initially photographed and basic identification is performed to determine the target table type. According to the target platform type, match the corresponding target platform model; perform environmental monitoring, obtain environmental monitoring data, create a virtual rendering environment based on the environmental monitoring data, and render and load the target platform model in the virtual rendering environment; The posture is adjusted and controlled during the movement of the sewing machine table; The sewing machine table is followed and the production line is paused after the posture adjustment control is performed. The posture image is captured. The posture of the captured image is recognized to determine the shooting posture, and the target platform model is adjusted accordingly in the virtual rendering environment to obtain a posture comparison image. Based on the posture comparison image, the posture captured image is subjected to online comparative detection and processing of processing quality.

2. The online inspection method for sewing machine table processing quality based on machine vision according to claim 1, characterized in that, The process of monitoring the flow of the production line, including initial photography and basic identification of the sewing machine table when it enters the sewing machine table to determine the target table type, specifically includes the following steps: Monitor the flow of the processing production line and obtain flow monitoring data; The flow monitoring data is identified to determine whether a sewing machine table has flowed into it; When a sewing machine table is being transported in, an initial photograph is taken of the sewing machine table to obtain an initial photographed image; The initial captured image is used for basic identification to determine the type of target platform.

3. The online inspection method for sewing machine table processing quality based on machine vision according to claim 2, characterized in that, The basic identification of the initial captured image to determine the target platform type specifically includes the following steps: Basic feature recognition is performed on the initial captured image to extract basic feature data; In a preset feature database, feature matching is performed on the basic feature data, and the feature matching results are recorded; Based on the feature matching results, the target table type of the sewing machine table to be transferred in is determined from the preset processing table types.

4. The online inspection method for sewing machine table processing quality based on machine vision according to claim 1, characterized in that, The steps of matching the corresponding target platform model according to the target platform type, performing environmental monitoring, acquiring environmental monitoring data, creating a virtual rendering environment based on the environmental monitoring data, and rendering and loading the target platform model in the virtual rendering environment specifically include the following steps: According to the target platform type, match the corresponding target platform model from the preset model database; Conduct environmental monitoring and obtain environmental monitoring data; Based on the environmental monitoring data, a corresponding virtual rendering environment is created; In the virtual rendering environment, the target platform model is rendered and loaded; The specific steps for creating a corresponding virtual rendering environment based on the environmental monitoring data are as follows: Illumination intensity and color temperature data that affect visual imaging are obtained from environmental monitoring data; using the illumination intensity and color temperature data, a basic rendering scene template is obtained by matching in a preset virtual rendering scene template library. Texture features are extracted from the background area in the initial captured image to obtain background texture data; the background texture data is used as texture mapping parameters to dynamically replace the background texture in the basic rendering scene template to generate a customized rendering scene. In the customized rendering scene, light intensity data, color temperature data, and preset production line light source layout parameters are used to construct a virtual lighting model with directional light sources, and the number and direction of the light sources are matched with the actual light source layout of the processing production line. The virtual lighting model is used to perform lighting rendering on the customized rendering scene to obtain a customized rendering scene with completed lighting rendering. Extract the material attribute data of the target platform model from its preset attributes; based on the material attribute data, in the customized rendering scene after completing the lighting rendering, parametrically adjust the reflection characteristics and shadow projection characteristics of the target platform model surface to obtain a high-fidelity virtual detection environment consistent with the visual performance of the sewing machine platform during posture shooting. In a high-fidelity virtual inspection environment, the target platform model is loaded, and the virtual lighting model and background texture data in the high-fidelity virtual inspection environment are used to render and load the target platform model to obtain the corresponding virtual rendering environment.

5. The online inspection method for sewing machine table processing quality based on machine vision according to claim 1, characterized in that, The posture adjustment control during the movement of the sewing machine table specifically includes the following steps: Determine the position for attitude adjustment; Real-time position of the sewing machine table during the transfer process; Based on the posture adjustment position, a matching analysis is performed on the real-time flow position to determine whether posture adjustment is required. When posture adjustment is required, perform the corresponding posture adjustment control.

6. The online inspection method for sewing machine table processing quality based on machine vision according to claim 5, characterized in that, The specific steps for performing attitude adjustment control when it is necessary include the following: Based on the target platform type, obtain the adjustment process parameters for attitude adjustment; When attitude adjustment is required, select the target adjustment parameter from the adjustment process parameters; Adjust the parameters according to the target and generate the corresponding attitude adjustment signal; Based on the attitude adjustment signal, perform corresponding attitude adjustment control.

7. The online inspection method for sewing machine table processing quality based on machine vision according to claim 5, characterized in that, The process of following the sewing machine table's rotation, adjusting its posture, pausing the production line, and capturing posture images specifically includes the following steps: The sewing machine table is moved and followed according to the real-time position; After the attitude adjustment control is completed, a production line pause command is generated; According to the production line stop command, the production line flow is stopped and controlled. During attitude adjustment control, corresponding attitude shooting is performed to acquire attitude shooting images.

