Method and system for extracting anti-counterfeiting features of automobile parts

Abstract

The present invention discloses a method and system for extracting anti-counterfeiting features of automobile parts. The method comprises the following steps: judging the texture type of a current texture based on two-dimensional frequency domain features; matching a corresponding analysis algorithm according to the texture type of the current texture to determine key attribute feature points; determining a microscopic visual feature recognition model for automobile parts based on the key attribute feature points and the training of a deep learning neural network framework; matching and classifying the microscopic visual feature recognition model for automobile parts with a priori feature library to determine the authenticity of automobile parts under anti-counterfeiting identification. At this time, judging the texture type of the current texture based on two-dimensional frequency domain features is convenient for matching the corresponding analysis algorithm for the texture type of the current texture, thereby being compatible with the analysis of continuous, discontinuous and contour texture features, and then utilizing artificial intelligence automatic matching technology to extract and identify anti-counterfeiting features of automobile parts based on microscopic visual image processing and a neural network hybrid algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of extracting anti-counterfeiting features of automobile parts, and in particular to a method and system for extracting anti-counterfeiting features of automobile parts. Background Art

[0002] With the development of science and technology, auto parts, as a part of the car, need to be identified when assembling the car to define the authenticity of the auto parts. In the prior art, by analyzing the characters printed on the surface of the parts, according to the detection needs, a set of auto parts character detection systems based on machine vision are designed to meet the company's automated production requirements. However, the prior art is not suitable for continuous and discontinuous texture feature extraction on the surface of various auto parts materials. Summary of the invention

[0003] The purpose of the present invention is to overcome the shortcomings of the prior art. The present invention provides a method and system for extracting anti-counterfeiting features of automobile parts. The texture type of the current texture is judged based on two-dimensional frequency domain features, so as to match the corresponding analysis algorithm for the texture type of the current texture, thereby being compatible with the analysis of continuous, discontinuous and contour texture features, and then utilizing artificial intelligence automatic matching technology to extract and identify anti-counterfeiting features of automobile parts based on microscopic visual image processing and neural network hybrid algorithms, so as to determine the authenticity of automobile parts under anti-counterfeiting identification.

[0004] In order to solve the above technical problems, an embodiment of the present invention provides a method for extracting anti-counterfeiting features of automobile parts, which is applied to the scenario of extracting anti-counterfeiting features of automobile parts; the method for extracting anti-counterfeiting features of automobile parts comprises:

[0005] Collect microscopic visual feature images of automotive parts;

[0006] Outputting two-dimensional frequency domain features according to image processing of microscopic visual feature images of automobile parts;

[0007] Determine the texture type of the current texture based on two-dimensional frequency domain features;

[0008] Match the corresponding analysis algorithm according to the texture type of the current texture to determine the key attribute feature points;

[0009] Determine the microscopic visual feature recognition model of automotive parts based on key attribute feature points and the training of deep learning neural network framework;

[0010] Based on the micro-visual feature recognition model of automobile parts and the prior feature library, matching and classification are carried out to determine the authenticity of automobile parts under anti-counterfeiting identification.

[0011] Optionally, the collecting of microscopic visual feature images of automobile parts includes:

[0012] Select the category of auto parts to be identified;

[0013] Based on the auto parts category, collect the auto parts products of this category that are mixed with authenticity;

[0014] The surface image of the genuine auto parts is observed through a microscopic electron microscope, and the microscopic visual feature image is recorded.

[0015] Optionally, the outputting of two-dimensional frequency domain features according to image processing of microscopic visual feature images of automobile parts includes:

[0016] Freeze the microscopic visual feature images of automotive parts;

[0017] Perform microscopic image segmentation processing on the microscopic visual feature images of automobile parts and form multiple regions;

[0018] Perform regional operations based on multiple regions and output two-dimensional frequency domain features.

[0019] Optionally, the outputting of two-dimensional frequency domain features according to image processing of microscopic visual feature images of automobile parts also includes:

[0020] Before filtering the microscopic visual feature images of automotive parts, the areas with obvious material property characteristics are first normalized;

[0021] The microscopic visual feature images of automobile parts are processed in a block-by-block manner, and I(i,j) is defined as the grayscale value at pixel (i,j). i , η i Block S i The mean and variance of i (x, y) is the normalized grayscale value at pixel (x, y). i All pixels in the normalized image are defined as:

[0022]

[0023] A0, η0 are the expected mean and variance, respectively. Normalization is performed uniformly on all pixels, so it does not change the texture structure of the microscopic visual feature image of automobile parts;

[0024] A 16×16 sub-block micro-feature image is selected from the micro-visual feature image of automobile parts. The mathematical expression of the two-dimensional Gabor function in the spatial domain is as follows:

[0025]

[0026] Among them, λ represents the wavelength, its value is greater than or equal to 2, in pixels, and cannot be greater than one-fifth of the input image size; θ represents the direction, with a value of 0 to 360 degrees, specifying the direction of the parallel stripes of the Gabor function; γ represents the aspect ratio (spatial aspect ratio), which determines the ellipticity of the Gabor function curve shape, and its value is 0.5; when γ=1, the curve shape is circular; when γ<1, its shape is continuously stretched along the parallel stripe direction; σ represents the standard deviation of the Gaussian part, and its value cannot be directly given, and is restricted by the center frequency of the filter; x, y represent the spatial position coordinates of the image pixel points;

[0027] Then the feature map of the microscopic visual feature image of automobile parts after image filtering and enhancement is expressed as:

[0028] H(x,y)=ζ i (x,y)*G(x,y).

