ETFE film light transmittance evaluation method, device and storage medium

The texture characteristics of the ETFE film are extracted through multi-band spectroscopy and imaging technology, combined with feature dimensionality reduction and fusion technology, and the gradient enhancement decision tree model is used to solve the accuracy and efficiency of the light transmittance evaluation of ETFE films in the existing technology, achieving the correctness and efficiency of the ETFE film. Fine evaluation and efficient evaluation of light transmission performance of ETFE membrane.

CN118711178BActive Publication Date: 2024-11-05深圳市烨兴智能空间技术有限公司
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202411189035.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-28
Publication Date
2024-11-05
Estimated Expiration
2044-08-28

AI Technical Summary

Technical Problem

The existing ETFE membrane light transmittance evaluation methods have problems such as insufficient feature extraction and serious loss of dimensionality reduction information, and it is difficult to achieve rapid and accurate evaluation of the structure of large-area ETFE membranes.

Method used

Using multi-band spectral scanning and imaging technology, texture features are extracted through quaternion discrete cosine transformation and binary processing, combined with local binary mode and entropy calculation, feature dimensionality reduction and fusion are performed, and the pre-trained gradient-enhancing decision tree evaluation model is evaluated. Perform light transmission evaluation.

Benefits of technology

The accuracy and generalization ability of ETFE membrane light transmittance evaluation are improved, and the refined evaluation of light transmittance performance is achieved, which enhances feature expression ability and prediction accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118711178B_ABST
    Figure CN118711178B_ABST
Patent Text Reader

Abstract

The present application relates to the technical field of image processing, and discloses an ETFE film transmittance evaluation method, device and storage medium. The method comprises: performing multi-band spectral scanning and transmittance analysis on a plurality of sample ETFE films to obtain a standardized transmittance vector; performing multi-spectral imaging on the ETFE film to be tested to obtain a multi-band image, and performing registration and preprocessing to obtain a preprocessed image and a target frequency domain coefficient; performing quaternion discrete cosine transform and binarization processing to obtain a target binary feature matrix, and performing local binary pattern and entropy calculation to obtain a texture entropy feature; performing feature dimension reduction to obtain a reduced dimension feature vector, and performing feature fusion on the reduced dimension feature vector and the standardized transmittance vector to obtain a target fused feature vector; and inputting the target fused feature vector into a pre-trained gradient boosting decision tree evaluation model to perform transmittance evaluation to obtain a transmittance evaluation result of the ETFE film to be tested, thereby improving the accuracy of the ETFE film transmittance evaluation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to an ETFE film transmittance evaluation method, device and storage medium. Background Art

[0002] As a new type of building material, ETFE membrane is widely used in modern architecture due to its light weight, high light transmittance and durability. However, the light transmittance of ETFE membrane is affected by many factors during actual use, such as environmental pollution, aging and mechanical stress, which leads to a gradual decline in its light transmittance. The traditional ETFE membrane light transmittance evaluation method mainly relies on laboratory testing, which is not only time-consuming and labor-intensive, but also difficult to achieve rapid and accurate evaluation of large-area ETFE membrane structures.

[0003] Existing evaluation methods often have problems such as insufficient feature extraction and serious information loss in the dimensionality reduction process when processing multispectral image data, which affects the accuracy and reliability of the evaluation results. In addition, how to effectively integrate multiple feature information and build an efficient evaluation model is also a challenge faced by current research. Summary of the invention

[0004] The present application provides an ETFE film transmittance evaluation method, device and storage medium, thereby improving the accuracy of ETFE film transmittance evaluation.

[0005] The first aspect of the present application provides a method for evaluating the light transmittance of an ETFE film, the method comprising:

[0006] Perform multi-band spectral scanning and transmittance analysis on multiple sample ETFE films to obtain standardized transmittance vectors;

[0007] Performing multispectral imaging on the ETFE film to be tested to obtain a multi-band image, and registering and preprocessing the multi-band image to obtain a preprocessed image and a target frequency domain coefficient;

[0008] Performing quaternion discrete cosine transformation and binarization processing on the target frequency domain coefficients to obtain a target binary feature matrix, and performing local binary pattern and entropy calculation on the preprocessed image to obtain texture entropy features;

[0009] Performing feature dimensionality reduction on the target binary feature matrix and the texture entropy feature to obtain a reduced-dimensional feature vector, and performing feature fusion on the reduced-dimensional feature vector and the normalized transmittance vector to obtain a target fused feature vector;

[0010] The target fusion feature vector is input into a pre-trained gradient boosting decision tree evaluation model to perform light transmittance evaluation, thereby obtaining a light transmittance evaluation result of the ETFE film to be tested.

[0011] The second aspect of the present application provides an ETFE film light transmittance evaluation device, the ETFE film light transmittance evaluation device comprising:

[0012] An analysis module is used to perform multi-band spectral scanning and transmittance analysis on multiple sample ETFE films to obtain a standardized transmittance vector;

[0013] An imaging module is used to perform multispectral imaging on the ETFE film to be tested to obtain a multi-band image, and to register and preprocess the multi-band image to obtain a preprocessed image and a target frequency domain coefficient;

[0014] A calculation module, used for performing quaternion discrete cosine transform and binarization processing on the target frequency domain coefficients to obtain a target binary feature matrix, and performing local binary pattern and entropy calculation on the preprocessed image to obtain texture entropy features;

[0015] A fusion module is used to perform feature dimension reduction on the target binary feature matrix and the texture entropy feature to obtain a reduced dimension feature vector, and perform feature fusion on the reduced dimension feature vector and the normalized transmittance vector to obtain a target fused feature vector;

[0016] The evaluation module is used to input the target fusion feature vector into a pre-trained gradient boosting decision tree evaluation model to perform light transmittance evaluation to obtain a light transmittance evaluation result of the ETFE film to be tested.

[0017] The third aspect of the present application provides an electronic device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the electronic device executes the above-mentioned ETFE film transmittance evaluation method.

[0018] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned method for evaluating light transmittance of an ETFE film.

[0019] Compared with the prior art, the present application has the following beneficial effects: by performing multi-band spectral scanning and standardization processing on multiple sample ETFE films, a standardized transmittance vector is obtained; a multi-band image of the ETFE film to be tested is obtained by using multi-spectral imaging technology, and registration and preprocessing are performed to effectively extract the spectral feature information of the ETFE film; a quaternion discrete cosine transform and binarization processing are performed on the frequency domain coefficients to obtain a binary feature matrix, which effectively captures the texture and structural features of the ETFE film and enhances the feature expression capability. The local binary pattern and entropy calculation are performed on the preprocessed image to obtain the texture entropy feature, further extract the microstructure information of the ETFE film surface, and enrich the feature representation. Through the feature dimension reduction and fusion technology, the binary feature matrix and the texture entropy feature are reduced in dimension and fused to obtain the target fused feature vector, which effectively reduces the data dimension while retaining key information. The pre-trained gradient boosting decision tree evaluation model is used for transmittance evaluation, which fully utilizes the advantages of the machine learning algorithm and improves the accuracy and generalization ability of the evaluation. By constructing a multi-layer decision tree structure and extracting and utilizing feature information layer by layer, a refined evaluation of the light transmittance of ETFE film is achieved, which improves the model's expressive power and prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] 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 labor.

[0021] The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size, without affecting the effects and purposes that can be achieved by the present invention, should still fall within the scope of the technical contents disclosed by the present invention.

[0022] Figure 1 1 is a schematic flow chart of a method for evaluating light transmittance of an ETFE film provided in an embodiment of the present invention;

[0023] Figure 2 It is a schematic block diagram of the structure of an ETFE film transmittance evaluation device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0024] 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 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.

