Automobile part intelligent identification method based on picture classification model

Abstract

The invention relates to the technical field of intelligent identification methods, in particular to an automobile part intelligent identification method based on a picture classification model. The method comprises the following steps: obtaining high-quality automobile part image data through image acquisition and optimization, and carrying out image optimization processing on the high-quality automobile part image data; according to the optimized image data, geometric feature recognition and structural stability evaluation of the automobile part are carried out, so that the durability of the part is estimated; performing function classification on the images based on the durability data and the geometric feature data, training an image classification model, and generating an automobile accessory image classification model; the method comprises the following steps: analyzing fault data of a running automobile, performing applicable accessory identification in combination with an image classification model, and optimizing the image classification model based on the fault data so as to generate a more accurate intelligent identification model; according to the invention, the method achieves the more accurate recognition of the automobile parts through the recognition and optimization of the automobile part image.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent recognition methods, and particularly to an intelligent recognition method for automotive parts based on an image classification model. Background Art

[0002] The types and complexities of automotive parts are constantly increasing. Especially in the context of the continuous improvement of intelligence and automation, the quality and maintenance management of automotive parts have become increasingly important. In aspects such as image classification and object detection, the performance of deep convolutional neural networks (CNNs) and neural network models has been continuously improved, significantly improving the accuracy and efficiency of image analysis. Image classification models can automatically identify the categories, shapes, and features of objects by learning from a large amount of image data, and have become the core technology of modern intelligent recognition systems. By performing deep learning processing on images of automotive parts, the accuracy and speed of part recognition are effectively improved, and vehicle maintenance efficiency is enhanced. Image classification models are trained using a large amount of labeled data and can identify the shapes, sizes, colors, and key features of automotive parts, such as the surface texture and external contour of the parts. However, there are problems with inaccurate assessment of the stability of automotive parts and inaccurate recognition of the adaptation of automotive parts in a traditional intelligent recognition method for automotive parts based on an image classification model. Summary of the Invention

[0003] Based on this, it is necessary to provide an intelligent recognition method for automotive parts based on an image classification model to solve at least one of the above technical problems.

[0004] To achieve the above object, an intelligent recognition method for automotive parts based on an image classification model includes the following steps:

[0005] Step S1: Collect images of automotive parts to obtain automotive part image data; optimize the automotive part image data to obtain optimized automotive part image data;

[0006] Step S2: Identify the geometric features of automotive parts based on the optimized automotive part image data to obtain automotive part geometric feature data; evaluate the structural stability of automotive parts based on the automotive part geometric feature data to obtain automotive part structural stability data; estimate the durability of automotive parts based on the automotive part structural stability data to obtain automotive part durability data;

[0007] Step S3: Classify the functions of the automotive part image based on the automotive part durability data and the automotive part geometric feature data to obtain automotive part image function classification data; train the image classification model based on the automotive part image function classification data to obtain an automotive part image classification model;

[0008] Step S4: Obtain the fault data of the moving vehicle; identify the applicable parts based on the vehicle parts image classification model for the fault data of the moving vehicle to obtain the vehicle fault applicable parts data; optimize the intelligent identification of vehicle parts for the vehicle parts image classification model based on the vehicle fault applicable parts data to generate an optimized vehicle parts image classification model.

[0009] The present invention ensures high-quality input data through image acquisition and optimization. By carefully collecting lighting data and identifying detail anomalies in automotive parts images, problems such as uneven lighting and overexposure can be eliminated at an early stage, which directly improves the image quality and ensures the accuracy of subsequent analysis. In addition, image optimization processing further enhances the image quality, reduces interference caused by image quality degradation, and helps improve the accuracy and robustness of subsequent recognition. Geometric feature recognition and structural stability assessment enable the precise capture of the geometric characteristics and structural states of parts, helping the system understand the functions and durability of the parts. Through geometric feature recognition, key information such as the external contours and internal structures of parts can be effectively extracted, providing data support for structural stability assessment and durability prediction. This precise geometric feature data not only improves the recognition accuracy of parts but also enables the early prediction of structural failures, thereby helping to prevent potential damage. In the image function classification step, an image classification model for parts images is formed through classification training of the images. This step can distinguish different types of parts and their functions through in-depth analysis of image function classification, providing strong support for subsequent fault diagnosis and part matching. At the same time, the training of the classification model also improves the recognition ability for new images, and as the amount of data increases, the model will gradually become more accurate. The fault recognition part analyzes the fault data of running vehicles and combines with the image classification model to identify the applicability of parts, which can effectively determine which parts have faults and perform intelligent recognition optimization. Through the spatial distribution analysis and concentration calculation of the fault areas, the system can discover the high-incidence areas of the fault areas, thereby predicting in advance the key parts affecting the vehicle performance. This data-driven optimization method makes fault recognition more accurate and helps improve the scientific nature of part management and maintenance decision-making. In addition, the calculation of the fault impact degree and the weighted assessment of severity contribute to providing more accurate guidance for maintenance. By comprehensively considering the concentration and impact degree of the faults, the system can perform a weighted assessment of the fault severity of the parts, thereby optimizing the maintenance process. The system will prioritize the processing of high-severity fault areas to ensure the reasonable allocation of resources, thereby effectively reducing the maintenance cost and enhancing the overall safety and reliability of the vehicle. Therefore, the present invention is an optimization of a traditional intelligent recognition method for automotive parts based on a picture classification model, solving the problems of inaccurate stability assessment of automotive parts and inaccurate adaptation recognition of automotive parts existing in the traditional intelligent recognition method for automotive parts based on a picture classification model. It improves the accuracy of the stability assessment of automotive parts and the accuracy of the adaptation recognition of automotive parts. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a schematic flow chart of the steps of an intelligent recognition method for automotive parts based on a picture classification model;

[0011] Figure 2 For Figure 1 a detailed implementation step flowchart diagram of step S2 in

[0012] Figure 3 For Figure 1 a detailed implementation step flowchart diagram of step S4 in

[0013] The realization, functional features and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. Specific implementation manners

[0014] The technical method of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0015] In addition, the accompanying drawings are only schematic diagrams of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings represent the same or similar parts, and thus the repeated description thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor methods and / or microcontroller methods.

[0016] It should be understood that although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed related items.

[0017] To achieve the above object, please refer to Figures 1 to 3 , an intelligent recognition method for automotive parts based on an image classification model, comprising the following steps:

[0018] Step S1: Collect images of automotive parts to obtain automotive part image data; optimize the automotive part image data to obtain optimized automotive part image data;

[0019] Step S2: Identify the geometric features of the auto parts based on the optimized data of the auto parts images to obtain the geometric feature data of the auto parts; evaluate the structural stability of the auto parts based on the geometric feature data of the auto parts to obtain the structural stability data of the auto parts; estimate the durability of the auto parts according to the structural stability data of the auto parts to obtain the durability data of the auto parts;

[0020] Step S3: Classify the functions of the auto parts images based on the durability data of the auto parts and the geometric feature data of the auto parts to obtain the function classification data of the auto parts images; train the image classification model based on the function classification data of the auto parts images to obtain the auto parts image classification model;

[0021] Step S4: Obtain the fault data of the moving vehicle; identify the applicable parts for the fault data of the moving vehicle according to the auto parts image classification model to obtain the applicable parts data for the vehicle faults; optimize the intelligent identification of the auto parts for the auto parts image classification model based on the applicable parts data for the vehicle faults to generate an optimized model for the auto parts image classification.

[0022] In the embodiment of the present invention, with reference to Figure 1 As shown, in this example, the method for intelligent identification of auto parts based on an image classification model includes the following steps:

[0023] Step S1: Collect images of the auto parts to obtain the image data of the auto parts; optimize the images of the auto parts for the image data of the auto parts to obtain the optimized data of the auto parts images;

[0024] In the embodiments of the present invention, a high-precision industrial camera is used to collect images of automotive parts. The detailed information of the automotive parts is captured, and high-definition image shooting is achieved through appropriate lens adjustment. Parameters such as the focal length, aperture, and shutter speed of the camera are automatically adjusted according to the shooting environment to ensure that the images are not blurred or distorted. The lighting conditions in the acquisition area are controlled by uniform LED lights to ensure that the lighting intensity in the entire area remains between 500 - 800 lux, so as to reduce image quality problems caused by shadows or overexposure. During the shooting process, the parts are placed on a rotating platform to be photographed from multiple angles to obtain comprehensive image data. After the image acquisition, the next step is image optimization. Image optimization software (such as Adobe Photoshop or OpenCV tool library) is used to process the collected image data, perform denoising processing, and reduce the random noise in the image using the Gaussian filtering algorithm. Then, the histogram equalization technique is adopted to improve the contrast of the image and ensure clear details. The image optimization data includes operations such as color correction and exposure adjustment. By adjusting the brightness, contrast, and color temperature of the image, the details of the image are made more prominent, especially the tiny texture and structural features on the surface of the automotive parts.