[0029] Optionally, judging the texture type of the current texture based on the two-dimensional frequency domain feature includes:

[0030] Collect two-dimensional frequency domain features;

[0031] In the two-dimensional frequency domain features, Fourier spectrum texture analysis converts the image from the spatial domain to the frequency domain through Fourier transform to obtain the frequency distribution of the microscopic visual feature image of automobile parts;

[0032] Defining texture features of the microscopic visual feature images of automobile parts according to the frequency distribution of the microscopic visual feature images of automobile parts;

[0033] The texture type of the current texture is defined according to the texture features of the microscopic visual feature image of the automobile parts.

[0034] Optionally, matching a corresponding analysis algorithm according to the texture type of the current texture to determine key attribute feature points includes:

[0035] Collect the texture type of the current texture;

[0036] Match the corresponding analysis algorithm based on the texture type of the current texture;

[0037] In the corresponding analysis algorithm, depth calculation is performed based on the current texture, and key attribute feature points are determined.

[0038] Optionally, matching a corresponding analysis algorithm according to the texture type of the current texture to determine the key attribute feature points further includes:

[0039] For continuous textures, the Poincare index algorithm is used;

[0040] For non-continuous textures, the gray level co-occurrence matrix (GLCM) method is used to extract and analyze texture features;

[0041] For contour texture, Freeman chain code and its derivative feature extraction algorithm are used.

[0042] Optionally, the method of determining the microscopic visual feature recognition model of automobile parts based on key attribute feature points and training of a deep learning neural network framework includes:

[0043] Determine the key attribute feature points;

[0044] Associate key attribute feature points with the deep learning neural network framework and trigger corresponding training;

[0045] A microscopic visual feature recognition model for automotive parts is determined based on training.

[0046] Optionally, the matching and classification based on the microscopic visual feature recognition model of automobile parts and the prior feature library to determine the authenticity of automobile parts under anti-counterfeiting recognition includes:

[0047] Collect microscopic visual feature recognition models for automotive parts;

[0048] Associating the microscopic visual feature recognition model of automobile parts with the prior feature library;

[0049] Each image to be verified in the prior feature library is input into the microscopic visual feature recognition model of automobile parts to determine the authenticity of automobile parts under anti-counterfeiting recognition.

[0050] Optionally, a system for extracting anti-counterfeiting features of automobile parts is provided, wherein the system for extracting anti-counterfeiting features of automobile parts is applied to the above-mentioned method for extracting anti-counterfeiting features of automobile parts, and the system for extracting anti-counterfeiting features of automobile parts comprises:

[0051] An acquisition module is used to acquire microscopic visual feature images of automotive parts;

[0052] A two-dimensional frequency domain feature module, used for outputting two-dimensional frequency domain features according to image processing of microscopic visual feature images of automobile parts;

[0053] A texture type module, used to determine the texture type of the current texture based on two-dimensional frequency domain features;

[0054] A key attribute feature point module is used to match the corresponding analysis algorithm according to the texture type of the current texture to determine the key attribute feature points;

[0055] The recognition module is used to determine the microscopic visual feature recognition model of automotive parts based on key attribute feature points and the training of the deep learning neural network framework;

[0056] The classification module is used to match and classify the microscopic visual feature recognition model of automobile parts with the prior feature library to determine the authenticity of automobile parts under anti-counterfeiting identification.

[0057] In an embodiment of the present invention, through the method in the embodiment of the present invention, a microscopic visual feature image of an automobile part is collected; two-dimensional frequency domain features are output based on image processing of the microscopic visual feature image of the automobile part; the texture type of the current texture is judged based on the two-dimensional frequency domain features; the corresponding analysis algorithm is matched according to the texture type of the current texture to determine the key attribute feature points; the microscopic visual feature recognition model of the automobile part is determined based on the key attribute feature points and the training of the deep learning neural network framework; matching and classification are performed based on the microscopic visual feature recognition model of the automobile part and the prior feature library to determine the authenticity of the automobile part under anti-counterfeiting identification. At this time, the texture type of the current texture is judged based on the two-dimensional frequency domain features, so as to match the corresponding analysis algorithm for the texture type of the current texture, so as to be compatible with the analysis of continuous, discontinuous and contour texture features, and then the anti-counterfeiting feature extraction and identification of automobile parts are performed based on the microscopic visual image processing and neural network hybrid algorithm using artificial intelligence automatic matching technology, so as to determine the authenticity of automobile parts under anti-counterfeiting identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. 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 creative work.

[0059] Figure 1 is a schematic flow chart of a method for extracting anti-counterfeiting features of automobile parts in an embodiment of the present invention;

[0060] Figure 2 is a schematic flow chart of S11 in the method for extracting anti-counterfeiting features of automobile parts in an embodiment of the present invention;

[0061] Figure 3 is a schematic flow chart of S12 in the method for extracting anti-counterfeiting features of automobile parts in an embodiment of the present invention;

[0062] Figure 4 is a schematic flow chart of S13 in the method for extracting anti-counterfeiting features of automobile parts in an embodiment of the present invention;

[0063] Figure 5 is a schematic flow chart of S14 in the method for extracting anti-counterfeiting features of automobile parts in an embodiment of the present invention;

[0064] Figure 6 is a schematic flow chart of S15 in the method for extracting anti-counterfeiting features of automobile parts in an embodiment of the present invention;

[0065] Figure 7 is a schematic flow chart of S16 in the method for extracting anti-counterfeiting features of automobile parts in an embodiment of the present invention;

[0066] Figure 8 Schematic diagram of the structure of the system for extracting anti-counterfeiting features of automobile parts in an embodiment of the present invention;

[0067] Fig. 9 The figure is a hardware diagram of an electronic device according to an exemplary embodiment. DETAILED DESCRIPTION

