Automobile wire harness detection system based on artificial intelligence and machine vision

Abstract

The invention discloses an automobile wire harness detection system based on artificial intelligence and machine vision, and relates to the technical field of intelligent detection, and the system comprises a feature extraction module which carries out the noise reduction and feature enhancement of hyperspectral reflection data through a low-rank sparse decomposition method, extracts the principal component features and local sparse features of an automobile wire harness sheath, and carries out the recognition of the local sparse features of the automobile wire harness sheath; spectral features are formed; the material identification module is used for identifying the category of the automobile wire harness sheath material according to the similarity distribution of the spectral characteristics in the multi-dimensional spectral space; the state analysis module is used for combining the spectral characteristics with the types of the automobile wire harness sheath materials, identifying the surface state of the automobile wire harness sheath and carrying out quantitative analysis to obtain an aging grade evaluation result; and the quality evaluation module is used for carrying out comprehensive quality evaluation on the automobile wire harness sheath according to the automobile wire harness sheath material category and the aging grade evaluation result, and generating a detection report. According to the invention, the capability of distinguishing aging modes of different materials is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent detection technology, in particular to an automobile wiring harness detection system based on artificial intelligence and machine vision. Background Art

[0002] In modern automotive manufacturing, wiring harnesses serve as the core transmission channel of the electrical system. Material consistency and surface integrity have a significant impact on the vehicle's electrical performance and operational safety. To ensure product consistency and service reliability, manufacturers typically perform visual inspections and material identification on automotive wiring harness sheaths before assembly and shipment. Currently, widely used wiring harness inspection methods primarily include visible light image recognition technology based on industrial cameras, near-infrared spectroscopy analysis, and manual sampling. The former often uses traditional image processing algorithms or convolutional neural network structures to identify the position, color, or defect morphology of the wiring harness in an image, while the latter estimates material composition by acquiring spectral response data.

[0003] Conventional single-band or low-spectral-resolution detection methods are particularly limited for tasks such as material identification and aging assessment, which rely heavily on detailed changes in spectral response. For example, visible light-based image recognition methods can only capture surface texture and color information, but have difficulty reflecting changes in the material's internal physical structure and spectral characteristics. While traditional spectral analysis methods can obtain information from some bands, they are difficult to deploy for rapid inspections on production sites due to high costs, large data volumes, and complex data processing processes. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an automobile wiring harness detection system based on artificial intelligence and machine vision to solve the problem that the surface aging status of automobile wiring harness sheaths is difficult to quantify and evaluate.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] The application provides an automobile wire harness detection system based on artificial intelligence and machine vision, which comprises a data acquisition module, a data reconstruction module, a feature extraction module, a material identification module, a state analysis module and a quality evaluation module.

[0008] As a preferred scheme of the automobile wire harness detection system based on artificial intelligence and machine vision, the generation of the diffraction signal data comprises the following steps,

[0009] Spectrum reflection intensity data of the automobile wire harness sheath surface are acquired and compressed sampling is performed to generate compressed image data.

[0010] The compressed image data are input into a diffraction response model for light field propagation calculation to generate diffraction signal data.

[0011] As a preferred scheme of the automobile wire harness detection system based on artificial intelligence and machine vision, the construction of the diffraction response model comprises the following steps,

[0012] The curvature radius, inclination angle parameters and material refractive index parameters of the microlens array are measured, and a three-dimensional optical structure model is established and focal length parameters and wavelength compensation coefficients are extracted by combining geometric layout and optical path characteristics.

[0013] Based on the three-dimensional optical structure model, the propagation and diffraction behavior of light waves are simulated by using the Fresnel diffraction method to form a preliminary diffraction response function.

[0014] The preliminary diffraction response function, the focal length parameters and the wavelength compensation coefficients are corrected and fitted to construct the diffraction response model.

[0015] As a preferred scheme of the automobile wire harness detection system based on artificial intelligence and machine vision, the generation of the hyperspectral reflection data comprises the following steps,

[0016] The diffraction signal data are filtered and normalized to obtain preprocessed diffraction signal data.

[0017] According to the pre-processed diffraction signal data, the basis functions are selected and the random sparse coefficient values ​​are initialized to construct a sparse representation model;

[0018] The pre-processed diffraction signal data is iteratively solved through the sparse representation model to obtain preliminary hyperspectral reflectance data;

[0019] The preliminary hyperspectral reflectance data are subjected to denoising enhancement and spectral correction processing to obtain hyperspectral reflectance data.

[0020] As a preferred solution of the automobile wiring harness detection system based on artificial intelligence and machine vision of the present invention, wherein: according to the pre-processed diffraction signal data, the basis function is selected and the random sparse coefficient value is initialized to construct a sparse representation model. The specific steps are as follows:

[0021] Analyze the frequency domain and spatial domain distribution characteristics of the pre-processed diffraction signal data and extract the modeling data sample set;

[0022] Selecting a sparse representation basis function set of diffraction signal characteristics from the modeling data sample set to obtain a candidate basis function library;

[0023] Based on the candidate basis function library, the matching pursuit algorithm is used to perform preliminary fitting of the sparse representation of the preprocessed diffraction signal data to generate the initial coefficient matrix;

[0024] The initial coefficient matrix is ​​optimized with sparse constraints, and the compressed sensing constraint method is used to iteratively converge to generate a sparse representation model.

[0025] As a preferred solution of the automobile wiring harness detection system based on artificial intelligence and machine vision of the present invention, the specific steps of forming the spectral characteristics are as follows:

[0026] Converting the hyperspectral reflectance data into a two-dimensional matrix to form a hyperspectral data matrix;

[0027] The hyperspectral data matrix is ​​decomposed using a low-rank sparse decomposition algorithm to obtain a low-rank component matrix and a sparse component matrix;

[0028] Extract the principal component features reflecting the overall material properties of the automotive wiring harness sheath from the low-rank component matrix;

[0029] Extract sparse features reflecting local detail changes from the sparse component matrix;

[0030] The principal component features are fused with the sparse features to form spectral features.

