Industrial scanning identification method and system suitable for high-reflection material

Abstract

The invention discloses an industrial scanning identification method and system suitable for a high-reflection material, and relates to the field of high-reflection material identification. The method comprises the following steps: performing surface scanning on a high-reflection material through industrial scanning equipment to obtain a radiance map and depth information of the surface of the material, and constructing a multi-modal data image; denoising the obtained multi-modal data image to obtain a denoised image; the quality of the denoised image is improved by estimating a non-uniform illumination field, and a normalized image is obtained; extracting features invariant to illumination and scale change according to the normalized image; and carrying out feature fusion according to the extracted features to obtain a fused feature vector, carrying out training through a supervised learning method, and carrying out scanning identification on the high-reflection material by using a trained model. Through multi-modal data fusion, illumination and scale invariant feature extraction and an efficient feature fusion and classification method, the precision, robustness and classification stability of high-reflection material identification are significantly improved.

Description

Technical Field

[0001] The present invention relates to the field of highly reflective material identification, and in particular to an industrial scanning and identification method and system suitable for highly reflective materials. Background Art

[0002] With the widespread use of highly reflective materials in industrial manufacturing processes, how to efficiently and accurately identify and classify highly reflective materials, especially in complex scanning environments, has become an important research topic. Highly reflective materials usually have very high reflectivity, which makes their surface easily interfered with by factors such as lighting, angle, and surface roughness during scanning, thereby affecting data acquisition and subsequent image processing. Traditional material recognition methods often rely on direct measurement of surface texture, color, or geometry. However, these methods perform poorly in the scanning process of highly reflective materials because the light reflection characteristics of highly reflective materials can lead to the loss or deformation of surface information, affecting the accuracy and stability of recognition.

[0003] To address the above issues, the present invention proposes an industrial scanning and identification method suitable for highly reflective materials. By acquiring multimodal data, including radiance maps and depth information, through high-precision scanning equipment, and combining techniques such as denoising, illumination field estimation, feature extraction, and fusion, the accuracy and robustness of highly reflective material identification can be effectively improved. The present method combines multiple data features to extract features that are invariant to changes in illumination and scale. These features are then trained using supervised learning methods to construct a highly reflective material identification model, ultimately achieving accurate classification and identification of highly reflective materials. Summary of the Invention

[0004] Based on the above-mentioned shortcomings of the prior art, the purpose of the present invention is to provide an industrial scanning and identification method and system suitable for highly reflective materials to solve the above-mentioned technical problems.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an industrial scanning and identification method suitable for highly reflective materials, comprising:

[0006] Use high-precision industrial scanning equipment to scan the surface of highly reflective materials, obtain the emissivity map and depth information of the surface of highly reflective materials, and construct a multimodal data image;

[0007] Denoising the acquired multimodal data image to obtain a denoised image;

[0008] The quality of the denoised image is improved by estimating the non-uniform illumination field and normalizing it to obtain a normalized image;

[0009] Extract features that are invariant to changes in illumination and scale based on the normalized image;

[0010] Feature fusion is performed based on the extracted features to obtain a fused feature vector. Based on the fused feature vector, a high-reflective material recognition model is constructed, which is trained through a supervised learning method, and the trained model is used to scan and identify high-reflective materials.

[0011] The present invention is further configured such that the surface of the highly reflective material is scanned by a high-precision industrial scanning device to obtain the emissivity map and depth information of the surface of the highly reflective material to form a multimodal data image, and the processing steps are as follows:

[0012] Get the radiance map by different exposure time t k Collect exposure image {I k}, synthesize the emissivity map L(x,y), the synthesis formula is: Among them, L(x,y) is the pixel value of the synthesized radiance map, the radiance at position (x,y), I k (x, y) is the pixel value of the k-th exposure image at position (x, y), t k is the exposure time of the kth exposure, w(I k (x,y)) is the weight function, which weights the different exposure images, and N is the number of exposures;

[0013] The depth information of the surface is obtained by laser scanning and triangulation. The depth formula is: Where d(x,y) represents the depth at the image pixel position (x,y), b is the baseline distance of the laser scanner, f is the focal length of the laser scanner, s(x,y) is the displacement of the laser spot at the corresponding position (x,y) in the image, and φ(x,y) is the surface reflection phase offset term.

[0014] By fusing the radiance map and depth information, a multimodal data image is obtained: M = [L, D], where M is the multimodal data image, L is the radiance map, and D is the depth information.