[0068] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0069] Example

[0070] See also Figures 1 to 9 , a method for extracting anti-counterfeiting features of automobile parts, applied to the scene of extracting anti-counterfeiting features of automobile parts; the method for extracting anti-counterfeiting features of automobile parts comprises:

[0071] Step S11: collecting microscopic visual feature images of automobile parts;

[0072] Step S12: outputting two-dimensional frequency domain features according to image processing of the microscopic visual feature images of automobile parts;

[0073] Step S13: determining the texture type of the current texture based on the two-dimensional frequency domain features;

[0074] Step S14: matching a corresponding analysis algorithm according to the texture type of the current texture to determine key attribute feature points;

[0075] Step S15: Determine the microscopic visual feature recognition model of automobile parts based on the key attribute feature points and the training of the deep learning neural network framework;

[0076] Step S16: Matching and classifying based on the microscopic visual feature recognition model of automobile parts and components and the prior feature library to determine the authenticity of automobile parts and components under anti-counterfeiting identification.

[0077] In an embodiment of the present invention, through the method in the embodiment of the present invention, the texture type of the current texture is judged based on the two-dimensional frequency domain features, so as to match the corresponding analysis algorithm for the texture type of the current texture, thereby being compatible with the analysis of continuous, discontinuous and contour texture features, and then using artificial intelligence automatic matching technology based on microscopic visual image processing and neural network hybrid algorithm to extract and identify anti-counterfeiting features of automobile parts, so as to determine the authenticity of automobile parts under anti-counterfeiting identification.

[0078] In step S11, a microscopic visual feature image of an automobile part is collected;

[0079] In the specific implementation process of the present invention, the specific steps may be:

[0080] S111: Select the category of auto parts to be identified;

[0081] S112: Based on the auto parts category, collect the auto parts products of this category that are mixed with authenticity;

[0082] S113: Observe the surface image of the genuine automobile parts through a micro electron microscope and record the micro visual feature image.

[0083] In an embodiment of the present application, an automobile parts category to be identified is selected to facilitate the management and control of automobile parts within the automobile parts category, thereby collecting mixed authentic and fake automobile parts products of this category based on the automobile parts category, and then observing the surface images of the authentic automobile parts products through a microscopic electron microscope, and recording the microscopic visual feature images.

[0084] At this point, first select the category of auto parts to be identified, collect a certain number of mixed genuine and fake auto parts products of this category, and distinguish and classify the genuine and fake products. Observe the surface images of the genuine and fake auto parts products through a microscopic electron microscope, and record the application type, part material classification, total product sample size, product production date and observed microscopic feature images of each genuine and fake auto parts product to prepare a microscopic feature image data set for the microscopic feature image preprocessing module described later.

[0085] In step S12, two-dimensional frequency domain features are output according to image processing of the microscopic visual feature image of the automobile parts;

[0086] In the specific implementation process of the present invention, the specific steps may be:

[0087] S121: Freeze the microscopic visual feature images of automotive parts;

[0088] S122: performing microscopic image segmentation processing on the microscopic visual feature image of the automobile parts and forming a plurality of regions;

[0089] S123: performing regional operations according to the multiple regions, and outputting two-dimensional frequency domain features.

[0090] In an embodiment of the present application, a microscopic visual feature image of an automobile part is frozen to facilitate image processing of the microscopic visual feature image of the automobile part. At this time, microscopic image segmentation processing is performed on the microscopic visual feature image of the automobile part to form a plurality of regions; regional operations are performed based on the plurality of regions, and two-dimensional frequency domain features are output, thereby ensuring the accuracy of the two-dimensional frequency domain features.

[0091] At this time, the microscopic visual feature image is subjected to image segmentation, the microscopic visual feature image direction map is obtained, the microscopic visual feature image is enhanced and filtered, and the microscopic visual feature image is binarized and refined. Since there is a probability that the background area exists in the region of the microscopic visual feature image of the automotive parts, which affects the recognition accuracy of the microscopic visual feature image of the automotive parts, the background area can be segmented and eliminated by image segmentation processing, and the observation noise generated by external influencing factors caused by observation tools such as microscopic electron microscopes is reduced. The influence on the recognition of the microscopic visual feature image of the automotive parts is reduced. Then its direction map is obtained. Since the microscopic visual feature image of the automotive parts is a multi-edge directed graph with a certain texture direction, the present invention utilizes the texture and directionality of the microscopic visual feature image of the automotive parts, and calculates the direction field as its direction map. Finally, a two-dimensional Gabor filter is used to enhance and filter the microscopic visual feature image of the automotive parts, remove image noise, make the characteristic texture of the microscopic image have obvious gray value differences, and retain the characteristic information of the obvious surface texture of the automotive parts material.

[0092] Specifically, the microscopic visual feature image of the automobile parts is subjected to microscopic image segmentation. The microscopic visual feature image f(i, j) of the automobile parts is stable in a small local area and can be described by the following model:

[0093] f(i,j)=m 1f (i,j)+δ f (i,j)w(i,j)

[0094] In the formula, m 1f (i,j),δ f (i,j) are the local means of (i,j) in a small neighborhood of point (i,j), and the variance w(i,j) is a Gaussian process with zero mean and unit variance.

[0095] For a pixel point (i, j) with a neighborhood size of (2Q+1)×(2Q+1):

[0096]

[0097] In the formula, m 2f 、c 2f They are the second-order moment and second-order cumulant at point (i, j) when τ = 0.

[0098]

[0099] m 3f 、c 3f They are the third-order moment and third-order cumulant at point (i, j) when τ1=τ2=0 respectively.

