Surface quality detection method and device for high-precision metal component

Abstract

The invention discloses a surface quality detection method and device for a high-precision metal component, and relates to the technical field of image processing. The method comprises the following steps: acquiring a metal component surface image, and acquisition control information and acquisition environment information corresponding to the metal component surface image; performing reference image registration on the metal component, and establishing a registration reference image domain; performing acquisition scene twinning correction on the registration reference image domain to obtain a corrected reference image domain; mapping pixel difference comparison is carried out on the surface image of the metal component, and an abnormal pixel sequence of the surface of the component is determined; inputting the component surface abnormal pixel sequence into a component surface quality detection multi-channel to obtain a component surface quality detection result; and when the surface quality detection result of the component is qualified, the metal component is transported to a qualified component warehouse. The technical problems that in the prior art, metal component surface quality detection is low in reliability and unstable in detection result are solved, and the technical effect of improving the detection accuracy is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a surface quality detection method and device for high-precision metal components. Background Art

[0002] The surface quality of metal components is crucial to the performance, service life and safety of the product. However, traditional surface quality detection methods often rely on manual detection or automatic detection based on simple image processing, which has problems such as low detection accuracy, large subjective errors, and poor environmental adaptability, making it difficult to meet the quality control requirements of high-precision manufacturing. With the development of industrial intelligence, automated detection technology based on image processing and artificial intelligence has gradually become an important direction for improving the level of surface quality detection of metal components. However, when faced with complex acquisition environments, different types of surface defects and high-precision detection requirements, existing technologies still have problems such as unstable detection results and high false detection rates. Summary of the invention

[0003] The present application provides a method and device for detecting the surface quality of high-precision metal components, which solves the technical problems of low reliability and unstable detection results of metal component surface quality detection in the prior art.

[0004] In a first aspect of the present application, a surface quality detection method for a high-precision metal component is provided, the method comprising: The metal component is imaged according to an industrial camera to obtain a surface image of the metal component, as well as acquisition control information and acquisition environment information corresponding to the surface image of the metal component; the component design information of the metal component is retrieved, and the metal component is subjected to reference image registration according to the component design information to establish a registration reference image domain; based on the acquisition control information and the acquisition environment information, the registration reference image domain is subjected to acquisition scene twin correction according to a scene adaptive image correction model to obtain a correction reference image domain; the surface image of the metal component is mapped and compared with pixel differences according to the correction reference image domain to determine a sequence of abnormal pixels on the component surface; the sequence of abnormal pixels on the component surface is input into a component surface quality detection multi-channel to obtain a component surface quality detection result; when the component surface quality detection result is qualified, the metal component is transported to a qualified component warehouse.

[0005] A second aspect of the present application provides a surface quality detection device for high-precision metal components, the device comprising: An image acquisition module is used to acquire images of metal components according to an industrial camera to obtain surface images of the metal components, as well as acquisition control information and acquisition environment information corresponding to the surface images of the metal components; an image registration module is used to retrieve component design information of the metal components, and perform reference image registration on the metal components according to the component design information to establish a registration reference image domain; a correction module is used to perform acquisition scene twin correction on the registration reference image domain according to a scene adaptive image correction model based on the acquisition control information and the acquisition environment information to obtain a correction reference image domain; a comparison module is used to perform mapping pixel difference comparison on the surface image of the metal components according to the correction reference image domain to determine the abnormal pixel sequence on the component surface; a detection module is used to input the abnormal pixel sequence on the component surface into a component surface quality detection multi-channel to obtain a component surface quality detection result; a judgment module is used to transport the metal components to a qualified component warehouse when the component surface quality detection result is qualified.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: First, the image of the metal component is acquired by an industrial camera to obtain the surface image of the metal component, as well as the acquisition control information and acquisition environment information corresponding to the surface image of the metal component. Next, the component design information of the metal component is retrieved, and the metal component is subjected to reference image registration according to the component design information to establish a registration reference image domain. Then, based on the acquisition control information and the acquisition environment information, the registration reference image domain is subjected to acquisition scene twin correction according to the scene adaptive image correction model to obtain a correction reference image domain. Further, the surface image of the metal component is mapped and compared with pixel differences according to the correction reference image domain to determine the abnormal pixel sequence on the component surface. Finally, the abnormal pixel sequence on the component surface is input into the component surface quality detection multi-channel to obtain the component surface quality detection result; when the component surface quality detection result is qualified, the metal component is transported to a qualified component warehouse. The technical problems of low reliability and unstable detection results of metal component surface quality detection in the prior art are solved, and the technical effect of improving detection accuracy is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0008] Figure 1 A schematic flow chart of a surface quality detection method for high-precision metal components provided in an embodiment of the present application; Figure 2 A schematic structural diagram of a surface quality detection device for high-precision metal components provided in an embodiment of the present application.

[0009] Explanation of the reference numerals: image acquisition module 11 , image registration module 12 , correction module 13 , comparison module 14 , detection module 15 , judgment module 16 . DETAILED DESCRIPTION

[0010] The present application solves the technical problems of low reliability and unstable detection results of metal component surface quality detection in the prior art by providing a method and device for surface quality detection of high-precision metal components.