[0015] The present invention is further configured such that the denoising of the acquired multimodal data image to obtain the denoised image is performed in the following steps:

[0016] Construct a tensor diffusion equation to remove noise and preserve image structure;

[0017] The tensor diffusion equation is iteratively solved using the explicit difference method until convergence, and the denoised image M is obtained. ′ .

[0018] The present invention is further configured such that the denoised image quality is improved by estimating the non-uniform illumination field, and the processing steps are as follows:

[0019] Estimate the illumination field using a Gaussian smoothing filter: Among them, B(x,y) represents the estimated illumination field, and represents the image M ′ The part of each pixel affected by the lighting, is the Gaussian kernel, σ is the illumination estimation scale, which is determined by the width W and height H of the image. The calculation formula is:

[0020] From the denoised image M ′ Extract the reflection component R(x,y) from the equation: Among them, R(x,y) is the extracted reflection component, reflecting the surface reflection characteristics of each pixel in the image, and δ is the adjustment parameter;

[0021] The reflection component is enhanced by nonlinear mapping enhancement, and the formula is: Among them, γ is used to control the contrast enhancement intensity to obtain the normalized image after enhancement processing.

[0022] The present invention is further configured such that the features that are invariant to illumination and scale changes are extracted based on the normalized image, and the features include: multi-scale response features, phase consistency features, and dynamic surface reflection features.

[0023] The present invention is further configured as follows: the multi-scale and multi-directional wavelet transform generates a complex analytic wavelet basis, and the complex analytic wavelet basis and the normalized image are used to generate the complex analytic wavelet basis. Data extraction multi-scale response features;

[0024] Generate complex analytic wavelet basis, the formula is: Where s and θ represent scale and direction respectively;

[0025] The multi-scale response characteristic formula is: Among them, W f (s,θ,x,y) is the multi-scale response feature, K is the number of channels of the image, is the kth channel value of the normalized image, φ(x,y) is the surface reflection phase offset term, is the gradient information of the image, and β is the interaction coefficient between the reflective surface and the scanning device.

[0026] The present invention is further configured as follows: the multi-scale response feature and the normalized image The phase consistency feature is extracted from the data, and the formula is: Among them, W f (s k ,θ,x,y) represents the k and multi-scale response characteristics under direction θ, is the value of the first channel of the normalized image, and δ and ∈ are adjustment parameters.

[0027] The present invention is further configured such that the dynamic surface reflection feature is calculated based on the radiance map, the depth information and the normalized image data, and the dynamic surface reflection feature formula is: Among them, K is the number of channels, τ is the adjustment constant for controlling the depth, β k Parameters for dynamically adjusting the reflection characteristic response.

[0028] The present invention is further configured to perform feature fusion based on the extracted features to obtain a fused feature vector, use the fused feature vector as input, output the category of high-reflective material, construct a high-reflective material recognition model, use a supervised learning method to train the model, and after the training is completed, use the trained model to scan and identify high-reflective materials;

[0029] The multi-scale response features, phase consistency features and dynamic surface reflection features are sequentially connected and fused into a feature vector, F = [W1, W2, ..., W n ,H1,H2,...,H n ,T1,T2,...,T n ], where W i is the multi-scale response feature, H i is the phase-consistent feature, T i is the dynamic surface reflection feature;

[0030] Classification is performed by support vector machine SVM, given a training set Among them, F i is the eigenvector of the i-th sample, N is the number of samples, y i is the corresponding high reflective material category label;

[0031] Support vector machine SVM separates samples of different categories by finding a hyperplane to maximize the interval between different categories. The hyperplane is defined as: T F + b = 0, where w is the normal vector of the hyperplane, T represents the transpose operation of the matrix, F is the eigenvector, and b is the bias term;

[0032] Find the optimal hyperplane in high-dimensional space and use the kernel function to map the data to high-dimensional space. The kernel function is defined as: K(F i ,F j )=φ(F i ) T φ(F j ), φ(·) is the mapping function, which converts the input sample F i and F j Mapping from the original input space to the high-dimensional feature space, φ(Fj ) and φ(F i ) is the eigenvector after mapping, φ(F i ) T φ(F j ) represents the inner product of samples in high-dimensional space. The inner product value represents the similarity between them. The larger the inner product, the higher the similarity between samples.

