Training method of chopped fiber order degree evaluation model in meta-aramid insulation paper and chopped fiber order degree evaluation method

Abstract

The embodiment of the invention relates to the field of material engineering, and provides a training method of a chopped fiber order degree evaluation model in meta-aramid insulation paper and a chopped fiber order degree evaluation method, and the method comprises the steps: obtaining a sample image of the meta-aramid insulation paper from a training sample of a data set; based on a scale invariant feature conversion algorithm, performing multi-scale feature extraction on the sample image; carrying out feature fusion on the sample features of each dimension; based on a principal component analysis algorithm, screening each sample feature in the sample comprehensive feature vector to obtain a sample dimensionality reduction feature vector; and training a chopped fiber order degree evaluation model constructed based on a support vector machine algorithm according to the sample label and the sample dimensionality reduction feature vector associated with the pre-labeled sample image. By adopting the method, the damage to the to-be-detected sample can be avoided, and the efficiency and accuracy of order evaluation are improved.

Description

Technical Field

[0001] This application relates to the field of materials engineering technology, and in particular to a training method and a method for evaluating the orderliness of short-cut fibers in meta-aramid insulating paper. Background Technology

[0002] In the field of materials engineering, chopped fibers are an effective material for preparing meta-aramid insulating paper. The degree of order of chopped fibers produced under different conditions varies greatly, and the degree of order directly affects its physical properties such as stress strength, filtration, porosity, and dielectric strength. Chopped fibers with different degrees of order cause differences in meta-aramid insulating paper in actual papermaking production. Therefore, the detection of order is an important part of the quality control of meta-aramid insulating paper in the production process.

[0003] However, there is currently no clear method for measuring the orderliness of chopped fibers. Traditional methods require sampling with relevant equipment or instruments (such as scanning electron microscopes) for observation and analysis. This method is cumbersome and labor-intensive. Sampling can cause some damage to the meta-aramid insulating paper, resulting in changes in the properties of the chopped fibers before and after sampling. This method of analyzing microscopic images under a microscope is difficult to provide a macroscopic interpretation of the orderliness of chopped fibers. Analyzing only a few fiber images makes the analysis results highly random and prone to errors, resulting in insufficient persuasiveness of the analysis results.

[0004] In summary, current traditional methods for assessing the orderliness of short-cut fibers in meta-aramid insulating paper suffer from low detection efficiency and accuracy. Summary of the Invention

[0005] This application provides a training method and a method for evaluating the orderliness of chopped fibers in meta-aramid insulating paper, as well as related devices and computer storage media, which can avoid damage to the sample to be tested and improve the efficiency and accuracy of orderliness evaluation.

[0006] Firstly, from the perspective of model training, embodiments of this application provide a training method for an evaluation model of the orderliness of short-cut fibers in meta-aramid insulating paper, comprising:

[0007] Sample images of meta-aramid insulating paper were obtained from the training samples of the dataset;

[0008] Based on the scale-invariant feature transformation algorithm, multi-scale feature extraction is performed on the sample image to obtain the sample features of each dimension corresponding to the sample image;

[0009] The sample features of each dimension are fused to obtain a fused comprehensive feature vector of sample with a preset length;

[0010] Based on the principal component analysis algorithm, the sample features in the sample comprehensive feature vector are filtered to obtain the sample dimensionality reduction feature vector corresponding to the sample comprehensive feature vector;

[0011] Based on the pre-labeled sample images associated with the sample labels and the sample dimensionality reduction feature vectors, the short-cut fiber orderliness evaluation model constructed based on the support vector machine algorithm is trained to obtain the trained short-cut fiber orderliness evaluation model.

[0012] Secondly, from the perspective of model application, embodiments of this application provide a method for evaluating the orderliness of short-cut fibers in meta-aramid insulating paper, the method comprising:

[0013] Obtain the image of the meta-aramid insulating paper to be evaluated;

[0014] Based on the scale-invariant feature transformation algorithm, multi-scale feature extraction is performed on the image to be evaluated to obtain the features to be evaluated in each dimension of the image to be evaluated.

[0015] The features to be evaluated in each dimension are fused to obtain a fused comprehensive feature vector to be evaluated with a preset length.

[0016] Based on the principal component analysis algorithm, each feature to be evaluated in the comprehensive feature vector to be evaluated is screened to obtain the dimension-reduced feature vector to be evaluated corresponding to the comprehensive feature vector to be evaluated.

[0017] The dimensionality-reduced feature vector to be evaluated is input into the trained short-cut fiber orderliness evaluation model as described in any of the first aspects above, so as to output the orderliness evaluation result associated with the image to be evaluated.