[0011] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0012] It should be noted that the terms "including" and "having" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or are inherent to these processes, methods, products or devices.

[0013] Embodiment 1, as Figure 1 As shown, the present application provides a surface quality detection method for high-precision metal components, wherein the method comprises: The image of the metal component is captured by an industrial camera to obtain a surface image of the metal component, as well as capture control information and capture environment information corresponding to the surface image of the metal component.

[0014] Specifically, select industrial cameras with high resolution, low noise and high dynamic range, and install them on stable brackets or mechanical arms according to the size and surface features of metal components to ensure constant shooting angle and field of view. Adjust the focal length of the camera to an appropriate position to ensure clear imaging, and configure appropriate light sources, such as ring light, coaxial light or low-angle oblique light, to enhance the visibility of metal surface defects, and use matte materials as background in the inspection area to reduce ambient light interference.

[0015] During the image acquisition process, the key parameters of the industrial camera need to be recorded synchronously as acquisition control information, including exposure time, gain, white balance, image resolution and frame rate. The exposure time is dynamically adjusted according to the reflective characteristics of the metal surface to avoid overexposure or underexposure. The gain parameter is optimized while ensuring the signal-to-noise ratio. The white balance parameter is adjusted according to the characteristics of the light source to ensure color accuracy. The image resolution and frame rate are set according to the detection requirements to balance the detection accuracy and processing speed.

[0016] After completing the camera settings, synchronously obtain the acquisition environment information, including ambient temperature, humidity, vibration status and light intensity. The infrared temperature sensor is used to monitor the ambient temperature in real time to ensure that temperature changes do not affect the camera's exposure and color performance; the humidity sensor is used to monitor the air humidity to prevent the lens from fogging or changes in the optical properties of the metal surface; the acceleration sensor is used to detect the vibration of the device to reduce the image blur caused by vibration; the light sensor is used to monitor the ambient light intensity and dynamically adjust the camera exposure compensation when necessary to ensure lighting consistency.

[0017] After completing the acquisition environment control, start the industrial camera and trigger the image acquisition signal. The triggering method can be synchronized with the assembly line through the PLC control system, or the shutter can be controlled by the detection software. When the camera shoots the surface of the metal component, it automatically stores the acquired surface image and records the acquisition control information and acquisition environment information corresponding to the image.

[0018] The component design information of the metal component is retrieved, and reference image registration is performed on the metal component according to the component design information to establish a registration reference image domain.

[0019] The component design information of the target metal component is retrieved from the component design database, which includes but is not limited to the component's geometric dimensions, surface morphology, material properties, processing parameters, and standard defect distribution model. The retrieved design information is parsed and structured to match the data format required for image registration. The collected surface image of the metal component is aligned and matched with the design information to establish a registration reference image domain for subsequent image analysis and comparison.

[0020] Furthermore, performing reference image registration on the metal component according to the component design information to establish a registration reference image domain includes: A component reference image library is obtained, wherein the component reference image library includes a plurality of component reference image samples and a plurality of component design data samples corresponding to the plurality of component reference image samples; a consistency evaluation is performed on the plurality of component design data samples according to the component design information to obtain a plurality of component design registration coefficients; based on the plurality of component design registration coefficients, the plurality of component reference image samples are optimally screened according to a component design registration threshold to generate the registration reference image domain.

[0021] The component reference image library pre-stores multiple component reference image samples and contains component design data corresponding to these samples, such as CAD models, 3D scanning data, standard process parameters, and historical inspection data.

[0022] According to the component design information of the target metal component, the consistency evaluation of multiple component design data samples in the component reference image library is performed to calculate the component design registration coefficient. Specifically, the key parameters of the component design information, such as dimensional tolerance, shape characteristics, surface texture characteristics, processing technology type, etc., are extracted and compared with the corresponding parameters in the reference image library; the comparison method can use a multi-dimensional feature matching algorithm, such as numerical similarity calculation based on Euclidean distance or Mahalanobis distance, or feature vector matching using a deep learning model (such as convolutional neural network CNN), so as to calculate the design registration coefficient between the target component and the sample in the library. The registration coefficient reflects the matching degree between the current metal component and different reference image samples. The higher the value, the higher the similarity.

[0023] After obtaining multiple component design registration coefficients, multiple component reference image samples are optimally screened based on the set component design registration threshold. Specifically, component reference image samples corresponding to component design registration coefficients greater than or equal to the component design registration threshold are recorded as registration reference images, and the screened reference image sample set constitutes a registration reference image domain, which includes M registration reference images.

[0024] Based on the acquisition control information and the acquisition environment information, the registration reference image domain is subjected to acquisition scene twin correction according to a scene adaptive image correction model to obtain a correction reference image domain.

[0025] The acquisition control information includes the camera parameters used in the image acquisition process, such as exposure time, gain, focal length, resolution, shutter speed, etc.; while the acquisition environment information includes environmental factors such as light intensity, temperature and humidity, and vibration status during acquisition.