[0100] When Q=4, 8×8 blocks are taken from the foreground and background areas of the microscopic visual feature image of automobile parts. By comparing the grayscale histogram, the grayscale of the foreground area of ​​the microscopic visual feature image of automobile parts is mainly distributed at the two ends of black (grayscale value 0) and white (grayscale value 255), showing a regular distribution with a large variance. The grayscale of the background area of ​​the microscopic visual feature image of automobile parts is mainly composed of noise, which can approximately satisfy the Gaussian distribution. By calculating the third-order cumulant, an appropriate threshold T is set according to the actual image grayscale value parameters. When c 3f >T, the block is the background area of ​​the microscopic visual feature image, and this part of the area is cut off.

[0101] Obtain the directional map of the microscopic visual feature image of the automobile parts. I(i,j) represents the grayscale value of the microscopic visual feature image of the automobile parts at (i,j). Divide the image into small blocks of size M×M and large blocks of size N×N. For the values ​​of M and N, The gradient Dx and Dy of each pixel I(i,j) in the large block on the x-axis and y-axis are calculated based on the Sobel operator.

[0102] D x (i,j)=[I(i-1,j-1)+2I(i-1,j)+I(i-1,j+1)]

[0103] -[I(i+1,j-1)+2I(i+1,j)+I(i+1,j+1)]

[0104] D y (i,j)=[I(i-1,j-1)+2I(i,j-1)+I(i+1,j-1)]

[0105] -[I(i-1,j+1)+2I(i,j+1)+I(i+1,j+1)]

[0106] Calculate the direction θ(i,j) of the large block centered at pixel (i,j).

[0107]

[0108] D x ≠0 and D y ≠0

[0109] When the number of zero Dx or Dy in a large block reaches a certain level, such as more than 80% of the total number of pixels in the large block, θ(i,j) should be directly set to 0 or π. The direction θ of the N×N large block is saved as the direction of the M×M small block, and the direction of the next block is calculated in a loop.

[0110] The microscopic visual feature images of automotive parts are enhanced and filtered. A two-dimensional Gabor filter with good directional and frequency selectivity is used to remove image noise and retain the texture structure of the microscopic image. Before filtering the microscopic visual feature images of automotive parts, the areas with obvious material property characteristics are normalized, and their mean and variance are fixed to constants. Normalization can eliminate sensor noise and gray value differences caused by different pressures.

[0111] In addition, before filtering the microscopic visual feature images of automotive parts, the regions with obvious material property characteristics are normalized;

[0112] The microscopic visual feature images of automobile parts are processed in a block-by-block manner, and I(i,j) is defined as the grayscale value at pixel (i,j). i , η i Block S i The mean and variance of i (x, y) is the normalized grayscale value at pixel (x, y). i All pixels in the normalized image are defined as:

[0113]

[0114] A0, η0 are the expected mean and variance, respectively. Normalization is performed uniformly on all pixels, so it does not change the texture structure of the microscopic visual feature image of automobile parts;

[0115] A 16×16 sub-block micro-feature image is selected from the micro-visual feature image of automobile parts. The mathematical expression of the two-dimensional Gabor function in the spatial domain is as follows:

[0116]

[0117] Among them, λ represents the wavelength, its value is greater than or equal to 2, in pixels, and cannot be greater than one-fifth of the input image size; θ represents the direction, with a value of 0 to 360 degrees, specifying the direction of the parallel stripes of the Gabor function; γ represents the aspect ratio (spatial aspect ratio), which determines the ellipticity of the Gabor function curve shape, and its value is 0.5; when γ=1, the curve shape is circular; when γ<1, its shape is continuously stretched along the parallel stripe direction; σ represents the standard deviation of the Gaussian part, and its value cannot be directly given, and is restricted by the center frequency of the filter; x, y represent the spatial position coordinates of the image pixel points;

[0118] Then the feature map of the microscopic visual feature image of automobile parts after image filtering and enhancement is expressed as:

[0119] H(x,y)=ζ i (x,y)*G(x,y).

[0120] In step S13, the texture type of the current texture is determined based on the two-dimensional frequency domain features;

[0121] In the specific implementation process of the present invention, the specific steps may be:

[0122] S131: collecting two-dimensional frequency domain features;

[0123] S132: In the two-dimensional frequency domain features, Fourier spectrum texture analysis converts the image from the spatial domain to the frequency domain through Fourier transform to obtain the frequency distribution of the microscopic visual feature image of the automobile parts;

[0124] S133: defining texture features of the microscopic visual feature image of the automobile part according to the frequency distribution of the microscopic visual feature image of the automobile part;

[0125] S134: defining a texture type of the current texture according to texture features of the microscopic visual feature image of the automobile parts.

[0126] In an embodiment of the present application, two-dimensional frequency domain features are collected and image processing is performed on the two-dimensional frequency domain features. At this time, Fourier spectral texture analysis converts the image from the spatial domain to the frequency domain through Fourier transform to obtain the frequency distribution of the microscopic visual feature image of the automobile parts; the texture features of the microscopic visual feature image of the automobile parts are defined according to the frequency distribution of the microscopic visual feature image of the automobile parts, so as to define the texture type of the current texture according to the texture features of the microscopic visual feature image of the automobile parts.

[0127] Specifically, the microscopic visual feature image of automobile parts is transformed from the spatial domain to the frequency domain. For the microscopic visual feature image of M×N automobile parts, if the texture feature of the microscopic visual feature image belongs to two-dimensional discrete data, the process of Fourier transform is:

[0128]

[0129] The microscopic visual feature image of automobile parts is transformed from the spatial domain to the frequency domain. If the texture features of the microscopic visual feature image of automobile parts belong to two-dimensional continuous data, the process of Fourier transform is:

[0130]

[0131] At this time, the texture features of the microscopic visual feature images of automobile parts are judged and classified as continuous or discontinuous in the two-dimensional frequency domain. Fourier spectrum texture analysis converts the image from the spatial domain to the frequency domain through Fourier transform, thereby obtaining the frequency distribution of the microscopic visual feature images of automobile parts. By analyzing the frequency distribution, the distribution of different frequency components in the microscopic visual feature images of automobile parts can be understood, thereby revealing the texture features of the microscopic visual feature images of automobile parts and correctly classifying them.

