Vehicle paint diagnosis method and device, diagnosis equipment and readable storage medium

Abstract

The invention relates to the technical field of vehicle detection, and discloses a vehicle paint diagnosis method and device, diagnosis equipment and a readable storage medium, and the vehicle paint diagnosis method comprises the steps: carrying out the signal collection of a vehicle paint surface, obtaining original optical image information and original light intensity information, carrying out the feature extraction of the original optical image information and the original light intensity information, and carrying out the feature extraction of the original optical image information and the original light intensity information; the method comprises the steps of obtaining optical image feature information and light intensity feature information, inputting the optical image feature information into a convolutional neural network model to obtain vehicle paint surface defect information, comparing the light intensity feature information with a spectrum parameter database, obtaining vehicle paint surface coating thickness distribution information based on a frequency domain inversion algorithm, and obtaining operation information of a target vehicle. According to the vehicle paint diagnosis method, the vehicle paint state can be comprehensively evaluated, the structured vehicle paint diagnosis report is generated, and the accuracy, efficiency and intelligent level of vehicle paint diagnosis are improved.

Description

Technical Field

[0001] The present application relates to the field of vehicle detection technology, and in particular to a vehicle paint diagnosis method, apparatus, diagnostic equipment, and readable storage medium. Background Art

[0002] During vehicle use, the paint, a crucial protective layer on the vehicle body, is susceptible to environmental factors (such as ultraviolet rays, acid rain, and stone impacts) as well as everyday scratches, causing various damages such as fading, scratches, dents, and bubbles. These damages not only affect the vehicle's aesthetics but can also degrade the paint's protective properties, accelerate corrosion of metal components, and shorten the vehicle's service life. Therefore, regular inspection and maintenance of the paint's condition is crucial.

[0003] At present, vehicle paint inspection mainly relies on manual visual observation. Although this traditional method is simple and easy to use, it has obvious limitations. (1) High subjectivity and low precision: Manual observation is greatly affected by personal experience and it is difficult to detect minor defects (for example, shallow bubbles or fine scratches). (2) Unable to accurately judge internal damage: For complex multi-layer vehicle paint structures (for example, primer, color paint, and varnish), manual observation cannot effectively assess the internal damage. (3) Limitations of professional instruments: Although some professional vehicle paint inspection instruments have high precision, they are usually large in size, complex to operate, and expensive, and are not suitable for general car repair shops or daily self-inspections by car owners.

[0004] Therefore, how to achieve automated, high-precision detection of vehicle paint surface and internal damage in different scenarios while reducing equipment costs and simplifying operating procedures has become an important issue that needs to be urgently addressed. Summary of the Invention

[0005] In view of this, the embodiments of the present application provide a vehicle paint diagnosis method, device, diagnostic equipment and readable storage medium, which can effectively solve the problems of low detection accuracy, high equipment cost, complex operation, insufficient intelligent analysis, poor adaptability to unknown vehicle paint and weak data fusion capability in the existing technology.

[0006] In a first aspect, an embodiment of the present application provides a vehicle paint diagnosis method, comprising:

[0007] Perform optical signal acquisition on the paint surface of the target vehicle to obtain original optical image information and original light intensity information;

[0008] performing feature extraction processing on the original optical image information and the original light intensity information respectively to obtain optical image feature information and light intensity feature information;

[0009] Inputting the optical image feature information into a convolutional neural network model to identify defect information on the vehicle paint surface;

[0010] Comparing the light intensity characteristic information with a preset spectral parameter database, and obtaining coating thickness distribution information of the vehicle paint surface based on a frequency domain inversion algorithm;

[0011] Obtaining the operating information of the target vehicle through the vehicle interface;

[0012] The operation information, the defect information and the coating thickness distribution information are correlated and fused to generate a vehicle paint diagnosis report.

[0013] In some embodiments, performing optical signal acquisition on the paint surface of the target vehicle to obtain raw optical image information and raw light intensity information includes:

[0014] Adjust the relative distance between the detection module and the vehicle paint surface to within a preset detection distance range, and measure and record the distance information in real time using a laser ranging module;

[0015] Based on the distance information and a preset detection mode, controlling the light source module to sequentially emit laser signals according to a predetermined wavelength set and angle set to illuminate the paint surface of the target vehicle;

[0016] The image acquisition unit and the light intensity acquisition unit of the detection module are synchronously driven to collect the image data and light intensity data reflected by the vehicle paint surface at each preset wavelength and each preset angle to obtain the original optical image information and the original light intensity information.

[0017] In some embodiments, performing feature extraction processing on the original optical image information and the original light intensity information to obtain optical image feature information and light intensity feature information respectively includes:

[0018] Performing grayscale processing on the original optical image information to obtain a grayscale image;

[0019] Performing multi-scale noise reduction filtering on the grayscale image to obtain a noise-reduced image;

[0020] Performing edge enhancement processing on the noise reduction image to obtain an edge enhanced image;

[0021] Extracting texture features, structural features and edge gradient information from the edge-enhanced image to construct the optical image feature information;

[0022] Performing fast Fourier transform processing on the original light intensity information to obtain a frequency domain amplitude spectrum;

[0023] A frequency band filtering algorithm is performed on the frequency domain amplitude spectrum to extract frequency domain eigenvalues ​​and construct the light intensity characteristic information.

[0024] In some embodiments, inputting the optical image feature information into a convolutional neural network model to identify defect information on the vehicle paint surface includes:

[0025] Inputting the optical image feature information into a pre-trained defect recognition model to extract a feature vector representing the semantic structure of the vehicle paint image;

[0026] Based on the feature vector, the defect types existing on the vehicle paint surface are identified, and the spatial positions, boundary contours, and size parameters of each type of defect are determined.

[0027] In some embodiments, comparing the light intensity characteristic information with a preset spectral parameter database and obtaining the coating thickness distribution information of the vehicle paint surface based on a frequency domain inversion algorithm includes:

[0028] Matching the light intensity characteristic information with a standard reflection spectrum corresponding to the vehicle paint surface in the preset spectral parameter database to extract reference spectrum parameters corresponding to the target detection area;

[0029] Constructing a frequency domain inversion analysis model based on the frequency domain amplitude difference and phase change characteristics between the light intensity characteristic information and the reference spectrum parameters;

[0030] Utilizing the frequency domain inversion analysis model to perform parameter back-propagation processing on the light intensity characteristic information, and calculating the coating thickness distribution information of each detection area;

[0031] The coating thickness distribution information is associated with the corresponding spatial position information to generate the coating thickness distribution information.

[0032] In some embodiments, obtaining the operating information of the target vehicle through the vehicle interface includes:

[0033] Obtaining the vehicle identification code, mileage, and current diagnosis time of the target vehicle through the vehicle interface;

[0034] Based on the vehicle identification code, searching for standard vehicle model parameters corresponding to the vehicle paint surface in a vehicle model information database and obtaining the standard vehicle model parameters; the standard vehicle model parameters include standard vehicle paint structure parameters and a coating reference thickness;

[0035] The vehicle identification code, the mileage, the current diagnosis time, and the standard vehicle model parameters are associated to form the operation information.