[0033] The present invention also provides an industrial scanning and identification system suitable for highly reflective materials, the system comprising:

[0034] Constructing a multimodal data image module: Using high-precision industrial scanning equipment to scan the surface of highly reflective materials, obtain the emissivity map and depth information of the highly reflective material surface, and construct a multimodal data image;

[0035] Obtaining denoised image module: denoising the acquired multimodal data image to obtain a denoised image;

[0036] Obtaining normalized image module: improves the quality of denoised image by estimating the non-uniform illumination field, performs normalization, and obtains a normalized image;

[0037] Feature extraction module: extracts features that are invariant to illumination and scale changes based on the normalized image;

[0038] Scanning and recognition module: Feature fusion is performed based on the extracted features to obtain a fused feature vector. Based on the fused feature vector, a high-reflective material recognition model is constructed, which is trained through supervised learning methods, and the trained model is used to scan and recognize high-reflective materials.

[0039] The present invention provides an industrial scanning and identification method and system for highly reflective materials. The method scans the surface of the highly reflective material using high-precision industrial scanning equipment to obtain a radiometric map and depth information of the surface of the highly reflective material, and constructs a multimodal data image; denoises the obtained multimodal data image to obtain a denoised image; improves the quality of the denoised image by estimating a non-uniform illumination field, and performs normalization to obtain a normalized image; extracts features that are invariant to illumination and scale changes from the normalized image; performs feature fusion based on the extracted features to obtain a fused feature vector; constructs a high-reflective material identification model based on the fused feature vector, trains the model using a supervised learning method, and uses the trained model to scan and identify the highly reflective material. The beneficial effects produced include:

[0040] 1. Improved recognition accuracy of highly reflective materials: Through multimodal data fusion, including radiance maps, depth information, and denoised images, the surface characteristics of highly reflective materials can be fully and accurately reflected. Through processing steps such as denoising, illumination field estimation, and feature extraction, the interference of surface illumination and noise on the scanned data of highly reflective materials is effectively reduced, improving image quality and feature reliability, thereby enhancing the accuracy of highly reflective material recognition.

[0041] 2. Extract features that are invariant to changes in illumination and scale: We propose invariant features such as multi-scale response features, phase-consistent features, and dynamic surface reflection features, which can effectively cope with changes in illumination, scale, and angle. The extraction of these features is unaffected by scanning angle, surface reflection characteristics, and illumination changes, enabling the recognition model to operate stably under different conditions.

[0042] 3. Efficient feature fusion and classification: By fusing multiple features to generate a comprehensive feature vector, this method provides strong support for subsequent classification and identification. Combined with supervised learning methods such as support vector machines for training and classification, it can effectively distinguish between different types of highly reflective materials, with stable and accurate classification results.

[0043] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without inventive efforts. In the drawings:

[0045] Figure 1 This is a flow chart showing an industrial scanning and identification method applicable to highly reflective materials according to an exemplary embodiment of the present invention;

[0046] Figure 2 The figure is a schematic structural diagram of an industrial scanning and identification system suitable for highly reflective materials, showing an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following describes the embodiments of the present invention with reference to the accompanying drawings and preferred embodiments. Those skilled in the art will readily appreciate the other advantages and benefits of the present invention from the disclosure herein. The present invention may also be implemented or applied through various other specific embodiments, and the various details in this specification may be modified or altered based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are intended only to illustrate the present invention and are not intended to limit the scope of protection of the present invention.

[0048] It should be noted that the illustrations provided in the following embodiments are merely schematic illustrations of the basic concept of the present invention. Therefore, the illustrations only show components related to the present invention and are not drawn according to the number, shape, and size of components in actual implementation. In actual implementation, the type, quantity, and proportion of each component may be changed arbitrarily, and the component layout may also be more complex.

[0049] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it will be apparent to those skilled in the art that the embodiments of the present invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring the embodiments of the present invention.

[0050] Example 1

[0051] An industrial scanning and recognition method suitable for highly reflective materials, such as Figure 1 As shown, including:

[0052] Use high-precision industrial scanning equipment to scan the surface of highly reflective materials, obtain the emissivity map and depth information of the surface of highly reflective materials, and construct a multimodal data image;

[0053] Denoising the acquired multimodal data image to obtain a denoised image;

[0054] The quality of the denoised image is improved by estimating the non-uniform illumination field and normalizing it to obtain a normalized image;

[0055] Extract features that are invariant to changes in illumination and scale based on the normalized image;

[0056] Feature fusion is performed based on the extracted features to obtain a fused feature vector. Based on the fused feature vector, a high-reflective material recognition model is constructed, which is trained through a supervised learning method, and the trained model is used to scan and identify high-reflective materials.