[0132] S14: matching a corresponding analysis algorithm according to the texture type of the current texture to determine key attribute feature points;

[0133] In the specific implementation process of the present invention, the specific steps may be:

[0134] S141: Collect the texture type of the current texture;

[0135] S142: Matching a corresponding analysis algorithm based on the texture type of the current texture;

[0136] S143: In the corresponding analysis algorithm, depth calculation is performed based on the current texture, and key attribute feature points are determined.

[0137] In an embodiment of the present application, the texture type of the current texture is collected so as to match the corresponding analysis algorithm based on the texture type of the current texture, so as to match the corresponding analysis algorithm for the texture type of the current texture, so as to be compatible with the analysis of continuous, discontinuous and contour texture features, and then use artificial intelligence automatic matching technology based on microscopic visual image processing and neural network hybrid algorithm to extract and identify anti-counterfeiting features of automobile parts, so as to determine the authenticity of automobile parts under anti-counterfeiting identification.

[0138] At this time, the Poincare index algorithm is used for continuous textures;

[0139] For continuous stripes, the Poincare index algorithm is used. A data point is randomly selected as the initial center point. The Poincare distance in non-Euclidean space is used in high-dimensional space to measure the distance between data points, and the geometric center of the data set is approximated in an iterative manner. The key attribute points of the microscopic visual feature images of different automotive parts have different Poincare indices. First, the position of the data point in the embedded space is randomly initialized, and then the position of the data point in the embedded space is adjusted by minimizing the Poincare distance so that the data point can better represent the relationship between the original data in the low-dimensional space. Finally, the optimization steps are iterated until the stopping condition is met (such as reaching the maximum number of iterations or error convergence). Since the block calculation method is used in the calculation of the texture direction field of the microscopic visual feature image, the 8-neighborhood area of ​​the corresponding block is taken near the key attribute point in the texture image direction field. The direction field of the microscopic visual feature image of the automotive parts is preprocessed to obtain θ(α), that is, α=β1,β2,β i …β8 (i=1,2,…8), and perform deep directional field normalization optimization processing on it, and its Poincare index is calculated as follows:

[0140]

[0141] Among them, γ and λ are learnable parameters, and ε is a value close to 0.

[0142]

[0143] α′=(α+1)mod8,α=β1,β2,β i …β8(i=1,2,…8)

[0144] Wherein, Dx and Dy are the gradients Dx and Dy based on the Sobel operator on the x-axis and y-axis of each pixel I(i,j) in the block.

[0145] In the part of the technology for identifying genuine and fake parts of microscopic visual feature images of automobile parts provided by the present invention, in the calculation of the Poincare index for feature extraction of microscopic visual feature images with continuous texture, θ(α) is obtained by taking the 8-neighborhood area of ​​the corresponding block near the key attribute point in the direction field of the texture image, and preprocessing the direction field of the microscopic visual feature image of the automobile parts, that is, α = β1, β2, β i …β8(i=1,2,…8), and perform deep directional field normalization optimization processing on it, and iteratively calculate the Poincare index of different key attribute points.

[0146] For non-continuous textures, the gray level co-occurrence matrix (GLCM) method is used to extract and analyze texture features;

[0147] The grayscale co-occurrence matrix algorithm used for discontinuous stripes uses the statistical information of the grayscale level of pixels in the microscopic image, describes the texture features of the microscopic image by calculating the spatial relationship between pixels, and distinguishes image areas with different textures. First, a grayscale offset direction is defined, which is horizontal, vertical, 45 degrees and 135 degrees. Then, for each pixel, the frequency of occurrence of the grayscale value pairs of the pixels in the specified offset direction adjacent to it is calculated. Finally, a grayscale co-occurrence matrix, namely GLCM, is constructed based on the calculation results. The rows and columns of the matrix represent the grayscale level, and each element in the matrix represents the frequency of occurrence of the corresponding grayscale level pair. Then, the similarity of the grayscale co-occurrence matrix is ​​calculated to study the similarity of the microscopic visual feature images of automotive parts. Two images are selected, defined as the target image F and the source image G, whose grayscale levels are both M, and the grayscale co-occurrence matrices are f and g respectively. The similarity value of the two is S. The base similarity D(f,g) is calculated based on the Bhattacharyya distance, and then the microscopic image feature similarity mixed reference factor Δ(f,g) of the energy and contrast of the grayscale co-occurrence matrix is ​​added. The calculation formula is as follows:

[0148]

[0149] In the formula, S represents the texture similarity between the target image and the source image. This similarity can quantify the similarity of the grayscale co-occurrence matrix of the two images. If S = 0, it means that the texture features of the target and the background are completely different; if S = 1, it means that the texture features of the target and the background are completely consistent; the larger S is, the smaller the difference between the target and the source texture features is, indicating that the texture similarity between the target image and the source feature image is better. e is the energy of the grayscale co-occurrence matrix, which reflects the uniformity of the grayscale distribution and the texture coarseness of the microscopic visual feature image. c is the contrast of the grayscale co-occurrence matrix, which reflects the depth of the image texture grooves.