[0036] In some embodiments, the correlating and fusing the operating information with the defect information and the coating thickness distribution information to generate a paint diagnosis report includes:

[0037] Mapping and integrating the defect information and the coating thickness distribution information according to the body structure area of ​​the target vehicle to form vehicle paint status data of each detection area;

[0038] Performing field association on the paint status data, the vehicle identification code, the mileage, and the diagnosis time to generate fused diagnosis data;

[0039] Based on preset defect classification rules and thickness deviation assessment rules, each inspection area in the diagnostic data is graded and assessed to obtain diagnostic assessment results and repair suggestions;

[0040] The paint diagnosis report is generated based on the diagnostic evaluation results and the repair suggestions.

[0041] In a second aspect, an embodiment of the present application provides a vehicle paint diagnostic device, comprising:

[0042] A signal acquisition module is used to perform optical signal acquisition on the paint surface of the target vehicle to obtain original optical image information and original light intensity information;

[0043] a data acquisition module, configured to perform feature extraction processing on the original optical image information and the original light intensity information, respectively, to obtain optical image feature information and light intensity feature information;

[0044] a defect information acquisition module, configured to input the optical image feature information into a convolutional neural network model to identify defect information on the vehicle paint surface;

[0045] a coating information processing module, configured to compare the light intensity characteristic information with a preset spectral parameter database and obtain coating thickness distribution information on the vehicle paint surface based on a frequency domain inversion algorithm;

[0046] An operation information acquisition module, used to obtain the operation information of the target vehicle through a vehicle-mounted interface;

[0047] The diagnostic report generation module is used to associate and fuse the operating information with the defect information and the coating thickness distribution information to generate a vehicle paint diagnostic report.

[0048] In a third aspect, an embodiment of the present application provides a diagnostic device, which includes a processor and a memory, wherein the memory stores a computer program, and the processor is used to execute the computer program to implement the vehicle paint diagnostic method of the first aspect mentioned above.

[0049] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, and when the computer program is executed on a processor, the vehicle paint diagnosis method of the first aspect described above is implemented.

[0050] The embodiments of the present application have the following beneficial effects: by collecting optical signals from the paint surface of the target vehicle, the original optical image information and light intensity information are obtained. Subsequently, these original information are subjected to feature extraction processing to obtain optical image feature information and light intensity feature information respectively. Among them, the optical image feature information is input into the convolutional neural network model to achieve accurate identification of the type, spatial position and size of defects on the paint surface; at the same time, the light intensity feature information is compared with the preset spectral parameter database, and the coating thickness is deduced based on the frequency domain inversion algorithm. In addition, the vehicle operation information is read through the on-board interface. Finally, the vehicle operation information, paint defect information and coating thickness distribution information are correlated and fused to generate a structured paint diagnosis report. The present invention can achieve synchronous, non-destructive and high-precision detection of paint defects and coating thickness, and provide personalized repair suggestions and intelligent valuation based on the actual use status of the vehicle, significantly improving the detection efficiency and result reliability, and providing a comprehensive and scientific decision-making basis for automobile repair assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0052] Figure 1 A flow chart of a vehicle paint diagnosis method according to an embodiment of the present application is shown;

[0053] Figure 2 A schematic diagram of a vehicle paint diagnosis report in a vehicle paint diagnosis method according to an embodiment of the present application is shown;

[0054] Figure 3 A structural schematic diagram of a vehicle paint diagnosis method in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0055] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments.

[0056] The components of the embodiments of the present application generally described and illustrated in the drawings herein may be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed application, but rather merely represents selected embodiments of the present application. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative effort are within the scope of protection of the present application.

[0057] Hereinafter, the terms "including", "having" and their cognates used in various embodiments of the present application are intended only to indicate specific features, numbers, steps, operations, elements, components or combinations of the aforementioned items, and should not be understood as excluding the existence of one or more other features, numbers, steps, operations, elements, components or combinations of the aforementioned items or adding the possibility of one or more features, numbers, steps, operations, elements, components or combinations of the aforementioned items. In addition, the terms "first", "second", "third" and the like are only used to distinguish descriptions and should not be understood as indicating or implying relative importance.

[0058] Unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by those skilled in the art to which the various embodiments of the present application belong. The terms (such as those defined in generally used dictionaries) will be interpreted as having the same meaning as in the context of the relevant technical field and will not be interpreted as having an idealized meaning or an overly formal meaning unless clearly defined in the various embodiments of the present application.

[0059] The following describes some embodiments of the present application in detail with reference to the accompanying drawings. In the absence of conflict, the following embodiments and features in the embodiments may be combined with each other.

[0060] Taking into account the problems of existing vehicle paint detection technology, such as strong subjectivity, low precision, insufficient ability to judge complex multi-layer structures, and poor adaptability to unknown vehicle paint types, a vehicle paint diagnosis method is proposed. Through multi-band light source irradiation, high-precision optical signal acquisition, intelligent data analysis and data fusion technology, convenient, efficient and accurate detection of vehicle paint surface and internal damage is achieved.

[0061] The vehicle paint diagnosis method is described below with reference to some specific embodiments.

[0062] Figure 1 A flow chart of a vehicle paint diagnosis method according to an embodiment of the present application is shown. Exemplarily, the vehicle paint diagnosis method includes the following steps:

[0063] Step S100 , performing optical signal acquisition on the paint surface of the target vehicle to obtain original optical image information and original light intensity information.

[0064] The purpose of optical signal acquisition is to obtain raw data that can be used for defect identification and coating thickness analysis. This acquisition process is achieved through two independent channels:

[0065] Reflected image data acquisition: Using a high-resolution image acquisition device (e.g., a CCD camera), the reflected image of the paint surface is captured under specific lighting conditions to generate "optical image information." This information is used to characterize the paint's color, gloss, texture, and the morphology of surface defects (e.g., scratches, bubbles, etc.).

[0066] Light intensity signal acquisition: A photoelectric detector records the light reflection intensity signal, generating "light intensity information." This information reflects the paint coating's absorption and reflection characteristics for light of different wavelengths, providing a basis for subsequent coating thickness estimation and color uniformity analysis.

[0067] For example, the entire optical signal acquisition process is based on a multi-band, multi-angle illumination mechanism, designed to fully stimulate the differences in the paint's response to light from different directions and wavelengths. This approach can capture detailed information such as tiny defects (such as shallow bubbles or minor scratches), interlayer structural anomalies (such as coating thickness deviations), or coating uniformity deviations, laying the foundation for subsequent defect identification and coating thickness analysis.

[0068] In an optional embodiment, step S100 includes:

[0069] Adjust the relative distance between the detection module and the car paint surface to within the preset detection distance range, and use the laser ranging module to measure and record the distance information in real time.

[0070] In practice, the detection module is positioned within a preset detection distance of 10–30 cm from the vehicle's paint surface. A laser ranging sensor with ±1 mm accuracy is integrated into the front of the detection module to monitor the distance to the paint surface in real time. If the measured distance deviates from the preset range (10–30 cm), the motor-driven telescopic bracket automatically adjusts the detection module's position to ensure that its relative distance to the paint surface remains within the target range. The final distance value is also recorded to ensure illumination uniformity and data acquisition stability.

[0071] Based on the distance information and the preset detection mode, the light source module is controlled to emit laser signals in sequence according to a predetermined wavelength set and angle set to illuminate the paint surface of the target vehicle.