[0031] As a preferred solution of the automobile wiring harness detection system based on artificial intelligence and machine vision of the present invention, the specific steps of identifying the material category of the automobile wiring harness sheath are as follows:

[0032] Arrange the spectral features into vectors according to the intensity values ​​of each wavelength channel and map them to the multidimensional spectral space to generate a high-dimensional feature vector;

[0033] Compare the high-dimensional feature vector with the spectral features of material categories in the spectral material database one by one to generate a material category matching score;

[0034] Based on the material category matching score, the category label with the highest score is filtered out from the spectral material database, and the material category of the automotive wiring harness sheath is identified.

[0035] As a preferred solution of the automobile wiring harness detection system based on artificial intelligence and machine vision of the present invention, the specific steps of combining spectral characteristics with automobile wiring harness sheath material categories to identify the surface state of automobile wiring harness sheaths and perform quantitative analysis to obtain aging grade evaluation results are as follows:

[0036] The spectral features are fused with the material categories of the automotive wiring harness sheath to obtain fused features;

[0037] Extracting spectral variation features reflecting surface states from fused features;

[0038] Based on the spectral variation characteristics, the system can identify the oil stains, photooxidation aging and fading areas on the surface of the automotive wiring harness sheath, generate surface degradation characteristics, and perform quantitative processing to form quantitative analysis data.

[0039] Based on the quantitative analysis data, a comprehensive evaluation is conducted on the surface degradation degree and spectral change characteristics of the automotive wiring harness sheath to obtain the aging grade evaluation results.

[0040] As a preferred solution of the automobile wiring harness detection system based on artificial intelligence and machine vision of the present invention, the following specific steps are used to identify oil stains, photooxidation aging and fading areas on the surface of the automobile wiring harness sheath according to the spectral change characteristics, form surface degradation characteristics, and perform quantitative processing to form quantitative analysis data.

[0041] Analyze the time series and amplitude changes of the spectral change characteristics, extract the wavelength offset data and reflection intensity change data, and form the spectral anomaly fragments;

[0042] Classify and identify abnormal spectral fragments and mark them as surface conditions of oil stains, photooxidation aging, and faded areas;

[0043] The sliding window statistical method is used to analyze the time series changes of spectral anomaly fragments corresponding to different surface states and extract the spectral change trend characteristics;

[0044] According to the spectral change trend characteristics, the spectral intensity change value, wavelength offset value and reflectivity gradient value are calculated, the surface degradation characteristics are generated and numerical standardization is performed to form quantitative analysis data.

[0045] As a preferred solution of the automobile wiring harness detection system based on artificial intelligence and machine vision of the present invention, the specific steps of generating the detection report are as follows:

[0046] Combined analysis of automotive wiring harness sheath material categories and aging grade evaluation results to obtain a comprehensive quality score;

[0047] Based on the comprehensive quality score and the grading threshold range, the overall quality level of the automotive wiring harness sheath is divided;

[0048] Integrate the overall quality status grade with quantitative analysis data to generate a test report.

[0049] The present invention achieves the following beneficial effects: By fusing high-dimensional spectral features with identified automotive wiring harness sheath material categories, it achieves joint modeling of intrinsic material properties and spectral response data, effectively enhancing the targetedness and accuracy of subsequent state identification. Spectral variation characteristics reflecting typical surface conditions, such as oil stains, photooxidative degradation, and faded areas, are extracted from the fused features, enabling precise classification and identification of localized degradation states in automotive wiring harness sheaths. By converting surface aging phenomena into quantifiable feature expressions, the ability to distinguish between different material aging patterns is significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0051] Figure 1 Schematic diagram of the automotive wiring harness detection system based on artificial intelligence and machine vision in the present invention.

[0052] Figure 2 Flowchart for generating diffraction signal data in the present invention.

[0053] Figure 3 This is a flow chart for constructing the diffraction response model in the present invention.

[0054] Figure 4 This is a flow chart for obtaining hyperspectral reflectance data in the present invention. DETAILED DESCRIPTION

[0055] In order to make the above objectives, features and advantages of the present application more apparent, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0056] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. The present application, however, can be practiced in a variety of ways beyond the specific embodiments described herein without departing from the scope of the present application, which is set forth in the appended claims. It should be recognized, therefore, that the specific embodiments disclosed are illustrative only and not limiting of the present application.

[0057] Secondly, the term "one embodiment" or "an embodiment" as used herein means that a particular implementation can include a particular feature, structure, or characteristic, but every embodiment can not necessarily include the particular feature, structure, or characteristic. Furthermore, the following claims can refer to "one embodiment" or "an embodiment" in the sense of claiming a particular feature, structure, or characteristic of more than one embodiment.

[0058] Reference will now be made to the drawings, in which Figures 1 to 4 For one embodiment of the present application, the embodiment provides an automobile wire harness detection system based on artificial intelligence and machine vision, comprising the following steps:

[0059] The data acquisition module acquires compressed image data of multi-band spectral reflectance information, converts the compressed image data through a diffraction response model, and generates diffraction signal data.

[0060] The spectral reflectance intensity data of the surface of the automobile wire harness sheath is acquired and compressed sampled to generate compressed image data.

[0061] Specifically, the linear movement device is controlled to move the automobile wire harness sheath sample on the fixed track at a constant speed, a spectral acquisition device is used, for example, to emit incident light at a wavelength interval of every 5 nm in the range of 370 nm to 1000 nm, and an imaging detector is used to synchronously record the reflection intensity signals corresponding to each group of wavelength channels; the reflection intensity data acquired by the spectral acquisition device is column-compressed, each group of wavelength channel data is compressed and encoded using a random sampling matrix, and is combined into two-dimensional compressed image data by frames, the compression sampling process is completed, and the compressed image data is generated.

[0062] The compressed image data is input into the diffraction response model for light field propagation calculation to generate diffraction signal data.