[0150] In the similarity calculation of the comparison of microscopic visual feature images with discontinuous textures, the similarity of the grayscale co-occurrence matrix of the microscopic visual feature images of automobile parts is calculated. Two microscopic visual feature images of automobile parts are selected, defined as the target image F and the source image G, whose grayscale levels are both M, and the grayscale co-occurrence matrices are f and g respectively. The similarity value of the two is S. The base similarity D(f,g) is calculated based on the Bhattacharyya distance, and then the energy of the grayscale co-occurrence matrix and the contrast of the microscopic image feature similarity mixed reference factor Δ(f,g) are added to calculate the similarity S of the two microscopic visual feature images of automobile parts.

[0151] For contour texture, Freeman chain code and its derivative feature extraction algorithm are used. At this time, Freeman chain code and its derivative feature extraction algorithm are used for contour texture to perform feature extraction processing of microscopic images. The point sequence on the contour of the microscopic image is represented by a series of digital codes, each number represents the moving direction from the current point to the next point, and 8 possible moving directions are represented by numbers from 0 to 7. The derivative feature extraction algorithm is a feature extraction method based on Freeman chain code. By processing the Freeman chain code sequence, more distinguishing feature information can be extracted. First, the contour is extracted to obtain the point sequence on the contour. The Freeman chain code is generated according to the point sequence on the contour. The moving direction of each point is encoded as a number from 0 to 7. The derivative operation is performed on the generated Freeman chain code sequence, and the difference between two adjacent numbers is calculated to obtain the derivative sequence. Then the derivative sequence is further processed, such as smoothing, normalization, etc. Finally, the extracted derivative features are used to perform tasks such as texture recognition and shape matching of microscopic visual feature images.

[0152] S15: Determine the microscopic visual feature recognition model of automotive parts based on key attribute feature points and the training of deep learning neural network framework;

[0153] In the specific implementation process of the present invention, the specific steps may be:

[0154] S151: freeze key attribute feature points;

[0155] S152: Associating key attribute feature points with a deep learning neural network framework and triggering corresponding training;

[0156] S153: Determine a microscopic visual feature recognition model for automotive parts based on training.

[0157] S16: Matching and classification based on the microscopic visual feature recognition model of automobile parts and the prior feature library to determine the authenticity of automobile parts under anti-counterfeiting identification;

[0158] In the specific implementation process of the present invention, the specific steps may be:

[0159] S161: Acquisition of microscopic visual feature recognition model for automotive parts;

[0160] S162: Associating the microscopic visual feature recognition model of automobile parts with the prior feature library;

[0161] S163: Each image to be verified in the prior feature library is input into the microscopic visual feature recognition model of automobile parts to determine the authenticity of the automobile parts under anti-counterfeiting recognition.

[0162] In an embodiment of the present application, key attribute feature points are frozen, and the key attribute feature points are associated with a deep learning neural network framework, and corresponding training is triggered; based on the training, a micro-visual feature recognition model for automotive parts is determined. At this time, a deep learning neural network framework CNN is built to learn and train the key attribute points of the micro-visual feature images of the above-mentioned automotive parts and the corresponding feature extraction algorithms, and a micro-visual feature recognition model for automotive parts is built, which is matched and classified with a priori feature library, and finally the authenticity result of anti-counterfeiting identification of automotive parts is obtained.

[0163] In an embodiment of the present invention, through the method in the embodiment of the present invention, a microscopic visual feature image of an automobile part is collected; two-dimensional frequency domain features are output based on image processing of the microscopic visual feature image of the automobile part; the texture type of the current texture is judged based on the two-dimensional frequency domain features; the corresponding analysis algorithm is matched according to the texture type of the current texture to determine the key attribute feature points; the microscopic visual feature recognition model of the automobile part is determined based on the key attribute feature points and the training of the deep learning neural network framework; matching and classification are performed based on the microscopic visual feature recognition model of the automobile part and the prior feature library to determine the authenticity of the automobile part under anti-counterfeiting identification. At this time, the texture type of the current texture is judged based on the two-dimensional frequency domain features, so as to match the corresponding analysis algorithm for the texture type of the current texture, so as to be compatible with the analysis of continuous, discontinuous and contour texture features, and then the anti-counterfeiting feature extraction and identification of automobile parts are performed based on the microscopic visual image processing and neural network hybrid algorithm using artificial intelligence automatic matching technology, so as to determine the authenticity of automobile parts under anti-counterfeiting identification.

[0164] The automatic matching technology of anti-counterfeiting feature extraction algorithm for automobile parts provided by the present invention can automatically match the texture features of different microscopic visual feature images of various genuine and fake automobile parts with a texture feature extraction algorithm that is more appropriately matched to the corresponding texture type.

[0165] In the part of the technology for identifying genuine and fake parts of microscopic visual feature images of automobile parts provided by the present invention, in the calculation of the Poincare index for feature extraction of microscopic visual feature images with continuous texture, θ(α) is obtained by taking the 8-neighborhood area of ​​the corresponding block near the key attribute point in the direction field of the texture image, and preprocessing the direction field of the microscopic visual feature image of the automobile parts, that is, α = β1, β2, β i …β8(i=1,2,…8), and perform deep directional field normalization optimization processing on it, and iteratively calculate the Poincare index of different key attribute points.

[0166] The technical part of the genuine parts identification of microscopic visual feature images of automobile parts provided by the present invention calculates the similarity of the grayscale co-occurrence matrix of the microscopic visual feature images of automobile parts in the similarity calculation of the comparison of microscopic visual feature images with discontinuous textures, selects two microscopic visual feature images of automobile parts, which are defined as the target image F and the source image G, and their grayscale levels are both M. The grayscale co-occurrence matrices are f and g respectively, and the similarity value of the two is S. The base similarity D(f,g) is calculated based on the Bhattacharyya distance, and then the microscopic image feature similarity mixed reference factor Δ(f,g) of the energy and contrast of the grayscale co-occurrence matrix is ​​added to calculate the similarity S of the two microscopic visual feature images of automobile parts.