[0026] According to these acquisition control information and acquisition environment information, the environmental features of the current acquisition scene are first extracted and compared with the existing reference image samples. Next, the scene adaptive image correction model is used to identify the difference between the current acquisition scene and the registration reference image domain. Specifically, the twin network technology can be used. The twin network predicts the deviation of the acquisition scene by comparing the current scene information with the environmental parameters of the reference image, and adjusts various parameters of the image (such as brightness, contrast, color correction, etc.) to compensate for these deviations, and finally obtains the corrected reference image domain.

[0027] Furthermore, based on the acquisition control information and the acquisition environment information, the registration reference image domain is subjected to acquisition scene twin correction according to a scene adaptive image correction model to obtain a correction reference image domain, including: According to the registration reference image domain, extract the mth registration reference image, where m is a positive integer; retrieve the acquisition control parameters and acquisition environment parameters of the mth registration reference image to obtain the mth acquisition scene information; take the acquisition control information and the acquisition environment information as the current acquisition scene information, and perform an acquisition scene twin test on the mth acquisition scene information according to the current acquisition scene information to obtain the mth acquisition scene twin test result; when the mth acquisition scene twin test result is failure, activate the scene adaptive image correction model; identify the deviation between the mth acquisition scene information and the current acquisition scene information to generate the mth scene correction factor; input the mth scene correction factor and the mth registration reference image into the scene adaptive image correction model to obtain the mth correction reference image, and add the mth correction reference image to the correction reference image domain.

[0028] In the process of performing twin calibration of the acquisition scene on the registration reference image domain according to the acquisition control information and the acquisition environment information, the mth registration reference image is first extracted from the established registration reference image domain. The mth registration reference image contains the image generated when the metal component is registered with the reference image according to the metal component design information, and the image serves as the basis for subsequent calibration. Next, the acquisition control parameters and acquisition environment parameters corresponding to the mth registration reference image are retrieved to form the mth acquisition scene information; the acquisition control parameters include the camera's exposure, gain, resolution and other settings, while the acquisition environment parameters include external environmental factors such as temperature, humidity, and light intensity. On this basis, the acquisition control information and acquisition environment information of the current acquisition scene are used as the current acquisition scene information, and compared with the mth acquisition scene information to perform twin calibration of the acquisition scene. Through algorithms such as twin neural networks, the similarity between the current acquisition scene and the mth acquisition scene is calculated to obtain the twin calibration result of the mth acquisition scene. The twin calibration process evaluates the similarity between the current acquisition conditions and the mth acquisition scene by analyzing the differences in various parameters (such as light, temperature, humidity, etc.) in the scene information.

[0029] If the twin inspection result of the mth acquisition scene fails, that is, the current acquisition scene differs greatly from the mth acquisition scene, the scene adaptive image correction model is activated. The model will identify the deviation based on the difference between the current acquisition scene information and the mth acquisition scene information, and generate the mth scene correction factor. The mth scene correction factor is used to adjust the image deviation caused by the difference in environment or control parameters during the image acquisition process, so as to ensure the accuracy of the image. The generated mth scene correction factor is input into the scene adaptive image correction model together with the mth registration reference image for further image correction. The scene adaptive image correction model adjusts the brightness, contrast, color and other features of the image according to the correction factor and the reference image, so that the image is closer to the ideal state, and finally obtains the mth correction reference image. Finally, the corrected mth correction reference image is added to the correction reference image domain, and the correction image domain is updated to provide a more accurate reference in subsequent quality inspections, wherein the correction reference image domain contains M correction reference images.

[0030] Further, performing a collection scene twin test on the mth collection scene information according to the current collection scene information to obtain the mth collection scene twin test result includes: Based on the twin neural network, the twin degree evaluation is performed on the current collection scene information and the mth collection scene information to obtain the mth scene twin coefficient; it is determined whether the mth scene twin coefficient is greater than or equal to the predetermined scene twin coefficient; if the mth scene twin coefficient is greater than or equal to the predetermined scene twin coefficient, the twin inspection result of the mth collection scene is passed; if the mth scene twin coefficient is less than the predetermined scene twin coefficient, the twin inspection result of the mth collection scene is failed.

[0031] In the process of performing twin inspection of the mth acquisition scene information according to the current acquisition scene information, the twin neural network is first used to evaluate the twin degree of the current acquisition scene information and the mth acquisition scene information. The twin neural network is a deep learning model specifically used to measure similarity, which can extract and compare the feature vectors of two input data to calculate the similarity between the two. Specifically, the control parameters (such as camera exposure, gain, resolution) and environmental parameters (such as temperature, humidity, and light intensity) of the current acquisition scene are input into the twin neural network, and the corresponding parameters of the mth acquisition scene are input at the same time. After feature extraction and comparison operations, the mth scene twin coefficient is generated. The mth scene twin coefficient is used to quantify the similarity between the current acquisition scene information and the mth acquisition scene information. The closer the value of the coefficient is to 1, the higher the similarity between the two scenes, and the closer the value is to 0, the greater the difference between the two. Subsequently, it is determined whether the mth scene twin coefficient is greater than or equal to the preset scene twin coefficient threshold. The threshold is a standard value set based on experimental data and experience, which is used to distinguish whether the acquisition scene can directly use the existing benchmark image for quality detection. If the twin coefficient of the mth scene is greater than or equal to the predetermined scene twin coefficient threshold, it means that the current acquisition scene is highly similar to the mth acquisition scene, that is, the acquisition environment and acquisition parameters have not changed much, so the system determines that the twin inspection result of the mth acquisition scene is passed. In this case, the registration reference image of the mth acquisition scene can be used directly for subsequent quality inspection without additional correction. On the contrary, if the twin coefficient of the mth scene is less than the predetermined scene twin coefficient threshold, it means that there is a large difference between the current acquisition scene and the mth acquisition scene, which may be due to changes in the external environment or adjustments to the acquisition parameters, resulting in a decrease in the matching degree between the reference image and the current acquisition image. In this case, the twin inspection result of the mth acquisition scene is determined to be unsuccessful, and the scene adaptive image correction process is entered to further calculate the correction factor to ensure the accuracy of quality inspection.