[0025] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0026] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0027] It should be further understood that the term "and / or" used in the specification and appended claims of this application refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations. Figure 1 In the embodiment of the present application, one embodiment of the method for evaluating the light transmittance of ETFE film includes:

[0028] Step 100, performing multi-band spectral scanning and transmittance analysis on a plurality of sample ETFE films to obtain a standardized transmittance vector;

[0029] It is understandable that the execution subject of the present application may be an ETFE film transmittance evaluation device, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject.

[0030] Specifically, a multi-band spectral scan is performed on multiple sample ETFE films to obtain an original spectral data matrix, and the spectral response data of each sample in different bands is obtained. According to the original spectral data matrix, the spectral data of each band is subjected to median filtering to obtain a filtered spectral data matrix. Median filtering is an effective denoising method that can remove noise in spectral data and improve the smoothness and reliability of data. The filtered spectral data matrix is ​​fitted with a least squares polynomial to obtain a spectral baseline. The spectral baseline is the background signal of the spectral signal, and the true characteristic signal in the spectrum is separated by fitting. According to the filtered spectral data matrix and the spectral baseline, a baseline correction is performed to obtain a corrected spectral data matrix. The purpose of baseline correction is to remove the baseline drift in the spectral data, so that the spectral signal is purer and more accurate. The corrected spectral data matrix is ​​subjected to a standard normal variable transformation to obtain a standardized spectral data matrix. Standard normal variable transformation is a commonly used data preprocessing method that can eliminate the scale differences between different samples and make the data more unified and standardized. According to the standardized spectral data matrix, the transmittance of each sample ETFE film in each band is calculated to obtain an initial transmittance matrix. Transmittance is an important indicator for evaluating the optical properties of ETFE film. By calculating the transmittance, the transmittance performance of each sample in different bands can be quantified. The initial transmittance matrix is ​​subjected to principal component analysis to obtain the principal component score matrix and the principal component load matrix. Principal component analysis is a dimensionality reduction method that projects high-dimensional data into a low-dimensional space through linear transformation to extract the main features of the data. According to the principal component score matrix and the principal component load matrix, the optimal number of principal components is selected to obtain the transmittance feature matrix after dimensionality reduction. The selection of the optimal number of principal components is to ensure that the information of the original data is retained to the greatest extent while reducing the data dimension. The transmittance feature matrix after dimensionality reduction is subjected to mean clustering analysis to obtain the cluster center vector. Mean clustering analysis is an unsupervised learning method that classifies samples with similar characteristics into one category through a clustering algorithm and extracts the representative features of each category of samples. According to the cluster center vector, the transmittance feature matrix after dimensionality reduction is standardized to obtain a standardized transmittance vector. The eigenvalues ​​of different cluster center vectors are normalized to make the features between different samples comparable.

[0031] Step 200, performing multispectral imaging on the ETFE film to be tested to obtain a multi-band image, and performing registration and preprocessing on the multi-band image to obtain a preprocessed image and a target frequency domain coefficient;

[0032] Specifically, multi-band spectral imaging is performed on the ETFE film to be tested to obtain a multi-band image, and each image in the multi-band image is subjected to histogram equalization processing to improve the contrast of the image and obtain an image set with enhanced contrast. A reference band image is selected according to the image set with enhanced contrast, and phase correlation registration is performed on other band images to obtain a preliminarily registered multi-band image set. Phase correlation registration is a commonly used image registration method, which aligns images of different bands by calculating phase differences. Affine transformation registration is performed on the preliminarily registered multi-band image set to obtain a precisely registered multi-band image set. Affine transformation registration makes the multi-band images completely aligned in space by rotating, scaling and translating the images. According to the precisely registered multi-band image set, the spectral curve of each pixel is calculated. The spectral curve can reflect the spectral information of each pixel in different bands and is the key data for evaluating light transmittance. The spectral curve is smoothed and filtered to obtain a filtered spectral data set. Smoothing filtering can remove noise in the spectral curve, making the spectral data smoother and more stable. The filtered spectral data set is subjected to minimum noise fraction transformation to obtain a denoised spectral data set. Minimum noise fraction transformation is a denoising method that can effectively reduce the noise in spectral data and improve the signal-to-noise ratio of the data. A multi-band image is reconstructed based on the denoised spectral data set to obtain a preprocessed image. A two-dimensional discrete Fourier transform is performed on each band of the preprocessed image to obtain a frequency domain representation. The frequency domain representation converts the image in the spatial domain to the frequency domain, which can better analyze the frequency characteristics of the image. The amplitude spectrum and phase spectrum are calculated based on the frequency domain representation to obtain the initial frequency domain coefficients. The amplitude spectrum and phase spectrum represent the intensity and phase information of the image, respectively, and are the basic data for frequency domain analysis. The initial frequency domain coefficients are logarithmically transformed and bandpass filtered to obtain the target frequency domain coefficients. Logarithmic transformation can compress the dynamic range of the frequency domain coefficients, making the frequency domain data smoother; bandpass filtering can remove high-frequency and low-frequency noise in the frequency domain data, retain the intermediate frequency information, and obtain more accurate and stable target frequency domain coefficients.

[0033] Step 300, performing quaternion discrete cosine transform and binarization processing on the target frequency domain coefficients to obtain a target binary feature matrix, and performing local binary pattern and entropy calculation on the preprocessed image to obtain texture entropy features;

[0034] It should be noted that the target frequency domain coefficients are converted to quaternion representation to obtain a quaternion target frequency domain coefficient matrix. Quaternion representation conversion is to expand the original frequency domain coefficients to the quaternion domain so that they can be processed using quaternion discrete cosine transform. Quaternion discrete cosine transform is performed on the quaternion target frequency domain coefficient matrix to obtain a quaternion DCT coefficient matrix. Quaternion amplitude calculation is performed on the quaternion DCT coefficient matrix to obtain an amplitude matrix. The amplitude matrix represents the intensity information of the frequency domain coefficients. The global threshold is calculated based on the amplitude matrix, and the amplitude matrix is ​​binarized to obtain an initial binary feature matrix. Binarization is to convert the elements in the amplitude matrix into a binary form in order to extract binary features. Morphological opening and closing operations are performed on the initial binary feature matrix to obtain a target binary feature matrix. Morphological opening and closing operations are performed to remove noise and isolated points in the binary feature matrix to make the binary features smoother and more continuous. At the same time, the image blocks are divided according to the preprocessed image, and the local binary pattern (LBP) features are calculated for each image block to obtain the LBP feature matrix. The LBP feature is a texture descriptor that extracts the local texture features of the image by calculating the binary pattern of each pixel in the image block. The LBP feature matrix is ​​subjected to rotation invariance processing to obtain a rotation invariant LBP feature matrix. This makes the LBP features consistent at different rotation angles and improves the robustness of the features. The local entropy value of each image block is calculated based on the rotation invariant LBP feature matrix to obtain a local entropy matrix. The local entropy value can reflect the distribution and complexity of pixels in the image block and is an important texture feature. The local entropy matrix is ​​subjected to Gaussian smoothing processing to remove the noise in the local entropy matrix, making the texture entropy feature smoother and more stable, and obtaining the texture entropy feature.