[0072] In specific implementation, after completing the distance calibration, according to the detection mode (for example, quick inspection, fine detection, custom detection), the light source module is controlled to emit laser signals in sequence according to a predetermined wavelength set (for example, 450nm blue light, 550nm green light, 650nm red light, 365nm ultraviolet light, etc.) and a predetermined set of incident angles (for example, 30°, 45°, 60°) to illuminate the vehicle paint surface.

[0073] In rapid inspection mode, the light source module illuminates the paint surface with white light at a 45° angle of incidence. A motorized, rotatable bracket allows the light source to cover a larger area of ​​the vehicle body, quickly capturing the overall paint gloss and obvious surface defects such as large areas of fading and deep scratches. At this point, the image acquisition unit (CCD camera) continuously captures reflected images of the paint surface at a high frame rate (up to 10 fps), generating a series of reflected image data for rapid assessment of the paint's overall condition.

[0074] In fine detection mode, the light source switches from multiple preset angles in sequence. For example, it first illuminates at 30°, collects a set of data, and then switches to 60° and 90°. Each switch pauses for 2-3 seconds to allow the optical acquisition module to collect data stably. At the same time, the diagnostic equipment dynamically adjusts the light source parameters for different car paint materials and colors: (1) Wavelength selection: Select the appropriate light source wavelength according to the car paint type. For example, when detecting metallic paint, the ultraviolet band (such as 365nm) is preferentially enabled to assist in detecting the distribution of pigment particles and enhance the ability to detect subtle defects. (2) Light source intensity adjustment: Dynamically adjust the light source intensity according to the depth of the car paint color. For dark car paint, appropriately increase the light intensity by 20%-30% to ensure that the reflected light signal is strong enough for easy collection and analysis. In specific implementation, after completing the distance calibration, according to the detection mode (for example, quick inspection, fine detection, custom detection), the light source module is controlled to emit laser signals in sequence according to a predetermined wavelength set (for example, 450nm blue light, 550nm green light, 650nm red light, 365nm ultraviolet light, etc.) and a predetermined set of incident angles (for example, 30°, 45°, 60°) to illuminate the vehicle paint surface.

[0075] Specifically, a laser ranging sensor first monitors and records the relative distance between the detection module and the vehicle paint surface in real time. If the measured distance exceeds a preset range (e.g., less than 10 cm or greater than 30 cm), the main control unit automatically adjusts the position of the detection module via a motor-driven telescopic bracket to ensure it remains within the target range (e.g., 10–30 cm) from the vehicle paint surface. Once this step is complete, the light source module is controlled to emit laser signals according to the preset detection mode.

[0076] In the rapid inspection mode, the light source module illuminates the paint surface with a single wavelength (such as white light, covering multiple visible light bands) and a fixed incident angle (such as 45 degrees). By electrically controlling the rotatable bracket, the light source covers a larger area of ​​the vehicle body, thereby quickly capturing the overall gloss of the paint and obvious surface defects (such as large-scale fading, deep scratches, etc.). At this time, the image acquisition unit (such as a CCD camera) continuously captures the reflected image of the paint surface at a high frame rate (up to 10fps or more), generating a series of reflected image data for rapid assessment of the overall condition of the paint.

[0077] In fine detection mode, the light source module switches from multiple preset angles (such as 30°, 45°, and 60°) in sequence, pausing for 2-3 seconds each time to ensure that the optical acquisition module can stably collect data. At the same time, the diagnostic equipment dynamically adjusts the light source parameters for different car paint materials and colors:

[0078] It is worth noting that the appropriate light source wavelength should be selected according to the type of paint. For example, when inspecting metallic paint, it is preferred to use the ultraviolet band (such as 365nm) to assist in the detection of pigment particle distribution and enhance the ability to detect subtle defects.

[0079] Dynamically adjust the light intensity based on the paint color. For dark paint, increase the light intensity appropriately (e.g., by 20%-30%) to ensure the reflected light signal is strong enough for subsequent collection and analysis.

[0080] Through the above method, the system can emit laser signals in sequence according to a predetermined wavelength set and angle set in different detection modes, thereby achieving comprehensive illumination and data collection of the vehicle paint surface.

[0081] The image acquisition unit and the light intensity acquisition unit of the synchronous drive detection module collect image data and light intensity data reflected from the vehicle paint surface at each preset wavelength and each preset angle to obtain original optical image information and original light intensity information.

[0082] In specific implementation, each time the light source emits a laser signal, the main control unit activates the image acquisition unit (CCD camera) and the light intensity acquisition unit (photodetector) through a synchronous trigger mechanism to ensure that the image data and light intensity data at each preset wavelength and each preset angle can be accurately matched.

[0083] In rapid inspection mode, the high-resolution CCD camera in the optical acquisition module continuously captures images of reflected light from the paint surface at a rate of over 10 frames per second, lasting approximately 5 seconds to capture sufficient sample images. This high-speed continuous shooting method efficiently covers a large area of ​​the vehicle body, capturing the overall gloss of the paint and obvious surface defects such as large areas of fading and deep scratches.

[0084] In fine inspection mode, the capture time is extended to 30 seconds, intensively sampling key areas. By continuously capturing reflection images of the paint surface, a series of reflection image data is generated to capture tiny defects and their details.

[0085] Simultaneously, the light intensity acquisition unit (photodetector) records changes in reflected light intensity in real time at a sampling frequency of 1kHz, acquiring three-dimensional time-angle-wavelength intensity data related to the current wavelength and incident angle. This data reflects the paint coating's absorption and reflection characteristics for light of different wavelengths, providing a basis for subsequent coating thickness estimation and color uniformity analysis.

[0086] Through a synchronized acquisition mechanism, the system ensures accurate matching of image data and light intensity data at each preset wavelength and angle. This matching ensures a high degree of consistency and integrity in the collected data, thereby obtaining complete raw optical image and light intensity information, laying the foundation for subsequent feature extraction and analysis.

[0087] Step S200 , performing feature extraction processing on the original optical image information and the original light intensity information respectively to obtain optical image feature information and light intensity feature information.

[0088] In this embodiment, in order to achieve joint diagnosis of vehicle paint surface defect identification and coating thickness analysis, it is necessary to perform feature extraction processing on the original optical image information and the original light intensity information respectively, so as to generate optical image feature information and light intensity feature information that can be used for model input or physical analysis.

[0089] Among them, optical image feature information is mainly used for the spatial positioning and classification identification of defects. It extracts important indicators such as texture, structure, and boundaries that reflect local changes in the image through classic image preprocessing methods (for example, grayscale, noise reduction, edge enhancement, etc.).

[0090] Light intensity characteristic information: reflects the reflection intensity distribution pattern of the car paint surface to multi-band lasers. Through fast Fourier transform (FFT) and frequency band filtering technology, the data is compressed in the spectral domain and the main features are extracted to obtain the frequency domain characteristics related to the coating thickness and uniformity.

[0091] Through the above feature extraction process, the initially collected raw data can be converted into structured high-quality feature representations, providing support for subsequent model recognition (e.g., convolutional neural network CNN) and numerical analysis (e.g., coating thickness inversion).

[0092] In an optional embodiment, step S200 includes:

[0093] Grayscale processing is performed on the original optical image information to obtain a grayscale image.