[0018] Thirdly, embodiments of this application provide a training device for evaluating the orderliness of chopped fibers in meta-aramid insulating paper, which has the function of implementing a training method corresponding to the evaluation model for the orderliness of chopped fibers in meta-aramid insulating paper provided in the first aspect above. This function can be implemented by hardware or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above function, and the modules can be software and / or hardware.

[0019] In one possible design, the device includes:

[0020] The image acquisition module is used to acquire sample images of meta-aramid insulating paper from the training samples of the dataset;

[0021] The feature extraction module is used to perform multi-scale feature extraction on the sample image based on the scale-invariant feature transformation algorithm to obtain the sample features of each dimension corresponding to the sample image.

[0022] The feature fusion module is used to fuse the sample features of each dimension to obtain a fused sample comprehensive feature vector with a preset length.

[0023] The feature filtering module is used to filter the sample features in the sample comprehensive feature vector based on the principal component analysis algorithm, so as to obtain the sample dimensionality reduction feature vector corresponding to the sample comprehensive feature vector.

[0024] The model training module is used to train the short-cut fiber orderliness evaluation model constructed based on the support vector machine algorithm according to the pre-labeled sample labels associated with the sample images and the sample dimensionality reduction feature vectors, so as to obtain the trained short-cut fiber orderliness evaluation model.

[0025] Fourthly, embodiments of this application provide a device for evaluating the orderliness of chopped fibers in meta-aramid insulating paper, which has the function of implementing the method for evaluating the orderliness of chopped fibers in meta-aramid insulating paper provided in the second aspect above. This function can be implemented by hardware or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above function, and the modules can be software and / or hardware.

[0026] In one possible design, the device includes:

[0027] The image acquisition module is used to acquire the image of the meta-aramid insulating paper to be evaluated.

[0028] The feature extraction module is used to extract features from the image to be evaluated at multiple scales based on the scale-invariant feature transformation algorithm, so as to obtain the features to be evaluated in each dimension of the image to be evaluated.

[0029] The feature fusion module is used to fuse the features to be evaluated in each dimension to obtain a fused comprehensive feature vector to be evaluated with a preset length.

[0030] The feature selection module is used to select each feature to be evaluated in the comprehensive feature vector to be evaluated based on the principal component analysis algorithm, so as to obtain the dimension reduction feature vector to be evaluated corresponding to the comprehensive feature vector to be evaluated.

[0031] The orderliness evaluation result output module is used to input the dimensionality reduction feature vector to be evaluated into the trained short-cut fiber orderliness evaluation model as described in any of the first aspects above, so as to output the orderliness evaluation result associated with the image to be evaluated.

[0032] In another aspect, this application provides a device for evaluating the orderliness of short-cut fibers in meta-aramid insulating paper and for training a model, which includes at least one connected processor and a memory, wherein the memory is used to store program code, and the processor is used to call the program code in the memory to execute the methods described in the above aspects.

[0033] In another aspect, this application provides a computer storage medium including instructions that, when executed on a computer, cause the computer to perform the methods described in the above aspects.

[0034] In another aspect, this application provides a computer program product containing instructions that, when run on a computer, cause the computer to perform the methods described in the above aspects.

[0035] Compared to traditional measurement methods in existing technologies, this embodiment obtains sample images of meta-aramid insulating paper from the training samples of the dataset, then performs multi-scale feature extraction based on the scale-invariant feature transformation algorithm, and then fuses the sample features of each dimension. Principal component analysis is then used to filter the sample features. Finally, a short-cut fiber order evaluation model constructed based on the support vector machine algorithm is trained based on the pre-labeled sample images, associated sample labels, and sample dimensionality-reduced feature vectors. This design provides a safer measurement method with fewer operational steps, avoiding damage to the sample during sampling and observation analysis, simplifying the process, reducing labor costs, and preventing damage to the meta-aramid insulating paper during sampling. Combining machine learning and image technology, the order of short-cut fibers is determined based on computed tomography images, providing a macroscopic interpretation of the order of short-cut fibers, avoiding the randomness of traditional methods, reducing observation errors, and improving the efficiency and accuracy of detection. Attached Figure Description

[0036] Figure 1 This is an application environment diagram of the training method for evaluating the orderliness of short-cut fibers in meta-aramid insulating paper in one embodiment;

[0037] Figure 2 This is a flowchart illustrating the training method for a model to evaluate the orderliness of short-cut fibers in meta-aramid insulating paper in one embodiment.

[0038] Figure 3 This is a comparison chart of the effects of image preprocessing in one embodiment;

[0039] Figure 4 A diagram illustrating key points in one embodiment;

[0040] Figure 5This is a histogram of orientation points for key points in one embodiment;

[0041] Figure 6 This is a visualization of the principal component analysis results in one embodiment;

[0042] Figure 7 This is a schematic diagram of the classification results of the model in one embodiment;

[0043] Figure 8 This is a flowchart illustrating the training method for the short-cut fiber order evaluation model in meta-aramid insulating paper in another embodiment.