[0035] Step 400: performing feature dimensionality reduction on the target binary feature matrix and the texture entropy feature to obtain a reduced-dimensional feature vector, and performing feature fusion on the reduced-dimensional feature vector and the normalized transmittance vector to obtain a target fused feature vector;

[0036] Specifically, singular value decomposition is performed on the target binary feature matrix to obtain a left singular matrix, a singular value matrix and a right singular matrix. Singular value decomposition is a matrix decomposition method that effectively extracts the main feature information in the matrix by decomposing the original matrix into three parts. The main singular values ​​are selected according to the singular value matrix, and the left singular matrix and the right singular matrix are truncated to obtain the target binary feature matrix after dimensionality reduction. Kernel principal component analysis is performed on the texture entropy feature to obtain a kernel matrix and a projection matrix. Kernel principal component analysis is a nonlinear dimensionality reduction method that maps the originally linearly inseparable data to a high-dimensional space by introducing a kernel function, so that the data is linearly separable in the high-dimensional space. The texture entropy feature is nonlinearly mapped according to the kernel matrix and the projection matrix to obtain the texture entropy feature vector after dimensionality reduction. The tensor product operation is performed on the target binary feature matrix and the texture entropy feature vector after dimensionality reduction to obtain a high-order feature tensor. The tensor product operation is to combine two dimensionality-reduced features together to form a high-order feature tensor. Tucker decomposition is performed on the high-order feature tensor to obtain a core tensor and a factor matrix. Tucker decomposition is a high-order tensor decomposition method that extracts the main feature information from the tensor by decomposing the high-order tensor into a core tensor and a factor matrix. The core tensor is flattened to obtain a reduced-dimensional feature vector. The flattening operation converts the multi-dimensional core tensor into a one-dimensional vector for subsequent feature processing. Nonlinear feature mapping is performed on the reduced-dimensional feature vector and the standardized transmittance vector to obtain an intermediate feature vector. The nonlinear mapping is to combine the reduced-dimensional feature vector and the standardized transmittance vector to form a new feature vector for subsequent feature fusion. Feature reconstruction and denoising are performed on the intermediate feature vector to obtain a reconstructed feature vector. A weighted sum operation is performed on the reconstructed feature vector and the standardized transmittance vector to combine the two feature vectors to obtain the target fused feature vector.

[0037] Step 500: Input the target fusion feature vector into the pre-trained gradient boosting decision tree evaluation model to perform light transmittance evaluation, and obtain the light transmittance evaluation result of the ETFE film to be tested.

[0038] Specifically, the target fusion feature vector is standardized to obtain a standardized input vector. The difference between feature dimensions is eliminated so that the input vector is compared and processed on a uniform scale. The standardized input vector is input into the pre-trained gradient boosting decision tree evaluation model. According to the standardized input vector, the first-layer decision tree set of the pre-trained gradient boosting decision tree evaluation model is feature segmented to obtain an initial node set. Feature segmentation refers to dividing each node of the decision tree according to the input feature to form multiple child nodes. The initial node set is traversed through the tree structure to obtain a first-layer leaf node prediction value set. Tree structure traversal refers to starting from the root node of the decision tree, traversing layer by layer until the leaf node, and obtaining the prediction value of each leaf node. The first-layer residual is calculated based on the first-layer leaf node prediction value set to obtain a first-layer residual vector. The residual vector refers to the difference between the predicted value and the actual value, which is used to guide the training of the next layer of decision trees. The first-layer residual vector and the standardized input vector are feature combined to obtain the second-layer input feature. Feature combination refers to combining the residual vector with the input vector to form a new input feature for the training of the next layer of decision trees. According to the second-layer input features, the second-layer decision tree set of the pre-trained gradient boosting decision tree evaluation model is feature segmented and node traversed to obtain the second-layer leaf node prediction value set. The second-layer leaf node prediction value is obtained by feature segmentation and traversal similar to the first-layer. The first-layer leaf node prediction value set and the second-layer leaf node prediction value set are weighted and accumulated to obtain the intermediate prediction result. Weighted accumulation means that the prediction values ​​of the leaf nodes are weighted and summed according to the weights of each layer of decision trees to obtain a comprehensive prediction result. The second-layer residual is calculated based on the intermediate prediction result and the actual transmittance to obtain the second-layer residual vector. The residual vector is calculated in the same way as the first layer, which is the difference between the intermediate prediction result and the actual transmittance. The second-layer residual vector, the standardized input vector and the intermediate prediction result are feature fused to obtain the final layer input feature. Multiple features are combined to form new input features to improve the prediction ability of the model. According to the final layer input features, the final layer decision tree set of the pre-trained gradient boosting decision tree evaluation model is subjected to feature segmentation, node traversal and prediction value accumulation to obtain the light transmittance evaluation result of the ETFE film to be tested. In the final layer decision tree, the prediction value of each leaf node is obtained by further segmenting and traversing the input features, and the final evaluation result is obtained by weighted accumulation.

[0039] The feature index mapping is performed on the input features of the final layer, and the input features are matched with the feature index in the decision tree model to obtain a feature index vector. According to the feature index vector, the corresponding decision tree structure is extracted from the final layer decision tree set to obtain the decision tree set to be evaluated. For each decision tree in the decision tree set to be evaluated, feature comparison and branch selection are performed starting from the root node to obtain a leaf node path. Feature comparison refers to comparing the value of the input feature with the node splitting condition and selecting the corresponding branch until the leaf node is reached. According to the leaf node path, the corresponding prediction value is extracted from the pre-stored leaf node value table to obtain a single tree prediction result. The leaf node value table is generated during the model training process and is used to store the prediction value of each leaf node. The single tree prediction results are weighted and summed based on the preset tree weights to obtain a weighted prediction result. The tree weight is determined during the model training process and is used to balance the contribution of different decision trees. The weighted prediction result is added to the global bias term to obtain the original output value of the final layer. The global bias term is a constant term in the model, which is used to adjust the baseline value of the prediction result. The activation function operation is performed on the original output value of the final layer to obtain the activation output of the final layer. The activation function is a nonlinear transformation used to increase the expressive power of the model so that it can fit complex nonlinear relationships. The activation output of the final layer is added to the cumulative result of the previous layer to obtain the cumulative prediction value of the current iteration. The cumulative result of the previous layer is the comprehensive output of the decision trees of the previous layers. By accumulating the activation output of the current layer, the true value is gradually approached. The inverse transformation function is operated on the cumulative prediction value, and the prediction result is mapped back to the original transmittance scale to obtain the mapped prediction result. The mapped prediction result is threshold constrained and normalized to obtain the transmittance evaluation result of the ETFE film to be tested. The threshold constraint is to limit the prediction result within a reasonable range to avoid the influence of outliers; the normalization process is to make the prediction results of different samples comparable.

[0040] In the embodiment of the present application, a standardized transmittance vector is obtained by performing multi-band spectral scanning and standardization on multiple sample ETFE films, and a multi-band image of the ETFE film to be tested is obtained by using multi-spectral imaging technology, and registration and pre-processing are performed to effectively extract the spectral feature information of the ETFE film, and the frequency domain coefficients are subjected to quaternion discrete cosine transformation and binarization to obtain a binary feature matrix, which effectively captures the texture and structural features of the ETFE film and enhances the expressiveness of the features. The local binary pattern and entropy calculation are performed on the pre-processed image to obtain the texture entropy feature, further extract the microstructure information of the ETFE film surface, and enrich the feature representation. Through feature dimension reduction and fusion technology, the binary feature matrix and the texture entropy feature are reduced in dimension and fused to obtain the target fused feature vector, which effectively reduces the data dimension while retaining key information. The pre-trained gradient boosting decision tree evaluation model is used for transmittance evaluation, which fully utilizes the advantages of the machine learning algorithm and improves the accuracy and generalization ability of the evaluation. By constructing a multi-layer decision tree structure and extracting and utilizing feature information layer by layer, a refined evaluation of the light transmittance of ETFE film is achieved, which improves the model's expressive power and prediction accuracy.