[0063] Specifically, the pixel coordinates of the compressed image data are normalized in two-dimensional space. Based on the microlens array parameters contained in the diffraction response model and the optical path setting in the three-dimensional optical structure model, the Fresnel diffraction method is used to calculate the complex amplitude contribution of each compressed image data pixel point to each diffraction point on the observation plane in the frequency domain; all pixel points are accumulated point by point to obtain the sum of the complex amplitudes at each position on the observation plane; the complex amplitude is then modulo-squared to obtain the light intensity distribution value; the light intensity distribution result is organized and output in the form of a two-dimensional matrix to generate diffraction signal data.

[0064] Furthermore, by measuring the curvature radius, inclination parameters and material refractive index parameters of the microlens array, combined with the geometric layout and optical path characteristics, a three-dimensional optical structure model is established and the focal length parameters and wavelength compensation coefficients are extracted.

[0065] Specifically, a profile interferometer is used to scan the curvature profile of each microlens unit in the microlens array to obtain curvature radius data; an angle measuring instrument is used to measure the angle between the normal direction of the microlens unit and the array substrate to obtain the inclination parameter; an ellipsometer is used to measure the complex refractive index of the microlens material at different wavelengths to obtain the material refractive index parameter; based on the obtained curvature radius, inclination parameter and material refractive index parameter, an optical path structure containing the geometric layout of the microlens array is constructed in three-dimensional space, and the focusing characteristics of the microlens array for different wavelengths are calculated through a joint modeling method of geometric optics and wave optics, and the corresponding focal length parameters are extracted; and based on the fitting relationship between wavelength and optical path difference, a wavelength compensation coefficient is obtained.

[0066] Based on the three-dimensional optical structure model, the Fresnel diffraction method is used to simulate the propagation and diffraction behavior of light waves to form a preliminary diffraction response function.

[0067] Specifically, the spatial position parameters, curvature radius, inclination parameters and material refractive index parameters of each microlens unit in the three-dimensional optical structure model are input into the Fresnel diffraction method, and the propagation distance between the observation plane and the microlens array is set. For example, the propagation distance is set to 5 mm, and the wavelength range of the light source is defined as 400 nm to 1000 nm. The complex amplitude distribution of the light field on the observation surface is calculated at each wavelength; the transfer function of the light field propagation is constructed in the frequency domain, and the fast Fourier transform is performed to obtain the frequency domain propagation characteristics. Then, the diffraction distribution in the spatial domain is restored by the inverse Fourier transform; the spatial distribution results at all wavelengths are normalized by complex amplitude to obtain the light wave response data of the corresponding band, and finally the response results of all wavelengths are summarized to form a preliminary diffraction response function.

[0068] The preliminary diffraction response function, focal length parameters and wavelength compensation coefficients are corrected and fitted to construct a diffraction response model.

[0069] Specifically, the complex amplitudes and focal length parameters corresponding to all wavelengths in the preliminary diffraction response function are corrected for image plane position, and the spatial distribution of the light field propagation results on the observation plane is adjusted to match the focal position with the focal length parameters; based on the light field response offset of each wavelength, a linear fitting method is applied to fit the wavelength response change trend with the wavelength compensation coefficient to correct the phase and intensity errors between different bands; for example, the least squares method can be used to fit the difference between the actual response and the ideal compensated target response, and the corresponding items in the preliminary diffraction response function are iteratively adjusted; after completing the steps, the corrected full-band response results are fused to form a response function that meets the focal length and wavelength compensation constraints, and a diffraction response model is constructed.

[0070] The data reconstruction module uses the compressed sensing decoding method to reconstruct the diffraction signal data to obtain hyperspectral reflectance data.

[0071] The diffraction signal data is filtered and normalized to obtain pre-processed diffraction signal data.

[0072] Specifically, the time domain or space domain sequence of the diffraction signal data is edge-cut to remove the noise interference area; the median filtering method is used to smooth the intensity sequence of each channel in the diffraction signal data to remove isolated abnormal points; a low-pass filter, such as a Butterworth filter, is used to perform frequency domain filtering on the diffraction signal data to retain the main frequency information and suppress high-frequency noise; after completing the frequency domain filtering, the maximum and minimum values ​​of the diffraction signal data in each channel are calculated respectively, and according to the maximum and minimum normalization method, all channel intensity values ​​are linearly transformed to, for example, the interval [0,1]; the normalized result is the pre-processed diffraction signal data.

[0073] According to the preprocessed diffraction signal data, basis functions are selected and random sparse coefficient values ​​are initialized to construct a sparse representation model.

[0074] Furthermore, the frequency domain and spatial domain distribution characteristics of the pre-processed diffraction signal data are analyzed to extract the modeling data sample set.

[0075] Specifically, the preprocessed diffraction signal data is sliced ​​by channel in the spatial dimension, and the distribution position of the local extreme points and the corresponding intensity values ​​of each channel are counted; the intensity sequence of each channel is fast Fourier transformed to obtain the frequency domain amplitude spectrum, and the frequency domain characteristic parameters such as the frequency domain main peak frequency, bandwidth and frequency domain energy concentration are extracted; the spatial domain extreme points are matched with the corresponding positions of the frequency domain principal components, and the diffraction signal data segments with significant spatial domain gradient changes and frequency domain principal component energy concentration are screened out; finally, the screened diffraction signal data segments are used as samples to form a modeling data sample set.

[0076] A sparse representation basis function set of diffraction signal characteristics is selected from the modeling data sample set to obtain a candidate basis function library.

[0077] Specifically, a Fourier transform is performed on each diffraction signal data segment in the modeling data sample set to extract the frequency domain principal component characteristics; according to the main peak position and frequency distribution law of the frequency domain characteristics, the types of basis functions disclosed in the existing sparse representation methods are matched, such as Gabor basis function, discrete cosine basis function and wavelet basis function, etc.; for different frequency domain structures, several groups of sparse representation basis functions that match the frequency domain principal components are selected respectively, and a set is formed by the sparse representation basis function expression and parameter configuration method to obtain a sparse representation basis function set of the diffraction signal characteristics and construct a candidate basis function library.