[0032] Mapping pixel differences of the metal component surface image are compared according to the correction reference image domain to determine abnormal pixel sequences on the component surface.

[0033] Using the calibrated reference image domain as a reference, the original metal component surface image is mapped and compared with pixel differences. Through the comparison, the abnormal pixel sequence on the metal component surface can be determined. These abnormal pixels may represent quality problems such as scratches, rust, and deformation.

[0034] Furthermore, mapping pixel differences of the metal component surface image is compared according to the correction reference image domain to determine the abnormal pixel sequence on the component surface, including: The metal component surface image is mapped and analyzed for pixel differences according to the correction reference image domain to obtain a plurality of pixel difference evaluation clusters; the plurality of pixel difference evaluation clusters are respectively calculated for centralization values ​​to obtain a plurality of pixel difference characteristic values; it is determined whether the plurality of pixel difference characteristic values ​​are greater than or equal to a predetermined pixel difference characteristic value to obtain a plurality of pixel difference judgment results; abnormal pixels are identified on the metal component surface image based on the plurality of pixel difference judgment results to generate an abnormal pixel sequence on the component surface.

[0035] In the process of mapping pixel difference comparison of metal component surface image based on the correction reference image domain to determine the abnormal pixel sequence on the component surface, it is first necessary to analyze the mapping pixel difference of the metal component surface image. Specifically, according to the standard image data of the correction reference image domain, the current metal component surface image is geometrically aligned and color space matched so that the pixels at the same position can be directly compared. After completing the mapping and matching, each pixel in the metal component surface image is traversed, and the corresponding reference mapping pixel is found in the correction reference image domain, and the pixel difference value between the two is calculated, including but not limited to brightness difference, color deviation, edge feature change and texture inconsistency. Subsequently, the difference values ​​of all pixels are classified and aggregated to form multiple pixel difference evaluation clusters, each of which represents a class of similar pixel difference patterns. Next, a centralized value calculation is performed on each pixel difference evaluation cluster to extract multiple pixel difference feature values, such as statistical features such as mean, median, range or variance, which are used to evaluate the degree to which the pixels of this category deviate from the standard image. After obtaining multiple pixel difference feature values, a threshold judgment is performed on them to compare whether each pixel difference feature value is greater than or equal to a preset pixel difference feature threshold to obtain multiple pixel difference judgment results. If a pixel difference characteristic value exceeds the set threshold, it indicates that the pixel may be abnormal; otherwise, the pixel is considered to be in the normal range. Finally, based on multiple pixel difference judgment results, the metal component surface image is identified for abnormal pixels, and the component pixels corresponding to the pixel difference characteristic values ​​greater than or equal to the predetermined pixel difference characteristic value in the metal component surface image are recorded as abnormal pixels on the component surface, and finally a component surface abnormal pixel sequence is generated. The abnormal pixel sequence not only contains the position information of the abnormal pixels, but can also be used for abnormal pattern analysis, providing data support for subsequent quality inspection, defect classification and production optimization.

[0036] Furthermore, mapping pixel difference analysis is performed on the surface image of the metal component according to the correction reference image domain to obtain a plurality of pixel difference evaluation clusters, including: The surface image of the metal component is traversed to extract the first pixel of the component; mapping pixel extraction is performed on the correction reference image domain according to the first pixel of the component to obtain M reference mapping pixels, where M is a positive integer greater than 1; according to the M reference mapping pixels, difference evaluation is performed on the first pixel of the component respectively to obtain M pixel difference evaluation coefficients; the M pixel difference evaluation coefficients are used as a first pixel difference evaluation cluster, and the first pixel difference evaluation cluster is added to the multiple pixel difference evaluation clusters.

[0037] In the process of mapping pixel difference analysis of the metal component surface image based on the correction reference image domain, all pixel points of the metal component surface image are first traversed, and a single component pixel is extracted in sequence, which is recorded as the first pixel of the component. For the first pixel of the component, according to its position and characteristics in the metal component surface image, the correction reference image domain is queried, and M reference mapping pixels of the pixel in the correction reference image domain are extracted by using geometric registration and color correction methods, where M is a positive integer greater than 1. These reference mapping pixels are used to provide multiple reference samples to enhance the robustness and accuracy of detection.