[0167] Example

[0168] See also Figure 8 , Figure 8 It is a schematic diagram of the structural composition of the system for extracting anti-counterfeiting features of automobile parts in an embodiment of the present invention.

[0169] like Figure 8 As shown, a system for extracting anti-counterfeiting features of automobile parts, the system for extracting anti-counterfeiting features of automobile parts comprises:

[0170] The acquisition module 21 is used to acquire microscopic visual feature images of automobile parts;

[0171] A two-dimensional frequency domain feature module 22, used to output two-dimensional frequency domain features according to image processing of microscopic visual feature images of automobile parts;

[0172] A texture type module 23, used to determine the texture type of the current texture based on the two-dimensional frequency domain features;

[0173] A key attribute feature point module 24 is used to match a corresponding analysis algorithm according to the texture type of the current texture to determine the key attribute feature points;

[0174] Recognition module 25, used to determine the microscopic visual feature recognition model of automobile parts based on key attribute feature points and the training of deep learning neural network framework;

[0175] The classification module 26 is used to perform matching and classification based on the microscopic visual feature recognition model of automobile parts and components and the prior feature library, so as to determine the authenticity of automobile parts and components under anti-counterfeiting recognition.

[0176] Example

[0177] See also Fig. 9 , refer to the following Fig. 9 The electronic device 40 according to this embodiment of the present invention will be described. Fig. 9The electronic device 40 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0178] like Fig. 9 As shown, the electronic device 40 is in the form of a general computing device. The components of the electronic device 40 may include but are not limited to: at least one processing unit 41, at least one storage unit 42, and a bus 43 connecting different system components (including the storage unit 42 and the processing unit 41).

[0179] The storage unit stores program codes, which can be executed by the processing unit 41, so that the processing unit 41 executes the steps according to various exemplary embodiments of the present invention described in the above “Embodiment Method” section of this specification.

[0180] The storage unit 42 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit 421 and / or a cache storage unit 422 , and may further include a read-only storage unit (ROM) 423 .

[0181] The storage unit 42 may also include a program / utility 424 having a set (at least one) of program modules 425, such program modules 425 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0182] Bus 43 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0183] The electronic device 40 may also communicate with one or more external devices (e.g., keyboards, pointing devices, Bluetooth devices, etc.), one or more devices that enable a user to interact with the electronic device 40, and / or any device that enables the electronic device 40 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 44. Furthermore, the electronic device 40 may also communicate with one or more networks (e.g., local area networks, wide area networks, and / or public networks, such as the Internet) via a network adapter 45. Fig. 9 As shown, the network adapter 45 communicates with other modules of the electronic device 40 via the bus 43. It should be understood that although Fig. 9Not shown, other hardware and / or software modules may be used in conjunction with the electronic device 40, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, tape drives, and data backup planning systems.

[0184] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0185] Those skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium can include: a read-only memory, a random access memory, a disk or an optical disk, etc. In addition, it stores computer program instructions, and when the computer program instructions are executed by a computer, the computer executes the above method.

[0186] In addition, the above method and system for extracting anti-counterfeiting features of automotive parts provided in the embodiments of the present invention are introduced in detail. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for extracting anti-counterfeiting features of automobile parts, characterized in that: The method is applied to extracting anti-counterfeiting features of automobile parts. The method comprises: Collect microscopic visual feature images of auto parts; select the category of auto parts to be identified; collect mixed genuine and fake auto parts products of this category based on the category of auto parts; observe the surface images of the genuine and fake auto parts products through a microscopic electron microscope and record the microscopic visual feature images; Outputting two-dimensional frequency domain features according to image processing of microscopic visual feature images of automobile parts; freezing the microscopic visual feature images of automobile parts; performing microscopic image segmentation processing on the microscopic visual feature images of automobile parts and forming a plurality of regions; performing regional operations according to the plurality of regions and outputting two-dimensional frequency domain features; Before filtering the microscopic visual feature images of automotive parts, the areas with obvious material property characteristics are first normalized; The microscopic visual feature images of automobile parts are processed in a block-by-block manner, and I(i,j) is defined as the grayscale value at pixel (i,j). i , η i Block S i The mean and variance of i (x, y) is the normalized grayscale value at pixel (x, y). i All pixels in the normalized image are defined as: A0, η0 are the expected mean and variance, respectively, and normalization is performed uniformly on all pixels; A 16×16 sub-block micro-feature image is selected from the micro-visual feature image of automobile parts, and a two-dimensional Gabor filter is used to perform image enhancement and filtering on the micro-visual feature image of automobile parts. The mathematical expression of the two-dimensional Gabor function in the spatial domain is as follows: Among them, λ represents the wavelength, its value is greater than or equal to 2, in pixels, and cannot be greater than one-fifth of the input image size; θ represents the direction, with a value of 0 to 360 degrees, specifying the direction of the parallel stripes of the Gabor function; γ represents the aspect ratio, which determines the ellipticity of the Gabor function curve shape, and its value is 0.5; when γ=1, the curve shape is circular; when γ<1, its shape is continuously stretched along the direction of the parallel stripes; σ represents the standard deviation of the Gaussian part, and its value cannot be directly given, which is restricted by the center frequency of the filter; x, y represent the spatial position coordinates of the image pixel points; Perform microscopic image enhancement and filtering on the microscopic visual feature images of automotive parts; Then the feature map of the microscopic visual feature image of automobile parts after image filtering and enhancement is expressed as: H(x,y)=ζ i (x,y)*G(x,y); Determine the texture type of the current texture based on two-dimensional frequency domain features; Match the corresponding analysis algorithm according to the texture type of the current texture to determine the key attribute feature points; Freeze key attribute feature points; associate key attribute feature points with the deep learning neural network framework and trigger corresponding training; determine the microscopic visual feature recognition model of automotive parts based on the training; Based on the micro-visual feature recognition model of automobile parts and the prior feature library, matching and classification are carried out to determine the authenticity of automobile parts under anti-counterfeiting identification.