[0057] The present invention is further configured such that the surface of the highly reflective material is scanned by a high-precision industrial scanning device to obtain the emissivity map and depth information of the surface of the highly reflective material to form a multimodal data image, and the processing steps are as follows:

[0058] Get the radiance map by different exposure time t k Collect exposure image {I k}, synthesize the emissivity map L(x,y), the synthesis formula is: Among them, L(x,y) is the pixel value of the synthesized radiance map, the radiance at position (x,y), I k (x, y) is the pixel value of the k-th exposure image at position (x, y), t k is the exposure time of the kth exposure, w(I k (x,y)) is a weight function, which is a normalized value ranging from 0 to 1. It weights the different exposure images. N is the number of exposures. The weighted synthesis of different exposure images can effectively eliminate the interference of illumination changes and reflection characteristics, obtain a more accurate radiance map, and improve the accuracy of radiance calculation under different lighting conditions.

[0059] The depth information of the surface is obtained by laser scanning and triangulation. The depth formula is: Where d(x,y) represents the depth at the image pixel position (x,y), that is, the actual distance from the pixel to the scanning device, b is the baseline distance of the laser scanner, f is the focal length of the laser scanner, s(x,y) is the displacement of the laser spot at the corresponding position (x,y) in the image, and φ(x,y) is the surface reflection phase offset term. It is affected by the surface material and surface texture. Laser scanning and triangulation technology can provide accurate depth information and is suitable for scanning fine structures on highly reflective surfaces. Considering the phase offset helps to accurately compensate for depth calculation errors caused by changes in surface material and illumination angle.

[0060] By fusing the radiance map and depth information, a multimodal data image is obtained: M = [L, D], where M is the multimodal data image, L is the radiance map, and D is the depth information. Multimodal data fusion can comprehensively reflect the illumination reflection characteristics and geometric shape characteristics of the object surface, providing rich multidimensional data for subsequent feature extraction and recognition. The fusion of the radiance map and depth information enables the recognition method to not only rely on the surface reflection characteristics, but also combine the geometric information of the object, thereby improving the recognition accuracy under different conditions.

[0061] The present invention is further configured such that the denoising of the acquired multimodal data image to obtain the denoised image is performed in the following steps:

[0062] Construct a tensor diffusion equation to remove noise and maintain the image structure. The tensor diffusion equation is expressed as: in, is the diffusion coefficient, defined as: Among them, κ is the gradient threshold, which is used to judge the noise, α is the nonlinear attenuation factor, which controls the diffusion nonlinearity, and div represents the divergence operation, which is used to calculate the diffusion of the tensor. Represents the gradient change of multimodal data image;

[0063] The tensor diffusion equation is iteratively solved using the explicit difference method until convergence, and the denoised image M is obtained. ′ , the formula is: ∥M (n+1) -M (n) ∥ F <∈, ∈ is the convergence criterion. The explicit difference method is a numerical solution method used to solve partial differential equations. It obtains the numerical solution of the equation by performing iterative calculations on the discretized space grid and time step.

[0064] Through the above steps, image noise is gradually eliminated while preserving the image's edges and texture. Ultimately, after multiple iterations, the image reaches a convergence state, resulting in a denoised image. The tensor diffusion equation adaptively adjusts the diffusion process based on the image's local structure, avoiding blurring of important edge and texture information. Compared to traditional denoising methods, such as mean filtering, tensor diffusion effectively preserves image detail and avoids oversmoothing. Through iterative solutions using explicit differencing, noise can be gradually removed, making the denoising process more refined and ultimately achieving a high-quality denoised image.

[0065] The present invention is further configured such that the denoised image quality is improved by estimating the non-uniform illumination field, and the processing steps are as follows:

[0066] In surface scanning of highly reflective materials, the lighting conditions are usually non-uniform, resulting in some areas of the image being over-illuminated or dimmed. These lighting differences will affect the image quality. In order to improve image quality, it is necessary to estimate the impact of non-uniform lighting and use Gaussian smoothing filtering to estimate the lighting field: Among them, B(x,y) represents the estimated illumination field, and represents the image M ′ The part of each pixel affected by the lighting, is the Gaussian kernel, σ is the illumination estimation scale, which is determined by the width W and height H of the image. The calculation formula is: Gaussian smoothing filter is used to remove high-frequency noise or texture in the image while retaining the lower-frequency structural features. For the denoised image M ′Smoothing is performed to estimate the part affected by illumination in the image. The scale parameter σ of the Gaussian kernel controls the intensity of the smoothing. The larger the value, the stronger the smoothing effect.