8. The online inspection method for sewing machine table processing quality based on machine vision according to claim 1, characterized in that, The process of performing posture recognition on the captured image to determine the shooting posture, and then adjusting the corresponding posture of the target platform model in the virtual rendering environment to obtain a posture comparison image specifically includes the following steps: The posture of the captured image is subjected to posture recognition to determine the shooting posture; According to the shooting posture, the target platform model is adjusted accordingly. Obtain the viewpoint parameters corresponding to the posture-captured image; According to the aforementioned viewpoint parameters, the target platform model after attitude adjustment is cropped accordingly to obtain an attitude comparison image. Specifically, according to the aforementioned viewpoint parameters, the target platform model after attitude adjustment is cropped accordingly to obtain an attitude comparison image. The specific steps are as follows: The position and orientation parameters of the virtual camera in three-dimensional space are determined based on the viewpoint parameters, and the virtual lighting model parameters corresponding to the pose-captured images are extracted from the virtual rendering environment. The pose of the virtual camera is set in the virtual rendering environment according to the position and orientation parameters, and the imaging parameters of the virtual camera are matched and adjusted using the virtual lighting model parameters to obtain the virtual camera after the settings are completed. The virtual camera, once set up, is used to capture images of the target platform model after its posture has been adjusted, in order to obtain an initial cropped image. Extract image resolution and image format parameters from the pose-captured image; based on the image resolution and image format parameters, perform resolution and format conversion on the initial cropped image to obtain an intermediate cropped image with the same image size and image encoding as the pose-captured image; use the intermediate cropped image as the pose reference image.

9. The online inspection method for sewing machine table processing quality based on machine vision according to claim 1, characterized in that, The online comparison detection and processing of the processing quality of the posture-captured image based on the posture comparison image specifically includes the following steps: Based on the posture comparison image, the posture image is subjected to online comparison and detection of processing quality to determine whether there are any processing quality defects. To determine if there are processing quality defects, obtain quality defect data; Based on the quality defect data, quality defects are located and defect location data is obtained; Based on the defect location data, the sewing machine table is marked with defects. The specific steps for locating quality defects and obtaining defect location data based on the aforementioned quality defect data are as follows: Extract defect type labels and defect location coordinates from quality defect data; use defect type labels to determine the preset detection area on the sewing machine table corresponding to the defect type labels, and obtain the image coordinates of the defect in the posture-captured image based on the defect location coordinates; Obtain the shooting posture and viewpoint parameters corresponding to the shooting posture and viewpoint parameters in the shooting posture and viewpoint image; based on the shooting posture and viewpoint parameters, retrieve the viewpoint image of the target platform model that matches the shooting posture and viewpoint parameters from the virtual rendering environment; extract the 3D model region corresponding to the preset detection area from the viewpoint image of the target platform model. Establish the spatial mapping relationship between the image coordinates of the defect in the posture-captured image and the 3D model region; use the spatial mapping relationship to convert the image coordinates into 3D spatial coordinates on the target platform model to obtain the preliminary positioning coordinates of the defect on the target platform model; Acquire adjacent posture images corresponding to the posture image; perform cross-validation on the adjacent posture images to obtain the cross-validation results; based on the cross-validation results, determine the image coordinates of the defect in the adjacent posture images, and correct the preliminary positioning coordinates to obtain the precise three-dimensional spatial coordinates of the defect on the target platform model. The precise three-dimensional spatial coordinates are associated with and stored with the defect type label, and the associated and stored precise three-dimensional spatial coordinates and defect type label are used as defect location data.

10. The online inspection method for sewing machine table processing quality based on machine vision according to claim 9, characterized in that, The online comparison and detection of processing quality of the posture-captured image based on the posture comparison image to determine whether there are processing quality defects specifically includes the following steps: Based on the posture comparison image, the structural dimensions of the posture-captured image are compared and detected online, and the first detection result is recorded. Based on the posture comparison image, online comparative detection of surface defects is performed on the posture captured image, and the second detection result is recorded; Based on the combined results of the first and second tests, a processing quality analysis is performed to determine whether any processing quality defects exist. Specifically, based on the posture comparison image, online comparative detection of surface defects is performed on the posture-captured image, and the second detection result is recorded. The specific steps are as follows: Obtain the shooting posture corresponding to the shooting posture image, and extract the posture reference image corresponding to the shooting posture from the posture reference image as the reference reference image; use the shooting posture to determine the target detection area exposed by the sewing machine table in the posture shooting image; In the reference reference image, the corresponding reference region image is obtained based on the target detection region; in the pose-captured image, the test region image is obtained by using the target detection region; the test region image and the reference region image are aligned pixel-level to obtain the aligned test region image and the aligned reference region image. The aligned test region image and the aligned reference region image are differentially processed to generate a difference image; connected component analysis is performed on the difference image to extract the difference region; based on the position of the difference region in the pose image, defect candidate image blocks are extracted from the pose image. The candidate image block of the defect is input into the preset surface defect classification model. The preset surface defect classification model is used to extract texture features and identify defect types of the candidate image block of the defect to obtain the defect type label and defect confidence score corresponding to the candidate image block of the defect. The defect confidence score is used to determine whether the candidate image block of the defect belongs to a real surface defect and obtain the defect judgment result. Based on the defect judgment results, when a candidate image block is determined to be a real surface defect, the defect type label, defect location coordinates, and defect size data corresponding to the candidate image block are recorded. The defect type label, defect location coordinates, and defect size data are then summarized to obtain the second detection result.

Images