Hub quality detection method and system

Abstract

The invention relates to the technical field of image processing and industrial detection, and discloses a hub quality detection method and system, and the method comprises the steps: collecting a hub coating image through an industrial camera; introducing a U-Net network to perform bubble region segmentation on the image in combination with an attention mechanism; a ResNet network is combined with an edge enhancement method to extract a coating peeling area; and outputting a hub coating quality evaluation report. Compared with the prior art that the defect detection of the hub surface uniformity is carried out mostly depending on manual visual inspection or a traditional image processing method, and particularly under the condition of uneven illumination or complex texture, the technical problem that high-precision automatic detection of the hub coating uniformity is difficult to realize is solved. And a fusion mechanism aiming at bubble, peeling and thickness error points is constructed, so that the problems of boundary false detection, misjudgment and incapability of quantitative output in a traditional method are avoided, and the accuracy and the automation degree of hub coating quality detection are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing and industrial detection, and in particular relates to a wheel hub quality detection method and system. Background Art

[0002] Currently, wheel coating quality inspection primarily relies on manual visual inspection and traditional image processing methods, such as edge detection and threshold segmentation techniques. However, these traditional methods have numerous shortcomings. For example, visual inspection relies on manual operation, which is not only subject to human error, but also slow and unsuitable for large-scale production. Traditional image processing methods, on the other hand, often rely on fixed thresholds or edge detection algorithms. These methods suffer from high rates of false positives and missed detections when dealing with defects such as complex coating backgrounds, irregular coating flaking, tiny bubbles, and uneven thickness. Existing technologies cannot fully meet the demand for high-precision, high-efficiency automated coating defect detection. This is particularly true in situations with complex coating textures, uneven illumination, or tiny defects (such as bubbles, flaking areas, and thickness errors). Traditional image processing methods still struggle to accurately analyze and detect coating uniformity. Therefore, there is an urgent need for a high-precision, automated, and comprehensive wheel coating quality inspection method that can handle complex coating textures, varying illumination, and the presence of tiny defects, thereby improving the efficiency, accuracy, and automation of quality control in industrial production. Summary of the Invention

[0003] In response to the above-mentioned technical deficiencies, the purpose of the present invention is to propose a wheel hub quality inspection method, which aims to solve the technical problem that the existing technology mostly relies on manual visual inspection or traditional image processing methods to detect defects in the uniformity of the wheel hub surface, especially under conditions of uneven lighting or complex textures, and it is difficult to achieve high-precision automatic detection of the uniformity of the wheel hub coating.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions: The present invention provides a wheel hub quality detection method,

[0005] The wheel hub quality detection method comprises:

[0006] Step S10: acquiring a first wheel hub coating image through an industrial camera, and performing denoising and illumination compensation processing on the first wheel hub coating image to obtain a second wheel hub coating image;

[0007] Step S20: Introducing a U-Net network combined with an attention mechanism to segment the second wheel hub coating image to obtain a first wheel hub segmentation image; and extracting a bubble region image using an adaptive threshold method based on the first wheel hub segmentation image;

[0008] Step S30: introducing a deep residual network combined with an edge enhancement method to segment the second wheel hub coating image to obtain a second wheel hub segmentation image; extracting a peeling area image based on the second wheel hub segmentation image combined with a gradient-based edge detection method;

[0009] Step S40: Scanning the wheel hub by laser to obtain coating surface height data, and obtaining coating surface texture data from the second wheel hub segmentation image; integrating the coating surface height data and the coating surface texture data, and using a deep regression network to perform coating thickness error point prediction processing to obtain a coating thickness error image;

[0010] Step S50: performing a comprehensive analysis of the hub coating uniformity based on the bubble area image, the peeling area image, and the coating thickness error image, and generating and outputting a hub coating quality report.

[0011] Preferably, in step S10, the step of performing denoising and illumination compensation on the first hub coating image to obtain the second hub coating image specifically includes: performing denoising on the first hub coating image using bilateral filtering; and performing illumination compensation on the first hub coating image using adaptive histogram equalization.