[0078] Based on the candidate basis function library, the matching pursuit algorithm is used to perform preliminary fitting of the sparse representation of the preprocessed diffraction signal data to generate the initial coefficient matrix.

[0079] Specifically, all sparse representation basis functions in the candidate basis function library are traversed one by one, and the inner product value between each basis function and the current diffraction signal vector is calculated; the sparse representation basis function with the largest absolute value of the inner product value is selected as the matching basis function of the current iteration, and the index position and the corresponding coefficient value are recorded; the sparse representation basis function and the corresponding coefficient value are used to generate the current fitting part, and the fitting part is subtracted from the current diffraction signal vector to obtain the residual signal; the sparse representation basis function indices and coefficient values ​​obtained in each round of iteration are arranged in order to form an initial coefficient matrix corresponding to the preprocessed diffraction signal data.

[0080] The initial coefficient matrix is ​​optimized with sparse constraints, and the compressed sensing constraint method is used to iteratively converge to generate a sparse representation model.

[0081] Specifically, the objective function form in the compressed sensing constraint method is selected, for example, the reconstruction error objective function that minimizes the L1 norm constraint is adopted; the initial coefficient matrix is ​​used as the starting value, and the sparse optimization method is used to calculate the gradient direction of the joint minimization of the sparse representation reconstruction error and the coefficient sparsity; then each coefficient value in the initial coefficient matrix is ​​updated along the gradient direction, and after each round of iteration, a soft threshold function is introduced to perform a sparsity constraint operation on the updated coefficient matrix to eliminate low-amplitude noise coefficients; the convergence of the reconstruction error is continuously evaluated during the iteration process until the iteration stop condition is met; the converged sparse coefficient matrix is ​​combined with the candidate basis function library to generate a sparse representation model.

[0082] The pre-processed diffraction signal data is iteratively solved through the sparse representation model to obtain preliminary hyperspectral reflectance data.

[0083] Specifically, the preprocessed diffraction signal data is represented as an observation vector in a sparse representation model, and the inversion expression is initialized in combination with the candidate basis function library and the converged sparse coefficient matrix; in each round of iteration, the sparse reconstruction algorithm is used to calculate the projection of the observation vector in the sparse representation space to obtain the reconstruction vector estimated in the current round; the current reconstruction vector is compared with the actual preprocessed diffraction signal data for error, the sparse coefficient is adjusted and the reconstruction vector is updated; the sparse coefficient matrix is ​​updated through round-by-round iterations to continuously reduce the reconstruction error; finally, the iteration is stopped when the error convergence condition is met to obtain preliminary hyperspectral reflectance data.

[0084] The preliminary hyperspectral reflectance data are subjected to denoising enhancement and spectral correction processing to obtain hyperspectral reflectance data.

[0085] Specifically, the wavelet threshold denoising method is used to process the preliminary hyperspectral reflectance data to remove high-frequency noise components and obtain denoised hyperspectral reflectance data; a local contrast enhancement operation is performed on the denoised hyperspectral reflectance data to enhance the response intensity of the material edge area in the spectral image; then the standard reference spectral data is used to perform band alignment and response amplitude adjustment, and the minimum mean square error criterion is used to calculate the offset value and gain coefficient of each band to complete the spectral correction; finally, the hyperspectral reflectance data that has completed denoising, enhancement and spectral correction processing is output.

[0086] The feature extraction module uses the low-rank sparse decomposition method to reduce noise and enhance features of hyperspectral reflectance data, extract the main component features and local sparse features of the automobile wiring harness sheath, and form spectral features.

[0087] The hyperspectral reflectance data is converted into a two-dimensional matrix to form a hyperspectral data matrix.

[0088] Specifically, the three-dimensional spectral data cube in the hyperspectral reflectance data is read, and the three-dimensional spectral data cube contains spatial dimensions (row index and column index) and spectral dimensions (wavelength index); it is expanded according to the wavelength dimension, and all wavelength reflectance values ​​corresponding to each spatial position are extracted as a column vector; the column vectors corresponding to all spatial positions are arranged as row vectors of a two-dimensional matrix in the order of row index and column index, forming a hyperspectral data matrix with pixels as rows and wavelength channels as columns; for example, if the spatial resolution is 100×100 and the number of wavelength channels is 200, the size of the generated hyperspectral data matrix is ​​10000×200.

[0089] The hyperspectral data matrix is ​​decomposed using a low-rank sparse decomposition algorithm to obtain a low-rank component matrix and a sparse component matrix.

[0090] Specifically, when processing the hyperspectral data matrix, the hyperspectral data matrix is ​​used as input data, and the low-rank component matrix and the sparse component matrix are initialized as all-zero matrices; the robust principal component analysis algorithm in the principal component tracking method is used to iteratively decompose the hyperspectral data matrix. In each round of iteration, the current sparse component matrix is ​​fixed, and the characteristic singular values ​​and the corresponding left and right singular vectors are extracted based on the singular value decomposition method to update the low-rank component matrix; the updated low-rank component matrix is ​​fixed, and the difference between the hyperspectral data matrix and the low-rank component matrix is ​​subjected to element-by-element soft threshold processing, and the elements below the preset noise suppression threshold are eliminated and set to zero, thereby updating the sparse component matrix; the preset noise suppression threshold is based on an estimated value of the noise energy in the hyperspectral data matrix, or an estimated value composed of the median and standard deviation in the global pixel value distribution of the hyperspectral data matrix; the iterative steps are continued until the residual between the combination result of the low-rank component matrix and the sparse component matrix and the original hyperspectral data matrix converges, and finally the low-rank component matrix and the sparse component matrix are output.