[0038] After obtaining M reference mapping pixels, the difference between them and the first pixel of the component is evaluated respectively, and the M pixel difference evaluation coefficients are calculated; the evaluation process can be based on multiple dimensions, including but not limited to pixel brightness difference, color channel deviation, local gradient change, texture similarity, and edge feature comparison. Specifically, the color information (such as RGB, HSV or Lab color space value), brightness value, texture feature (such as local binary pattern LBP, gradient direction histogram HOG) and edge feature of the first pixel of the component are extracted; then, corresponding to the M reference mapping pixels, the deviation between them and the first pixel of the component in each feature dimension is calculated respectively, including color deviation (such as Euclidean distance or cosine similarity), brightness difference, gradient direction change rate, texture pattern similarity, etc. In the calculation process, weighted summation, principal component analysis (PCA) or deep learning model (such as convolutional neural network CNN or twin neural network) can be used to perform comprehensive normalization processing on multiple features to ensure the consistency of different feature scales and reduce noise interference. Finally, a pixel difference evaluation coefficient is generated for each reference mapping pixel, which reflects the matching degree between the reference mapping pixel and the first pixel of the component in the overall feature space. M reference mapping pixels correspond to M pixel difference evaluation coefficients respectively, and these coefficients are combined to form the difference evaluation result of the first pixel of the component, which is stored in the pixel difference evaluation cluster to support subsequent abnormal pixel detection and component surface quality analysis.

[0039] Next, the calculated M pixel difference evaluation coefficients are combined into a first pixel difference evaluation cluster and stored in a set of multiple pixel difference evaluation clusters. The set is used for subsequent statistical analysis to determine the global pixel difference pattern and further screen out possible abnormal areas on the component surface, providing a reliable basis for the final quality inspection.

[0040] The abnormal pixel sequence on the component surface is input into the component surface quality detection multi-channel to obtain the component surface quality detection result.

[0041] The component surface quality detection multi-channel includes a component surface quality assessment channel and a component surface quality inspection channel. In the component surface quality assessment channel, based on multiple pre-trained component surface quality assessment models, feature extraction and abnormal classification are performed on the abnormal pixel sequence on the component surface. Each assessment model analyzes different types of surface defects (such as scratches, pits, burrs, corrosion, pollution, etc.) and generates corresponding component surface quality assessment coefficients. The component surface quality assessment coefficient is input into the component surface quality inspection channel, and a comprehensive judgment is made on it according to the component surface quality inspection mechanism. If the component surface quality assessment coefficient is greater than or equal to the set component surface quality assessment threshold, the output component surface quality inspection result is qualified; if the component surface quality assessment coefficient is less than the set threshold, the output component surface quality inspection result is unqualified.

[0042] Furthermore, the abnormal pixel sequence on the component surface is input into the component surface quality detection multi-channel to obtain the component surface quality detection result, including: The component surface quality detection multi-channel includes a component surface quality assessment channel and a component surface quality inspection channel; the component surface abnormal pixel sequence is input into the component surface quality assessment channel to obtain a component surface quality assessment coefficient, wherein the component surface quality assessment channel includes K component surface quality assessment models, K is a positive integer greater than 1; the component surface quality assessment coefficient is input into the component surface quality inspection channel, and the component surface quality inspection result is output, wherein the component surface quality inspection channel includes a component surface quality inspection mechanism, and the component surface quality inspection mechanism includes that if the component surface quality assessment coefficient is greater than or equal to a component surface quality assessment threshold, the component surface quality inspection result is qualified, and if the component surface quality assessment coefficient is less than the component surface quality assessment threshold, the component surface quality inspection result is unqualified.

[0043] Based on the abnormal pixel sequence on the component surface, it is first input into the component surface quality detection multi-channel, wherein the component surface quality detection multi-channel includes a component surface quality assessment channel and a component surface quality inspection channel. In the component surface quality assessment channel, K component surface quality assessment models are used to perform multi-dimensional assessment on the abnormal pixel sequence on the component surface, and the corresponding K component surface quality assessment sub-coefficients are calculated, wherein K is a positive integer greater than 1. Through weighted average calculation, the K component surface quality assessment sub-coefficients are fused to generate the component surface quality assessment coefficient. Subsequently, the component surface quality assessment coefficient is input into the component surface quality inspection channel, and the quality is determined by the component surface quality inspection mechanism, wherein the component surface quality inspection mechanism includes comparing the component surface quality assessment coefficient with the component surface quality assessment threshold, if the component surface quality assessment coefficient is greater than or equal to the component surface quality assessment threshold, then the output component surface quality detection result is qualified; if the component surface quality assessment coefficient is less than the component surface quality assessment threshold, then the output component surface quality detection result is unqualified, and the abnormal detection result is recorded for further analysis and optimization of the detection model.

[0044] When the surface quality inspection result of the component is qualified, the metal component is transported to a qualified component warehouse.

[0045] When the surface quality inspection result of the component is qualified, the automated conveying equipment or robotic arm is dispatched to transport the metal component from the inspection station to the qualified component warehouse.