[0094] In specific implementation, grayscale processing is performed on the original optical image information, that is, the three-channel RGB image is converted into a single-channel grayscale image. This process usually adopts the weighted average method or the brightness component extraction method to enhance the response to local brightness differences by reducing the image dimension, thereby improving the contrast of the defect edge area. For example, a weighted average grayscale conversion algorithm is used to weight the RGB three channels according to the human eye's perception sensitivity (generally 0.3R+0.59G+0.11B) to generate a single-channel grayscale image. This processing not only reduces the input dimension of the subsequent convolution model, but also improves the clarity of the expression of key morphological features such as boundaries and textures in the image.

[0095] Perform multi-scale denoising filtering on the grayscale image to obtain a denoised image.

[0096] In practice, a multi-scale noise reduction algorithm is applied to the grayscale image for smoothing, eliminating background noise and photoreceptor interference generated during image acquisition. Noise reduction methods can include Gaussian filtering, median filtering, or bilateral filtering, performed in parallel at multiple window sizes. This removes noise while preserving edges and details at different scales.

[0097] For example, a Gaussian filter is applied to the entire image to smooth background noise, and a median filter is applied to local, small-scale regions to suppress sharp isolated points and measurement interference. In this way, a denoised image with a uniform background and stable edges is ultimately generated.

[0098] Perform edge enhancement processing on the denoised image to obtain an edge enhanced image.

[0099] For example, edge enhancement is performed on the denoised image to emphasize the outlines and positional variations of possible defects. This process can utilize edge detection algorithms such as the Sobel operator, Laplacian operator, or Canny operator. By calculating first- or second-order derivative responses, it highlights areas of boundary variation, thereby improving defect recognizability.

[0100] For example, the Sobel operator is used to perform gradient detection in the horizontal and vertical directions respectively, and the gradient amplitude image is combined with non-maximum suppression and double threshold processing to effectively highlight local change areas such as tiny scratches on the paint surface, corner depressions, etc., and finally generate an enhanced image with clear edge structure.

[0101] Texture features, structural features and edge gradient information are extracted from edge-enhanced images to construct optical image feature information.

[0102] In specific implementation, based on the edge enhanced image, a joint extraction operation of texture, structure and edge gradient is performed.

[0103] Texture features are obtained through local binary patterns (LBP) or Gabor filters to describe the micro-texture characteristics of the car paint surface;

[0104] The structural features are calculated through the gray-level co-occurrence matrix (GLCM) to calculate the contrast, entropy and energy index between blocks, reflecting the overall structural changes of the car paint surface;

[0105] The edge gradient information is obtained through the histogram statistics of pixel gradient direction and amplitude, highlighting the boundary change characteristics of the defect area.

[0106] Finally, the extracted texture features, structural features and edge gradient information are spliced ​​together to form complete optical image feature information, which serves as input data for the convolutional neural network (CNN) defect recognition model.

[0107] The original light intensity information is processed by fast Fourier transform to obtain the frequency domain amplitude spectrum.

[0108] For example, a Fast Fourier Transform (FFT) is performed on the raw light intensity information, converting it from a time-domain signal to a frequency-domain signal, thereby obtaining a frequency-domain amplitude spectrum. This conversion process can reveal the periodicity or amplitude variation patterns implicit in the reflected light intensity, and is particularly suitable for extracting interference fringes or frequency-domain response shifts caused by coating thickness variations.

[0109] The frequency domain amplitude spectrum is subjected to frequency band filtering algorithm to extract frequency domain eigenvalues ​​and construct light intensity characteristic information.

[0110] In specific implementations, bandpass filtering or envelope analysis is performed on the frequency-domain amplitude spectrum obtained through a fast Fourier transform (FFT) across multiple frequency ranges to extract stable and physically meaningful frequency eigenvalues. These eigenvalues, including metrics such as the main peak frequency, spectral width, and frequency offset, quantify the energy concentration of the reflected light, thereby reflecting differences in the thickness of the paint coating or material properties. Ultimately, the extracted frequency-domain eigenvalues ​​are integrated to form complete light intensity characteristic information.

[0111] In step S300, the optical image feature information is input into a convolutional neural network model to identify defect information on the vehicle paint surface.

[0112] The Convolutional Neural Network (CNN) is a typical deep learning architecture with powerful image semantic understanding capabilities. It detects complex texture changes and local structural anomalies through the following core operations: convolutional layers extract local spatial features; pooling layers achieve dimensionality reduction and compression; and fully connected layers perform feature fusion and classification prediction. Compared to traditional defect detection methods based on edge operators or manual rules, CNNs can autonomously learn discriminative features from data, maintaining high robustness under varying lighting conditions, paint colors, and camera angles.

[0113] In practice, to automatically identify common paint surface defects (such as scratches, dents, blisters, and peeling), the pre-processed optical image feature information is fed into a pre-set CNN model to perform defect recognition. The CNN model analyzes the input image features, extracts spatial texture distribution information containing semantic structure, and outputs the corresponding defect type label along with structured parameters such as location, boundary, and area in the image coordinate system.

[0114] To ensure the CNN model is adaptable to a variety of real-world inspection scenarios, training data is collected from a variety of sources, including the following: 1. Real-world paint sample data: Paint samples from a variety of mainstream car models on the market are collected, covering different colors (e.g., black, white, silver, red, etc.) and types (solid paint, metallic paint, pearlescent paint, etc.). For each car model and paint type, data is collected under different lighting conditions (e.g., strong light, low light, side light, etc.) and angles to simulate the complex conditions encountered in real-world inspections. In addition, paint samples with various defects are collected or fabricated, including scratches (of varying lengths, widths, and depths), pits (of varying diameters and depths), bubbles (of varying sizes and distributions), and particles (of varying sizes and densities). These samples can be sourced from actual car repair processes or simulated manufacturing in a laboratory environment. 2. Simulated data: Utilizing computer graphics techniques, a large number of simulated paint images are generated. By adjusting parameters such as lighting and material properties, diverse simulated data is generated to compensate for the lack of real-world data, particularly for rare defect types or difficult-to-obtain scenarios. 3. Data augmentation: Perform data augmentation operations on the actual collected car paint images, such as rotation, flipping, scaling, brightness adjustment, and contrast adjustment, to further increase the diversity of training data and improve the generalization ability of the model.

[0115] To ensure the quality of the training data, manual annotation is used to accurately annotate defects in car paint images. In the early stages of model training, due to the lack of sufficient automated annotation tools and prior knowledge, professional annotators use image annotation tools (e.g., LabelImg, RectLabel, etc.) to annotate defects in car paint images. The annotation content includes the type of defect (e.g., scratches, pits, bubbles, etc.), location (annotated by rectangular or polygonal boxes), and size (e.g., length, width, diameter, etc.). To ensure the accuracy and consistency of the annotation, another professional annotator or domain expert reviews and corrects the annotation results to ensure that the annotation quality meets the requirements.

[0116] In the field of industrial visual inspection, CNN models for tasks like surface defect detection typically achieve an accuracy of around 90%-95% on a test set. Given the complexity and diversity of automotive paint defect detection, as well as the high accuracy requirements in practical applications, this CNN model is expected to achieve a mid-to-high accuracy on the test set, around 92%-95%.

[0117] In an optional embodiment, step S300 includes:

[0118] The optical image feature information is input into a pre-trained defect recognition model to extract feature vectors representing the semantic structure of the car paint image.