[0012] Preferably, in step S20, a U-Net network is introduced in combination with an attention mechanism to perform segmentation processing on the second wheel hub coating image to obtain a first wheel hub segmentation image; the step of extracting a bubble area image based on the first wheel hub segmentation image using an adaptive threshold method specifically includes: introducing a U-Net network, and adding an attention mechanism to the U-Net network, training the U-Net network with the attention mechanism using a cross-entropy loss function to obtain a pre-trained U-Net network, inputting the second wheel hub coating image into the pre-trained U-Net network to extract a first wheel hub segmentation image including bubble position features, bubble shape features and bubble size features; calculating an adaptive threshold within a local window based on the first wheel hub segmentation image using an adaptive threshold method, and extracting a bubble area image based on the adaptive threshold within the local window.

[0013] Preferably, in step S20, the formula used for the cross entropy loss function is: Among them, L seg is the cross entropy loss function, y is the true label of the bubble area, is the bubble area label predicted by the U-Net network; the formula used in the adaptive threshold method is: Where T(x,y) is the adaptive threshold in the local window where the pixel point (x,y) is located, which is used to determine whether the current pixel belongs to the bubble area; N is the total number of pixels in the local window; i is the horizontal offset relative to the center pixel; j is the vertical offset relative to the center pixel; k is the radius of the local window.

[0014] Preferably, in step S30, a deep residual network is introduced in combination with an edge enhancement method to segment the second hub coating image to obtain a second hub segmentation image; and the step of extracting a peeling area image based on the second hub segmentation image in combination with a gradient-based edge detection method specifically includes:

[0015] Step S301: using a ResNet network to perform feature encoding on the second wheel hub coating image to obtain a multi-scale deep texture feature map;

[0016] Step S302: using the first-order and second-order gradient information of the image to enhance the multi-scale deep texture feature map to obtain an edge enhancement feature map of the peeling area;

[0017] Step S303: inputting the edge enhancement feature map of the peeling area into the lightweight segmentation network head FCN to output a second hub segmentation image;

[0018] Step S304: extracting a peeling area image based on the second hub segmentation image and a gradient-based edge detection method.

[0019] Preferably, in step S40, the step of synthesizing the coating surface height data and the coating surface texture data and using a deep regression network to perform coating thickness error point prediction processing to obtain a coating thickness error image specifically includes:

[0020] Combining coating surface height data and coating surface texture data to construct a high-dimensional fusion input data pair;

[0021] The high-dimensional fusion input data pairs are fed into a deep regression network to obtain the mapping relationship features between texture change and height error. Based on the mapping relationship features, the thickness error of each pixel in the second wheel hub coating image is predicted and calculated.

[0022] The thickness error of each pixel is spatially mapped to obtain a coating thickness error image.

[0023] Preferably, in step S50, the step of performing a comprehensive analysis of the hub coating uniformity on the bubble area image, the peeling area image, and the coating thickness error image, and generating and outputting a hub coating quality report specifically includes:

[0024] Step S501: aligning the bubble area image outputted in step S20, the peeling area image outputted in step S30, and the thickness error image outputted in step S40, and generating a quality analysis matrix by logical mask superposition;

[0025] Step S502: calculating hub coating evaluation indicators of the hub coating based on the quality analysis matrix, where the hub coating evaluation indicators include a total defect area ratio, a maximum thickness error point, and a local abnormality density;

[0026] Step S503: matching the quality grade according to the wheel hub coating evaluation index to generate a quality rating result;

[0027] Step S504: Output the final hub coating quality report based on the quality rating result. The hub coating quality report includes the total defect area ratio, the maximum thickness error point, the local abnormality density, the quality rating result, the inspection time and the hub number.

[0028] The present invention also provides a wheel hub quality detection system comprising:

[0029] An image acquisition and processing module is used to acquire a first wheel hub coating image through an industrial camera, and perform denoising and illumination compensation on the first wheel hub coating image to obtain a second wheel hub coating image;

[0030] The bubble region image extraction module is used to segment the second wheel hub coating image using a U-Net network combined with an attention mechanism to obtain a first wheel hub segmentation image; and to extract the bubble region image using an adaptive threshold method based on the first wheel hub segmentation image.