[0046] Furthermore, when the surface quality inspection result of the component is qualified, the metal component is transported to a qualified component warehouse, including: The geometric features of the metal component are verified according to the component design information to obtain the component geometric features verification results; when the component surface quality inspection results and the component geometric features verification results are both qualified, a component qualified signal is generated; according to the component qualified signal, the metal component is transported to the qualified component warehouse.

[0047] When the surface quality inspection result of the component is qualified, the geometric features of the metal component are first checked according to the component design information, the key geometric parameters are extracted, and compared with the standard design parameters to obtain the component geometric feature verification results. Subsequently, it is determined whether the component surface quality inspection results and the component geometric feature verification results are both qualified. If both are qualified, a component qualified signal is generated and sent to the automated transportation control module. After receiving the component qualified signal, the automated conveying equipment or robotic arm executes the transportation instruction to transport the metal component from the inspection station to the qualified component warehouse, and at the same time updates the inventory information of the warehouse management system to ensure that the storage records of qualified components are complete and traceable.

[0048] Furthermore, when the component surface quality detection result is unqualified, a component surface quality warning signal is generated.

[0049] When the component surface quality inspection result is unqualified, the component surface quality early warning mechanism is immediately triggered and a component surface quality early warning signal is generated. The component surface quality early warning signal will be notified to the operator or relevant personnel through the alarm system or monitoring interface and recorded in the quality management system for subsequent tracking and processing.

[0050] In summary, the embodiments of the present application have at least the following technical effects: First, the image of the metal component is acquired by an industrial camera to obtain the surface image of the metal component, as well as the acquisition control information and acquisition environment information corresponding to the surface image of the metal component. Next, the component design information of the metal component is retrieved, and the metal component is subjected to reference image registration according to the component design information to establish a registration reference image domain. Then, based on the acquisition control information and the acquisition environment information, the registration reference image domain is subjected to acquisition scene twin correction according to the scene adaptive image correction model to obtain a correction reference image domain. Further, the surface image of the metal component is mapped and compared with pixel differences according to the correction reference image domain to determine the abnormal pixel sequence on the component surface. Finally, the abnormal pixel sequence on the component surface is input into the component surface quality detection multi-channel to obtain the component surface quality detection result; when the component surface quality detection result is qualified, the metal component is transported to a qualified component warehouse. The technical problems of low reliability and unstable detection results of metal component surface quality detection in the prior art are solved, and the technical effect of improving detection accuracy is achieved.

[0051] Embodiment 2 is based on the same inventive concept as the surface quality detection method for high-precision metal components in the above embodiment. Figure 2 As shown, the present application provides a surface quality detection device for high-precision metal components, wherein the device comprises: An image acquisition module 11 is used to acquire images of metal components according to an industrial camera to obtain surface images of the metal components, as well as acquisition control information and acquisition environment information corresponding to the surface images of the metal components; an image registration module 12 is used to retrieve component design information of the metal components, and perform reference image registration on the metal components according to the component design information to establish a registration reference image domain; a correction module 13 is used to perform acquisition scene twin correction on the registration reference image domain according to a scene adaptive image correction model based on the acquisition control information and the acquisition environment information to obtain a correction reference image domain; a comparison module 14 is used to perform mapping pixel difference comparison on the surface image of the metal components according to the correction reference image domain to determine the abnormal pixel sequence on the component surface; a detection module 15 is used to input the abnormal pixel sequence on the component surface into a component surface quality detection multi-channel to obtain a component surface quality detection result; a judgment module 16 is used to transport the metal components to a qualified component warehouse when the component surface quality detection result is qualified.

[0052] Furthermore, the correction module 13 is used to perform the following method: According to the registration reference image domain, extract the mth registration reference image, where m is a positive integer; retrieve the acquisition control parameters and acquisition environment parameters of the mth registration reference image to obtain the mth acquisition scene information; take the acquisition control information and the acquisition environment information as the current acquisition scene information, and perform an acquisition scene twin test on the mth acquisition scene information according to the current acquisition scene information to obtain the mth acquisition scene twin test result; when the mth acquisition scene twin test result is failure, activate the scene adaptive image correction model; identify the deviation between the mth acquisition scene information and the current acquisition scene information to generate the mth scene correction factor; input the mth scene correction factor and the mth registration reference image into the scene adaptive image correction model to obtain the mth correction reference image, and add the mth correction reference image to the correction reference image domain.

[0053] Furthermore, the correction module 13 is used to perform the following method: Based on the twin neural network, the twin degree evaluation is performed on the current collection scene information and the mth collection scene information to obtain the mth scene twin coefficient; it is determined whether the mth scene twin coefficient is greater than or equal to the predetermined scene twin coefficient; if the mth scene twin coefficient is greater than or equal to the predetermined scene twin coefficient, the twin inspection result of the mth collection scene is passed; if the mth scene twin coefficient is less than the predetermined scene twin coefficient, the twin inspection result of the mth collection scene is failed.