[0091] It should also be explained that when presetting the noise suppression threshold, the background area or low signal area is selected as the reference area in the hyperspectral data matrix, and the spectral intensity values ​​of all pixels in the reference area are extracted; the global median and standard deviation of the spectral intensity values ​​in the reference area are calculated, and a noise estimation index is constructed; the median and standard deviation are weighted and combined to obtain an estimated value reflecting the noise energy level; finally, the estimated value is used as the noise suppression threshold.

[0092] The principal component features reflecting the overall material properties of the automotive wiring harness sheath are extracted from the low-rank component matrix.

[0093] Specifically, singular value decomposition is performed on the low-rank component matrix to obtain singular values ​​and corresponding singular vectors; the singular values ​​are sorted according to their sizes, and the singular vectors corresponding to the first several largest singular values ​​are selected as principal component eigenvectors; the low-rank component matrix is ​​projected into the selected principal component eigenvector space to obtain principal component features that reflect the overall material properties of the automotive wiring harness sheath; for example, the singular vectors corresponding to the first three largest singular values ​​are selected as principal component features, a three-dimensional feature space is constructed, and the principal component feature extraction is completed.

[0094] Sparse features reflecting local detail changes are extracted from the sparse component matrix.

[0095] Specifically, when processing the sparse component matrix, all elements in the matrix are judged one by one according to the set sparse threshold, elements with amplitudes less than the sparse threshold are set to zero, elements with amplitudes greater than or equal to the sparse threshold are retained, and local areas with significant amplitude changes are highlighted; a variance evaluation-based method is used to select sparse feature points whose amplitude change ranks high among all non-zero elements from the retained non-zero positions, for example, points with variances greater than the median of the variances of all non-zero elements are selected as representative sparse feature points, and a sparse feature vector is formed in the order of arrangement; then the sparse feature vector is normalized to obtain sparse features reflecting the changes in local details of the automotive wiring harness sheath.

[0096] It should also be explained that the setting of the sparse threshold includes: selecting a sparse threshold higher than the noise amplitude based on the noise level of the actual spectral reflectance data of the automotive wiring harness sheath to effectively distinguish signal features from noise interference; combining statistical analysis of historical detection data to determine the sparse threshold range that can highlight local detail changes without excessively losing effective information; for example, after many experiments, setting the sparse threshold to 0.05 can take into account both signal fidelity and noise suppression to achieve feature extraction effects.

[0097] The principal component features are fused with the sparse features to form spectral features.

[0098] Specifically, the principal component features extracted from the low-rank component matrix and the sparse features extracted from the sparse component matrix are normalized separately. After unifying the feature dimensions, they are weighted and fused according to the weight coefficient to generate a spectral feature vector; the weight coefficient is set according to the feature importance, for example, the principal component feature weight is 0.7, and the sparse feature weight is 0.3. The fused spectral features contain overall material characteristics and local detail change information, forming the final spectral features.

[0099] The material recognition module identifies the material category of automotive wiring harness sheaths based on the similarity distribution of spectral features in multidimensional spectral space.

[0100] The spectral features are arranged into vectors according to the intensity values ​​of each wavelength channel and mapped to the multidimensional spectral space to generate a high-dimensional feature vector.

[0101] Specifically, the intensity values ​​corresponding to each wavelength channel in the spectral feature are arranged in sequence according to the wavelength order to form a one-dimensional spectral feature vector; according to the coordinate mapping rule of the multidimensional spectral space, the one-dimensional spectral feature vector is mapped to the multidimensional spectral space through linear transformation or nonlinear mapping method to generate a high-dimensional feature vector.

[0102] The high-dimensional feature vector is compared one by one with the spectral features of the material categories in the spectral material database to generate a material category matching score.

[0103] Specifically, the high-dimensional feature vector is compared with the spectral feature vectors corresponding to each material category in the spectral material database one by one, and the matching degree between the spectral feature vector of each material category and the high-dimensional feature vector is calculated by using a Euclidean distance, a cosine similarity or a correlation coefficient as a measurement method, to obtain a corresponding matching score; for example, the cosine similarity of the high-dimensional feature vector and the example spectral feature in the database is calculated to obtain the matching scores of multiple material categories, which constitute the material category matching scores.

[0104] It should also be noted that the spectral material database is derived from a public spectral database and experimental measurement data, and contains standard spectral feature information of multiple common and specific material categories in the visible light to near-infrared band. The spectral material data is obtained by a high-precision spectrometer in a controlled environment, and is pre-processed and corrected to ensure the accuracy and consistency of the spectral material data; for example, part of the data of the spectral material database is obtained from the spectral standard library and the spectral data in the related public literature, and the sample spectrum obtained by actual measurement in the field is combined to form a complete spectral material database.

[0105] According to the material category matching scores, the highest scoring category label is selected from the spectral material database, and the material category of the automobile wire harness sheath is identified.

[0106] Specifically, according to the material category matching scores, the material category labels and corresponding scores are read from the spectral material database, all scores are compared, the highest scoring material category label is selected, and the material category corresponding to the highest scoring material category label is determined as the identification result; for example, the category label with the highest score "polyvinyl chloride" is selected from the score list, and is used as the identification result of the material category of the automobile wire harness sheath.

[0107] The state analysis module combines the spectral feature with the material category of the automobile wire harness sheath, identifies the surface state of the automobile wire harness sheath and performs quantitative analysis, and obtains an aging grade evaluation result.

[0108] The spectral feature and the material category of the automobile wire harness sheath are fused to obtain a fused feature.

[0109] Specifically, the spectral feature and the material category of the automobile wire harness sheath are weighted according to a preset weight coefficient, and the corresponding elements are added one by one by using a weighted fusion method to generate a fused feature.

[0110] It should also be explained that when presetting the weight coefficient, the initial weight ratio is determined based on the relative importance of the spectral characteristics and the material category of the automotive wiring harness sheath in the identification task, and multiple groups of experiments are conducted in combination with historical weight data or verification sets. The weight value is adjusted through performance evaluation indicators such as accuracy and recall rate, and the cross-validation method is used to compare the performance of each combination in material recognition accuracy. Finally, the combination with the highest recognition accuracy is selected as the optimal weight coefficient.