[0041] In a specific embodiment, the process of executing step 100 may specifically include the following steps:

[0042] Perform multi-band spectral scanning on multiple sample ETFE films to obtain an original spectral data matrix, and perform median filtering on the spectral data of each band according to the original spectral data matrix to obtain a filtered spectral data matrix;

[0043] Performing least square polynomial fitting on the filtered spectral data matrix to obtain a spectral baseline, and performing baseline correction based on the filtered spectral data matrix and the spectral baseline to obtain a corrected spectral data matrix;

[0044] The corrected spectral data matrix is ​​transformed into a standard normal variable to obtain a standardized spectral data matrix, and the transmittance of each sample ETFE film in each band is calculated according to the standardized spectral data matrix to obtain an initial transmittance matrix;

[0045] The principal components of the initial transmittance matrix are extracted to obtain the principal component score matrix and the principal component load matrix, and the optimal number of principal components is selected according to the principal component score matrix and the principal component load matrix to obtain the transmittance feature matrix after dimensionality reduction;

[0046] The transmittance feature matrix after dimensionality reduction is subjected to mean clustering analysis to obtain a cluster center vector, and the transmittance feature matrix after dimensionality reduction is standardized according to the cluster center vector to obtain a standardized transmittance vector.

[0047] Specifically, multiple sample ETFE films are placed in a spectral scanner, and spectral data in different bands are obtained through multi-band spectral scanning. samples, each sample in The spectral data under the bands form a The original spectral data matrix ,in Indicates The sample in Spectral data under bands. For the original spectral data matrix The spectral data of each band is processed by median filtering. Median filtering is an effective denoising method that can eliminate sharp noise in spectral data. For the The spectral data of each band is processed by median filtering to obtain the filtered spectral data matrix . Perform the least squares polynomial fitting on the filtered spectral data matrix to obtain the spectral baseline. The spectral baseline is the low-frequency component in the spectral signal, and the least squares polynomial fitting can effectively fit this baseline. Assuming that The fitting equation can be expressed as:

[0048] ;

[0049] in, Indicates The spectral baseline of the band, is the wavelength, are the polynomial coefficients. According to the filtered spectral data matrix and spectral baseline , perform baseline correction and obtain the corrected spectral data matrix . Subtract the corresponding spectral baseline from the filtered spectral data, that is:

[0050] ;

[0051] The corrected spectral data matrix Perform standard normal variable transformation to obtain standardized spectral data matrix The formula for standard normal variable transformation is:

[0052] ;

[0053] in, and Respectively represent The mean and standard deviation of the spectral data of each band. According to the standardized spectral data matrix , calculate the transmittance of each sample ETFE film in each band and get the initial transmittance matrix . Light transmittance Indicates The sample in The transmittance in each band can be obtained by calculating the relative transmittance of the spectral data. Perform principal component analysis to extract the main feature information. Through principal component analysis, we can get the principal component score matrix And the principal component loading matrix The mathematical expression of principal component analysis is:

[0054] ;

[0055] in, yes The principal component score matrix of yes The principal component loading matrix of is the number of principal components selected. According to the principal component score matrix And the principal component loading matrix , choose the optimal number of principal components , to reduce the dimension of the data and obtain the transmittance feature matrix after dimensionality reduction The selection of the optimal number of principal components can be based on the cumulative variance contribution rate or other evaluation indicators. Perform mean cluster analysis to identify the cluster centers of the data. Mean cluster analysis distributes data points to Cluster centers, and get the center vector of each cluster The goal of mean clustering is to minimize the sum of squared distances from samples to their nearest cluster centers. The formula is:

[0056] ;

[0057] in, is the indicator function, when The samples belong to When clustering, , otherwise it is 0. The transmittance feature matrix after dimension reduction is standardized according to the cluster center vector to obtain a standardized transmittance vector.

[0058] In a specific embodiment, the process of executing step 200 may specifically include the following steps:

[0059] Performing multi-band spectral imaging on the ETFE film to be tested to obtain a multi-band image, and performing histogram equalization processing on each image in the multi-band image to obtain an image set with enhanced contrast;

[0060] Select a reference band image according to the contrast-enhanced image set, perform phase correlation registration on other band images, and obtain a preliminarily registered multi-band image set;

[0061] Performing affine transformation registration on the preliminarily registered multi-band image set to obtain a precisely registered multi-band image set, and calculating the spectral curve of each pixel point based on the precisely registered multi-band image set, and performing smoothing filtering on the spectral curve to obtain a filtered spectral data set;

[0062] Performing minimum noise fraction transformation on the filtered spectral data set to obtain a denoised spectral data set, and reconstructing a multi-band image based on the denoised spectral data set to obtain a preprocessed image;

[0063] A two-dimensional discrete Fourier transform is performed on each band of the preprocessed image to obtain a frequency domain representation, and the amplitude spectrum and phase spectrum are calculated based on the frequency domain representation to obtain the initial frequency domain coefficients. The initial frequency domain coefficients are logarithmically transformed and bandpass filtered to obtain the target frequency domain coefficients.

[0064] Specifically, the image data of the ETFE film to be tested in different bands are obtained by multi-band spectral imaging technology. Assuming that image data is collected in five different bands, a multi-band image set containing five images is obtained. ,in Indicates In order to enhance the image contrast, each image is processed by histogram equalization. Histogram equalization is an image processing technology that adjusts the grayscale distribution of the image to make the image contrast more uniform and enhance the detail performance. The processed image set is denoted as . Based on the image set with enhanced contrast , select a reference band image (For example, select the image of band 1 As a reference), phase correlation registration is performed on other band images. Phase correlation registration calculates the phase information of the image, finds the best matching position, and aligns the images of different bands. A preliminarily registered multi-band image set is obtained through phase correlation registration. . Based on the preliminary registration, affine transformation registration is performed on the multi-band image set to improve the registration accuracy. Affine transformation is a geometric transformation, including translation, rotation, scaling and shearing, to accurately align images. Through affine transformation registration, the registered multi-band image set is obtained. Based on a precisely registered multi-band image set , calculate the spectral curve of each pixel. The spectral curve represents the spectral response of each pixel in different bands and is an important feature of multi-band images. Assume Indicates The pixel in The spectral response in each band, the spectral curve can be expressed as In order to remove noise, the spectral curve is smoothed and filtered to obtain the filtered spectral data set . For the filtered spectral dataset Perform minimum noise fraction transformation to reduce noise and improve signal quality. Minimum noise fraction transformation minimizes noise by maximizing the signal-to-noise ratio and projecting the data into a new space. After the minimum noise fraction transformation, the spectral data set after noise reduction is obtained. According to the denoised spectral dataset , reconstruct the multi-band image and obtain the preprocessed image set . Perform a two-dimensional discrete Fourier transform on each band of the preprocessed image to obtain a frequency domain representation. The two-dimensional discrete Fourier transform converts the image from the spatial domain to the frequency domain, making it easier to analyze the frequency characteristics of the image. The two-dimensional discrete Fourier transform of can be expressed as:

[0065] ;

[0066] in, Indicates The representation of the band image in the frequency domain is: is the frequency coordinate, are spatial coordinates, and are the width and height of the image respectively. According to the frequency domain representation , calculate the amplitude spectrum and phase spectrum. The amplitude spectrum represents the intensity of each frequency component in the frequency domain, while the phase spectrum represents the phase information of each frequency component. and phase spectrum Can be expressed as:

[0067] ;

[0068] ;

[0069] in, Indicates plural The model, Indicates plural The phase angle of the initial frequency domain coefficients. After obtaining the initial frequency domain coefficients, the initial frequency domain coefficients are logarithmically transformed and bandpass filtered. The logarithmic transformation can compress the dynamic range of the amplitude spectrum, making the frequency domain characteristics more obvious. The logarithmic transformation can be expressed as:

[0070] ;

[0071] in, represents the amplitude spectrum after logarithmic transformation. Bandpass filtering uses a filter to retain only the components within a certain frequency range in the frequency domain and remove low-frequency and high-frequency noise. The frequency domain coefficient after bandpass filtering is expressed as the target frequency domain coefficient Assume that the frequency range of the bandpass filter is , the filtering process can be expressed as:

[0072] ;

[0073] in, is the transfer function of the bandpass filter, when hour, ,otherwise

[0074] In a specific embodiment, the process of executing step 300 may specifically include the following steps:

[0075] The target frequency domain coefficients are converted into quaternion representation to obtain a quaternion target frequency domain coefficient matrix, and a quaternion discrete cosine transform is performed on the quaternion target frequency domain coefficient matrix to obtain a quaternion DCT coefficient matrix;

[0076] The quaternion amplitude is calculated for the quaternion DCT coefficient matrix to obtain an amplitude matrix, a global threshold is calculated based on the amplitude matrix, and the amplitude matrix is ​​binarized to obtain an initial binary feature matrix;

[0077] Perform morphological opening and closing operations on the initial binary feature matrix to obtain the target binary feature matrix;

[0078] Divide the image blocks according to the preprocessed image, calculate the local binary pattern features for each image block to obtain the LBP feature matrix, and perform rotation invariance processing on the LBP feature matrix to obtain the rotation invariant LBP feature matrix;

[0079] The local entropy value of each image block is calculated according to the rotation invariant LBP feature matrix to obtain the local entropy matrix, and the local entropy matrix is ​​Gaussian smoothed to obtain the texture entropy feature.

[0080] Specifically, the target frequency domain coefficient matrix is ​​expanded to the quaternion domain. A quaternion is a hypercomplex number consisting of a real part and three imaginary parts, usually expressed as ,in is a real number, is an imaginary unit, satisfying Assume that the target frequency domain coefficient matrix is , which can be represented as a quaternion matrix ,in:

[0081] ;

[0082] Perform quaternion discrete cosine transform on the quaternion target frequency domain coefficient matrix to obtain the quaternion DCT coefficient matrix The calculation formula of quaternion discrete cosine transform is similar to DCT in real number domain, but it is performed in quaternion domain. The matrix , the formula for quaternion discrete cosine transform is as follows:

[0083] ;

[0084] in, are the elements in the quaternion DCT coefficient matrix, is the element in the input matrix. For the quaternion DCT coefficient matrix Perform quaternion amplitude calculation to obtain the amplitude matrix The amplitude calculation of quaternion is similar to that of complex number, the formula is:

[0085] ;

[0086] in, Quaternions The real part and three imaginary parts of , calculate the global threshold and binarize the amplitude matrix. The global threshold can be calculated using Otsu's method, which finds the optimal threshold by maximizing the variance between classes. The binarization process converts the elements in the amplitude matrix into Convert to 0 or 1 to obtain the initial binary feature matrix. Perform morphological opening and closing operations on the initial binary feature matrix to remove noise and fill holes. The morphological opening operation is to corrode first and then expand, while the closing operation is to expand first and then corrode. Through these operations, a smoother and more continuous target binary feature matrix is ​​obtained. The image blocks are divided according to the preprocessed image, and the local binary pattern features are calculated for each image block to obtain the LBP feature matrix. LBP is an effective texture descriptor that generates a binary pattern by comparing the grayscale values ​​of each pixel with its neighboring pixels. The specific calculation formula is as follows:

[0087] ;

[0088] in, is the gray value of the center pixel, is the gray value of the neighborhood pixel. The LBP feature matrix is ​​processed for rotation invariance to obtain the rotation invariant LBP feature matrix. The local entropy value of each image block is calculated based on the rotation invariant LBP feature matrix to obtain the local entropy matrix. Entropy is an indicator of the amount of information, and the local entropy represents the complexity of the texture in the image block. The local entropy matrix is ​​Gaussian smoothed to remove noise and obtain the texture entropy feature.

[0089] In a specific embodiment, the process of executing step 400 may specifically include the following steps:

[0090] Perform singular value decomposition on the target binary feature matrix to obtain a left singular matrix, a singular value matrix and a right singular matrix, select the main singular values ​​according to the singular value matrix, truncate the left singular matrix and the right singular matrix, and obtain the target binary feature matrix after dimensionality reduction;

[0091] Performing kernel principal component analysis on the texture entropy features to obtain a kernel matrix and a projection matrix, and performing nonlinear mapping on the texture entropy features according to the kernel matrix and the projection matrix to obtain a texture entropy feature vector after dimensionality reduction;

[0092] Perform tensor product operation on the target binary feature matrix and texture entropy feature vector after dimension reduction to obtain a high-order feature tensor, and perform Tucker decomposition on the high-order feature tensor to obtain a core tensor and a factor matrix;

[0093] Flatten the core tensor to obtain a reduced-dimensional feature vector, and perform feature nonlinear mapping based on the reduced-dimensional feature vector and the normalized transmittance vector to obtain an intermediate feature vector;

[0094] The intermediate feature vector is subjected to feature reconstruction and denoising to obtain a reconstructed feature vector, and a weighted sum operation is performed on the reconstructed feature vector and the normalized transmittance vector to obtain a target fused feature vector.

[0095] Specifically, singular value decomposition is performed on the target binary feature matrix to obtain the left singular matrix, singular value matrix and right singular matrix. Assume that the target binary feature matrix is , the formula for singular value decomposition is:

[0096] ;

[0097] in, is a left singular matrix, is the singular value matrix, is a right singular matrix, Represents the transpose of a matrix. Singular value matrix It is a diagonal matrix, and its diagonal elements are singular values. According to the singular value matrix , select the main singular value. The selection of the main singular value can be based on a threshold, such as retaining the singular value with cumulative energy reaching 95%. and the right singular matrix Truncate and retain the columns corresponding to the main singular values ​​to obtain the truncated matrix and The target binary feature matrix after dimensionality reduction is It can be expressed as:

[0098] ;

[0099] in, is the truncated singular value matrix. Perform kernel principal component analysis on the texture entropy feature. Kernel principal component analysis is a nonlinear dimensionality reduction method that maps data to a high-dimensional space through kernel techniques and then performs principal component analysis in the high-dimensional space. Assume that the texture entropy feature matrix is , calculate the kernel matrix , whose elements are:

[0100] ;

[0101] in, is a kernel function, such as a Gaussian kernel function. Calculate the kernel matrix The eigenvalues ​​and eigenvectors of get the projection matrix . Through the projection matrix Perform nonlinear mapping on the texture entropy features to obtain the texture entropy feature vector after dimensionality reduction After obtaining the target binary feature matrix after dimensionality reduction And the texture entropy feature vector after dimensionality reduction After that, a tensor product operation is performed to obtain a high-order feature tensor The tensor product operation can perform the outer product of each element of two matrices to form a high-order tensor. Perform Tucker decomposition. Tucker decomposition is a high-order tensor decomposition method that decomposes a tensor into a core tensor and a set of factor matrices. Assume that the high-order eigentensor The Tucker decomposition is:

[0102] ;

[0103] in, is the core tensor, is a factor matrix, Represents the tensor product on the nth mode. For the core tensor Perform a flattening operation and convert it into a vector to obtain a reduced-dimensional feature vector According to the reduced dimensionality feature vector and the normalized transmittance vector Perform nonlinear mapping of features to obtain the intermediate feature vector For the intermediate eigenvector Perform feature reconstruction and noise reduction to obtain the reconstructed feature vector . Feature reconstruction can be achieved through reverse mapping and noise reduction algorithms, such as low-rank approximation. According to the reconstructed feature vector and the normalized transmittance vector Perform weighted sum operation to obtain the target fusion feature vector The formula for weighted summation is:

[0104] ;

[0105] in, and is a weight parameter used to adjust the contribution of the reconstructed eigenvector and the normalized transmittance vector.