[0025] Step S2: Identify the geometric features of the automotive parts based on the optimized data of the automotive parts images to obtain the geometric feature data of the automotive parts; evaluate the structural stability of the automotive parts based on the geometric feature data of the automotive parts to obtain the structural stability data of the automotive parts; estimate the durability of the automotive parts based on the structural stability data of the automotive parts to obtain the durability data of the automotive parts;

[0026] In the embodiments of the present invention, based on the optimized image data, image processing technology based on deep learning is used for geometric feature recognition. Specifically, a convolutional neural network (CNN) is used to extract the geometric features of automotive parts. The network is trained to recognize key geometric information such as the size, shape, and hole positions of the parts. Each automotive part in the image undergoes convolutional feature extraction in a local area to obtain data such as the length, width, height, and angle of the part. At this time, the geometric feature data of the part is extracted as a series of numerical values, such as the contour curve, surface area, and geometric dimensions of each part of the part. Based on these geometric feature data, the physical structure of the automotive part is evaluated through a structural stability analysis model. Using finite element analysis (FEA) technology, the stress distribution and deformation of the part under the working load are simulated. This process requires the material properties (such as density, elastic modulus, etc.) of the part as input parameters, and static or dynamic load analysis is used to evaluate the stability of the automotive part under different working conditions. For example, by using different load simulations (such as compression, tension), the deformation of the part under the force condition is calculated to obtain its structural stability data. Based on the structural stability data, the durability prediction of the part is carried out. Through the cumulative damage theory, fatigue life prediction model, etc., combined with the actual working conditions and historical failure data of the part, the service life of the part is estimated. The durability data includes key parameters such as the expected working life and maintenance cycle of the part.

[0027] Step S3: Classify the optimized data of the automotive part image according to the durability data of the automotive part and the geometric feature data of the automotive part to obtain the functional classification data of the automotive part image; based on the functional classification data of the automotive part image, train the image classification model to obtain the automotive part image classification model;

[0028] In the embodiments of the present invention, the obtained durability data of automotive parts is combined with geometric feature data to classify the functions of automotive part images. At this time, the support vector machine (SVM) in the machine learning method is used for function classification training. By annotating data of different types of parts (such as engine parts, body parts, etc.), a classification model is trained to identify parts with different functions. By extracting geometric features in the image data, these features are used to classify the types of parts. For example, if a part has geometric features such as a large size, complex curved surfaces, and high strength, the system will identify it as a body structure part; while a part with a small size and high-precision hole positions is classified as an electronic control part. After completing the function classification, the system trains the picture classification model based on these classification data, and uses convolutional neural network (CNN) or transfer learning technology to optimize the image classification model. The input of the model is the optimized data of the image, and the output is the category label of the part. During the training process, the cross-validation method is used to evaluate the classification accuracy of the model, and the model weights are adjusted through the backpropagation algorithm until the classification accuracy reaches the preset target, obtaining an optimized automotive part image classification model that can efficiently classify the functions of the input images.

[0029] Step S4: Obtain the fault data of the running vehicle; identify the applicable parts for the fault data of the running vehicle according to the automotive part image classification model to obtain the automotive fault applicable part data; optimize the intelligent identification of automotive parts for the automotive part image classification model based on the automotive fault applicable part data to generate an optimized automotive part image classification model.

[0030] In the embodiments of the present invention, the fault data of the running vehicle is obtained. These data are obtained through the on-board diagnostic system (OBD) and record the fault information of the vehicle in different states, including but not limited to fault codes, fault locations, fault occurrence frequencies, etc. These fault data are collected in vehicle repair stations or intelligent monitoring systems and are processed to form a standardized data set. After obtaining the fault data, based on the existing part image classification model, the applicable parts for the automotive fault data are identified. Specifically, by matching the automotive fault data with the part categories, the relationship between the fault occurrence area and the corresponding part types is judged. For example, if the fault code indicates engine overheating, the system will identify the relevant parts in the engine cooling system, such as water pumps, radiators, etc. The fault applicability of the parts is determined by combining with the output results of the part function classification model to determine the specific applicable parts for each fault point. Based on the automotive fault applicable part data, the intelligent identification of automotive parts is optimized. By analyzing the fault data and the part matching degree, the image classification model is further optimized. During this process, the historical repair information and part replacement data in the fault data are used to retrain the part classification model to enhance its identification ability under complex fault conditions.

[0031] Preferably, step S1 includes the following steps:

[0032] Step S11: Obtain the automotive parts and the illumination data of the image acquisition area of the automotive parts; perform image acquisition on the automotive parts to obtain the image data of the automotive parts;

[0033] Step S12: Identify image detail anomalies in the automotive parts image data and the automotive parts according to the illumination data of the image acquisition area of the automotive parts, so as to obtain the image detail anomaly data of the automotive parts;

[0034] Step S13: Perform automotive parts image quality attenuation detection according to the image detail anomaly data of the automotive parts to obtain the automotive parts image quality attenuation data;

[0035] Step S14: Optimize the automotive parts image with the automotive parts image quality attenuation data to obtain the optimized automotive parts image data.

[0036] In the embodiments of the present invention, an industrial camera is used to collect images of automotive parts. The camera used is required to have a resolution of at least over 40 million pixels to ensure that the collected images have sufficient clarity and details. To ensure the quality of the images, the lighting conditions in the part image collection area need to be precisely controlled. By configuring the light source system and adopting an LED uniform light source, the lighting intensity is guaranteed to be 700 lux, avoiding areas that are too bright or too dark from affecting the image quality. The camera should be set to a fixed focal length and an automatic exposure mode, with the exposure time controlled within 1 / 100 second and the shutter speed at 1 / 500 second, ensuring that the image is not affected by motion blur or overexposure and thus the accuracy of the image data. In addition, to obtain multi-angle images, the parts are placed on a rotating platform, and the rotating speed of the rotating platform is 90 degrees per second, capturing comprehensive details of the automotive parts from different angles. After the image collection is completed, the lighting data of the collection area is recorded and stored. Based on the automotive part image data collected in step S11 and the lighting data of the collection area, image detail anomaly recognition is performed. Image detail anomalies generally refer to image quality problems caused by factors such as uneven lighting, strong reflection, overexposure, and shadows. To identify these anomalies, an edge detection algorithm (such as the Canny algorithm) is used to detect whether there are overexposed areas in the image or detail loss due to strong light reflection. For the brightness anomalies in the image, the histogram of the image is analyzed. If there are areas that are overly dark or bright, the lighting difference is calculated to obtain the lighting anomaly data for each area. At the same time, by comparing the actual color of the automotive parts in the image with the standard color spectrum, color deviation is detected, thereby further identifying the image detail anomaly areas. Through these technical means, the obvious detail anomaly areas in the image are marked to generate image detail anomaly data. This data includes the coordinates of each anomaly area and the anomaly type (such as overexposure, reflection, light spot, etc.). For the automotive part image detail anomaly data obtained in step S12, image quality attenuation detection is performed. The image is processed using the frequency domain analysis method of the image, converting the image into a frequency domain representation and analyzing its high-frequency components. If the high-frequency components of the image are lost or damaged, it indicates that the image has become blurred or detail loss has occurred. The fast Fourier transform (FFT) is used to process the image, calculating the energy change of the high-frequency components in the spectrum. If the energy drops significantly, it indicates that there is a problem with image quality attenuation. Then, the texture of the image is detected through granularity analysis. If the granularity of the texture becomes coarser, it indicates that the image has become distorted or the quality has decreased. By comparing the image detail anomaly data with the standard model of image attenuation, the specific areas of image quality attenuation in the image are detected and calibrated to generate image quality attenuation data.This data includes the degree of image attenuation, regional distribution, and type of attenuation (such as blurring, grain coarsening, detail loss, etc.). Based on the automotive parts image quality attenuation data obtained in step S13, the automotive parts image is optimized. According to the blurring information in the image attenuation data, the deconvolution algorithm is used to restore the blurred image and improve the image clarity. For images with overly coarse grain, the wavelet transform technology is used to denoise the image, remove noise points, and restore details. Subsequently, through the color correction algorithm, the color saturation of the image is adjusted to avoid color distortion or excessive saturation and maintain the natural and true color of the image. Further, the image sharpening algorithm (such as Laplacian sharpening) is used to enhance the edges of the image, making the contours of the parts more obvious and improving the visibility of details. The image optimization process combines the above multiple technical means to gradually restore the image quality and generate optimized image data. This optimized data includes the clear image after denoising, the detailed image after sharpening, and the standard image after color adjustment.

[0037] Preferably, step S12 includes the following steps:

[0038] Step S122: Estimate the direct light irradiation of the automotive parts based on the light data of the automotive parts image acquisition area to obtain the direct light irradiation data of the automotive parts;

[0039] Step S123: Analyze the overly strong reflected light of the automotive parts based on the direct light irradiation data of the automotive parts to obtain the overly strong reflected light data of the automotive parts;

[0040] Step S124: Detect the overexposure of the automotive parts image data based on the overly strong reflected light data of the automotive parts to obtain the overexposure data of the automotive parts image;

[0041] Step S125: Estimate the overly high color saturation of the automotive parts image based on the overexposure data of the automotive parts image to obtain the overly high color saturation data of the automotive parts image;

[0042] Step S126: Identify the abnormal details of the automotive parts image data based on the overly high color saturation data of the automotive parts image and the overexposure data of the automotive parts image, so as to obtain the abnormal details data of the automotive parts image.