[0111] Spectral change features reflecting the surface state are extracted from the fused features.

[0112] Specifically, principal component analysis is performed on the fused features, and the principal component vectors are extracted by calculating the covariance matrix of the fused features and adopting the singular value decomposition method; the first-order difference processing is performed on the intensity values ​​of each wavelength channel in each principal component vector according to the channel order, and the intensity change values ​​between adjacent channels are calculated to form a change rate vector reflecting the intensity fluctuation trend; in the change rate vector, the sliding window method is used to identify the change amplitudes exceeding the set standard deviation threshold interval within the shortwave band, mediumwave band and longwave band, and the wavelength channel intensity difference within the standard deviation threshold interval is extracted as the spectral change feature reflecting the change of the surface state of the automobile wiring harness sheath.

[0113] It should also be explained that the specific steps for setting the standard deviation threshold interval are: statistical analysis is performed on the intensity change values ​​of each principal component vector in the fusion feature in all wavelength channels, and the full sample mean and standard deviation of the intensity change rate of the wavelength channel corresponding to each principal component vector are calculated; the standard deviation multiplication factor is preset according to the target sensitivity requirements, for example, it is set to 1.5 or 2.0 in the example, and the standard deviation multiplication factor is multiplied by the standard deviation to obtain the upper and lower fluctuation threshold range; finally, the mean plus or minus the fluctuation threshold range is used as the standard deviation threshold interval, which is used to identify the wavelength channels in the shortwave band, medium wave band and long wave band where the spectral intensity change rate deviates significantly from the mean.

[0114] Based on the spectral change characteristics, the oil stains, photo-oxidation aging and fading areas on the surface of the automotive wiring harness sheath are identified, the surface degradation characteristics are generated, and quantified to form quantitative analysis data.

[0115] Furthermore, the spectral variation characteristics are analyzed in terms of time series and amplitude variation, and the wavelength offset data and reflection intensity variation data are extracted to form spectral anomaly segments.

[0116] Specifically, a time series sequence is constructed for the spectral change characteristics according to the arrangement order of the wavelength channels, and the difference method is used to calculate the difference in reflection intensity corresponding to two consecutive wavelength channels, and data reflecting the amplitude change are extracted; then the sliding window method is used to analyze the intensity change trend in the continuous wavelength interval, and the offset of the center position of the wavelength channel in each sliding interval relative to the initial channel position is calculated, and the wavelength offset data is extracted; finally, the wavelength offset data is combined with the reflection intensity change data of the corresponding wavelength channel. When the offset of multiple consecutive wavelength channels exceeds twice the statistical fluctuation mean, and the reflection intensity change amplitude of the corresponding channel is greater than 1.5 times the average intensity difference of adjacent channels, it is marked as a spectral anomaly segment.

[0117] It should also be explained that the specific process of the sliding window method for analyzing the intensity change trend within a continuous wavelength interval is as follows: in the extracted change rate vector, a sliding window with a length equal to the number of fixed wavelength channels is set, for example, the window length is set to 5 wavelength channels; the window is slid on the change rate vector in sequence according to the wavelength channel order, moving one channel position each time; within each window interval, the mean and standard deviation of all change rate values ​​in the window are calculated to characterize the average intensity change degree and fluctuation amplitude of the current band; the calculated results are compared with the historical average change pattern to identify the band area where the change rate significantly increases or decreases, and then the intensity change trend within the continuous wavelength interval is judged.

[0118] Abnormal spectral fragments are classified and identified, and marked as surface conditions of oil stains, photooxidation aging, and faded areas.

[0119] Specifically, the reflection intensity values ​​of two adjacent sampling moments or two adjacent positions in the spectral anomaly segment under the same wavelength channel are read in turn, and the first-order difference method is used for calculation, that is, the reflection intensity value of the current point is subtracted from the reflection intensity value of the previous position or the previous moment, and divided by the time interval or spatial interval between the two points to obtain the reflection intensity change rate of the corresponding wavelength channel; the operation is repeated for all wavelength channels, and finally the reflection intensity change rate corresponding to each wavelength channel is arranged in wavelength order to form a reflection intensity change rate feature vector; based on the typical reflection intensity change rate feature vector, the similarity score of the feature vector of the spectral anomaly segment to be classified and each typical feature vector is calculated one by one; according to the category label with the highest similarity score, the spectral anomaly segment is marked and identified as the surface state of the oil stain area, photo-oxidation aging area or faded area.

[0120] The sliding window statistical method is used to analyze the time series changes of spectral anomaly fragments corresponding to different surface states and extract the spectral change trend characteristics.

[0121] Specifically, a sliding window statistical method is used to divide the spectral anomaly fragments corresponding to different surface states into multiple windows of fixed length in chronological order, for example, the window length is 10 sampling points, and the window sliding step is 1 sampling point; within each sliding window, the mean and variance of the reflection intensity change rate of each wavelength channel are calculated to form the statistical characteristics within the window; the sliding window is moved sequentially along the time axis, and the statistical characteristics of each window are recorded to construct a time series change sequence of the spectral anomaly fragments; combined with the time series change sequence, the spectral change trend characteristics are extracted.

[0122] According to the spectral change trend characteristics, the spectral intensity change value, wavelength offset value and reflectivity gradient value are calculated, the surface degradation characteristics are generated and numerical standardization is performed to form quantitative analysis data.

[0123] Specifically, according to the spectral change trend characteristics, the difference method is used to calculate the spectral intensity change value, the peak detection method is used to determine the wavelength offset value, and the gradient calculation method is used to obtain the reflectivity gradient value; the spectral intensity change value, the wavelength offset value and the reflectivity gradient value are combined into a surface degradation feature vector; the surface degradation feature vector is normalized, and the minimum-maximum normalization method is used to map the spectral intensity change value, the wavelength offset value and the reflectivity gradient value to a numerical interval such as [0,1] to form quantitative analysis data.