[0067] The reflection component in the denoised image is affected by the illumination, which leads to inaccurate extraction of the reflection information. In order to accurately extract the reflection component, the estimated illumination field B(x,y) is used to correct the illumination effect, thereby restoring the pure reflection information. ′ Extract the reflection component R(x,y) from the equation: Among them, R(x,y) is the extracted reflection component, which represents the surface reflection part in the image and reflects the surface reflection characteristics of each pixel in the image. δ is the adjustment parameter;

[0068] After the reflection component is extracted, due to the non-uniformity of illumination, the contrast of the reflection component is low and needs to be enhanced to better highlight the surface details. The reflection component is enhanced by nonlinear mapping enhancement. The formula is: Among them, γ is used to control the contrast enhancement intensity to obtain the normalized image after enhancement processing. After the nonlinear mapping enhancement processing of the reflected component, the obtained image has higher contrast in visual effect, the surface features are more prominent, and the image quality is significantly improved.

[0069] The present invention is further configured such that the features that are invariant to illumination and scale changes are extracted based on the normalized image, and the features include: multi-scale response features, phase consistency features, and dynamic surface reflection features.

[0070] The present invention is further configured as follows: the multi-scale and multi-directional wavelet transform generates a complex analytic wavelet basis, and the complex analytic wavelet basis and the normalized image are used to generate the complex analytic wavelet basis. Data extraction multi-scale response features;

[0071] The complex analytic wavelet basis can simultaneously capture the details and structural changes in the image. For different scales and directions, the complex analytic wavelet basis will generate multiple different waveforms, thereby performing multi-scale and multi-directional local analysis of the image. In this way, multiple layers of information with different scale and direction characteristics can be extracted, which helps to capture different details of the image and generate a complex analytic wavelet basis. The formula is: Among them, s and θ represent scale and direction respectively. s is the scale factor, which indicates the degree of scaling of the wavelet basis. It is a positive number and determines the degree of refinement of the transformation. When the scale is small, the resolution of the wavelet basis is higher and it can capture more details. θ is the direction angle, which indicates the rotation angle of the wavelet basis. It usually takes a value in the range of [0,2π] to control the directionality of the wavelet basis. ψ is the original wavelet function, which defines the basic form of the wavelet basis. s -1 is the scale normalization factor, used to adjust the scale size;

[0072] By using complex analytical wavelet basis for multi-scale and multi-directional analysis, different levels and details of the image can be captured. The response of each scale and direction will give a local feature map, which is synthesized to form a multi-scale response feature. To enhance the extraction effect, the response feature is also weighted with the surface reflection phase offset term and the gradient information of the image. The formula for the multi-scale response feature is: Among them, W f (s,θ,x,y) is a multi-scale response feature, which represents the response of the image at position (x,y), scale s, and direction θ. K is the number of channels of the image. is the kth channel value of the normalized image, φ(x,y) is the surface reflection phase offset term, is the gradient information of the image, β is the interaction coefficient between the reflective surface and the scanning device, It is the square of the image gradient, which indicates the degree of change of the image at that position.

[0073] The present invention is further configured as follows: the multi-scale response feature and the normalized image Data extraction phase consistency features, phase consistency is the consistency of local features in the image, especially in the edge or object contour area. In image processing and feature extraction, phase consistency is used to describe the consistency of the structure in the image, especially whether the response of the local structure remains consistent at different scales and directions. By combining multi-scale response features with normalized image data, the phase consistency features of the image can be extracted. By weighted summation of the multi-scale response features and weighting the first channel of the normalized image, the phase consistency of the image at different scales and directions can be calculated. This feature helps to better describe the surface characteristics of the object, especially in the surface analysis of highly reflective materials. The formula is: Among them, H(x,y,k) is the extracted phase consistency feature, which represents the phase consistency of the image at position (x,y) and channel k, and W f (s k ,θ,x,y) represents the k and multi-scale response characteristics under direction θ, is the value of the first channel of the normalized image, δ and ∈ are adjustment parameters, ∑ s,θ (|W f (s,θ,x,y)| 2 ) is the weighted sum of multiscale response features across all scales and directions, reflecting the overall response strength of the image at different scales and directions. By combining multiscale response features with normalized image data, more accurate phase consistency features can be extracted from the image. This helps to more stably identify object outlines, edges, or surface features in the image, especially in surface analysis of highly reflective materials, reducing the impact of illumination changes.