[0043] In the embodiments of the present invention, the illumination data of the image acquisition area of automotive parts is obtained. This process is measured using a precise illumination sensor, and the relevant data of the illumination intensity and the light source position are recorded. Specifically, the illumination sensor should be able to provide illumination intensity data with a range covering from 100 lux to 1000 lux, and its illumination measurement accuracy is ±5 lux. By performing sub-region illumination acquisition on the image acquisition area of automotive parts, the illumination distribution of each region is determined. Then, based on the acquired regional illumination data, the radiation transfer model is used to calculate the propagation path of light, and the direct illumination intensity received by the automotive parts under specific illumination is estimated. During the estimation process, considering the angle, distance of the light source and the attenuation law of light, through the illumination simulation algorithm, the direct illumination intensity of each part of the accessory surface by the light source is determined. The obtained direct light irradiation data includes information such as the irradiation intensity and illumination angle of each light source on different positions of the accessory surface. Using the direct light irradiation data of automotive parts obtained in step S122, the reflected light situation of automotive parts is analyzed. This process extracts the reflected light characteristics on the accessory surface in the image to determine whether it is too strong. Excessively strong reflected light usually leads to distortion of the details of the accessory in the image. To achieve this, the reflected light intensity analysis technology is used. Based on the surface material of the automotive parts (such as metal, plastic, etc.), the Fresnel reflection model is used to predict the reflected light intensity. By analyzing the relationship between the intensity of the reflected light and the angle of the incident light, the reflected light intensity at different positions on the surface of the automotive parts is obtained. When the intensity of the reflected light exceeds a certain threshold (for example, exceeding 80% of the illumination intensity), it is considered that the reflected light at that position is too strong. The data of excessively strong reflected light will include information such as the reflected light intensity, the position of the over-strong area, and the angle of the reflected light source. Using the data of excessively strong reflected light of automotive parts obtained in step S123, further overexposure detection is performed on the image data of automotive parts. To detect whether the image is overexposed, the histogram analysis method is adopted to analyze the brightness distribution of the image. Specifically, by calculating the brightness histogram of the image, it is judged whether there are a large number of pixels distributed in the high-brightness end (i.e., the area where the brightness is close to 255) of the image. If the brightness histogram of the image shows that more than 50% of the pixels are concentrated in the high-brightness area, the image is determined to be overexposed. In addition, based on the data of excessively strong reflected light, by comparing the light intensity of the reflected light source with the brightness value of the image area, it can be further judged whether the local area of the image is overexposed due to excessively strong reflected light. Based on the overexposure data of the automotive parts image obtained in step S124, an estimation of excessively high color saturation is performed. Excessively high color saturation usually occurs in the case of too strong light, and some areas of the image will have color bleeding and distortion. To solve this problem, the color space conversion technology is adopted to convert the image from the RGB color space to the HSV (hue, saturation, value) color space.After conversion, by extracting the saturation value of each pixel, the saturation distribution of the entire image is calculated. If the saturation in the image exceeds a certain threshold (for example, exceeding 90%), it is determined that the color saturation is too high. In addition, combined with the overexposure data of the image, it is further determined whether the area with too high color saturation coincides with the overexposed area. If there is a match, it means that the high saturation in these areas is caused by overexposure. The obtained data of too high color saturation includes the position of the high saturation area, the saturation value, and the threshold judgment related to exposure. Based on the data of too high color saturation of the automotive parts image obtained in step S125 and the overexposure data of the image obtained in step S124, the details abnormality recognition of the automotive parts image is carried out. Image details abnormality usually manifests as phenomena such as image blurring, distortion, or color bleeding. These abnormalities will affect the accuracy of subsequent image analysis and recognition. Combined with the overexposure data of the image, edge detection is performed on the overexposed area to determine whether there is detail loss in this area. Secondly, using the data of too high color saturation, abnormal calibration is carried out on the area with color bleeding in the image. By detecting the color change in these areas, it is evaluated whether it exceeds the normal range. In order to further improve the detection accuracy, wavelet transform is used to perform multi-scale analysis on the image to capture the change of detail information in the image. Through the analysis of the abnormal areas in the image, the details abnormality data of the automotive parts image is obtained, including the boundary of the abnormal area, the type of abnormality (such as blurring, distortion, color bleeding, etc.), and the severity of the abnormality.

[0044] Preferably, step S13 includes the following steps:

[0045] Step 131: Detect the loss of high-frequency information of the automotive parts image according to the details abnormality data of the automotive parts image, and obtain the data of the loss of high-frequency information of the automotive parts image;

[0046] Step S132: Estimate the coarsening of the granularity of the automotive parts image according to the data of the loss of high-frequency information of the automotive parts image, and obtain the data of the coarsening of the granularity of the automotive parts image;

[0047] Step S133: Perform fuzzy analysis on the automotive parts image based on the data of the coarsening of the granularity of the automotive parts image and the data of the loss of high-frequency information of the automotive parts image, and obtain the fuzzy data of the automotive parts image;

[0048] Step S134: Detect the color distortion of the automotive parts image according to the details abnormality data of the automotive parts image, and obtain the data of the color distortion of the automotive parts image;

[0049] Step S135: Estimate the loss of texture of the automotive parts image based on the data of the color distortion of the automotive parts image and the fuzzy data of the automotive parts image, and obtain the data of the loss of texture of the automotive parts image;

[0050] Step S136: Based on the texture loss data of the automotive parts image and the color distortion data of the automotive parts image, perform defect recognition on the automotive parts image to obtain the defect data of the automotive parts image;

[0051] Step S137: According to the defect data of the automotive parts image, perform quality attenuation detection on the automotive parts image to obtain the quality attenuation data of the automotive parts image.

[0052] In the embodiments of the present invention, based on the abnormal data of the details of the automotive parts image obtained in the previous step, the detection of the loss of high-frequency information in the automotive parts image is carried out. The loss of high-frequency information usually manifests as the blurring or loss of the detailed parts (such as edges, textures, etc.) in the image. For this reason, the frequency-domain analysis method is used to convert the image into a frequency-domain representation. The image is converted into the frequency domain through the discrete Fourier transform (DFT) or the discrete cosine transform (DCT). In the converted spectrum, the high-frequency components mainly contain detailed information (such as edges and textures), while the low-frequency components contain the main structure of the image. By analyzing the energy distribution of the high-frequency components in the image spectrum, the loss of the high-frequency components is calculated. If the energy ratio of the high-frequency part is lower than a certain set threshold (such as lower than 10%), it is determined that the image has a loss of high-frequency information. The high-frequency information loss data obtained in this step includes information such as the proportion of high-frequency energy loss and the position of the loss area. Based on the high-frequency information loss data of the automotive parts image obtained in step S131, the change in the granularity of the automotive parts image is estimated. The coarsening of the granularity usually manifests as the blurring of the detailed parts in the image, resulting in a rough or granular appearance on the image surface. In order to estimate the change in granularity, the change in granularity is detected through the local contrast analysis of the image. The local contrast calculation method, such as the contrast metric based on Gaussian filtering, is used to analyze the contrast change in each region of the image. When the loss of high-frequency information is severe, the local contrast will decrease and the granularity of the image will increase. The change in granularity is estimated through the degree of contrast reduction. For example, when the contrast reduction exceeds 30%, it is determined that the granularity has coarsened. The coarsening data of the granularity contains information such as the position of the region where the granularity changes and the change amplitude. Based on the coarsening data of the granularity of the automotive parts image obtained in step S132 and the high-frequency information loss data obtained in step S131, the image blur analysis is carried out. The image blur analysis quantifies the sharpness of the image to determine whether the image is blurred. Through the image gradient method, the gradient amplitude of each pixel in the image is calculated (for example, the horizontal and vertical gradients are calculated using the Sobel operator), and the overall sharpness of the image is evaluated. If the gradient amplitude is lower than a certain threshold (for example, the average gradient amplitude is less than 10), it is determined that the image is blurred. Combining with the coarsening data of the granularity, when the change amplitude of the granularity is large and the overall gradient of the image is low, it is further confirmed that the image has a blur problem. The blur analysis result will output the position of the blurred area in the image, the degree of blur, and the correlation with the change in granularity. Based on the abnormal data of the details of the automotive parts image obtained in step S131, the color distortion detection of the automotive parts image is carried out. Color distortion usually manifests as the deviation or unnaturalness of the colors in the image. For this reason, by performing color space conversion on the image, it is converted from the RGB color space to the HSV or Lab color space. In these color spaces, the hue (H), saturation (S), and value (V) or lightness (L) components of the image are analyzed independently. In the hue component of the image, the distortion usually manifests as the deviation of the hue.By comparing the difference between the hue in the image and the standard color model, it is determined whether the image has color distortion. For example, when the hue value of a certain area in the image deviates by more than ±15°, it can be determined as color distortion. This step outputs color distortion data, specifically including the position of the distorted area, the degree of distortion (deviation angle), and the source of color distortion (such as light change or white balance distortion of the camera). Based on the color distortion data of the automotive parts image obtained in step S134 and the image blur data obtained in step S133, the texture loss estimation of the automotive parts image is carried out. Texture loss usually manifests as the absence or blur of surface texture in the image. To estimate texture loss, the Local Binary Pattern (LBP) is used to perform texture analysis on the image. The LBP method generates a binary pattern by comparing the value of each pixel with the values of its neighboring pixels, reflecting local texture information. When the image is blurred or has color distortion, the contrast and consistency of the LBP texture features will be significantly reduced. Therefore, based on the blur data and color distortion data, the degree of texture loss in the image is further evaluated. The specific method includes calculating the entropy value and consistency index of the LBP features. When the entropy value of texture loss is greater than a certain threshold (such as 0.5), texture loss is determined. The texture loss data output by this step includes the position of the lost area, the degree of loss, and the correlation with blur and color distortion. Based on the texture loss data of the automotive parts image obtained in step S135 and the image color distortion data obtained in step S134, the defect recognition of the automotive parts image is carried out. Image defect recognition identifies the defective areas in the image by comparing the texture features of the normal image and the distorted image. This process uses a defect detection method based on feature matching, extracts the local features of the image (such as texture, edge, shape, etc.), and matches them with the features of the standard template or the normal image. When the features of the texture loss and color distortion areas are significantly different from the standard template, it can be determined that there are defects in this area. Through feature difference analysis, the types of defects in the image (such as color distortion, blur, texture loss, etc.) are identified. The identified defect data includes the type, position of the defect, and the size of the area affected by it. According to the defect data of the automotive parts image obtained in step S136, the quality attenuation detection of the automotive parts image is carried out. Quality attenuation detection comprehensively evaluates the number and severity of image defects to determine the degree of image quality attenuation, and counts the number of defective areas in the image and the influence range of each defect. For each type of defect (such as color distortion, texture loss, blur, etc.), a weighted evaluation is carried out according to the area it occupies in the image and its severity. If the area of the defective area in the image exceeds 30% of the total image area, it is determined that the image quality attenuation is relatively serious. The attenuation detection results include the distribution of the attenuation area, the degree of attenuation, and the overall quality score.