[0119] Among them, the defect recognition model is built based on the convolutional neural network (CNN) architecture, which includes multiple convolutional layers, activation layers, pooling layers, and fully connected layers. The model is trained with a large number of labeled real car paint samples, covering different car series, paint colors, and defect types (for example, scratches, dents, bubbles, etc.), so it has strong generalization capabilities. For example, in the model processing process: the first few convolutional layers extract low-level features such as local edge changes and texture directionality; the deep network aggregates the structural information in the feature map and extracts high-order semantic features; and finally generates a complete feature vector representation, which can effectively distinguish different types of car paint defects. In this way, the model can extract semantically meaningful feature vectors from the input optical image feature information, providing a basis for subsequent defect classification and positioning.

[0120] Based on the feature vector, the defect types on the paint surface are identified, and the spatial location, boundary contours and size parameters of each type of defect are determined.

[0121] Exemplarily, based on the extracted feature vectors, the model performs multi-class defect recognition and spatial position regression processing, outputting the type of paint surface defect and its spatial location, boundary contour, and size parameters. In this embodiment, the output of the convolutional neural network (CNN) model is divided into two branches: Classification branch: Outputs defect type labels (e.g., scratches, dents, bubbles, peeling, etc.) through a Softmax structure to achieve accurate identification of defect types. Positioning branch: Uses a regression structure to estimate the center position and boundary contour (e.g., a rectangular box or polygonal boundary) of the defect in the image coordinate system, and calculates the geometric size parameters of the defect (e.g., length, width, or diameter).

[0122] For example, for scratches, the model can accurately extract the scratch's starting and ending points, calculating its length and width with an accuracy of up to 0.1mm. For bubble defects, the model identifies their circular boundaries, measures their diameter and area, and, based on light reflection characteristics, further infers their depth (e.g., within the varnish layer or the paint layer), with an error of no more than ±0.2mm.

[0123] The final output is structured defect information data containing the following elements: defect type, location information (e.g., bounding box coordinates), geometric dimensions (e.g., length, width, or diameter), and boundary shape. This structured defect information data is combined with the coating thickness estimation results to construct a paint condition diagnosis model.

[0124] In step S400 , the light intensity characteristic information is compared with a preset spectral parameter database, and the coating thickness distribution information of the vehicle paint surface is obtained based on a frequency domain inversion algorithm.

[0125] In specific implementation, in order to accurately evaluate the coating thickness of the vehicle paint surface, based on the extracted light intensity characteristic information and combined with the preset paint spectral parameter database, a comparative analysis is performed and the frequency domain inversion algorithm is applied to estimate the thickness.

[0126] Among them, the light intensity characteristic information is the reflected signal collected by a high-sensitivity photoelectric detector under multi-angle, variable wavelength laser irradiation, and the spectral characteristics formed after fast Fourier transform (FFT) processing, which is used to characterize the response characteristics of car paint to different frequency light.

[0127] The frequency domain inversion algorithm is a mathematical derivation process based on frequency domain amplitude and phase differences. It can combine the optical properties of the coating material (for example, refractive index, extinction coefficient) with the standard reflection curve in the spectral database to infer the thickness values ​​of each layer on the car paint surface.

[0128] Compared to traditional time-domain signal fitting methods, the frequency-domain inversion algorithm offers the following advantages: high computational efficiency, reducing the time cost of complex signal fitting; strong anti-interference capabilities, providing enhanced robustness against noise and environmental interference; and strong adaptability to multi-layer paint structures, effectively addressing the thickness estimation needs of multi-layer structures such as primer, basecoat, and clearcoat. Ultimately, coating thickness estimation errors can be achieved within ±5μm, ensuring high accuracy and reliability.

[0129] In an optional embodiment, step S400 includes:

[0130] In step S410 , the light intensity characteristic information is matched with a standard reflection spectrum corresponding to the vehicle paint surface in a preset spectral parameter database, and reference spectrum parameters corresponding to the target detection area are extracted.

[0131] The spectral parameter database is constructed by integrating data from multiple sources, covering the optical properties of paints for mainstream vehicle models with high accuracy and adaptability. The data sources include laboratory measurement data, data from automotive industry collaborations, and publicly available standard data, accounting for 60%, 30%, and 10%, respectively. Laboratory measurement data is the core component of the database. Various types of standard paint samples (e.g., solid color, metallic, and pearlescent) are prepared on standard metal substrates using industrial-grade spray equipment. The thickness of each layer is precisely controlled in 5μm increments within a 20–200μm range. The reflectance (R), transmittance (T), and absorptivity (A) are measured over the wavelength range of 200–1000nm using a UV-Vis-NIR spectrometer (e.g., PerkinElmer Lambda 1050). The spectral response characteristics are also recorded at different incident angles (20°, 45°, and 60°) and polarization states (s-polarization / p-polarization). Combining the Fresnel formula with multilayer film interference theory, the refractive index (n), extinction coefficient (k), and thickness (d) of the sample are inferred, and a parameter mapping relationship between wavelength, angle, and optical constants is constructed, providing high-precision model support for subsequent spectral matching.

[0132] Furthermore, data from automaker collaborations further expands the database's scope. By establishing data-sharing mechanisms with major OEMs such as Toyota, Volkswagen, and Mercedes-Benz, the system has acquired original paint formulation information (e.g., resin base, pigment type, and metal particle distribution), coating structure parameters (primer / basecoat / clearcoat thickness), and reflectance spectra under standard processes for multiple vehicle models. This data covers over 500 mainstream vehicle models launched between 2020 and 2025, providing reliable and stable optical constants and structural parameters for specialized finishes such as matte and gradient paint.

[0133] To complement reference data for specific operating conditions, the database also integrates publicly available standards, including automotive paint optical performance standards and industry-standard reflectance models published by organizations such as ASTM and ISO. This data provides a universal reference for the model and is also a useful supplement when actual measurements or manufacturer data are unavailable for specific paint types.

[0134] Based on the spectral parameter database constructed by the aforementioned multi-source data fusion, the actual matching process combines the vehicle identification information, incident light wavelength, and angle conditions of the target detection area to retrieve the most closely matching standard reflectance spectrum from the database. This spectrum serves as the input reference for the subsequent frequency domain inversion model. Guided by this standard spectrum, the system achieves high-precision fitting of the actual optical behavior of the paint surface and thickness estimation, providing a solid data foundation for non-contact thickness analysis across multiple regions of the entire vehicle.

[0135] Step S420 : constructing a frequency domain inversion analysis model based on the frequency domain amplitude difference and phase change characteristics between the light intensity characteristic information and the reference spectrum parameters.

[0136] In specific implementation, after extracting the reference spectrum, a frequency-domain inversion analysis model is constructed based on the frequency-domain amplitude differences and phase variations between the light intensity characteristics and the reference spectrum parameters. A complex frequency-domain matching algorithm is used to calculate the amplitude difference (ΔA) and phase difference (Δφ) in the dominant frequency region between the measured light intensity spectrum and the standard reference spectrum. Combining the Fresnel equations with a multilayer film interference model, an inversion parameter expression is derived, and a "frequency-phase-thickness" mapping relationship is established, ultimately forming a frequency-domain inversion function model suitable for the target vehicle model.

[0137] Step S430 , performing parameter back-propagation processing on the light intensity characteristic information using the frequency domain inversion analysis model, and calculating the coating thickness distribution information of each detection area.