[0074] The present invention is further configured such that the dynamic surface reflection feature is calculated based on the radiance map, the depth information and the normalized image data, and the dynamic surface reflection feature formula is: Among them, K is the number of channels, τ is the adjustment constant for controlling the depth, β k To dynamically adjust the parameters of the reflectance feature response, T(x,y) represents the dynamic surface reflectance feature, defining the reflection intensity and characteristics at position (x,y). D0(x,y) is the reference depth value. By introducing depth information and a depth adjustment constant τ, this method can dynamically adjust the response of the surface reflectance feature to better adapt to surface reflectance characteristics under different depth conditions. The sensitivity of the reflectance feature is adjusted as the depth changes, avoiding the excessive influence of depth information on the reflectance feature calculation. Combining the dynamic surface reflectance features of the radiance map, depth information, and normalized image data provides a comprehensive, dynamic reflectance model, enabling a more refined description of complex surfaces. It can provide more accurate reflection intensity estimates for different depths and surface morphologies.

[0075] The present invention is further configured to perform feature fusion based on the extracted features to obtain a fused feature vector, use the fused feature vector as input, output the category of high-reflective material, construct a high-reflective material recognition model, use a supervised learning method to train the model, and after the training is completed, use the trained model to scan and identify high-reflective materials;

[0076] The multi-scale response features, phase consistency features and dynamic surface reflection features are sequentially connected and fused into a feature vector, F = [W1, W2, ..., W n ,H1,H2,...,H n ,T1,T2,...,T n ], where W i is the multi-scale response feature, H i is the phase-consistent feature, T i is the dynamic surface reflection feature;

[0077] Classification is performed by support vector machine SVM, given a training set Among them, F i is the eigenvector of the i-th sample, N is the number of samples, y i is the corresponding high reflective material category label;

[0078] Support vector machine SVM separates samples of different categories by finding a hyperplane to maximize the interval between different categories. The hyperplane is defined as: T F + b = 0, where w is the normal vector of the hyperplane, T represents the transpose operation of the matrix, F is the eigenvector, and b is the bias term;

[0079] The training of SVM can be solved by the following optimization problem: So that the constraints are satisfied: y i (w T F i +b)≥1, Among them, y i is a label representing different categories. Through this optimization method, SVM will try to maximize the interval between different categories, thereby improving classification accuracy;

[0080] Find the optimal hyperplane in high-dimensional space and use the kernel function to map the data to high-dimensional space. The kernel function is defined as: K(F i ,F j )=φ(F i ) T φ(F j ), φ(·) is the mapping function, which converts the input sample F i and F j Mapping from the original input space to the high-dimensional feature space, φ(F j ) and φ(F i ) is the eigenvector after mapping, φ(F i ) T φ(F j ) represents the inner product of samples in high-dimensional space. The inner product value represents the similarity between them. The larger the inner product, the higher the similarity between samples.

[0081] Post-processing further optimizes the classification results to achieve higher recognition accuracy and robustness. Geometric shape adjustment uses geometric shape features to fine-tune the recognition results. Image segmentation uses image segmentation methods such as threshold segmentation, region growing, or deep learning-based segmentation methods to improve the accuracy of object extraction.

[0082] By extracting multi-scale response features, phase consistency features, and dynamic surface reflectance features, and combining them with support vector machines for classification and recognition, efficient and accurate object recognition can be achieved. The support vector machine ensures good classification performance by maximizing class margins, while the use of kernel functions enables effective classification in higher-dimensional spaces. Post-processing further optimizes recognition results, making the system more stable and accurate in practical applications.