[0053] Preferably, step S2 includes the following steps:

[0054] Step S21: Identify the geometric features of auto parts based on the optimized data of auto part images to obtain the geometric feature data of auto parts;

[0055] Step S22: Identify the local details of auto parts based on the geometric feature data of auto parts and the optimized data of auto part images to obtain the local detail data of auto parts;

[0056] Step S23: Evaluate the structural stability of auto parts based on the local detail data of auto parts and the geometric feature data of auto parts to obtain the structural stability data of auto parts;

[0057] Step S24: Estimate the durability of auto parts based on the structural stability data of auto parts and the local detail data of auto parts to obtain the durability data of auto parts.

[0058] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes:

[0059] Step S21: Identify the geometric features of auto parts based on the optimized data of auto part images to obtain the geometric feature data of auto parts;

[0060] In the embodiment of the present invention, the basic geometric shapes and structures of auto parts are identified through image analysis algorithms, such as circles, squares, rectangles, curved edges, etc. The specific operation method is to use edge detection techniques, such as Canny edge detection or Sobel operator, to identify the significant contour lines in the image. Then, contour tracking and shape analysis techniques are used to extract the geometric information of auto parts. For example, if the image shows an auto wheel hub, the algorithm detects its circular edge and identifies its geometric shape through a circle fitting algorithm (such as Hough transform). This step can also further improve the accuracy of geometric features by combining feature point detection techniques, such as Harris corner detection. Through geometric feature extraction, data such as the size, shape, and symmetry of each auto part are obtained.

[0061] Step S22: Identify the local details of auto parts based on the geometric feature data of auto parts and the optimized data of auto part images to obtain the local detail data of auto parts;

[0062] In the embodiments of the present invention, the goal of local detail recognition is to further extract subtle geometric information on the surface of the parts, such as small holes, concave-convex textures, and fine cracks. High-pass filtering or Laplace operators are used to enhance the details of the image, highlighting the local changes in the image. Through the processed image, local feature descriptors, such as Scale-Invariant Feature Transform (SIFT) or Speeded Up Robust Features (SURF), are used to detect and describe local feature points. In particular, in the areas of subtle defects (such as cracks or wear) on the surface of the parts, the algorithm accurately locates and analyzes the details by matching local textures and edge features. This process uses a deep learning network (such as Convolutional Neural Network CNN) to further enhance the recognition ability, especially in the recognition of complex local structures. The local detail data output by the steps includes the position, size, type (such as cracks, wear, etc.) of the detail area and the corresponding geometric features.

[0063] Step S23: Based on the local detail data of the automotive parts and the geometric feature data of the automotive parts, conduct a structural stability assessment of the automotive parts to obtain the structural stability data of the automotive parts;

[0064] In the embodiments of the present invention, based on the geometric feature data of the automotive parts obtained in step S21 and the local detail data of the automotive parts obtained in step S22, a structural stability assessment of the automotive parts is conducted to obtain the structural stability data of the automotive parts. The goal of the structural stability assessment is to determine whether structural failures or malfunctions occur during the use of the automotive parts. The finite element analysis (FEA) method is used to build a structural model of the parts and simulate the stress conditions under different working conditions. Based on the geometric feature data, a three-dimensional model of the automotive parts is generated (for example, using computer-aided design (CAD) software or 3D scan data). In the model, different mechanical loads (such as vibration, pressure, etc.) are applied and stress analysis is conducted to evaluate whether the structure of the parts is stable. Local detail data (such as micro-cracks or surface wear) will affect the load-bearing capacity of the structure. Therefore, these local defects are used as initial defect points for further fatigue analysis or fracture analysis. Through these simulation calculations, the stress distribution and deformation of each part under different working conditions are obtained.

[0065] Step S24: Based on the structural stability data of the automotive parts and the local detail data of the automotive parts, conduct a durability prediction of the automotive parts to obtain the durability data of the automotive parts.

[0066] In the embodiments of the present invention, based on the structural stability data of automotive parts obtained in step S23 and the local detail data of automotive parts obtained in step S22, durability prediction of automotive parts is carried out to obtain the durability data of automotive parts. The goal of durability prediction is to predict the lifespan of automotive parts under long-term use or harsh conditions. For this purpose, through algorithms based on fatigue life, such as the Miner's rule or the S-N curve based on materials science, fatigue analysis is carried out in combination with the structural stability data of the parts. Specifically, the force cycle frequency of the parts under normal use is calculated, and combined with the fatigue limit of the material, the fatigue life of the parts is predicted. In addition, information on micro-cracks or surface defects in the local detail data will be incorporated into the prediction model, which are risk factors leading to early failure. On this basis, the Monte Carlo simulation method is used to conduct multiple rounds of simulations to simulate the durability performance under different environmental conditions (such as temperature, humidity, load, etc.). Through these simulation results, the expected lifespan and durability grade of automotive parts are obtained. The durability data output in this step includes the predicted lifespan, lifespan prediction under the use environmental conditions, and key durability indicators (such as fatigue life, wear life, etc.).

[0067] Preferably, step S21 includes the following steps:

[0068] Step S211: Perform automotive part image segmentation processing on the optimized data of automotive part images to obtain automotive part image segmentation data;

[0069] Step S212: Based on the automotive part image segmentation data, perform automotive part outer boundary refinement processing to obtain automotive part outer boundary refinement data;

[0070] Step S213: According to the automotive part outer boundary refinement data, draw the outer contour of the automotive part to obtain automotive part outer contour data;

[0071] Step S214: According to the automotive part image segmentation data, collect the deep-level internal features of the automotive part to obtain automotive part deep-level internal feature data;

[0072] Step S215: Based on the automotive part deep-level internal feature data, measure the internal structural shape of the automotive part to obtain automotive part internal structural shape data;

[0073] Step S216: Based on the automotive part internal structural shape data, detect the internal structural spatial layout of the automotive part to obtain automotive part internal structural spatial layout data;

[0074] Step S217: According to the automotive part outer contour data and the automotive part internal structural spatial layout data, identify the geometric features of the automotive part to obtain automotive part geometric feature data.