[0138] In specific implementation, a digital signal processing module (e.g., DSP) calls a fast Fourier transform (FFT) algorithm to convert the time-domain light intensity signal into a frequency-domain signal and extract characteristic indicators such as the main peak frequency, energy spectrum distribution, and phase difference. These characteristics are then input into a pre-built frequency-domain inversion analysis model. Through parameter back-propagation, the total coating thickness or layer thickness (e.g., varnish layer, color paint layer, primer layer) of each detection area is calculated. Ultimately, the model generates an estimate of the coating thickness with an error range of less than ±5μm.

[0139] The coating thickness distribution information is associated with the corresponding spatial position information to generate coating thickness distribution information.

[0140] In practice, the calculated coating thickness distribution information is field-bound with the spatial location information acquired during the acquisition phase to generate structured coating thickness distribution information. Each data unit contains fields such as "detection coordinate point," "thickness value," "light source wavelength," and "illumination angle," and is stored by vehicle body structural region, thus forming a complete coating thickness distribution profile.

[0141] Step S500: Acquire the target vehicle's operating information through the vehicle interface.

[0142] In practice, to combine paint inspection results with vehicle identity and operating status, the system acquires the target vehicle's operating information through an onboard interface. This information includes basic vehicle status parameters and vehicle attributes, such as the vehicle identification number (VIN), mileage, and current inspection time.

[0143] In an optional embodiment, step S500 includes:

[0144] The vehicle identification number, mileage and current diagnosis time of the target vehicle are obtained through the vehicle interface.

[0145] In specific implementation, a communication connection is established with the target vehicle through the OBD interface, and the vehicle identification number (VIN code), mileage, and current diagnosis time are collected in real time. The vehicle identification number (VIN code) is a unique vehicle identification code that is not only used to retrieve the corresponding paint structure and thickness benchmark in the vehicle model information database, but also obtains the frame number as a unique identifier through the OBD interface. At the same time, the inspection time is recorded (accurate to the second). The paint inspection data (such as defect location, coating thickness, etc.) and OBD data (such as mileage, engine model, and production date) are mapped one to one through "frame number + inspection time" to ensure the data is from the same source.

[0146] Mileage reflects vehicle usage intensity and is highly correlated with factors such as paint aging and thickness deviation. Furthermore, key parameters are extracted from OBD data, including but not limited to mileage, engine operating conditions (idle / high speed), maintenance records (read via the ECU), and accident repair history (if the vehicle is equipped with the relevant module). Maintenance records are particularly closely linked to the degree of paint aging. For example, a vehicle that has not been waxed for a long time may experience a 10% reduction in the clearcoat thickness. Diagnostic time is used to establish a data synchronization mechanism to ensure accuracy and data consistency at the time of detection, providing reliable timestamp support for subsequent analysis.

[0147] Based on the vehicle identification code, standard vehicle model parameters corresponding to the vehicle paint surface are retrieved from the vehicle model information database, and the standard vehicle model parameters are obtained.

[0148] In specific implementation, based on the collected vehicle identification number (VIN code), the corresponding vehicle model information is retrieved from the preset vehicle model information database, and the standard vehicle paint structure parameters and coating reference thickness of the vehicle model are extracted for subsequent deviation judgment. For example, for a 2024 vehicle model, after matching the VIN code, its standard varnish thickness (for example, 30μm), base paint thickness (20μm), and three-layer coating structure configuration can be retrieved, and supporting parameters such as the reflectivity and spectral absorption characteristics of the vehicle model under standard lighting conditions are provided.

[0149] The vehicle identification code, mileage, current diagnosis time and standard vehicle model parameters are associated to form operation information.

[0150] In specific implementation, the vehicle identification number, mileage, and current diagnosis time are field-linked with standard vehicle model parameters to construct a unified operational information dataset. This operational information dataset is used to assist in analyzing the causes of coating deviations, defect formation mechanisms, or abnormal conditions. For example, if the mileage exceeds 100,000 kilometers and the clearcoat thickness decreases by more than 30%, it can be inferred that the vehicle is experiencing natural wear and tear due to paint aging. If the basecoat thickness far exceeds the original manufacturer's standard and there is no corresponding mileage accumulation, it may be the result of repainting or accident repairs.

[0151] Ultimately, this operational information dataset will be combined with defect identification information and thickness analysis results for report generation, enabling personalized output of diagnostic conclusions.

[0152] Step S600: Correlate and fuse the operation information, defect information, and coating thickness distribution information to generate a paint diagnosis report.

[0153] In specific implementation, in order to achieve structured expression and comprehensive evaluation of vehicle paint inspection results, the operation information, defect information and coating thickness distribution information are correlated and fused to generate a vehicle paint diagnosis report, such as Figure 2 This report not only includes the types and distribution of paint defects in various areas of the vehicle, but also comprehensively analyzes coating thickness deviations and vehicle operating conditions (e.g., mileage, age, etc.). It then outputs repair recommendations, vehicle condition assessments, and potential valuation suggestions, providing a scientific basis for subsequent repair decisions or used vehicle transactions.

[0154] In an optional embodiment, step S600 includes:

[0155] The defect information and coating thickness distribution information are mapped and integrated according to the body structure area of ​​the target vehicle to form the paint status data of each inspection area.

[0156] In practice, defect information and coating thickness distribution information are mapped to a pre-set 3D vehicle model based on the vehicle body structure. For example, in a 3D vehicle model, the paint surface is divided into multiple regions, such as the roof, body, and doors. Defect information (e.g., type, size, location, etc.) and coating thickness distribution information (e.g., thickness values ​​of the clearcoat and basecoat layers) within each region are collected and matched independently.

[0157] Through this mapping integration, the "paint status data" of each inspection area is formed, which is a collection of coating thickness and defect information of each area, providing basic data support for the subsequent generation of paint diagnostic reports.

[0158] The paint status data is field-associated with the vehicle identification number, mileage, and diagnosis time to generate fused diagnostic data.

[0159] In specific implementation, the corresponding standard vehicle model parameters (for example, standard paint structure configuration, original coating thickness range, etc.) are retrieved from the preset vehicle model information database through the vehicle identification code to provide a benchmark reference for the paint condition; the mileage is used to evaluate the degree of aging of the paint or the possibility of abnormal wear. For example, the mileage is combined to determine whether the coating thickness deviation is caused by natural aging or whether there are traces of repainting; the diagnosis time is used as a timestamp to ensure the time consistency of each test data, which is convenient for historical data comparison and trend analysis. Through the above-mentioned field association process, the fused diagnostic data is generated. This dataset not only contains the paint status information of each detection area, but also integrates the basic attributes of the vehicle (for example, vehicle identification code) and usage status (for example, mileage, diagnosis time), thereby providing comprehensive and accurate data support for the subsequent generation of paint diagnosis reports, further improving the scientificity and practicality of the diagnosis results.

[0160] Based on the preset defect classification rules and thickness deviation assessment rules, each inspection area in the diagnostic data is graded and evaluated to obtain diagnostic assessment results and repair suggestions.