[0031] The peeling area image extraction module is used to introduce a deep residual network combined with an edge enhancement method to segment the second wheel hub coating image to obtain a second wheel hub segmentation image; the peeling area image is extracted based on the second wheel hub segmentation image combined with a gradient-based edge detection method;

[0032] The coating thickness error image analysis module is used to obtain coating surface height data by scanning the wheel hub with a laser, and obtain coating surface texture data from the second wheel hub segmentation image. The coating surface height data and coating surface texture data are integrated, and a deep regression network is used to predict the coating thickness error points to obtain a coating thickness error image.

[0033] The wheel hub coating quality report generation module is used to perform a comprehensive analysis of the wheel hub coating uniformity based on the bubble area image, peeling area image and coating thickness error image, and generate and output the wheel hub coating quality report.

[0034] The present invention also provides a computer program product, comprising a wheel hub quality detection program, wherein the wheel hub quality detection program implements the wheel hub quality detection method when executed by a processor.

[0035] The beneficial effects of this invention lie in its use of deep learning technology based on image recognition, combined with U-Net and ResNet networks, to accurately segment areas of bubbles, peeling, and coating thickness errors. This avoids the misjudgment and missed detection issues common in traditional methods when detecting complex backgrounds, uneven lighting, or minor defects. Through automated image recognition, manual intervention is reduced, the accuracy and efficiency of coating defect detection are improved, and the high-precision requirements for coating quality inspection in modern industry are met.

[0036] By integrating image recognition technology with laser scanning height data, this method not only analyzes minute coating surface defects but also accurately identifies issues such as uneven thickness and flaking, thereby enabling multi-dimensional, comprehensive coating quality inspection. This inspection method, based on the fusion of image and physical data, overcomes the limitations of traditional single-image processing methods, improving the robustness and adaptability of inspection, and is suitable for coating quality control across a wide range of coating complexities and in diverse production environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 This is a flow chart of a first embodiment of a wheel hub quality detection method according to the present invention.

[0039] Figure 2 A schematic diagram of a heat map expression of a coating thickness error image provided by a wheel hub quality detection method of the present invention.

[0040] Figure 3 This is a schematic diagram of equipment for a wheel hub quality detection method according to the present invention. DETAILED DESCRIPTION

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0042] Example 1: Figure 1 FIG. 1 is a flow chart of the first embodiment of the wheel hub quality detection method according to the present invention, and provides the first embodiment of the wheel hub quality detection method according to the present invention.

[0043] In a first embodiment, the wheel hub quality detection method includes:

[0044] Step S10: acquiring a first wheel hub coating image through an industrial camera, and performing denoising and illumination compensation processing on the first wheel hub coating image to obtain a second wheel hub coating image;

[0045] It should be noted that in step S10, the step of performing denoising and illumination compensation on the first hub coating image to obtain the second hub coating image specifically includes: performing denoising on the first hub coating image using bilateral filtering; and performing illumination compensation on the first hub coating image using adaptive histogram equalization.

[0046] As you can understand, in this processing step, bilateral filtering is used to denoise the coating image. By considering spatial distance and pixel value similarity within the image, bilateral filtering effectively removes noise while preserving image edges and details. This technique is particularly suitable for processing subtle noise and texture in wheel coating images. To address image brightness differences caused by uneven lighting, adaptive histogram equalization is employed. This method adaptively enhances the brightness of local areas, allowing the coating image to display sufficient detail even in uneven lighting conditions, especially small defects on the wheel surface (such as bubbles and peeling).

[0047] It should be understood that compared to traditional global histogram equalization, which may lead to excessive enhancement of the image or even the appearance of overly bright or dark areas, bilateral filtering can automatically adjust the contrast of the coating image under different lighting conditions, especially when there is local reflection or lighting difference on the coating surface, which can improve the visibility of tiny defects in the image.