[0075] In the embodiments of the present invention, the main purpose of image segmentation is to extract the area of automotive parts from the original image and remove background and interference information. The threshold segmentation algorithm is adopted. By calculating the gray value differences of different regions in the image, a suitable threshold is selected to divide the image into the parts area and the background area. Specifically, the Otsu algorithm is used to automatically calculate the optimal threshold, so as to effectively separate the automotive parts from the background area. In addition, the region growing algorithm is combined to perform more refined segmentation on the automotive parts, gradually expanding from a seed pixel point until a predetermined region size or threshold is reached. Edge detection is performed on the image, especially Canny edge detection, to further enhance the accuracy of the segmentation boundary. The obtained image segmentation data of the automotive parts contains clear contours and boundary information of the parts area. Based on the image segmentation data of the automotive parts, the external boundary refinement of the automotive parts is processed to obtain the refined data of the external boundary of the automotive parts. The purpose of external boundary refinement is to further improve the accuracy of the image segmentation boundary and make it closer to the actual contour of the automotive parts. Refinement algorithms such as the Zhang-Suen refinement algorithm or the Hilditch refinement algorithm are applied to remove the redundant pixels of the boundary and only retain the most accurate contour lines. These refinement algorithms analyze the edge points through repeated iterations to ensure that each edge point converges towards the true shape. During this process, the coherence and stability of the boundary line are checked pixel by pixel to eliminate small noise points or irregular shapes that appear during the segmentation process. The refined external boundary data will contain a more accurate parts contour and be ready to enter the next step of external contour drawing. According to the refined data of the external boundary of the automotive parts, the external contour of the automotive parts is drawn to obtain the external contour data of the automotive parts. The purpose of external contour drawing is to accurately draw the external contour of the automotive parts through the refined boundary data. By using linear interpolation or spline curve fitting methods, the refined boundary points are connected into continuous curves or polygons. For parts with complex shapes, Bezier curves or B-spline curves are used for high-precision fitting to ensure that the boundary curves are smooth and conform to the geometric shape of the actual parts. In some cases, if the parts have relatively regular geometric shapes (such as circles, rectangles, etc.), the external contour data output by directly fitting the contour through geometric formulas contains a detailed description of the external shape of the parts. According to the image segmentation data of the automotive parts, the deep internal features of the automotive parts are collected to obtain the deep internal feature data of the automotive parts. The goal of deep internal feature collection is to extract the structural details inside the automotive parts, such as holes, notches, reinforcement stripes, etc. The internal area of the parts is obtained through image segmentation, and then the image is processed using convolutional filters to enhance the visibility of the deep features. High-pass filters or gradient enhancement methods are used to highlight the subtle structural changes in the image. For example, the Sobel operator is used to perform edge detection on the image, or the Gabor filter is used to extract texture information.Deep feature acquisition not only includes features in terms of shape, but also details such as the contrast and texture patterns of the image. The output deep feature data contains various structural information inside the parts. Based on the deep feature data of the internal structure of automotive parts, the internal structure shape measurement of automotive parts is carried out to obtain the internal structure shape data of automotive parts. The goal of the internal structure shape measurement is to accurately measure the geometric features of the internal structure of automotive parts, such as the diameter, depth, angle, etc. of holes. Using the deep feature data, each part inside the part (such as holes, grooves, etc.) is extracted. Through geometric shape analysis methods, the specific dimensions of these structures are calculated. The Hough transform is used to detect the radius of circular holes, and the line fitting algorithm is used to measure the length and width of elongated notches. For internal structures with irregular shapes, the least squares method is used to fit their approximate geometric shapes, thereby obtaining their accurate dimension data. Based on the internal structure shape data of automotive parts, the internal structure spatial layout detection of automotive parts is carried out to obtain the internal structure spatial layout data of automotive parts. The goal of the spatial layout detection is to evaluate the relative positions and spatial relationships between the various structures inside the part. Based on the internal structure shape data, a three-dimensional model of the internal structure of the part is established. Using coordinate transformations in three-dimensional space, the distances, angles, and arrangement patterns between the various structures are analyzed. For complex structures, structure grouping is carried out through distance-based clustering algorithms (such as the K-means or DBSCAN algorithms) to detect whether the spatial arrangement of the internal structure meets the design specifications. In addition, combined with the design requirements of the part, a geometric constraint algorithm is used to detect whether there are potential interferences or conflicts inside the structure. The output spatial layout data contains the spatial and positional relationships between the internal structures of the part. Based on the external contour data of automotive parts and the internal structure spatial layout data of automotive parts, the geometric feature recognition of automotive parts is carried out to obtain the geometric feature data of automotive parts. The goal of geometric feature recognition is to extract the specific geometric features of the part from the external contour and internal structure data. Combining the external contour data, a shape matching algorithm is used to compare the part with a predefined geometric template to identify the shape type of the part (such as circular, rectangular, irregular, etc.). Then, combined with the internal structure spatial layout data, a spatial relationship analysis method is used to further analyze the geometric features inside the part. For example, by analyzing the distribution of holes, the shape of notches, etc., the functional features of the part (such as the flow holes of hydraulic system parts, the reinforcement grooves of bracket parts, etc.) are identified. The obtained geometric feature data includes information such as the external shape category, dimensions, hole diameter, and the relationship between the internal and external structures of the part, and can comprehensively describe the geometric characteristics of the part.

[0076] Preferably, step S23 includes the following steps:

[0077] Step S231: Measure the thickness at the connection of the automotive part according to the geometric feature data of the automotive part to obtain the thickness data at the connection of the automotive part;

[0078] Step S232: Identify the support structure of the auto parts based on the geometric feature data of the auto parts to obtain the support structure data of the auto parts;

[0079] Step S233: Collect the surface texture features of the auto parts based on the local detail data of the auto parts to obtain the surface texture feature data of the auto parts;

[0080] Step S234: Collect the force distribution points of the auto parts based on the surface texture feature data of the auto parts to obtain the force distribution point data of the auto parts;

[0081] Step S235: Perform a load simulation on the support structure data of the auto parts, the thickness data of the connections of the auto parts, and the force distribution point data of the auto parts to obtain the load simulation data of the auto parts;

[0082] Step S236: Evaluate the structural stability of the auto parts based on the load simulation data of the auto parts to obtain the structural stability data of the auto parts.

[0083] In the embodiments of the present invention, the thickness of the connection part of the automotive parts is measured according to the geometric feature data of the automotive parts to obtain the thickness data of the connection part of the automotive parts. The goal of this operation is to accurately measure the thickness of the connection part of the automotive parts to ensure that it meets the design requirements. Using the geometric feature data obtained in the previous steps, the connection area of the automotive parts is analyzed. By selecting a suitable measurement method, laser ranging technology or an ultrasonic thickness gauge is used to directly measure the thickness of the connection part. These measurement technologies can quickly and accurately obtain the thickness value of the connection part in a non-contact situation. To further improve the accuracy, 3D reconstruction technology is adopted. By collecting the 3D data of the parts, the thickness of the connection part is calculated using computer vision algorithms. During the processing, an interpolation algorithm is used to smooth the data and eliminate measurement errors. The obtained thickness data of the connection part will reflect the specific thickness of each connection part. According to the geometric feature data of the automotive parts, the support structure of the automotive parts is identified to obtain the support structure data of the automotive parts. The purpose of support structure identification is to analyze the geometric structure of the parts and identify the support parts therein. Based on the geometric feature data in step S231, a shape matching algorithm is used to analyze the geometric structure of the parts and identify the parts with a support function. The Hough transform is used to detect the support shapes (such as cylinders, reinforcing ribs, etc.) in the parts, and their positions and structures are analyzed in combination with geometric constraint conditions. For parts with relatively complex support shapes (such as complex inclined planes, reinforcing beams, etc.), feature extraction is performed on the images through deep learning algorithms (such as convolutional neural networks) to identify the specific shapes and positions of the support structures. Through multi-angle image analysis and in combination with a 3D reconstruction model, the support structure data of the parts is accurately extracted. According to the local detail data of the automotive parts, the surface texture features of the automotive parts are collected to obtain the surface texture feature data of the automotive parts. The goal of this step is to extract useful texture features from the surface of the automotive parts and analyze its surface state. Through image segmentation technology, the surface area of the automotive parts is separated from the parts. Then, a Gabor filter or wavelet transform is applied to extract features from the surface area to obtain its texture information. In particular, the Gabor filter effectively captures texture features at different scales and directions, while wavelet transform extracts surface details in different frequency ranges. Then, texture analysis algorithms (such as gray-level co-occurrence matrix) are used to calculate indicators such as the contrast, energy, and correlation of the surface texture. These surface texture feature data reflect the roughness, smoothness, and wear conditions of the part surface. According to the surface texture feature data of the automotive parts, the force distribution points of the automotive parts are collected to obtain the force distribution point data of the automotive parts. The goal of this step is to infer the force distribution of the parts based on the surface texture features of the parts. By the change of the surface texture, especially the non-uniformity of the texture, the force distribution of the automotive parts during use is inferred. Combining with the texture feature data extracted in the previous step, the relationship between the surface texture and the stress distribution is analyzed using a mechanical model.The geometric model of the fitting is combined with the stress state using the finite element analysis method (FEA). By simulating different load conditions, the stress distribution points are calculated. Local changes in the texture are usually related to stress concentration points. Therefore, by analyzing the texture changes and stress concentration areas, the stress distribution of the automotive fitting is deduced. The obtained stress distribution point data helps analyze the high-stress areas that occur during the actual use of the fitting. Load simulation is performed on the automotive fitting support structure data, the thickness data at the connection of the automotive fitting, and the stress distribution point data of the automotive fitting to obtain the automotive fitting load simulation data. The purpose of the load simulation is to evaluate the performance of the fitting under working conditions by applying different loads to various parts of the fitting. Combining the support structure data, a finite element model of the automotive fitting is established, and different loads are applied to each connection point of the fitting based on the thickness data at the connection. During the simulation process, load types such as tension, compression, and bending are used to simulate the stress conditions of the fitting during actual use. At the same time, combining the stress distribution point data, the simulation results are combined with the surface texture characteristics to further refine the load distribution. Through dynamic simulation, the stress and deformation conditions of the fitting under different loads are analyzed to obtain the load simulation data. Based on the automotive fitting load simulation data, the structural stability of the automotive fitting is evaluated to obtain the automotive fitting structural stability data. The purpose of this step is to evaluate the structural stability of the automotive fitting under different load conditions. By analyzing the load simulation data, the deformation of the fitting under various stress actions is detected. Using the yield criterion and ultimate bearing capacity analysis methods, combined with the theory of material mechanics, it is judged whether the fitting exhibits yield, failure, or instability. During this process, the Monte Carlo method is used for multiple simulations to analyze the stability under different load conditions to ensure the reliability of the evaluation results. Using the stress-strain curve, it is judged whether the fitting will undergo permanent deformation during operation or whether there is a risk of fatigue failure after long-term use. The obtained structural stability data reflects the bearing capacity and long-term durability of the automotive fitting during actual use.