2. The method for extracting anti-counterfeiting features of automobile parts according to claim 1, characterized in that: The determining the texture type of the current texture based on the two-dimensional frequency domain feature includes: Collect two-dimensional frequency domain features; In the two-dimensional frequency domain features, Fourier spectrum texture analysis converts the image from the spatial domain to the frequency domain through Fourier transform to obtain the frequency distribution of the microscopic visual feature image of automobile parts; Defining texture features of the microscopic visual feature images of automobile parts according to the frequency distribution of the microscopic visual feature images of automobile parts; The texture type of the current texture is defined according to the texture features of the microscopic visual feature image of the automobile parts.

3. The method for extracting anti-counterfeiting features of automobile parts according to claim 2, characterized in that: The matching of the corresponding analysis algorithm according to the texture type of the current texture to determine the key attribute feature points includes: Collect the texture type of the current texture; Match the corresponding analysis algorithm based on the texture type of the current texture; In the corresponding analysis algorithm, depth calculation is performed based on the current texture, and key attribute feature points are determined.

4. The method for extracting anti-counterfeiting features of automobile parts according to claim 3, characterized in that: The matching of the corresponding analysis algorithm according to the texture type of the current texture to determine the key attribute feature points also includes: For continuous textures, the Poincare index algorithm is used; For non-continuous textures, the gray level co-occurrence matrix (GLCM) method is used to extract and analyze texture features; For contour texture, Freeman chain code and its derivative feature extraction algorithm are used.

5. The method for extracting anti-counterfeiting features of automobile parts according to claim 4, characterized in that: The matching and classification based on the microscopic visual feature recognition model of automobile parts and the prior feature library to determine the authenticity of automobile parts under anti-counterfeiting recognition includes: Collect microscopic visual feature recognition models for automotive parts; Associating the microscopic visual feature recognition model of automobile parts with the prior feature library; Each image to be verified in the prior feature library is input into the microscopic visual feature recognition model of automobile parts to determine the authenticity of automobile parts under anti-counterfeiting recognition.

6. A system for extracting anti-counterfeiting features of automobile parts, characterized in that: The system for extracting anti-counterfeiting features of automobile parts is applied to the method for extracting anti-counterfeiting features of automobile parts as claimed in any one of claims 1 to 5, and the system for extracting anti-counterfeiting features of automobile parts comprises: The acquisition module is used to acquire microscopic visual feature images of auto parts; select the category of auto parts to be identified; collect the mixed genuine and fake auto parts products of this category based on the category of auto parts; observe the surface images of the genuine and fake auto parts products through a microscopic electron microscope and record the microscopic visual feature images; A two-dimensional frequency domain feature module is used to output two-dimensional frequency domain features according to image processing of microscopic visual feature images of automobile parts; freeze the microscopic visual feature images of automobile parts; perform microscopic image segmentation processing on the microscopic visual feature images of automobile parts and form multiple regions; perform regional operations according to the multiple regions and output two-dimensional frequency domain features; Before filtering the microscopic visual feature images of automotive parts, the areas with obvious material property characteristics are first normalized; The microscopic visual feature images of automobile parts are processed in a block-by-block manner, and I(i,j) is defined as the grayscale value at pixel (i,j). i , η i Block S i The mean and variance of i (x, y) is the normalized grayscale value at pixel (x, y). i All pixels in the normalized image are defined as: A0, η0 are the expected mean and variance, respectively, and normalization is performed uniformly on all pixels; A 16×16 sub-block micro-feature image is selected from the micro-visual feature image of automobile parts, and a two-dimensional Gabor filter is used to perform image enhancement and filtering on the micro-visual feature image of automobile parts. The mathematical expression of the two-dimensional Gabor function in the spatial domain is as follows: Among them, λ represents the wavelength, its value is greater than or equal to 2, in pixels, and cannot be greater than one-fifth of the input image size; θ represents the direction, with a value of 0 to 360 degrees, specifying the direction of the parallel stripes of the Gabor function; γ represents the aspect ratio, which determines the ellipticity of the Gabor function curve shape, and its value is 0.5; when γ=1, the curve shape is circular; when γ<1, its shape is continuously stretched along the direction of the parallel stripes; σ represents the standard deviation of the Gaussian part, and its value cannot be directly given, which is restricted by the center frequency of the filter; x, y represent the spatial position coordinates of the image pixel points; Perform microscopic image enhancement and filtering on the microscopic visual feature images of automotive parts; Then the feature map of the microscopic visual feature image of automobile parts after image filtering and enhancement is expressed as: H(x,y)=ζ i (x,y)*G(x,y); A texture type module, used to determine the texture type of the current texture based on two-dimensional frequency domain features; A key attribute feature point module is used to match the corresponding analysis algorithm according to the texture type of the current texture to determine the key attribute feature points; The recognition module is used to freeze the key attribute feature points; associate the key attribute feature points with the deep learning neural network framework and trigger the corresponding training; determine the microscopic visual feature recognition model of automobile parts based on the training; The classification module is used to match and classify the microscopic visual feature recognition model of automobile parts with the prior feature library to determine the authenticity of automobile parts under anti-counterfeiting identification.

Images