[0161] Among them, the defect grading rules classify defects into different levels, such as minor (L1), moderate (L2), and severe (L3), based on the severity and location of the defect and its impact on the paint function (for example, scratch depth, bubble size, etc.). The specific grading logic is as follows: If the defect type is "bubble" and its diameter is greater than 1mm, and the depth exceeds 30% of the thickness of the varnish layer, it is judged as a "severe" defect (L3), and sheet metal repair is usually required; if the defect type is "scratch" and its length exceeds 5cm, and the primer is visible, it is judged as a "moderate" defect (L2), and local touch-up is recommended; for other defects that do not meet the above conditions, they are judged as "minor" defects (L1) and can be resolved through polishing.

[0162] The thickness deviation assessment rule evaluates the deviation range of the coating by comparing the thickness of each coating with the standard value. The specific logic is as follows: First, calculate the deviation rate, and the formula is (measured coating thickness-standard value) / standard value. If the deviation rate of the clear coat layer is less than -20%, it is determined that the paint in this area is severely aged, and it is recommended to perform a full-car crystal coating treatment to improve the protection effect; if the deviation rate of the color paint layer is greater than +15%, it may be caused by later touch-up paint (non-original paint thickness is usually thicker). At this time, the area is marked as "suspected accident repair", prompting users to pay attention to potential repair history. If the deviation of other coatings exceeds ±15%, it will also be marked as a suspicious touch-up area or indicate the risk of structural aging, providing a basis for subsequent repair decisions.

[0163] Generate a paint diagnostic report based on the diagnostic assessment results and repair recommendations.

[0164] In specific implementation, a structured paint diagnostic report is generated based on defect assessment results, coating thickness deviation data, and repair suggestions. The report contains the following main contents:

[0165] Defect Type: Clearly lists the type of paint defect detected (e.g., scratches, dents, bubbles, etc.).

[0166] Defect Severity: Each defect is assigned a severity level (minor, moderate, severe, etc.) based on defect classification rules. Furthermore, incorporating the multi-dimensional weighted calculation method used in the used car valuation model, the defect severity score is incorporated into the assessment of the paint impact factor, with a weight of 0.6. For example, if there are no defects and the coating thickness meets the requirements, the paint condition can contribute 5% to the residual value; however, if there are severe defects, the residual value may decrease by 15%, especially if the mileage exceeds 100,000 kilometers.

[0167] Coating Thickness Deviation Analysis: This tool records the deviation of coating thickness from the standard value in each test area to help determine coating degradation or repair needs. The coating thickness deviation rate is weighted 0.3 within the paint impact coefficient. A clearcoat deviation rate of less than -20% indicates severe degradation; a basecoat deviation rate greater than +15% indicates the presence of touch-ups or accident repairs, both of which can directly impact vehicle valuation.

[0168] Repair recommendations: Based on the defect type and coating thickness assessment, provide specific repair solutions (e.g., touch-up, sanding, replacement, etc.).

[0169] Vehicle Information: This system associates and displays relevant vehicle information, including vehicle identification number, mileage, diagnostic time, and a mapping of paint condition and operating information. Combining mileage from OBD data with the average residual value of similar models in the market (sourced from a cloud database), comprehensive analysis generates a valuation recommendation with an accuracy of ±5%.

[0170] This paint diagnostic report not only provides users with a detailed analysis of the vehicle's paint condition, serving as a crucial basis for vehicle management and maintenance decisions, but also provides accurate paint maintenance guidance for subsequent vehicle use through a scientific used vehicle valuation reference. Furthermore, a multi-dimensional weighted valuation model further enhances the report's value and credibility in practical applications.

[0171] In an optional embodiment, before performing optical signal acquisition on the target vehicle's paint surface and obtaining raw optical image information and raw light intensity information, the following preparatory steps need to be completed to ensure the normal operation of the equipment and the accuracy of the test results:

[0172] 1. Invoke the initialization program of the diagnostic device and detection module to check for hardware damage or abnormalities. If any physical damage or functional abnormality is detected, the user is prompted to replace or repair the device.

[0173] 2. After powering on, the device enters self-test mode, automatically invoking a self-test routine to verify the proper functioning of each functional module (e.g., light source module, photodetector, image acquisition unit, etc.). Simultaneously, the battery voltage is collected to check the charge status. The specific criteria are: a fully charged voltage of 4.2V corresponds to 100% charge, while a voltage below 2.75V is considered depleted (corresponding to 1%). The system will only allow subsequent operations to proceed if the battery voltage is above 2.75V.

[0174] 3. Start the Wi-Fi module, scan and connect to the unique Wi-Fi hotspot bound to the detection module (each detection module has a unique SN serial number as its identifier). By matching the SN serial number with the pre-stored binding record, the correct hotspot is automatically selected for connection, ensuring a successful communication channel between the diagnostic device and the detection module.

[0175] 4. After the connection is successful, start the light source calibration procedure of the detection module, including wavelength calibration and intensity calibration:

[0176] Wavelength calibration: The system controls the detection module to align with a standard whiteboard and sequentially emit single-wavelength light sources (e.g., 450nm blue light, 550nm green light, 650nm red light, and 365nm ultraviolet light). A photodetector collects the reflected light intensity data and compares it with the theoretical value to calculate the wavelength offset Δλ. The system then sends instructions through the MCU to adjust the light source driver circuit (e.g., LED driver chip) and uses a stepper motor to control the wavelength adjustment knob until Δλ ≤ 1nm.

[0177] Intensity calibration: Fix the wavelength of the light source (for example, 550nm white light), adjust the light source intensity to the maximum nominal value (for example, 1000lux), and illuminate a standard whiteboard. The photodetector collects the actual light intensity value, Imeasured, and compares it with the standard value, Istandard (calibrated by an illuminance meter). The intensity error ε is calculated as (Imeasured - Istandard) / Istandard. If the error ε is greater than 5%, the system adjusts the LED drive current through the PWM signal to converge the error to within ±5%. At the same time, the system records the calibration coefficients for different intensity levels (from 10% to 100%, in steps of 10%) to facilitate subsequent precise control of the light source output.

[0178] The above preparatory steps ensure the normal operation of the equipment functions and the reliability of data acquisition, providing the necessary guarantee for subsequent optical signal acquisition.

[0179] Figure 3 A schematic diagram of the structure of a vehicle paint diagnostic device according to an embodiment of the present application is shown. Exemplarily, the vehicle paint diagnostic device 100 includes:

[0180] The signal acquisition module 110 is used to perform optical signal acquisition on the paint surface of the target vehicle to obtain original optical image information and original light intensity information;

[0181] The data acquisition module 120 is used to perform feature extraction processing on the original optical image information and the original light intensity information to obtain optical image feature information and light intensity feature information;

[0182] The defect information acquisition module 130 is used to input the optical image feature information into a convolutional neural network model to identify and obtain defect information on the vehicle paint surface;

[0183] The coating information processing module 140 is used to compare the light intensity characteristic information with a preset spectral parameter database and obtain the coating thickness distribution information of the vehicle paint surface based on a frequency domain inversion algorithm;

[0184] An operation information acquisition module 150 is used to acquire the operation information of the target vehicle through a vehicle-mounted interface;

[0185] The diagnostic report generating module 160 is used to associate and fuse the operating information with the defect information and the coating thickness distribution information to generate a vehicle paint diagnostic report.

[0186] It can be understood that the apparatus of this embodiment corresponds to the method of the above embodiment, and the options in the above embodiment are also applicable to this embodiment, so they will not be described again here.