[0054] Furthermore, the comparison module 14 is used to perform the following method: The metal component surface image is mapped and analyzed for pixel differences according to the correction reference image domain to obtain a plurality of pixel difference evaluation clusters; the plurality of pixel difference evaluation clusters are respectively calculated for centralization values ​​to obtain a plurality of pixel difference characteristic values; it is determined whether the plurality of pixel difference characteristic values ​​are greater than or equal to a predetermined pixel difference characteristic value to obtain a plurality of pixel difference judgment results; abnormal pixels are identified on the metal component surface image based on the plurality of pixel difference judgment results to generate an abnormal pixel sequence on the component surface.

[0055] Furthermore, the comparison module 14 is used to perform the following method: The surface image of the metal component is traversed to extract the first pixel of the component; mapping pixel extraction is performed on the correction reference image domain according to the first pixel of the component to obtain M reference mapping pixels, where M is a positive integer greater than 1; according to the M reference mapping pixels, difference evaluation is performed on the first pixel of the component respectively to obtain M pixel difference evaluation coefficients; the M pixel difference evaluation coefficients are used as a first pixel difference evaluation cluster, and the first pixel difference evaluation cluster is added to the multiple pixel difference evaluation clusters.

[0056] Furthermore, the detection module 15 is used to perform the following method: The component surface quality detection multi-channel includes a component surface quality assessment channel and a component surface quality inspection channel; the component surface abnormal pixel sequence is input into the component surface quality assessment channel to obtain a component surface quality assessment coefficient, wherein the component surface quality assessment channel includes K component surface quality assessment models, K is a positive integer greater than 1; the component surface quality assessment coefficient is input into the component surface quality inspection channel, and the component surface quality inspection result is output, wherein the component surface quality inspection channel includes a component surface quality inspection mechanism, and the component surface quality inspection mechanism includes that if the component surface quality assessment coefficient is greater than or equal to a component surface quality assessment threshold, the component surface quality inspection result is qualified, and if the component surface quality assessment coefficient is less than the component surface quality assessment threshold, the component surface quality inspection result is unqualified.

[0057] Furthermore, the image registration module 12 is used to perform the following method: A component reference image library is obtained, wherein the component reference image library includes a plurality of component reference image samples and a plurality of component design data samples corresponding to the plurality of component reference image samples; a consistency evaluation is performed on the plurality of component design data samples according to the component design information to obtain a plurality of component design registration coefficients; based on the plurality of component design registration coefficients, the plurality of component reference image samples are optimally screened according to a component design registration threshold to generate the registration reference image domain.

[0058] Furthermore, the judging module 16 is used to execute the following method: The geometric features of the metal component are verified according to the component design information to obtain the component geometric features verification results; when the component surface quality inspection results and the component geometric features verification results are both qualified, a component qualified signal is generated; according to the component qualified signal, the metal component is transported to the qualified component warehouse.

[0059] Furthermore, the judging module 16 is used to execute the following method: When the component surface quality detection result is unqualified, a component surface quality warning signal is generated.

[0060] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0061] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0062] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A surface quality detection method for high-precision metal components, characterized in that: The method comprises: Capturing the image of the metal component using an industrial camera to obtain a surface image of the metal component, as well as acquisition control information and acquisition environment information corresponding to the surface image of the metal component; Retrieving component design information of the metal component, and performing reference image registration on the metal component according to the component design information to establish a registration reference image domain; Based on the acquisition control information and the acquisition environment information, performing acquisition scene twin correction on the registration reference image domain according to a scene adaptive image correction model to obtain a correction reference image domain; Performing a mapping pixel difference comparison on the surface image of the metal component according to the correction reference image domain to determine an abnormal pixel sequence on the surface of the component; Inputting the abnormal pixel sequence on the component surface into the component surface quality detection multi-channel to obtain the component surface quality detection result; When the surface quality inspection result of the component is qualified, the metal component is transported to a qualified component warehouse.

2. A surface quality detection method for high-precision metal components as claimed in claim 1, characterized in that: Based on the acquisition control information and the acquisition environment information, performing acquisition scene twin correction on the registration reference image domain according to a scene adaptive image correction model to obtain a correction reference image domain, including: Extracting the mth registration reference image according to the registration reference image domain, where m is a positive integer; Retrieving acquisition control parameters and acquisition environment parameters of the mth registration reference image to obtain the mth acquisition scene information; Taking the acquisition control information and the acquisition environment information as current acquisition scene information, and performing an acquisition scene twin test on the mth acquisition scene information according to the current acquisition scene information to obtain an mth acquisition scene twin test result; When the twin inspection result of the mth acquisition scene fails, activating the scene adaptive image correction model; Identify deviations between the mth acquisition scene information and the current acquisition scene information to generate an mth scene correction factor; The mth scene correction factor and the mth registration reference image are input into the scene adaptive image correction model to obtain an mth correction reference image, and the mth correction reference image is added to the correction reference image domain.

3. A surface quality detection method for high-precision metal components as claimed in claim 2, characterized in that: Performing a collection scene twin test on the mth collection scene information according to the current collection scene information to obtain the mth collection scene twin test result includes: Based on the twin neural network, the twin degree evaluation is performed on the current collection scene information and the mth collection scene information to obtain the mth scene twin coefficient; Determining whether the mth scene twin coefficient is greater than or equal to a predetermined scene twin coefficient; If the m-th scene twin coefficient is greater than or equal to the predetermined scene twin coefficient, the m-th acquisition scene twin test result is passed; If the twin coefficient of the mth scene is less than the predetermined twin coefficient of the scene, the twin inspection result of the mth acquisition scene is failed.