[0124] Based on the quantitative analysis data, a comprehensive evaluation is conducted on the surface degradation degree and spectral change characteristics of the automotive wiring harness sheath to obtain the aging grade evaluation results.

[0125] Specifically, based on the quantitative analysis data, the surface degradation index and spectral change characteristic index of the automobile wiring harness sheath are extracted, and each index is normalized to ensure that the value is within a unified dimension. According to the importance of each index in the overall aging evaluation, a corresponding weight coefficient is assigned, and the normalized degradation index and the spectral change characteristic index are calculated using a weighted summation method to obtain a comprehensive evaluation index. According to the aging grade scoring interval, for example, a comprehensive evaluation index in the range of 0 to 0.3 is judged as mild aging, a range of 0.3 to 0.7 is judged as moderate aging, and a range of 0.7 to 1.0 is judged as severe aging. The aging grade evaluation result of the automobile wiring harness sheath is determined according to the aging grade scoring interval in which the comprehensive evaluation index is located.

[0126] The quality assessment module conducts a comprehensive quality assessment of automotive wiring harness sheaths based on their material category and aging grade evaluation results, and generates a test report.

[0127] The automotive wiring harness sheath material category and aging grade evaluation results are jointly analyzed to obtain a comprehensive quality score.

[0128] Specifically, according to the evaluation results of the material category and aging grade of the automobile wire harness sheath, the normalized processing is carried out respectively, the respective weight coefficients are given, and the weighted summation method is used to calculate the comprehensive quality score; specifically, the material category score after normalization and the material weight coefficient are calculated, the aging grade score after normalization and the aging weight coefficient are calculated, and the comprehensive quality score is obtained; the weight coefficient is based on the relative importance of the material category and the aging grade in quality evaluation, for example, the material weight coefficient is 0.6, and the aging weight coefficient is 0.4; finally, according to the size of the comprehensive quality score, the subsequent quality judgment or grading is carried out.

[0129] According to the comprehensive quality score, combined with the grading threshold range, the overall quality condition grade of the automobile wire harness sheath is determined.

[0130] It should be noted that the basis for setting the grading threshold range includes historical quality evaluation data, industry standards and actual application requirements; the steps are: collecting a large number of comprehensive quality scores of automobile wire harness sheaths and corresponding actual quality grades as reference samples; calculating the distribution characteristics of the scores by statistical analysis method, such as mean, variance and quantile, to determine the reasonable range of each grade score interval; combined with relevant industry standards, adjust the boundaries of each grade interval to meet the actual judgment accuracy and application effect; use sample verification or cross-validation method to optimize and adjust the grading threshold range, to ensure that the grading threshold can accurately reflect the overall quality condition; for example: the excellent grade is the interval with the comprehensive quality score not less than 0.8 and not more than 1.0, the good grade is the interval with the comprehensive quality score greater than or equal to 0.6 and less than 0.8, the medium grade is the interval with the comprehensive quality score greater than or equal to 0.4 and less than 0.6, and the poor grade is the interval with the comprehensive quality score less than 0.4;

[0131] For the automobile wire harness sheath to be evaluated, the comprehensive quality score is obtained by weighting and summing the automobile wire harness sheath material category and aging grade according to the weight coefficient, and the score is compared with each grading threshold interval one by one; the judgment steps include: if the comprehensive quality score is in the excellent grade interval, it is divided into excellent quality condition; if the comprehensive quality score is in the good grade interval, it is divided into good quality condition; if the comprehensive quality score is in the medium grade interval, it is divided into medium quality condition; if the comprehensive quality score is in the poor grade interval, it is divided into poor quality condition; finally, the determined overall quality condition grade is output as the quality grade of the automobile wire harness sheath, and the division process is completed.

[0132] The overall quality condition grade is integrated with the quantitative analysis data to generate a test report.

[0133] Specifically, based on the overall quality status level and the corresponding quantitative analysis data, including spectral change characteristics, reflection intensity change rate characteristic vectors, aging level evaluation results and material category identification results; then establish an information arrangement table, and classify and organize all relevant results by field; according to the report template sequence, place the overall quality status level in a prominent position in the report, and fill in the quantitative analysis data into the material identification item, aging analysis item and spectral characteristic analysis item respectively; finally, generate the test report text content in a unified format to generate a test report.

[0134] In summary, this invention achieves joint modeling of intrinsic material properties and spectral response data by fusing high-dimensional spectral features with identified automotive wiring harness sheath material categories, effectively enhancing the targetedness and accuracy of subsequent state identification. Spectral variation signatures reflecting typical surface conditions, such as oil stains, photooxidative degradation, and fading, are extracted from the fused features, enabling precise classification and identification of localized degradation states in automotive wiring harness sheaths. This conversion of surface aging phenomena into quantifiable feature expressions significantly improves the ability to distinguish between different material aging patterns.

[0135] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An automotive wiring harness detection system based on artificial intelligence and machine vision, characterized by: include, The data acquisition module collects compressed images of multi-band spectral reflectance information, converts the compressed images by constructing a diffraction response model, and generates diffraction signal data; The data reconstruction module uses the compressed sensing decoding method to reconstruct the diffraction signal data to obtain hyperspectral reflectance data; The feature extraction module uses a low-rank sparse decomposition method to reduce noise and enhance features of hyperspectral reflectance data, extracting the principal component features and local sparse features of the automotive wiring harness sheath to form spectral features; The material recognition module identifies the material category of automotive wiring harness sheaths based on the similarity distribution of spectral features in multidimensional spectral space; The state analysis module combines spectral characteristics with the material classification of automotive wiring harness sheaths to identify and quantitatively analyze the surface state of automotive wiring harness sheaths and obtain aging grade evaluation results; The quality assessment module conducts a comprehensive quality assessment of automotive wiring harness sheaths based on their material category and aging grade evaluation results, and generates a test report.