[0084] Preferably, step S3 includes the following steps:

[0085] Step S31: Calculate the basic parameters of the automotive fitting based on the automotive fitting durability data and the automotive fitting structural stability data to obtain the basic parameters of the automotive fitting;

[0086] Step S32: Classify the automotive fitting image optimization data according to the automotive fitting basic parameters and the automotive fitting geometric feature data to obtain the automotive fitting image function classification data;

[0087] Step S33: Perform automotive fitting image tagging annotation on the automotive fitting image function classification data to obtain the automotive fitting image tagged data;

[0088] Step S34: Train the image classification model with the image tagging data of auto parts and the functional classification data of auto parts images to obtain the auto parts image classification model.

[0089] In the embodiments of the present invention, the basic parameters of auto parts are calculated based on the durability data of auto parts and the structural stability data of auto parts to obtain the basic parameters of auto parts. The goal of this operation is to obtain the basic parameters of auto parts, such as the strength, deformation performance, service life, etc. of the material, by integrating the analysis results of durability and structural stability. Based on the durability data, the fatigue characteristics of the material of the auto parts are analyzed to calculate the fatigue life of the material under different loads. On the basis of the structural stability data, the stress-strain analysis method is used to further evaluate the structural performance of the parts during long-term use. Combining these data with the actual operating environment, the basic parameters are deduced through mathematical models (such as linear regression or non-linear fitting) to obtain the core performance parameters of auto parts. For example, the durability data reveals the potential fatigue failure risk of a certain part of the parts, while the structural stability data shows the deformation degree of some parts under specific stress conditions. According to the basic parameters of auto parts and the geometric feature data of auto parts, the image optimization data of auto parts is classified according to the functions of auto parts to obtain the image function classification data of auto parts. The purpose of this step is to classify the image data of auto parts according to their basic parameters and geometric features, so as to provide structured information for subsequent image analysis. Combining the basic parameters obtained in step S31, the material properties, structural characteristics, etc. of the parts are analyzed to determine their functions in the vehicle. Then, the image is classified through geometric feature data, such as the shape, size, and structural complexity of the parts. Using image feature extraction techniques, such as SIFT (Scale-Invariant Feature Transform) or SURF (Speeded-Up Robust Features), the key features of the parts image are extracted, and then clustering analysis is performed in combination with the geometric data of the parts. By clustering the parts with different functions, multiple function types (such as support parts, transmission parts, etc.) are divided. The image function classification data of auto parts is labeled to obtain the labeled data of auto parts images. The goal of this step is to label the image data in detail to provide accurate training samples for the training of the image classification model. According to the image function classification result in step S32, corresponding labels are assigned to each type of image. For example, the image of the support part category is labeled as "support", and the transmission part category is labeled as "transmission". The functional areas of each image are accurately labeled, and the labeling content includes but is not limited to the functional category of the parts, the specific use parts, etc. During the labeling process, image segmentation technology is used to divide the parts image into different regions and mark them according to the functions of the regions. These labels are stored in a standard data format (such as YOLO or COCO format), and the label of each image will include its functional category and additional information (such as the size, material, etc. of the parts). When performing labeled annotation, an appropriate annotation tool, such as LabelMe or VGG Image Annotator (VIA), is selected according to the complexity of the data.Based on the labeled data of automotive parts images and the functional classification data of automotive parts images, the image classification model is trained to obtain an automotive parts image classification model. The core objective of this step is to use the labeled image data to train an efficient and accurate image classification model. Using the labeled data obtained in step S33 and combining it with the functional classification data in step S32, each labeled parts image and its corresponding functional label are input into the deep learning model for training. The convolutional neural network (CNN) is selected as the main model architecture because of its excellent performance in image recognition. During the training process, the cross-entropy loss function is used to optimize the model, minimizing the error between each prediction result and the actual label. Through multiple iterations, the model continuously adjusts its weights to improve the classification accuracy. A common CNN architecture such as ResNet, VGG, or Inception model is selected, or the model is customized according to specific requirements. During the training process, data augmentation techniques such as rotation, scaling, and cropping are adopted to increase the robustness of the model. When training, hyperparameters such as the learning rate, batch size, and optimizer (such as Adam or SGD) are set according to the specific application scenario.

[0090] Preferably, step S4 includes the following steps:

[0091] Step S41: Obtain the fault data of the running vehicle;

[0092] Step S42: Analyze the severity of the parts faults based on the fault data of the running vehicle to obtain the data on the severity of automotive parts faults;

[0093] Step S43: Identify the applicable parts based on the data on the severity of automotive parts faults using the automotive parts image classification model to obtain the data on applicable parts for vehicle faults;

[0094] Step S44: Calculate the matching degree of automotive parts based on the data on applicable parts for vehicle faults to obtain the data on the matching degree of applicable automotive parts;

[0095] Step S45: Optimize the intelligent identification of automotive parts for the automotive parts image classification model based on the data on the matching degree of applicable automotive parts to generate an optimized automotive parts image classification model.

[0096] As an example of the present invention, refer to Figure 3 As shown, in this example, step S4 includes:

[0097] Step S41: Obtain the fault data of the running vehicle;

[0098] In an embodiment of the present invention, vehicle fault data during driving is obtained. The goal of this step is to collect and organize vehicle fault information that occurs during actual driving. This data is obtained through various channels, such as data recorded by the on-board diagnostic system (OBD), maintenance records, fault codes, sensor data, etc. The OBD system records information such as vehicle speed, engine load, temperature, etc. together with fault codes by real-time monitoring the operating status of each vehicle component, forming a fault data set. The on-board diagnostic device usually provides diagnostic trouble codes (DTCs), which are identifiers generated by the vehicle electronic control unit (ECU) when detecting abnormalities. Further, GPS devices are used to obtain vehicle position data to understand the environmental conditions where the faults occur.

[0099] Step S42: Analyze the severity of component faults based on the vehicle fault data during driving to obtain vehicle component fault severity data;

[0100] In an embodiment of the present invention, the severity of component faults of vehicle fault data is analyzed based on the vehicle fault data during driving to obtain vehicle component fault severity data. The purpose of this step is to conduct an in-depth analysis of vehicle component faults, identify and quantify the severity of the faults. According to the fault data obtained in step S41, each fault information is analyzed through data mining techniques to extract key features related to components (such as fault type, occurrence frequency, fault location, fault code, etc.). Systematic evaluation of each vehicle component is carried out using fault tree analysis (FTA) or root cause analysis (RCA) methods to identify which components have more severe faults. Data mining algorithms, such as clustering analysis or decision tree models, further help to discover the patterns of component faults and classify the component fault situations into different levels according to the occurrence frequency and impact degree of the faults. For example, if a certain component has a high fault frequency and directly affects vehicle safety, the severity level of its fault is higher, and the obtained vehicle component fault severity data forms a multi-dimensional database.

[0101] Step S43: Identify applicable components based on the vehicle component fault severity data using the vehicle component image classification model to obtain vehicle fault applicable component data;

[0102] In the embodiment of the present invention, the applicable parts are identified for the data of the severity of the automotive parts failure according to the automotive parts image classification model, and the applicable parts data for the automotive failure are obtained. This step aims to associate the parts failure situation with the applicable parts, provide a parts identification solution for specific failures, combine the parts failure severity data in step S42 with the parts image data, and use the pre-trained automotive parts image classification model to perform image classification and identification on the failure situation. In this process, the image classification model processes and classifies the images of the parts through deep learning algorithms such as convolutional neural networks (CNNs). The parts images are input into the model, and the model will identify the parts that need to be replaced according to the failure type and the part form. For example, when the engine sensor fails, the image classification model identifies the image of the part and matches it with the failure type to determine the appropriate replacement part. Through the output of the image classification model, the applicable parts data corresponding to the failure are generated.