[0083] Example 2

[0084] See also Figure 2 , the exemplary industrial scanning and identification system applicable to highly reflective materials includes:

[0085] Constructing a multimodal data image module: Using high-precision industrial scanning equipment to scan the surface of highly reflective materials, obtain the emissivity map and depth information of the highly reflective material surface, and construct a multimodal data image;

[0086] Obtaining denoised image module: denoising the acquired multimodal data image to obtain a denoised image;

[0087] Obtaining normalized image module: improves the quality of denoised image by estimating the non-uniform illumination field, performs normalization, and obtains a normalized image;

[0088] Feature extraction module: extracts features that are invariant to illumination and scale changes based on the normalized image;

[0089] Scanning and recognition module: Feature fusion is performed based on the extracted features to obtain a fused feature vector. Based on the fused feature vector, a high-reflective material recognition model is constructed, which is trained through supervised learning methods, and the trained model is used to scan and recognize high-reflective materials.

[0090] It should be noted that the industrial scanning and identification system for highly reflective materials provided in the above embodiment and the industrial scanning and identification method for highly reflective materials provided in the above embodiment are based on the same concept, and the specific manner in which each module and unit performs operations has been described in detail in the method embodiment and will not be repeated here. In actual applications, the industrial scanning and identification system for highly reflective materials provided in the above embodiment can allocate the above functions to different functional modules as needed, that is, divide the internal structure of the system into different functional modules to complete all or part of the functions described above, and this is not limited here.

[0091] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0092] It should be understood that the term "and / or" as used herein simply describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. A and B can be singular or plural. Furthermore, the character " / " as used herein generally indicates an "or" relationship between the associated objects, but it may also indicate an "and / or" relationship. For specific understanding, please refer to the context.

[0093] In this application, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or plural.

[0094] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0095] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0096] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0097] In the several embodiments provided in this application, it should be understood that the disclosed system can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0098] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0099] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0100] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0101] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. An industrial scanning and identification method suitable for highly reflective materials, characterized in that: include: Use high-precision industrial scanning equipment to scan the surface of highly reflective materials, obtain the emissivity map and depth information of the surface of highly reflective materials, and construct a multimodal data image; Denoising the acquired multimodal data image to obtain a denoised image; The quality of the denoised image is improved by estimating the non-uniform illumination field and normalizing it to obtain a normalized image; Extract features that are invariant to changes in illumination and scale based on the normalized image; Feature fusion is performed based on the extracted features to obtain a fused feature vector. Based on the fused feature vector, a high-reflective material recognition model is constructed, which is trained through a supervised learning method, and the trained model is used to scan and identify high-reflective materials.

2. The industrial scanning and identification method for highly reflective materials according to claim 1, characterized in that: Use high-precision industrial scanning equipment to scan the surface of highly reflective materials, obtain the emissivity map and depth information of the highly reflective material surface, and form a multimodal data image. The processing steps are as follows: Get the radiance map by different exposure time t k Collect exposure image {I k }, synthesize the emissivity map L(x,y), the synthesis formula is: Among them, L(x,y) is the pixel value of the synthesized radiance map, the radiance at position (x,y), I k (x, y) is the pixel value of the k-th exposure image at position (x, y), t k is the exposure time of the kth exposure, w(I k (x,y)) is the weight function, which weights the different exposure images, and N is the number of exposures; The depth information of the surface is obtained by laser scanning and triangulation. The depth formula is: Where d(x,y) represents the depth at the image pixel position (x,y), b is the baseline distance of the laser scanner, f is the focal length of the laser scanner, s(x,y) is the displacement of the laser spot at the corresponding position (x,y) in the image, and φ(x,y) is the surface reflection phase offset term. By fusing the radiance map and depth information, a multimodal data image is obtained: M = [L, D], where M is the multimodal data image, L is the radiance map, and D is the depth information.

3. The industrial scanning and identification method for highly reflective materials according to claim 2, characterized in that: Denoise the acquired multimodal data image to obtain a denoised image. The processing steps are as follows: Construct a tensor diffusion equation to remove noise and preserve image structure; The tensor diffusion equation is iteratively solved using the explicit difference method until convergence, and the denoised image M is obtained. ′ .

4. The industrial scanning and identification method for highly reflective materials according to claim 3, characterized in that: The quality of denoised images is improved by estimating the non-uniform illumination field. The processing steps are as follows: Estimate the illumination field using a Gaussian smoothing filter: Among them, B(x,y) represents the estimated illumination field, and represents the image M ′ The part of each pixel affected by the lighting, is the Gaussian kernel, σ is the illumination estimation scale, which is determined by the width W and height H of the image. The calculation formula is: From the denoised image M ′ Extract the reflection component R(x,y) from the equation: Among them, R(x,y) is the extracted reflection component, reflecting the surface reflection characteristics of each pixel in the image, and δ is the adjustment parameter; The reflection component is enhanced by nonlinear mapping enhancement, and the formula is: Among them, γ is used to control the contrast enhancement intensity to obtain the normalized image after enhancement processing.