[0048] For example, in experiments, when the first round of coating images had uneven lighting, bilateral filtering significantly reduced salt-and-pepper noise (such as black or white spots) in the image, while preserving coating edge details. After illumination compensation, coating details in low-light areas were significantly enhanced, and the contrast in bubble areas increased from 30% to 80%. This significantly improved subsequent bubble area extraction and reduced the false detection rate by 15%.

[0049] Step S20: Introducing a U-Net network combined with an attention mechanism to segment the second wheel hub coating image to obtain a first wheel hub segmentation image; and extracting a bubble region image using an adaptive threshold method based on the first wheel hub segmentation image;

[0050] It should be noted that in step S20, the formula used by the cross entropy loss function is: Among them, L seg is the cross entropy loss function, y is the true label of the bubble area, is the bubble area label predicted by the U-Net network; the formula used in the adaptive threshold method is: Where T(x,y) is the adaptive threshold in the local window where the pixel point (x,y) is located, which is used to determine whether the current pixel belongs to the bubble area; N is the total number of pixels in the local window; i is the horizontal offset relative to the center pixel; j is the vertical offset relative to the center pixel; k is the radius of the local window.

[0051] It is understandable that the U-Net network has an encoder-decoder structure. When combined with the attention mechanism, it can more effectively focus on areas in the image with local abnormal texture changes, thereby enhancing the response of the bubble area, suppressing interference from non-target areas, and improving image recognition accuracy. The use of the cross-entropy loss function during training can prompt the model to maximize its approximation to the true bubble boundary labels, ensuring that the model learns an accurate segmentation strategy. Compared to the fixed global threshold method, the adaptive threshold method can dynamically calculate the threshold based on the brightness mean of the local area around each pixel, thereby stably extracting the bubble area even when facing uneven coating illumination or background texture changes.

[0052] For example, in a test of a batch of industrial wheel hub images, the traditional fixed threshold method showed obvious missed detection in darkly lit areas, with a false detection rate of up to 22%. However, the method of the present invention first performs preliminary segmentation through the U-Net network, and uses the attention mechanism to enhance the response to small bubble areas, accurately identifying bubbles with a diameter of less than 2mm in the test set. Subsequently, the boundaries are further refined through the adaptive threshold method, and the final false detection rate is reduced to 5.7%, and the missed detection rate is reduced to 3.2%. In addition, through the optimization of the cross-entropy loss function, the convergence speed of the training process is improved by 28%, and a segmentation accuracy of 85.6% IoU (intersection over union) is achieved in the test set.

[0053] Step S30: introducing a deep residual network combined with an edge enhancement method to segment the second wheel hub coating image to obtain a second wheel hub segmentation image; extracting a peeling area image based on the second wheel hub segmentation image combined with a gradient-based edge detection method;

[0054] It should be noted that in step S30, a deep residual network is introduced in combination with an edge enhancement method to segment the second wheel hub coating image to obtain a second wheel hub segmentation image; and the step of extracting the peeling area image based on the second wheel hub segmentation image in combination with a gradient-based edge detection method specifically includes:

[0055] Step S301: using a ResNet network to perform feature encoding on the second wheel hub coating image to obtain a multi-scale deep texture feature map;

[0056] Step S302: using the first-order and second-order gradient information of the image to enhance the multi-scale deep texture feature map to obtain an edge enhancement feature map of the peeling area;

[0057] Step S303: inputting the edge enhancement feature map of the peeling area into the lightweight segmentation network head FCN to output a second hub segmentation image;

[0058] Step S304: extracting a peeling area image based on the second hub segmentation image and a gradient-based edge detection method.

[0059] It should be understood that when ResNet network is performing image feature extraction, its residual connection structure can effectively alleviate the gradient vanishing problem of deep networks, thereby extracting richer and more recognizable coating texture features. Especially when the edges of the peeling area are blurred and the texture is discontinuous, ResNet can retain the high-order information of the coating structure, providing a more discriminative deep expression for subsequent edge enhancement and segmentation. At the same time, the introduction of the first-order gradient of the image (such as the Sobel edge) and the second-order gradient (such as the Laplacian) to significantly enhance the deep features can further improve the model's response sensitivity to the peeling boundary and improve the accuracy of boundary positioning.