[0103] Step S44: Calculate the matching degree of automotive parts according to the applicable parts data for automotive failures, and obtain the matching degree data of applicable automotive parts;

[0104] In the embodiment of the present invention, the matching degree of automotive parts is calculated according to the applicable parts data for automotive failures, and the matching degree data of applicable automotive parts are obtained. The core objective of this step is to calculate the matching degree between automotive parts and the failure situation to ensure that the replacement parts meet the vehicle requirements. Based on the applicable parts data in step S43, a parts matching model is constructed. This model calculates the matching degree between each applicable part and the failed part by comparing the technical parameters, appearance features, and functional requirements of the current failed part. Use methods based on vector space models (VSMs) or similarity calculation methods, such as cosine similarity, Euclidean distance, or Manhattan distance, to perform matching calculations on the part characteristics. For example, input the characteristics such as the size, material, and interface type of the part, and compare them with the feature vectors of the failed part to obtain its matching degree score. The parts with higher matching degree scores will be recommended as replacement parts. Through this matching degree calculation process, the applicability evaluation data of automotive parts are obtained.

[0105] Step S45: Optimize the intelligent identification of automotive parts for the automotive parts image classification model based on the matching degree data of applicable automotive parts, and generate an optimized model for automotive parts image classification.

[0106] In the embodiments of the present invention, the automotive parts image classification model is optimized for intelligent recognition of automotive parts based on the automotive parts matching degree data, and an optimized automotive parts image classification model is generated. The goal of this step is to optimize the existing image classification model through the parts matching degree data to improve the accuracy and practicality of the model. According to the applicable parts matching degree data in step S44, a set of parts images with high matching degrees and corresponding labels (such as fault types, parts features, etc.) are collected. Using these high-quality data, the existing image classification model is retrained. Through transfer learning or fine-tuning techniques, the pre-trained model (such as ResNet or VGG) is optimized. During the training process, by adjusting hyperparameters such as the learning rate and batch size, the adaptability and accuracy of the model on the new dataset are ensured. In addition, to further improve the model performance, data augmentation techniques, such as image rotation, translation, and cropping, are adopted to increase the diversity of the training set. Through model optimization, the generated optimized automotive parts image classification model can more accurately identify the parts types.

[0107] Preferably, step S42 includes the following steps:

[0108] Step S421: Analyze the spatial distribution of automotive fault areas based on the driving automotive fault data to obtain automotive parts fault area data;

[0109] Step S422: Calculate the concentration degree of the parts fault area based on the automotive parts fault area data to obtain the parts fault area concentration degree data;

[0110] Step S423: Calculate the influence degree of the automotive parts fault on the parts fault area concentration degree data and the automotive parts fault area data to obtain the automotive parts fault influence degree data;

[0111] Step S424: Conduct a weighted evaluation of the severity of the automotive parts fault based on the automotive parts fault influence degree data and the parts fault area concentration degree data to obtain the weighted data of the severity of the automotive parts fault;

[0112] Step S425: According to the weighted data of the severity of the automotive parts fault and the automotive parts fault influence degree data, the automotive parts fault influence degree data.

[0113] In the embodiments of the present invention, for the spatial distribution analysis of vehicle fault areas, the purpose is to accurately identify the specific areas where faults occur by analyzing the fault data of moving vehicles, so as to provide basic data for subsequent parts diagnosis and optimization. Specifically, it is necessary to obtain the vehicle fault data during driving, and the data sources are the data collected by the on-board diagnostic system (OBD) of the vehicle and various sensors, such as engine temperature, brake system pressure, vehicle speed, etc. These data are recorded in real time and uploaded to the cloud or a local database. During this process, the quality and accuracy of the data should be ensured, and data preprocessing techniques (such as noise removal, data smoothing, etc.) are used to clean unnecessary outliers. Then, according to the GPS positioning data of the vehicle and the time stamp of the fault event occurrence, the fault data is paired with the geographical location information of the vehicle. In order to further analyze the fault distribution, spatial data analysis techniques, such as spatial clustering algorithms (such as the DBSCAN algorithm), are used to identify the areas where the fault occurrence frequency is higher and generate a spatial distribution map of the fault areas. During this process, according to the structural characteristics of the vehicle, the fault information is associated with different component areas of the vehicle, such as the engine compartment, chassis, brake system, etc., and the data is integrated to generate fault area data. Based on the results of the spatial distribution analysis, the fault hot areas of different vehicle parts are clearly shown. The concentration degree of the parts fault areas is calculated based on the parts fault area data. The goal is to quantify the concentration degree of the fault areas. The area where each fault event occurs is extracted from the fault area data obtained in step S421, and the fault concentration degree of different areas is measured according to the occurrence frequency of the fault events. To achieve this goal, spatial analysis techniques, such as the kernel density estimation (KDE) method, are used. Kernel density estimation is to generate a density map of the fault area by setting a window to weight the data points. Specifically, when implementing, a suitable bandwidth parameter needs to be determined, and this parameter controls the influence range of each fault point on its surrounding area. Selecting a suitable bandwidth is very important. Too small a bandwidth results in a too discrete distribution map, while too large a bandwidth will cover up the actual high-concentration areas. By calculating the number of fault points in different areas, a concentration index of the fault area is obtained. This index reflects the frequency and density of fault occurrences in a certain area, and the higher the value, the more concentrated the faults in that area. These data provide an important reference basis for the subsequent calculation of the fault impact degree. The areas with higher concentration are often the priority objects for fault repair, and these areas are related to high-risk or high-frequency faults. The impact degree of vehicle parts faults is calculated based on the concentration data of the parts fault areas and the vehicle parts fault area data. The purpose is to comprehensively consider the concentration of the fault areas and the impact of the faults on the vehicle functions, and evaluate the impact degree of the faults in each area. According to the concentration data in step S422 and the area fault information obtained in step S421, the impact degree of the fault areas is quantified by means of weighted calculation.In this process, a weight coefficient needs to be assigned to each failure area, and this coefficient is determined according to the type of failure and the automotive systems involved. For example, failures in the engine system usually have a greater impact on the safety and performance of the whole vehicle, so a higher weight should be given. While failures in the in-vehicle electrical system have a smaller impact on the overall performance of the vehicle, and a lower weight is given. After the weight assignment, the weighted average or weighted sum method is used to combine the concentration of the failure area with the influence of the failure type to calculate the total influence degree of each failure area. Areas with a higher influence degree usually refer to those areas where there are both highly concentrated failures and are directly related to the safety of vehicle operation. Failures in these areas have a greater impact on the overall performance of the vehicle. By calculating and sorting the influence degrees of each failure area, data on the influence degree of automotive parts failures is obtained. Based on the data on the influence degree of automotive parts failures and the data on the concentration of parts failure areas, a weighted assessment of the severity of automotive parts is carried out. The purpose is to evaluate and weight the severity of each parts failure, so as to provide a priority ranking for subsequent maintenance work. To achieve this assessment process, a severity weighting coefficient needs to be assigned to each failure area, and this coefficient is determined by comprehensively considering the influence degree and concentration of the failure. Specifically, a weighted formula is used to calculate the weighted severity of each failure area, and the formula is as follows: Weighted severity = w1 × Influence degree + w2 × Concentration. Where w1 and w2 are the weight coefficients representing the influence degree of the failure and the concentration of the failure respectively. The setting of the weight coefficients is adjusted according to the contribution of each vehicle part to the overall performance. For example, the importance of the engine and braking systems is relatively high, and larger weights are set. Through this weighted method, the influence of the failure area is combined with its concentration, so as to obtain a comprehensive score reflecting the severity of the failure. These scores will determine which areas of failure need to be processed first. The weighted severity data will be organized into a table or list, and the failure areas with higher rankings will be the priority items for maintenance or replacement of parts. Based on the weighted data of the severity of automotive parts failures and the data on the influence degree of automotive parts failures, the influence degree of automotive parts failures is further calculated and evaluated to provide a basis for intelligent identification and matching of parts. The goal of this step is to refine the analysis of the influence of automotive parts failures by deeply analyzing the relationship between the weighted severity data and the influence degree data of failures, and to provide support for parts matching and intelligent identification. When operating specifically, it is necessary to combine the weighted severity data obtained in step S424, evaluate the influence degree of each failure area, and deeply analyze the influence of different failure areas. Based on these analysis results, the influence rating of each failure area is further adjusted using a weighted algorithm. For example, the adjustment coefficient of its influence degree is determined according to the type of parts failure (such as affecting safety, affecting power transmission, etc.) and the actual scope of the influence (such as local damage, global systematic failure).