5. The industrial scanning and identification method for highly reflective materials according to claim 1, characterized in that: Based on the normalized image, features that are invariant to illumination and scale changes are extracted. The features include: multi-scale response features, phase consistency features, and dynamic surface reflection features.

6. The industrial scanning and identification method for highly reflective materials according to claim 5, characterized in that: Through multi-scale and multi-directional wavelet transform, complex analytical wavelet basis is generated. According to the complex analytical wavelet basis and normalized image Data extraction multi-scale response features; Generate complex analytic wavelet basis, the formula is: Where s and θ represent scale and direction respectively; The multi-scale response characteristic formula is: Among them, W f (s,θ,x,y) is the multi-scale response feature, K is the number of channels of the image, is the kth channel value of the normalized image, φ(x,y) is the surface reflection phase offset term, (x, y, k) is the gradient information of the image, and β is the interaction coefficient between the reflective surface and the scanning device.

7. The industrial scanning and identification method for highly reflective materials according to claim 5, characterized in that: Based on multi-scale response features and normalized image The phase consistency feature is extracted from the data, and the formula is: Among them, W f (s k ,θ,x,y) represents the k and multi-scale response characteristics under direction θ, is the value of the first channel of the normalized image, and δ and ∈ are adjustment parameters.

8. The industrial scanning and identification method for highly reflective materials according to claim 5, characterized in that: The dynamic surface reflection feature is calculated based on the radiance map, depth information and normalized image data. The dynamic surface reflection feature formula is: Among them, K is the number of channels, τ is the adjustment constant for controlling the depth, β k Parameters for dynamically adjusting the reflection characteristic response.

9. The industrial scanning and identification method applicable to highly reflective materials according to claim 5, characterized in that: Perform feature fusion based on the extracted features to obtain a fused feature vector. Use the fused feature vector as input to output the category of high-reflective materials, build a high-reflective material recognition model, and use supervised learning methods to train the model. After the training is completed, use the trained model to scan and identify high-reflective materials. The multi-scale response features, phase consistency features and dynamic surface reflection features are sequentially connected and fused into a feature vector, F = [W1, W2, ..., W n ,H1,H2,...,H n ,T1,T2,...,T n ], where W i is the multi-scale response feature, H i is the phase-consistent feature, T i is the dynamic surface reflection feature; Classification is performed by support vector machine SVM, given a training set Among them, F i is the eigenvector of the i-th sample, N is the number of samples, y i is the corresponding high reflective material category label; Support vector machine SVM separates samples of different categories by finding a hyperplane to maximize the interval between different categories. The hyperplane is defined as: T F + b = 0, where w is the normal vector of the hyperplane, T represents the transpose operation of the matrix, F is the eigenvector, and b is the bias term; Find the optimal hyperplane in high-dimensional space and use the kernel function to map the data to high-dimensional space. The kernel function is defined as: K(F i ,F j )=φ(F i ) T φ(F j ), φ(·) is the mapping function, which converts the input sample F i and F j Mapping from the original input space to the high-dimensional feature space, φ(F j ) and φ(F i ) is the eigenvector after mapping, φ(F i ) T φ(F j ) represents the inner product of samples in high-dimensional space. The inner product value represents the similarity between them. The larger the inner product, the higher the similarity between samples.

10. An industrial scanning and identification system for highly reflective materials, used to implement the industrial scanning and identification method for highly reflective materials according to any one of claims 1 to 9, characterized in that: include: Constructing a multimodal data image module: Using high-precision industrial scanning equipment to scan the surface of highly reflective materials, obtain the emissivity map and depth information of the highly reflective material surface, and construct a multimodal data image; Obtaining denoised image module: denoising the acquired multimodal data image to obtain a denoised image; Obtaining normalized image module: improves the quality of denoised image by estimating the non-uniform illumination field, performs normalization, and obtains a normalized image; Feature extraction module: extracts features that are invariant to illumination and scale changes based on the normalized image; Scanning and recognition module: Feature fusion is performed based on the extracted features to obtain a fused feature vector. Based on the fused feature vector, a high-reflective material recognition model is constructed, which is trained through supervised learning methods, and the trained model is used to scan and recognize high-reflective materials.

Images