[0060] For example, in a certain test sample, the boundaries of the peeling area in the original wheel hub image are blurred, and the transition with the normal coating texture is slow. When the traditional Sobel operator is used for edge detection, the peeling area is only partially extracted, and the missed detection rate is as high as 35%. Using the method of the present invention, after the deep texture features are extracted by the ResNet network, the gradient enhancement is performed and then input into the FCN decoder for segmentation, which can clearly extract the peeling area. Finally, after combining with the gradient edge correction, the edge of the peeling area is improved by an average of 21.4%, and the missed detection rate is reduced to 6.3%. In addition, the edge enhancement module also shows stronger robustness in the extraction of peeling in strong reflective areas, reducing the false detection rate in the highlight area by more than 18%, significantly improving the accuracy and stability of the overall defect recognition.

[0061] Step S40: Scanning the wheel hub by laser to obtain coating surface height data, and obtaining coating surface texture data from the second wheel hub segmentation image; integrating the coating surface height data and the coating surface texture data, and using a deep regression network to perform coating thickness error point prediction processing to obtain a coating thickness error image;

[0062] It should be noted that in step S40, the coating surface height data and the coating surface texture data are integrated, and the coating thickness error point prediction processing is performed using a deep regression network to obtain a coating thickness error image. The steps specifically include: integrating the coating surface height data and the coating surface texture data to construct a high-dimensional fusion input data pair; inputting the high-dimensional fusion input data pair into the deep regression network to obtain a mapping relationship feature between texture change and height error, and predicting and calculating the thickness error of each pixel point in the second wheel hub coating image based on the mapping relationship feature; and spatially mapping the thickness error of each pixel point to obtain a coating thickness error image.

[0063] Understandably, minute fluctuations in coating thickness are often difficult to accurately identify using texture images alone. However, relying solely on laser height data lacks contextual understanding of the coating texture's changing trends. Therefore, jointly modeling the image's texture information with the actual height data helps establish a nonlinear mapping relationship between texture changes and thickness fluctuations, enabling more accurate predictions of thickness error points. Furthermore, the use of deep regression networks enables complex mapping learning between high-dimensional inputs and regression labels, outperforming traditional linear fitting or local filtering methods, particularly when processing thickness data with a certain degree of continuity, trend, and contextual dependence.

[0064] It should be understood that the thickness error prediction in the present invention is not just a direct deviation judgment of the height, but is based on the collaborative characterization characteristics of texture features and height information, and the nonlinear relationship between their combined features and thickness error is learned through a deep network. It is applicable to a variety of complex structures and irregularly changing coating surfaces. Figure 2 As shown in the figure, a schematic diagram of the thermal map expression of a coating thickness error image is shown. By outputting the error value at the pixel level and spatially mapping it into an image form, a thickness error image can be directly formed to achieve high-resolution visualization of thickness anomalies, which is suitable for subsequent automatic diagnosis and zoning evaluation.

[0065] For example, on a typical wheel hub sample, the standard thickness range measured by laser scanning is 0.18–0.22mm. Traditional visual estimation methods can only identify error areas greater than ±0.05mm. However, the method proposed in this paper, by fusing images with height data and training a regression network, achieved an average error of 0.008mm in the test set, an accuracy improvement of over 60%. In one data set, the predicted thickness error image clearly identified three local thickness anomalies with diameters less than 3mm, which were not captured by traditional methods. The maximum position of the error map deviated from the actual measured error point by less than 1.2 pixels, verifying the advantages of this method in spatial positioning accuracy and anomaly perception.

[0066] Step S50: performing a comprehensive analysis of the hub coating uniformity based on the bubble area image, the peeling area image, and the coating thickness error image, and generating and outputting a hub coating quality report.