[0114] The above are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather will conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. An intelligent identification method for automobile parts based on an image classification model, characterized in that: The following steps are involved: Step S1: performing image acquisition on automobile parts to obtain automobile parts image data; performing automobile parts image optimization on the automobile parts image data to obtain automobile parts image optimization data; Step S2: performing geometric feature recognition of automobile parts according to the automobile parts image optimization data to obtain geometric feature data of automobile parts; performing structural stability evaluation of automobile parts based on the geometric feature data of automobile parts to obtain structural stability data of automobile parts; performing durability estimation of automobile parts according to the structural stability data of automobile parts to obtain durability data of automobile parts; Step S3: performing automobile parts image function classification on the automobile parts image optimization data according to the automobile parts durability data and the automobile parts geometric feature data to obtain automobile parts image function classification data; performing image classification model training on the picture classification model based on the automobile parts image function classification data to obtain an automobile parts image classification model; Step S4: Acquire the running vehicle fault data; identify the applicable parts for the running vehicle fault data according to the automobile parts image classification model to obtain the applicable parts data for the vehicle fault; optimize the automobile parts intelligent recognition of the automobile parts image classification model based on the applicable parts data for the vehicle fault to generate the automobile parts image classification optimization model.

2. The method for intelligent identification of automobile parts based on image classification model according to claim 1 is characterized in that: Step S1 includes the following steps: Step S11: Acquire the illumination data of the automobile parts and the automobile parts image acquisition area; perform image acquisition on the automobile parts, thereby obtaining the automobile parts image data; Step S12: performing image detail abnormality recognition on the automobile parts image data and the automobile parts according to the illumination data of the automobile parts image acquisition area, thereby obtaining automobile parts image detail abnormality data; Step S13: performing automobile parts image quality attenuation detection according to the automobile parts image detail abnormality data to obtain automobile parts image quality attenuation data; Step S14: Optimizing the automobile parts image data by using the automobile parts image quality attenuation data to obtain automobile parts image optimization data.

3. The method for intelligent identification of automobile parts based on image classification model according to claim 2 is characterized in that: Step S12 includes the following steps: Step S122: estimating the direct light irradiation of the automobile parts according to the illumination data of the automobile parts image acquisition area, and obtaining the direct light irradiation data of the automobile parts; Step S123: analyzing excessive reflected light from automobile parts according to the direct light irradiation data of automobile parts, and obtaining excessive reflected light data from automobile parts; Step S124: performing image overexposure detection on the image data of the automobile parts according to the data of excessively strong reflected light from the automobile parts, to obtain the image overexposure data of the automobile parts; Step S125: estimating the color saturation of the automobile parts image based on the overexposure data of the automobile parts image, and obtaining the color saturation of the automobile parts image data; Step S126: performing image detail anomaly recognition on the automobile parts image data based on the automobile parts image color oversaturation data and the automobile parts image overexposure data, thereby obtaining automobile parts image detail anomaly data.

4. The method for intelligent identification of automobile parts based on image classification model according to claim 2 is characterized in that: Step S13 includes the following steps: Step 131: performing high-frequency information loss detection on the automobile parts image according to the abnormal details data of the automobile parts image to obtain high-frequency information loss data on the automobile parts image; Step S132: Predicting the coarsening of the granularity of the automobile parts image according to the high-frequency information loss data of the automobile parts image, and obtaining the coarsening data of the granularity of the automobile parts image; Step S133: performing fuzzy analysis on the automobile parts image based on the coarsening data of the automobile parts image granularity and the high-frequency information loss data of the automobile parts image to obtain fuzzy data of the automobile parts image; Step S134: performing color distortion detection on the automobile parts image according to the abnormal detail data of the automobile parts image to obtain color distortion data on the automobile parts image; Step S135: performing automobile parts image texture loss estimation on the automobile parts image color distortion data and the automobile parts image blur data to obtain automobile parts image texture loss data; Step S136: performing automobile part image defect recognition based on the automobile part image texture loss data and the automobile part image color distortion data to obtain automobile part image defect data; Step S137: Performing automobile parts image quality attenuation detection according to the automobile parts image defect data to obtain automobile parts image quality attenuation data.

5. The method for intelligent identification of automobile parts based on image classification model according to claim 1 is characterized in that: Step S2 includes the following steps: Step S21: performing geometric feature recognition of automobile parts according to the automobile parts image optimization data to obtain geometric feature data of automobile parts; Step S22: recognizing local details of the automobile parts according to the geometric feature data of the automobile parts and the image optimization data of the automobile parts, and obtaining local detail data of the automobile parts; Step S23: evaluating the structural stability of the automobile parts based on the local detail data of the automobile parts and the geometric feature data of the automobile parts to obtain the structural stability data of the automobile parts; Step S24: performing durability estimation of the automobile part according to the structural stability data of the automobile part and the local detail data of the automobile part to obtain the durability data of the automobile part.

6. The method for intelligently identifying automobile parts based on an image classification model according to claim 5, characterized in that: Step S21 includes the following steps: Step S211: performing automobile parts image segmentation processing on the automobile parts image optimization data to obtain automobile parts image segmentation data; Step S212: performing automobile part outer boundary refinement processing based on the automobile part image segmentation data to obtain automobile part outer boundary refinement data; Step S213: drawing the outer contour of the automobile part according to the outer boundary refinement data of the automobile part to obtain the outer contour data of the automobile part; Step S214: collecting deep-level features inside the automobile parts according to the automobile parts image segmentation data to obtain deep-level feature data inside the automobile parts; Step S215: measuring the internal structure shape of the automobile part based on the internal deep-level feature data of the automobile part to obtain the internal structure shape data of the automobile part; Step S216: performing internal structure spatial layout detection of the automobile accessory based on the internal structure shape data of the automobile accessory to obtain the internal structure spatial layout data of the automobile accessory; Step S217: performing geometric feature recognition of the automobile part according to the external contour data of the automobile part and the internal structure space layout data of the automobile part to obtain geometric feature data of the automobile part.

7. The method for intelligently identifying automobile parts based on an image classification model according to claim 5, characterized in that: Step S23 includes the following steps: Step S231: measuring the thickness of the connection of the automobile parts according to the geometric feature data of the automobile parts to obtain the thickness data of the connection of the automobile parts; Step S232: identifying the supporting structure of the automobile part according to the geometric feature data of the automobile part to obtain the supporting structure data of the automobile part; Step S233: collecting surface texture features of the automobile parts according to the local detail data of the automobile parts to obtain surface texture feature data of the automobile parts; Step S234: collecting stress distribution points of the automobile parts according to the surface texture feature data of the automobile parts to obtain stress distribution point data of the automobile parts; Step S235: performing load simulation on the automobile parts support structure data, the automobile parts connection thickness data and the automobile parts force distribution point data to obtain automobile parts load simulation data; Step S236: Evaluate the structural stability of the automobile part according to the automobile part load simulation data to obtain the structural stability data of the automobile part.

8. The method for intelligently identifying automobile parts based on an image classification model according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: Calculating basic parameters of the automobile parts according to the durability data of the automobile parts and the structural stability data of the automobile parts to obtain basic parameters of the automobile parts; Step S32: performing automobile part image function classification on the automobile part image optimization data according to the basic parameters of the automobile part and the geometric feature data of the automobile part to obtain automobile part image function classification data; Step S33: labeling the automobile parts images for the automobile parts image function classification data to obtain automobile parts image labeling data; Step S34: performing image classification model training on the picture classification model based on the automobile parts image labeling data and the automobile parts image function classification data to obtain an automobile parts image classification model.

9. The method for intelligent identification of automobile parts based on image classification model according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: Acquire the running vehicle fault data; Step S42: Analyze the severity of the fault of the accessories according to the fault data of the running vehicle to obtain the severity of the fault of the accessories; Step S43: identifying suitable accessories for the automobile parts failure severity data according to the automobile parts image classification model to obtain suitable accessories data for the automobile failure; Step S44: Calculating the matching degree of automobile parts according to the data of suitable parts for automobile failure to obtain the matching degree data of suitable parts for automobile; Step S45: Based on the matching degree data of applicable automobile accessories, the automobile accessories image classification model is optimized for intelligent automobile accessories recognition to generate an automobile accessories image classification optimization model.

10. The method for intelligent identification of automobile parts based on image classification model according to claim 9, characterized in that: Step S42 includes the following steps: Step S421: Performing a spatial distribution analysis of automobile fault areas based on the running automobile fault data to obtain automobile parts fault area data; Step S422: Calculating the concentration of the parts failure regions based on the automobile parts failure region data to obtain the parts failure region concentration data; Step S423: calculating the impact degree of automobile parts failure on the parts failure area concentration data and the automobile parts failure area data to obtain automobile parts failure impact degree data; Step S424: performing a weighted evaluation of the severity of automobile parts failures according to the automobile parts failure impact degree data and the parts failure regional concentration data to obtain automobile parts failure severity weighted data; Step S425: weighting the automobile parts failure severity data and the automobile parts failure impact degree data.

Images