[0187] The present application also provides a diagnostic device. Exemplarily, the diagnostic device includes a processor and a memory, wherein the memory stores a computer program, and the processor runs the computer program to enable the diagnostic device to execute the functions of each module in the above method or the above apparatus.

[0188] Among them, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, including a central processing unit (CPU), a graphics processing unit (GPU) and a network processor (NP), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or at least one of other programmable logic devices, discrete gate or transistor logic devices, and discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc., which can implement or execute the disclosed methods, steps and logic block diagrams in the embodiments of the present application.

[0189] The memory may be, but is not limited to, a random access memory (RAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), etc. The memory is used to store a computer program, and the processor may execute the computer program accordingly after receiving an execution instruction.

[0190] The present application also provides a computer-readable storage medium for storing the computer program used in the above-mentioned diagnostic device. For example, the computer-readable storage medium may include, but is not limited to, various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0191] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and structure diagrams in the accompanying drawings show the possible architectures, functions and operations of the devices, methods and computer program products according to the multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and the module, program segment or a part of the code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in an alternative implementation, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the structure diagram and / or flowchart, and the combination of boxes in the structure diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or can be implemented using a combination of dedicated hardware and computer instructions.

[0192] In addition, the functional modules or units in the various embodiments of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0193] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a computer diagnostic device (which can be a smart phone, personal computer, server, or network diagnostic device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application.

[0194] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A vehicle paint diagnosis method, characterized in that: The method comprises: Perform optical signal acquisition on the paint surface of the target vehicle to obtain original optical image information and original light intensity information; performing feature extraction processing on the original optical image information and the original light intensity information respectively to obtain optical image feature information and light intensity feature information; Inputting the optical image feature information into a convolutional neural network model to identify defect information on the vehicle paint surface; Comparing the light intensity characteristic information with a preset spectral parameter database, and obtaining coating thickness distribution information of the vehicle paint surface based on a frequency domain inversion algorithm; Obtaining the operating information of the target vehicle through the vehicle interface; The operation information, the defect information and the coating thickness distribution information are correlated and fused to generate a vehicle paint diagnosis report.

2. The vehicle paint diagnosis method according to claim 1, characterized in that: The performing of optical signal acquisition on the paint surface of the target vehicle to obtain original optical image information and original light intensity information includes: Adjust the relative distance between the detection module and the vehicle paint surface to within a preset detection distance range, and measure and record the distance information in real time using a laser ranging module; Based on the distance information and a preset detection mode, controlling the light source module to sequentially emit laser signals according to a predetermined wavelength set and angle set to illuminate the paint surface of the target vehicle; The image acquisition unit and the light intensity acquisition unit of the detection module are synchronously driven to collect the image data and light intensity data reflected by the vehicle paint surface at each preset wavelength and each preset angle to obtain the original optical image information and the original light intensity information.

3. The vehicle paint diagnosis method according to claim 1, characterized in that: The performing feature extraction processing on the original optical image information and the original light intensity information respectively to obtain optical image feature information and light intensity feature information includes: Performing grayscale processing on the original optical image information to obtain a grayscale image; Performing multi-scale noise reduction filtering on the grayscale image to obtain a noise-reduced image; Performing edge enhancement processing on the noise reduction image to obtain an edge enhanced image; Extracting texture features, structural features and edge gradient information from the edge-enhanced image to construct the optical image feature information; Performing fast Fourier transform processing on the original light intensity information to obtain a frequency domain amplitude spectrum; A frequency band filtering algorithm is performed on the frequency domain amplitude spectrum to extract frequency domain eigenvalues ​​and construct the light intensity characteristic information.

4. The vehicle paint diagnosis method according to claim 1, characterized in that: Inputting the optical image feature information into a convolutional neural network model to identify defect information on the vehicle paint surface includes: Inputting the optical image feature information into a pre-trained defect recognition model to extract a feature vector representing the semantic structure of the vehicle paint image; Based on the feature vector, the defect types existing on the vehicle paint surface are identified, and the spatial positions, boundary contours, and size parameters of each type of defect are determined.

5. The vehicle paint diagnosis method according to claim 1, characterized in that: The comparing the light intensity characteristic information with a preset spectral parameter database and obtaining the coating thickness distribution information of the vehicle paint surface based on a frequency domain inversion algorithm includes: Matching the light intensity characteristic information with a standard reflection spectrum corresponding to the vehicle paint surface in the preset spectral parameter database to extract reference spectrum parameters corresponding to the target detection area; Constructing a frequency domain inversion analysis model based on the frequency domain amplitude difference and phase change characteristics between the light intensity characteristic information and the reference spectrum parameters; Utilizing the frequency domain inversion analysis model to perform parameter back-propagation processing on the light intensity characteristic information, and calculating the coating thickness distribution information of each detection area; The coating thickness distribution information is associated with the corresponding spatial position information to generate the coating thickness distribution information.

6. The vehicle paint diagnosis method according to claim 1, characterized in that: The obtaining the operating information of the target vehicle through the vehicle interface includes: Obtaining the vehicle identification code, mileage, and current diagnosis time of the target vehicle through the vehicle interface; Based on the vehicle identification code, searching for standard vehicle model parameters corresponding to the vehicle paint surface in a vehicle model information database and obtaining the standard vehicle model parameters; the standard vehicle model parameters include standard vehicle paint structure parameters and a coating reference thickness; The vehicle identification code, the mileage, the current diagnosis time, and the standard vehicle model parameters are associated to form the operation information.

7. The vehicle paint diagnosis method according to claim 6, characterized in that: The correlating and fusing the operation information with the defect information and the coating thickness distribution information to generate a paint diagnosis report includes: Mapping and integrating the defect information and the coating thickness distribution information according to the body structure area of ​​the target vehicle to form vehicle paint status data of each detection area; Performing field association on the paint status data, the vehicle identification code, the mileage, and the diagnosis time to generate fused diagnosis data; Based on preset defect classification rules and thickness deviation assessment rules, each inspection area in the diagnostic data is graded and assessed to obtain diagnostic assessment results and repair suggestions; The paint diagnosis report is generated based on the diagnostic evaluation results and the repair suggestions.

8. A vehicle paint diagnostic device, characterized in that: include: A signal acquisition module is used to perform optical signal acquisition on the paint surface of the target vehicle to obtain original optical image information and original light intensity information; a data acquisition module, configured to perform feature extraction processing on the original optical image information and the original light intensity information, respectively, to obtain optical image feature information and light intensity feature information; a defect information acquisition module, configured to input the optical image feature information into a convolutional neural network model to identify defect information on the vehicle paint surface; a coating information processing module, configured to compare the light intensity characteristic information with a preset spectral parameter database and obtain coating thickness distribution information on the vehicle paint surface based on a frequency domain inversion algorithm; An operation information acquisition module, used to obtain the operation information of the target vehicle through a vehicle-mounted interface; The diagnostic report generation module is used to associate and fuse the operating information with the defect information and the coating thickness distribution information to generate a vehicle paint diagnostic report.

9. A diagnostic device, characterized in that The diagnostic device includes a processor and a memory, wherein the memory stores a computer program, and the processor is configured to execute the computer program to implement the vehicle paint diagnostic method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The device stores a computer program, which, when executed on a processor, implements the vehicle paint diagnosis method according to any one of claims 1 to 7.

Images