[0067] It should be noted that in step S50, the step of performing a comprehensive analysis of the hub coating uniformity based on the bubble area image, the peeling area image, and the coating thickness error image, and generating and outputting the hub coating quality report specifically includes:

[0068] Step S501: aligning the bubble area image outputted in step S20, the peeling area image outputted in step S30, and the thickness error image outputted in step S40, and generating a quality analysis matrix by logical mask superposition;

[0069] Step S502: calculating hub coating evaluation indicators of the hub coating based on the quality analysis matrix, where the hub coating evaluation indicators include a total defect area ratio, a maximum thickness error point, and a local abnormality density;

[0070] Step S503: matching the quality grade according to the wheel hub coating evaluation index to generate a quality rating result;

[0071] Step S504: Output the final hub coating quality report based on the quality rating result. The hub coating quality report includes the total defect area ratio, the maximum thickness error point, the local abnormality density, the quality rating result, the inspection time and the hub number.

[0072] It can be understood that, through the fusion analysis of three types of defect images (bubbles, peeling, thickness errors), the quality analysis matrix constructed by the present invention realizes a unified representation of coating uniformity defects in the spatial dimension. Compared with the traditional sub-item detection method, this matrix analysis method can realize pixel-level comparison and comprehensive evaluation, providing a high-resolution, visual data basis for subsequent indicator calculation and grade matching. The total defect area ratio in the quality assessment index can quantify the overall defect degree of the coating, the maximum thickness error point reflects the local structural safety hazard, and the local abnormal density reflects the defect concentration trend in a small range. The combination of these multi-dimensional indicators can comprehensively characterize the quality status of the coating uniformity.

[0073] It should be understood that the evaluation logic adopted by the present invention not only has quantitative characteristics, but also has both interpretability and scalability. Various evaluation indicators are independent of each other and complement each other to form a multi-dimensional comprehensive analysis model. In addition, the quality grade matching part can flexibly set standards (such as national standards, enterprise standards or industry classifications) according to user needs, and supports adaptive judgment through rule trees or lightweight machine learning models (such as KNN, XGBoost), with good engineering deployability. At the same time, the output coating quality report has structured and standardized characteristics, which can not only visually display defective areas in the form of images, but also can be docked with the enterprise quality management system (QMS) through data format to achieve closed-loop quality control of detection-rating-tracing.

[0074] For example, during actual inspection, after analyzing a batch of wheel hubs for bubbles, spalling, and thickness errors, the quality analysis matrix identified significant defect areas covering approximately 3.8% of the total area, including 1.2% for bubbles, 0.8% for spalling, and 1.8% for thickness errors greater than ±0.02mm. Based on the predefined rules, the wheel hub was rated "B (Good)." A wheel hub coating quality report was automatically generated, including: Total defect area percentage: 3.8%, Maximum thickness error: 0.037mm, Local anomaly density: 2.1%, Quality rating: B, Inspection time: 2025-04-10 10:32:48, Wheel number: WZ20250410-015. The report was exported in both PDF and JSON formats and automatically pushed to downstream production management systems, enabling traceability and visualization of quality inspection results.

[0075] Embodiment 2: Furthermore, the present invention provides a wheel hub quality inspection system that utilizes the wheel hub quality inspection method described in the above embodiment to address the technical issues surrounding wheel hub quality inspection. Compared to the prior art, the wheel hub quality inspection system provided by the present invention achieves the same beneficial effects as the wheel hub quality inspection method described in the above embodiment. Other technical features of the wheel hub quality inspection system are the same as those disclosed in the above embodiment and are not further elaborated here.

[0076] Example 3: The present invention provides a wheel hub quality detection device, please refer to Figure 3A wheel quality testing device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform a wheel quality testing method according to the first embodiment. A wheel quality testing device in an embodiment of the present invention may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. A wheel quality testing device is merely an example and should not limit the functionality or scope of use of the embodiments of the present invention. A wheel quality testing device may include a processing device 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which may perform various appropriate actions and processes based on a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. RAM 1004 also stores various programs and data required for the operation of the wheel quality inspection device. The processing device 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems may be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, hard disk, etc.; and a communication device 1009. The communication device 1009 allows the wheel quality inspection device to communicate with other devices wirelessly or by wire to exchange data. Although the figure illustrates a wheel quality inspection device having various systems, it should be understood that implementation or presence of all the illustrated systems is not required. More or fewer systems may alternatively be implemented or present.