[0106] In a specific embodiment, the process of executing step 500 may specifically include the following steps:

[0107] The target fusion feature vector is normalized to obtain a normalized input vector, and the normalized input vector is input into the pre-trained gradient boosting decision tree evaluation model;

[0108] According to the standardized input vector, feature segmentation is performed on the first-layer decision tree set of the pre-trained gradient boosting decision tree evaluation model to obtain the initial node set;

[0109] Traverse the tree structure of the initial node set to obtain the first-layer leaf node prediction value set;

[0110] Calculate the first-layer residual according to the first-layer leaf node prediction value set to obtain the first-layer residual vector;

[0111] Perform feature combination on the first layer residual vector and the standardized input vector to obtain the second layer input features;

[0112] According to the second-layer input features, feature segmentation and node traversal are performed on the second-layer decision tree set of the pre-trained gradient boosting decision tree evaluation model to obtain a second-layer leaf node prediction value set;

[0113] Perform weighted accumulation on the first-layer leaf node prediction value set and the second-layer leaf node prediction value set to obtain an intermediate prediction result;

[0114] The second layer residual is calculated according to the intermediate prediction result and the actual transmittance, and the second layer residual vector is obtained;

[0115] The second layer residual vector, the standardized input vector and the intermediate prediction result are subjected to feature fusion to obtain the final layer input feature;

[0116] According to the final layer input features, feature segmentation, node traversal and prediction value accumulation are performed on the final layer decision tree set of the pre-trained gradient boosting decision tree evaluation model to obtain the transmittance evaluation result of the ETFE film to be tested.

[0117] Specifically, the target fusion feature vector is standardized, and each element in the feature vector is converted into a standard normal distribution with a mean of 0 and a variance of 1 to obtain a standardized input vector. The standardized input vector is input into the pre-trained gradient boosting decision tree evaluation model. The first-layer decision tree set is feature segmented to obtain an initial node set. Each decision tree will be segmented according to the input features, and the best features and thresholds will be selected for division to maximize the purity of each node. The initial node set contains the root nodes of all decision trees. The tree structure of the initial node set is traversed, and by comparing the input features with the node segmentation threshold, it is decided to move to the left child node or the right child node until a leaf node is reached. The value of the leaf node is the predicted value of the decision tree for the input feature. The leaf node prediction values ​​of all decision trees are combined to obtain the first-layer leaf node prediction value set. The first-layer residual is calculated based on the first-layer leaf node prediction value set. The residual is the difference between the actual value and the predicted value. Assume that the actual transmittance is , the prediction value of the first layer leaf node is , the first layer residual The calculation formula is as follows:

[0118] ;

[0119] The first layer residual vector and the standardized input vector are combined to form a new feature set. After obtaining the second layer input features, they are input into the second layer decision tree set of the pre-trained gradient boosting decision tree model for feature segmentation and node traversal to obtain the second layer leaf node prediction value set. The first layer leaf node prediction value set and the second layer leaf node prediction value set are weighted and accumulated to obtain the intermediate prediction result. Assume that the first layer leaf node prediction value is , the prediction value of the second layer leaf node is , the weighted accumulation formula is as follows:

[0120] ;

[0121] in, and is the weight parameter. The second layer residual is calculated based on the intermediate prediction result and the actual transmittance to obtain the second layer residual vector. Assume that the intermediate prediction result is , the second layer residual The calculation formula is as follows:

[0122] ;

[0123] The second layer residual vector, the standardized input vector and the intermediate prediction results are feature fused to form the final layer input features. The final layer input features are input into the final layer decision tree set of the pre-trained gradient boosting decision tree model for feature segmentation, node traversal and prediction value accumulation. Each decision tree performs feature segmentation based on the final layer input features, obtains the prediction value through node traversal, and accumulates the prediction values ​​of all decision trees to obtain the transmittance evaluation result of the ETFE film to be tested.

[0124] In a specific embodiment, the execution step performs feature segmentation, node traversal, and prediction value accumulation on the final layer decision tree set of the pre-trained gradient boosting decision tree evaluation model according to the final layer input features, and the process of obtaining the light transmittance evaluation result of the ETFE film to be tested can specifically include the following steps:

[0125] Perform feature index mapping on the input features of the final layer to obtain a feature index vector, and extract the corresponding decision tree structure from the decision tree set of the final layer according to the feature index vector to obtain the decision tree set to be evaluated;

[0126] For each decision tree in the decision tree set to be evaluated, feature comparison and branch selection are performed starting from the root node to obtain the leaf node path;

[0127] According to the leaf node path, the corresponding prediction value is extracted from the pre-stored leaf node value table to obtain the single tree prediction result, and the single tree prediction result is weighted summed based on the preset tree weight to obtain the weighted prediction result;

[0128] Add the weighted prediction result to the global bias term to get the original output value of the final layer, and perform an activation function operation on the original output value of the final layer to get the activation output of the final layer;

[0129] The activation output of the final layer is added to the cumulative result of the previous layer to obtain the cumulative prediction value of the current iteration, and the inverse transformation function is performed on the cumulative prediction value to map the prediction result back to the original transmittance scale to obtain the mapped prediction result;

[0130] The predicted results after mapping are subjected to threshold constraints and normalization to obtain the transmittance evaluation results of the ETFE film to be tested.

[0131] Specifically, the input features are encoded to obtain a feature index vector. The feature index vector is a vector used to locate the position of the feature in the decision tree. Assume that the final layer input feature is , the feature index vector is obtained through feature index mapping . Based on the feature index vector The corresponding decision tree structure is extracted from the final layer decision tree set of the pre-trained gradient boosting decision tree model to obtain the decision tree set to be evaluated. For each decision tree in the decision tree set to be evaluated, feature comparison and branch selection are performed starting from the root node. By comparing the input feature with the node's segmentation threshold, it is decided to move to the left child node or the right child node until the leaf node is reached. Each decision tree will be segmented according to the input feature, and the best feature and threshold will be selected for division to maximize the purity of each node. The leaf node path is a path from the root node to the leaf node. According to this path, the corresponding prediction value is extracted from the pre-stored leaf node value table to obtain the single tree prediction result. The single tree prediction results are weighted and summed based on the preset tree weights to obtain the weighted prediction result. Assume that the single tree prediction result is , the tree weight is , weighted prediction results The calculation formula is as follows:

[0132] ;

[0133] in, is the number of decision trees, It is The weight of a decision tree, It is The weighted prediction result is added to the global bias term to get the original output value of the final layer. Assume that the global bias term is , the final layer original output value

[0134] ;

[0135] The activation function is operated on the original output value of the final layer to obtain the activation output of the final layer. Commonly used activation functions include sigmoid function, tanh function and ReLU function. Assuming that the sigmoid function is used, the formula for the activation function operation is as follows:

[0136] ;

[0137] Add the final layer activation output to the previous layer cumulative result to get the cumulative prediction value of the current iteration. Assume that the previous layer cumulative result is The cumulative prediction value of the current iteration

[0138] ;

[0139] The inverse transformation function is performed on the accumulated prediction value, and the prediction result is mapped back to the original transmittance scale to obtain the mapped prediction result. The inverse transformation function can be a simple linear transformation. Assume that the coefficient of the linear transformation is and , the formula for the inverse transformation function operation is as follows:

[0140] ;