2. The automotive wiring harness detection system based on artificial intelligence and machine vision as claimed in claim 1, characterized in that: The specific steps of generating diffraction signal data are as follows: Collect spectral reflection intensity data of the surface of the automotive wiring harness sheath, perform compression sampling, and generate compressed image data; The compressed image data is input into the diffraction response model to perform light field propagation calculations and generate diffraction signal data.

3. The automotive wiring harness detection system based on artificial intelligence and machine vision according to claim 1, characterized in that: The specific steps of constructing the diffraction response model are as follows: By measuring the curvature radius, inclination parameters and material refractive index parameters of the microlens array, combined with the geometric layout and optical path characteristics, a three-dimensional optical structure model is established and the focal length parameters and wavelength compensation coefficient are extracted; Based on the three-dimensional optical structure model, the Fresnel diffraction method is used to simulate the propagation and diffraction behavior of light waves to form a preliminary diffraction response function; The preliminary diffraction response function, focal length parameters and wavelength compensation coefficients are corrected and fitted to construct a diffraction response model.

4. The automotive wiring harness detection system based on artificial intelligence and machine vision as claimed in claim 1, characterized in that: The specific steps of obtaining the hyperspectral reflectance data are as follows: Filtering and normalizing the diffraction signal data to obtain pre-processed diffraction signal data; According to the pre-processed diffraction signal data, the basis functions are selected and the random sparse coefficient values ​​are initialized to construct a sparse representation model; The pre-processed diffraction signal data is iteratively solved through the sparse representation model to obtain preliminary hyperspectral reflectance data; The preliminary hyperspectral reflectance data are subjected to denoising enhancement and spectral correction processing to obtain hyperspectral reflectance data.

5. The automotive wiring harness detection system based on artificial intelligence and machine vision as claimed in claim 4, characterized in that: According to the pre-processed diffraction signal data, the basis function is selected and the random sparse coefficient value is initialized to construct the sparse representation model. The specific steps are as follows: Analyze the frequency domain and spatial domain distribution characteristics of the pre-processed diffraction signal data and extract the modeling data sample set; Selecting a sparse representation basis function set of diffraction signal characteristics from the modeling data sample set to obtain a candidate basis function library; Based on the candidate basis function library, the matching pursuit algorithm is used to perform preliminary fitting of the sparse representation of the preprocessed diffraction signal data to generate the initial coefficient matrix; The initial coefficient matrix is ​​optimized with sparse constraints, and the compressed sensing constraint method is used to iteratively converge to generate a sparse representation model.

6. The automotive wiring harness detection system based on artificial intelligence and machine vision as claimed in claim 1, characterized in that: The specific steps of forming the spectral characteristics are as follows: Converting the hyperspectral reflectance data into a two-dimensional matrix to form a hyperspectral data matrix; The hyperspectral data matrix is ​​decomposed using a low-rank sparse decomposition algorithm to obtain a low-rank component matrix and a sparse component matrix; Extract the principal component features reflecting the overall material properties of the automotive wiring harness sheath from the low-rank component matrix; Extract sparse features reflecting local detail changes from the sparse component matrix; The principal component features are fused with the sparse features to form spectral features.

7. The automotive wiring harness detection system based on artificial intelligence and machine vision as claimed in claim 1, characterized in that: The specific steps for identifying the material category of the automotive wiring harness sheath are as follows: Arrange the spectral features into vectors according to the intensity values ​​of each wavelength channel and map them to the multidimensional spectral space to generate a high-dimensional feature vector; Compare the high-dimensional feature vector with the spectral features of material categories in the spectral material database one by one to generate a material category matching score; Based on the material category matching score, the category label with the highest score is filtered out from the spectral material database, and the material category of the automotive wiring harness sheath is identified.

8. The automotive wiring harness detection system based on artificial intelligence and machine vision as claimed in claim 1, characterized in that: The spectral characteristics are combined with the material category of the automobile wiring harness sheath to identify the surface state of the automobile wiring harness sheath and perform quantitative analysis to obtain the aging grade evaluation result. The specific steps are as follows: The spectral features are fused with the material categories of the automotive wiring harness sheath to obtain fused features; Extracting spectral variation features reflecting surface states from fused features; Based on the spectral variation characteristics, the system can identify the oil stains, photooxidation aging and fading areas on the surface of the automotive wiring harness sheath, generate surface degradation characteristics, and perform quantitative processing to form quantitative analysis data. Based on the quantitative analysis data, a comprehensive evaluation is conducted on the surface degradation degree and spectral change characteristics of the automotive wiring harness sheath to obtain the aging grade evaluation results.

9. The automotive wiring harness detection system based on artificial intelligence and machine vision according to claim 8, characterized in that: According to the spectral change characteristics, the oil stains, photooxidation aging and fading areas on the surface of the automotive wiring harness sheath are identified to form surface degradation characteristics, and quantitative processing is performed to form quantitative analysis data. The specific steps are as follows: Analyze the time series and amplitude changes of the spectral change characteristics, extract the wavelength offset data and reflection intensity change data, and form the spectral anomaly fragments; Classify and identify abnormal spectral fragments and mark them as surface conditions of oil stains, photooxidation aging, and faded areas; The sliding window statistical method is used to analyze the time series changes of spectral anomaly fragments corresponding to different surface states and extract the spectral change trend characteristics; According to the spectral change trend characteristics, the spectral intensity change value, wavelength offset value and reflectivity gradient value are calculated, the surface degradation characteristics are generated and numerical standardization is performed to form quantitative analysis data.

10. The automotive wiring harness detection system based on artificial intelligence and machine vision according to claim 1, characterized in that: The specific steps for generating the test report are as follows: Combined analysis of automotive wiring harness sheath material categories and aging grade evaluation results to obtain a comprehensive quality score; Based on the comprehensive quality score and the grading threshold range, the overall quality level of the automotive wiring harness sheath is divided; Integrate the overall quality status grade with quantitative analysis data to generate a test report.

Images