[0077] Example 4: The present invention also provides a computer program product, comprising a computer program. When executed by a processor, the computer program implements the steps of the aforementioned wheel hub quality testing method. The computer program product provided by the present invention can solve the technical problem of wheel hub quality testing. Compared with the prior art, the beneficial effects of the computer program product provided by the present invention are the same as those of the wheel hub quality testing method provided by the aforementioned embodiment, and are not further elaborated here.

[0078] In particular, according to the embodiments disclosed in the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments disclosed in the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program comprising program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present invention are performed.

[0079] It should be understood that the various parts disclosed in the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any appropriate manner in any one or more embodiments or examples.

[0080] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A wheel hub quality detection method, characterized in that: Methods include: Step S10: acquiring a first wheel hub coating image through an industrial camera, and performing denoising and illumination compensation processing on the first wheel hub coating image to obtain a second wheel hub coating image; Step S20: Introducing a U-Net network combined with an attention mechanism to segment the second wheel hub coating image to obtain a first wheel hub segmentation image; and extracting a bubble region image using an adaptive threshold method based on the first wheel hub segmentation image; Step S30: introducing a deep residual network combined with an edge enhancement method to segment the second wheel hub coating image to obtain a second wheel hub segmentation image; extracting a peeling area image based on the second wheel hub segmentation image combined with a gradient-based edge detection method; Step S40: Scanning the wheel hub by laser to obtain coating surface height data, and obtaining coating surface texture data from the second wheel hub segmentation image; integrating the coating surface height data and the coating surface texture data, and using a deep regression network to perform coating thickness error point prediction processing to obtain a coating thickness error image; Step S50: performing a comprehensive analysis of the hub coating uniformity based on the bubble area image, the peeling area image, and the coating thickness error image, and generating and outputting a hub coating quality report.

2. A wheel hub quality detection method according to claim 1, characterized in that: In step S10, the step of performing denoising and illumination compensation on the first hub coating image to obtain the second hub coating image specifically includes: performing denoising on the first hub coating image using bilateral filtering; and performing illumination compensation on the first hub coating image using adaptive histogram equalization.

3. A wheel hub quality detection method according to claim 1, characterized in that: In step S20, a U-Net network is introduced in combination with an attention mechanism to perform segmentation processing on the second wheel hub coating image to obtain a first wheel hub segmentation image; the step of extracting a bubble area image based on the first wheel hub segmentation image using an adaptive threshold method specifically includes: introducing a U-Net network, and adding an attention mechanism to the U-Net network, training the U-Net network with the attention mechanism using a cross-entropy loss function to obtain a pre-trained U-Net network, inputting the second wheel hub coating image into the pre-trained U-Net network to extract a first wheel hub segmentation image including bubble position features, bubble shape features, and bubble size features; calculating an adaptive threshold within a local window based on the first wheel hub segmentation image using an adaptive threshold method, and extracting a bubble area image based on the adaptive threshold within the local window.

4. A wheel hub quality detection method according to claim 3, characterized in that: In step S20, the formula used for the cross entropy loss function is: Among them, L seg is the cross entropy loss function, y is the true label of the bubble area, is the bubble area label predicted by the U-Net network; the formula used in the adaptive threshold method is: Where T(x,y) is the adaptive threshold in the local window where the pixel point (x,y) is located, which is used to determine whether the current pixel belongs to the bubble area; N is the total number of pixels in the local window; i is the horizontal offset relative to the center pixel; j is the vertical offset relative to the center pixel; k is the radius of the local window.