[0141] The predicted results after mapping are subjected to threshold constraints and normalization processing to obtain the transmittance evaluation results of the ETFE film to be tested. The threshold constraint is to ensure that the predicted results are within a reasonable range, and the normalization processing is to scale the predicted results to the [0, 1] interval. Assume that the threshold range is , the normalization formula is as follows:

[0142] ;

[0143] The above describes the method for evaluating the light transmittance of an ETFE film in the embodiment of the present application. The following describes the device 10 for evaluating the light transmittance of an ETFE film in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the ETFE film transmittance evaluation device 10 includes:

[0144] An analysis module 11 is used to perform multi-band spectrum scanning and transmittance analysis on multiple sample ETFE films to obtain a standardized transmittance vector;

[0145] The imaging module 12 is used to perform multi-spectral imaging on the ETFE film to be tested to obtain a multi-band image, and to register and pre-process the multi-band image to obtain a pre-processed image and a target frequency domain coefficient;

[0146] The calculation module 13 is used to perform quaternion discrete cosine transform and binarization processing on the target frequency domain coefficients to obtain the target binary feature matrix, and perform local binary pattern and entropy calculation on the preprocessed image to obtain texture entropy features;

[0147] A fusion module 14 is used to perform feature dimension reduction on the target binary feature matrix and the texture entropy feature to obtain a reduced dimension feature vector, and to perform feature fusion on the reduced dimension feature vector and the normalized transmittance vector to obtain a target fused feature vector;

[0148] The evaluation module 15 is used to input the target fusion feature vector into the pre-trained gradient boosting decision tree evaluation model to perform light transmittance evaluation, and obtain the light transmittance evaluation result of the ETFE film to be tested.

[0149] Through the cooperation of the above components, the standardized transmittance vector is obtained by multi-band spectral scanning and standardization of multiple sample ETFE films. The multi-band images of the ETFE film to be tested are obtained by multi-spectral imaging technology, and the registration and preprocessing are performed to effectively extract the spectral feature information of the ETFE film. The frequency domain coefficients are subjected to quaternion discrete cosine transformation and binarization to obtain a binary feature matrix, which effectively captures the texture and structural characteristics of the ETFE film and enhances the feature expression ability. The local binary pattern and entropy calculation of the preprocessed image are performed to obtain the texture entropy feature, which further extracts the microstructure information of the ETFE film surface and enriches the feature representation. Through feature dimension reduction and fusion technology, the binary feature matrix and texture entropy features are reduced and fused to obtain the target fused feature vector, which effectively reduces the data dimension while retaining key information. The pre-trained gradient boosting decision tree evaluation model is used for transmittance evaluation, which fully utilizes the advantages of machine learning algorithms and improves the accuracy and generalization ability of the evaluation. By constructing a multi-layer decision tree structure and extracting and utilizing feature information layer by layer, a refined evaluation of the light transmittance of ETFE film is achieved, which improves the model's expressive power and prediction accuracy.

[0150] The present application also provides an electronic device, which includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor executes the steps of the ETFE film transmittance evaluation method in the above-mentioned embodiments.

[0151] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the ETFE film transmittance evaluation method.

[0152] Those skilled in the art can 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.

[0153] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or 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, including several instructions to enable an electronic device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0154] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for evaluating the light transmittance of an ETFE film, characterized in that: The ETFE film light transmittance evaluation method comprises: The method comprises the following steps: performing multi-band spectral scanning and transmittance analysis on a plurality of sample ETFE films to obtain a standardized transmittance vector; specifically, performing multi-band spectral scanning on a plurality of sample ETFE films to obtain an original spectral data matrix, and performing median filtering on the spectral data of each band according to the original spectral data matrix to obtain a filtered spectral data matrix; performing least squares polynomial fitting on the filtered spectral data matrix to obtain a spectral baseline, and performing baseline correction according to the filtered spectral data matrix and the spectral baseline to obtain a corrected spectral data matrix; and performing standard normal variable transformation on the corrected spectral data matrix. , obtain a standardized spectral data matrix, and calculate the transmittance of each sample ETFE film in each band according to the standardized spectral data matrix to obtain an initial transmittance matrix; perform principal component extraction on the initial transmittance matrix to obtain a principal component score matrix and a principal component load matrix, and select the optimal number of principal components according to the principal component score matrix and the principal component load matrix to obtain a transmittance feature matrix after dimensionality reduction; perform mean clustering analysis on the transmittance feature matrix after dimensionality reduction to obtain a cluster center vector, and perform standardization on the transmittance feature matrix after dimensionality reduction according to the cluster center vector to obtain a standardized transmittance vector; Performing multispectral imaging on the ETFE film to be tested to obtain a multi-band image, and registering and preprocessing the multi-band image to obtain a preprocessed image and a target frequency domain coefficient; Performing quaternion discrete cosine transformation and binarization processing on the target frequency domain coefficients to obtain a target binary feature matrix, and performing local binary pattern and entropy calculation on the preprocessed image to obtain texture entropy features; Performing feature dimensionality reduction on the target binary feature matrix and the texture entropy feature to obtain a reduced-dimensional feature vector, and performing feature fusion on the reduced-dimensional feature vector and the normalized transmittance vector to obtain a target fused feature vector; The target fusion feature vector is input into a pre-trained gradient boosting decision tree evaluation model to perform light transmittance evaluation, thereby obtaining a light transmittance evaluation result of the ETFE film to be tested.

2. The ETFE film light transmittance evaluation method according to claim 1, wherein: The method of performing multispectral imaging on the ETFE film to be tested to obtain a multi-band image, and performing registration and preprocessing on the multi-band image to obtain a preprocessed image and a target frequency domain coefficient includes: Performing multi-band spectral imaging on the ETFE film to be tested to obtain a multi-band image, and performing histogram equalization processing on each image in the multi-band image to obtain an image set with enhanced contrast; Selecting a reference band image according to the contrast-enhanced image set, performing phase correlation registration on other band images, and obtaining a preliminarily registered multi-band image set; Performing affine transformation registration on the preliminarily registered multi-band image set to obtain a precisely registered multi-band image set, calculating a spectral curve of each pixel point according to the precisely registered multi-band image set, and performing smoothing filtering on the spectral curve to obtain a filtered spectral data set; Performing minimum noise fraction transformation on the filtered spectral data set to obtain a denoised spectral data set, and reconstructing a multi-band image based on the denoised spectral data set to obtain a preprocessed image; A two-dimensional discrete Fourier transform is performed on each band of the preprocessed image to obtain a frequency domain representation, and an amplitude spectrum and a phase spectrum are calculated based on the frequency domain representation to obtain initial frequency domain coefficients, and the initial frequency domain coefficients are logarithmically transformed and bandpass filtered to obtain target frequency domain coefficients.

3. The ETFE film light transmittance evaluation method according to claim 1, wherein: The target frequency domain coefficients are subjected to quaternion discrete cosine transformation and binarization processing to obtain a target binary feature matrix, and the preprocessed image is subjected to local binary pattern and entropy calculation to obtain texture entropy features, including: Performing quaternion representation conversion on the target frequency domain coefficients to obtain a quaternion target frequency domain coefficient matrix, and performing quaternion discrete cosine transform on the quaternion target frequency domain coefficient matrix to obtain a quaternion DCT coefficient matrix; Performing quaternion amplitude calculation on the quaternion DCT coefficient matrix to obtain an amplitude matrix, calculating a global threshold according to the amplitude matrix, and binarizing the amplitude matrix to obtain an initial binary feature matrix; Performing morphological opening and closing operations on the initial binary feature matrix to obtain a target binary feature matrix; Dividing the image blocks according to the preprocessed image, calculating the local binary pattern feature of each image block to obtain an LBP feature matrix, and performing rotation invariance processing on the LBP feature matrix to obtain a rotation invariant LBP feature matrix; The local entropy value of each image block is calculated according to the rotation-invariant LBP feature matrix to obtain a local entropy matrix, and Gaussian smoothing is performed on the local entropy matrix to obtain a texture entropy feature.