4. A surface quality detection method for high-precision metal components as claimed in claim 1, characterized in that: Mapping pixel differences of the metal component surface image is compared according to the correction reference image domain to determine abnormal pixel sequences on the component surface, including: Performing a mapping pixel difference analysis on the surface image of the metal component according to the correction reference image domain to obtain a plurality of pixel difference evaluation clusters; performing concentrated value calculations on the multiple pixel difference evaluation clusters respectively to obtain multiple pixel difference feature values; Determine whether the plurality of pixel difference characteristic values ​​are greater than or equal to a predetermined pixel difference characteristic value, and obtain a plurality of pixel difference determination results; Based on the plurality of pixel difference judgment results, abnormal pixels are identified on the surface image of the metal component to generate an abnormal pixel sequence on the component surface.

5. A surface quality detection method for high-precision metal components as claimed in claim 4, characterized in that: Mapping pixel difference analysis is performed on the surface image of the metal component according to the correction reference image domain to obtain a plurality of pixel difference evaluation clusters, including: Traversing the surface image of the metal component and extracting the first pixel of the component; Extracting mapping pixels from the correction reference image domain according to the first pixel of the component to obtain M reference mapping pixels, where M is a positive integer greater than 1; According to the M reference mapping pixels, respectively perform difference evaluation on the first pixel of the component to obtain M pixel difference evaluation coefficients; The M pixel difference evaluation coefficients are used as a first pixel difference evaluation cluster, and the first pixel difference evaluation cluster is added to the plurality of pixel difference evaluation clusters.

6. A surface quality detection method for high-precision metal components as claimed in claim 1, characterized in that: Inputting the abnormal pixel sequence on the component surface into the component surface quality detection multi-channel to obtain the component surface quality detection result, including: The component surface quality detection multi-channel includes a component surface quality assessment channel and a component surface quality inspection channel; Inputting the abnormal pixel sequence on the component surface into the component surface quality assessment channel to obtain a component surface quality assessment coefficient, wherein the component surface quality assessment channel includes K component surface quality assessment models, where K is a positive integer greater than 1; The component surface quality assessment coefficient is input into the component surface quality inspection channel, and the component surface quality inspection result is output, wherein the component surface quality inspection channel includes a component surface quality inspection mechanism, and the component surface quality inspection mechanism includes that if the component surface quality assessment coefficient is greater than or equal to a component surface quality assessment threshold, the component surface quality inspection result is qualified, and if the component surface quality assessment coefficient is less than the component surface quality assessment threshold, the component surface quality inspection result is unqualified.

7. A surface quality detection method for high-precision metal components as claimed in claim 1, characterized in that: Performing reference image registration on the metal component according to the component design information to establish a registration reference image domain includes: Obtaining a component reference image library, wherein the component reference image library includes a plurality of component reference image samples and a plurality of component design data samples corresponding to the plurality of component reference image samples; Performing consistency evaluation on the plurality of component design data samples according to the component design information to obtain a plurality of component design registration coefficients; Based on the multiple component design registration coefficients, the multiple component reference image samples are optimally screened according to the component design registration threshold to generate the registration reference image domain.

8. A surface quality detection method for high-precision metal components as claimed in claim 1, characterized in that: When the surface quality inspection result of the component is qualified, the metal component is transported to a qualified component warehouse, including: Performing geometric feature verification on the metal component according to the component design information to obtain a component geometric feature verification result; When the component surface quality inspection result and the component geometric feature verification result are both qualified, a component qualified signal is generated; According to the component qualification signal, the metal component is transported to the qualified component warehouse.

9. A surface quality detection method for high-precision metal components as claimed in claim 1, characterized in that: When the component surface quality detection result is unqualified, a component surface quality warning signal is generated.

10. A surface quality detection device for high-precision metal components, characterized in that: The device is used to implement a surface quality detection method for high-precision metal components according to any one of claims 1 to 9, comprising: An image acquisition module is used to acquire images of metal components using an industrial camera to obtain a surface image of the metal component, as well as acquisition control information and acquisition environment information corresponding to the surface image of the metal component; An image registration module, used to retrieve component design information of the metal component, and perform reference image registration on the metal component according to the component design information to establish a registration reference image domain; A correction module, configured to perform a twin correction of the registration reference image domain based on the acquisition control information and the acquisition environment information according to a scene adaptive image correction model to obtain a correction reference image domain; A comparison module, used for performing a mapping pixel difference comparison on the surface image of the metal component according to the correction reference image domain, and determining an abnormal pixel sequence on the component surface; A detection module, used for inputting the abnormal pixel sequence on the component surface into the component surface quality detection multi-channel to obtain the component surface quality detection result; The judgment module is used to transport the metal component to a qualified component warehouse when the surface quality inspection result of the component is qualified.

Images