5. A wheel hub quality detection method according to claim 1, characterized in that: In step S30, a deep residual network is introduced in combination with an edge enhancement method to segment the second wheel hub coating image to obtain a second wheel hub segmentation image; The step of extracting the peeling area image based on the second hub segmentation image and the gradient-based edge detection method specifically includes: Step S301: using a ResNet network to perform feature encoding on the second wheel hub coating image to obtain a multi-scale deep texture feature map; Step S302: using the first-order and second-order gradient information of the image to enhance the multi-scale deep texture feature map to obtain an edge enhancement feature map of the peeling area; Step S303: inputting the edge enhancement feature map of the peeling area into the lightweight segmentation network head FCN to output a second hub segmentation image; Step S304: extracting a peeling area image based on the second hub segmentation image and a gradient-based edge detection method.

6. A wheel hub quality detection method according to claim 1, characterized in that: In step S40, the coating surface height data and the coating surface texture data are integrated, and a deep regression network is used to perform coating thickness error point prediction processing to obtain a coating thickness error image, which specifically includes: Combining coating surface height data and coating surface texture data to construct a high-dimensional fusion input data pair; The high-dimensional fusion input data pairs are fed into a deep regression network to obtain the mapping relationship features between texture change and height error. Based on the mapping relationship features, the thickness error of each pixel in the second wheel hub coating image is predicted and calculated. The thickness error of each pixel is spatially mapped to obtain a coating thickness error image.

7. A wheel hub quality detection method according to claim 1, characterized in that: In step S50, a comprehensive analysis of the hub coating uniformity is performed on the bubble area image, the peeling area image, and the coating thickness error image to generate and output a hub coating quality report, specifically including: Step S501: aligning the bubble area image outputted in step S20, the peeling area image outputted in step S30, and the thickness error image outputted in step S40, and generating a quality analysis matrix by logical mask superposition; Step S502: calculating hub coating evaluation indicators of the hub coating based on the quality analysis matrix, where the hub coating evaluation indicators include a total defect area ratio, a maximum thickness error point, and a local abnormality density; Step S503: matching the quality grade according to the wheel hub coating evaluation index to generate a quality rating result; Step S504: Output the final hub coating quality report based on the quality rating result. The hub coating quality report includes the total defect area ratio, the maximum thickness error point, the local abnormality density, the quality rating result, the inspection time and the hub number.

8. A wheel hub quality inspection system, applied to a wheel hub quality inspection method according to any one of claims 1 to 7, characterized in that: The wheel hub quality detection system includes: An image acquisition and processing module is used to acquire a first wheel hub coating image through an industrial camera, and perform denoising and illumination compensation on the first wheel hub coating image to obtain a second wheel hub coating image; The bubble region image extraction module is used to segment the second wheel hub coating image using a U-Net network combined with an attention mechanism to obtain a first wheel hub segmentation image; and to extract the bubble region image using an adaptive threshold method based on the first wheel hub segmentation image. The peeling area image extraction module is used to introduce a deep residual network combined with an edge enhancement method to segment the second wheel hub coating image to obtain a second wheel hub segmentation image; the peeling area image is extracted based on the second wheel hub segmentation image combined with a gradient-based edge detection method; The coating thickness error image analysis module is used to obtain coating surface height data by scanning the wheel hub with a laser, and obtain coating surface texture data from the second wheel hub segmentation image. The coating surface height data and coating surface texture data are integrated, and a deep regression network is used to predict the coating thickness error points to obtain a coating thickness error image. The wheel hub coating quality report generation module is used to perform a comprehensive analysis of the wheel hub coating uniformity based on the bubble area image, peeling area image and coating thickness error image, and generate and output the wheel hub coating quality report.

9. A wheel hub quality inspection device, characterized in that: The wheel hub quality detection device includes: a memory, a processor, and a wheel hub quality detection program stored in the memory and executable on the processor. When the wheel hub quality detection program is executed by the processor, a wheel hub quality detection method according to any one of claims 1 to 7 is implemented.

10. A computer program product, characterized in that The computer program product includes a wheel hub quality detection program, and when the wheel hub quality detection program is executed by a processor, a wheel hub quality detection method according to any one of claims 1 to 7 is implemented.

Images