4. The ETFE film light transmittance evaluation method according to claim 1, wherein: The step of performing feature dimensionality reduction on the target binary feature matrix and the texture entropy feature to obtain a reduced-dimensional feature vector, and performing feature fusion on the reduced-dimensional feature vector and the standardized transmittance vector to obtain a target fused feature vector includes: Performing singular value decomposition on the target binary feature matrix to obtain a left singular matrix, a singular value matrix and a right singular matrix, selecting main singular values ​​according to the singular value matrix, truncating the left singular matrix and the right singular matrix, and obtaining a target binary feature matrix after dimensionality reduction; Performing a kernel principal component analysis on the texture entropy feature to obtain a kernel matrix and a projection matrix, and performing a nonlinear mapping on the texture entropy feature according to the kernel matrix and the projection matrix to obtain a texture entropy feature vector after dimensionality reduction; Performing a tensor product operation on the target binary feature matrix after dimension reduction and the texture entropy feature vector to obtain a high-order feature tensor, and performing Tucker decomposition on the high-order feature tensor to obtain a core tensor and a factor matrix; Flattening the core tensor to obtain a reduced-dimensional feature vector, and performing feature nonlinear mapping based on the reduced-dimensional feature vector and the normalized transmittance vector to obtain an intermediate feature vector; Feature reconstruction and noise reduction are performed on the intermediate feature vector to obtain a reconstructed feature vector, and a weighted sum operation is performed based on the reconstructed feature vector and the standardized transmittance vector to obtain a target fused feature vector.

5. The ETFE film light transmittance evaluation method according to claim 1, characterized in that: The step of inputting the target fusion feature vector into a pre-trained gradient boosting decision tree evaluation model to perform light transmittance evaluation to obtain a light transmittance evaluation result of the ETFE film to be tested includes: Standardizing the target fusion feature vector to obtain a standardized input vector, and inputting the standardized input vector into a pre-trained gradient boosting decision tree evaluation model; According to the standardized input vector, feature segmentation is performed on a first-layer decision tree set of a pre-trained gradient boosting decision tree evaluation model to obtain an initial node set; Performing a tree structure traversal on the initial node set to obtain a first-layer leaf node prediction value set; Calculate the first layer residual according to the first layer leaf node prediction value set to obtain the first layer residual vector; Performing feature combination on the first layer residual vector and the standardized input vector to obtain second layer input features; According to the second-layer input features, feature segmentation and node traversal are performed on the second-layer decision tree set of the pre-trained gradient boosting decision tree evaluation model to obtain a second-layer leaf node prediction value set; Performing weighted accumulation on the first layer leaf node prediction value set and the second layer leaf node prediction value set to obtain an intermediate prediction result; Calculate the second layer residual according to the intermediate prediction result and the actual transmittance to obtain a second layer residual vector; Performing feature fusion on the second layer residual vector, the standardized input vector and the intermediate prediction result to obtain a final layer input feature; According to the final layer input features, feature segmentation, node traversal and prediction value accumulation are performed on the final layer decision tree set of the pre-trained gradient boosting decision tree evaluation model to obtain the light transmittance evaluation result of the ETFE film to be tested.

6. The method for evaluating light transmittance of ETFE film according to claim 5, wherein: According to the final layer input features, feature segmentation, node traversal and prediction value accumulation are performed on the final layer decision tree set of the pre-trained gradient boosting decision tree evaluation model to obtain the light transmittance evaluation result of the ETFE film to be tested, including: Performing feature index mapping on the final layer input features to obtain a feature index vector, and extracting a corresponding decision tree structure from the final layer decision tree set according to the feature index vector to obtain a decision tree set to be evaluated; For each decision tree in the decision tree set to be evaluated, feature comparison and branch selection are performed starting from the root node to obtain a leaf node path; According to the leaf node path, a corresponding prediction value is extracted from a pre-stored leaf node value table to obtain a single tree prediction result, and a weighted sum is performed on the single tree prediction result based on a preset tree weight to obtain a weighted prediction result; Adding the weighted prediction result to the global bias term to obtain the final layer original output value, and performing an activation function operation on the final layer original output value to obtain the final layer activation output; Adding the activation output of the final layer to the cumulative result of the previous layer to obtain the cumulative prediction value of the current iteration, and performing an inverse transformation function operation on the cumulative prediction value to map the prediction result back to the original transmittance scale to obtain a mapped prediction result; Threshold constraint and normalization processing are performed on the mapped prediction results to obtain the light transmittance evaluation result of the ETFE film to be tested.

7. An ETFE film light transmittance evaluation device, characterized in that: Used to perform the ETFE film light transmittance evaluation method according to any one of claims 1 to 6, the ETFE film light transmittance evaluation device comprises: The analysis module is used for performing multi-band spectral scanning and transmittance analysis on multiple sample ETFE films to obtain a standardized transmittance vector; specifically comprising: performing multi-band spectral scanning on multiple sample ETFE films to obtain an original spectral data matrix, and performing median filtering on the spectral data of each band according to the original spectral data matrix to obtain a filtered spectral data matrix; performing least squares polynomial fitting on the filtered spectral data matrix to obtain a spectral baseline, and performing baseline correction according to the filtered spectral data matrix and the spectral baseline to obtain a corrected spectral data matrix; performing standard normal transformation on the corrected spectral data matrix. The method comprises the following steps: performing a principal component extraction on the initial transmittance matrix to obtain a principal component score matrix and a principal component load matrix, and selecting an optimal number of principal components according to the principal component score matrix and the principal component load matrix to obtain a transmittance feature matrix after dimensionality reduction; performing a mean clustering analysis on the transmittance feature matrix after dimensionality reduction to obtain a cluster center vector, and performing a standardization process on the transmittance feature matrix after dimensionality reduction according to the cluster center vector to obtain a standardized transmittance vector; An imaging module is used to perform multispectral imaging on the ETFE film to be tested to obtain a multi-band image, and to register and preprocess the multi-band image to obtain a preprocessed image and a target frequency domain coefficient; A calculation module, used for performing quaternion discrete cosine transform and binarization processing on the target frequency domain coefficients to obtain a target binary feature matrix, and performing local binary pattern and entropy calculation on the preprocessed image to obtain texture entropy features; A fusion module, used for performing feature dimension reduction on the target binary feature matrix and the texture entropy feature to obtain a reduced dimension feature vector, and performing feature fusion on the reduced dimension feature vector and the normalized transmittance vector to obtain a target fused feature vector; The evaluation module is used to input the target fusion feature vector into a pre-trained gradient boosting decision tree evaluation model to perform light transmittance evaluation to obtain a light transmittance evaluation result of the ETFE film to be tested.

8. An electronic device, characterized in that: The electronic device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the electronic device to execute the ETFE film transmittance evaluation method according to any one of claims 1 to 6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the method for evaluating the light transmittance of the ETFE film according to any one of claims 1 to 6 is implemented.

Citation Information

Patent Citations

  • Seed cotton mulching film identification and classification method based on hyperspectral image pixel block machine learning

    CN114882291A

  • Dust concentration detection method based on multi-feature fusion

    CN117470725A

  • Quantum dot optical film damage detection method and system based on machine vision

    CN117557820A

  • Image processing apparatus and method, data processing apparatus and method, and program and recording medium

    US20110026849A1

  • Quaternion multi-degree-of-freedom neuron-based multispectral welding image recognition method

    US20220414857A1