[0044] Figure 9 This is a flowchart illustrating a method for evaluating the orderliness of short-cut fibers in meta-aramid insulating paper in one embodiment.

[0045] Figure 10 This is a technical roadmap for analyzing the orderliness of chopped fibers in one embodiment;

[0046] Figure 11 This is a structural block diagram of a training device for evaluating the orderliness of short-cut fibers in meta-aramid insulating paper in one embodiment.

[0047] Figure 12 This is a structural block diagram of a device for evaluating the orderliness of short-cut fibers in meta-aramid insulating paper in one embodiment;

[0048] Figure 13 This is an internal structural diagram of a device for evaluating the orderliness of short-cut fibers in meta-aramid insulating paper and training a model in one embodiment.

[0049] Figure 14 This is an internal structural diagram of the device for evaluating the orderliness of short-cut fibers in meta-aramid insulating paper and training a model, as shown in another embodiment. Detailed Implementation

[0050] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or modules is not necessarily limited to those explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products, or devices. The module divisions appearing in the embodiments of this application are merely logical divisions; in actual applications, there may be other division methods. For example, multiple modules may be combined into or integrated into another system, or some features may be ignored or not performed. Furthermore, the shown or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces, and the indirect couplings or communication connections between modules may be electrical or other similar forms. These are not limited in the embodiments of this application. Moreover, modules or sub-modules described as separate components may or may not be physically separated, may or may not be physical modules, or may be distributed across multiple circuit modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the embodiments of this application.

[0051] This application provides a training method for evaluating the orderliness of short-cut fibers in meta-aramid insulating paper, which can be applied to applications such as... Figure 1 In the application scenario shown, terminal 102 communicates with server 104 via a network.

[0052] The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle systems. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0053] It should be specifically noted that the terminal 102 involved in this application embodiment can be a wired terminal or a wireless terminal. It can be a device that provides voice and / or data connectivity to a user, a handheld device with wireless connectivity, or other processing devices connected to a wireless modem. The wireless terminal can communicate with one or more core networks via a Radio Access Network (RAN). The wireless terminal can be a mobile terminal, such as a mobile phone (or "cellular" phone) and a computer with a mobile terminal. For example, it can be a portable, pocket-sized, handheld, computer-embedded, or vehicle-mounted mobile device that exchanges voice and / or data with the radio access network. Examples include Personal Communication Service (PCS) phones, cordless phones, Session Initiation Protocol (SIP) phones, Wireless Local Loop (WLL) stations, and Personal Digital Assistants (PDAs). Wireless terminals can also be referred to as systems, subscriber units, subscriber stations, mobile stations, mobile stations, remote stations, access points, remote terminals, access terminals, user terminals, terminal equipment, user agents, user devices, or user equipment.

[0054] Please refer to Figure 2 The following describes a method for training a model for evaluating the orderliness of short-cut fibers in meta-aramid insulating paper, as provided in the embodiments of this application. From the perspective of model training, the embodiments of this application mainly include:

[0055] S201, Obtain sample images of meta-aramid insulating paper from the training samples of the dataset.

[0056] The dataset consists of training and testing samples, used for training and testing the chopped fiber order evaluation model. The sample images are images of meta-aramid insulating paper, an industrial product whose chopped fiber order is closely related to the product's quality and performance.

[0057] For example, a dataset of 200 micro-CT (Computed Tomography) scan images of meta-aramid insulating paper was obtained by computed tomography, which included 150 training samples and 50 test samples. The image size can be 1929×1222 pixels, the horizontal and vertical resolution can be 96 dpi, and the bit depth can be 24.

[0058] S202, based on the scale-invariant feature transformation algorithm, performs multi-scale feature extraction on the sample image to obtain the sample features of each dimension corresponding to the sample image.

[0059] In the feature extraction process, the Scale-Invariant Feature Transform (SIFT) algorithm is used. The sample features of each dimension can be obtained by representing and quantizing the key points of the image respectively.

[0060] S203, fuse the sample features of each dimension to obtain a fused sample comprehensive feature vector with a preset length.

[0061] Feature fusion, also known as feature splicing, has a preset length that is a fixed length of the spliced ​​or fused vector. The sample comprehensive feature vector is the vector obtained after fusion.

[0062] S204, based on the principal component analysis algorithm, filters the features of each sample in the sample comprehensive feature vector to obtain the sample dimensionality-reduced feature vector corresponding to the sample comprehensive feature vector.

[0063] In the feature selection process, the Principal Component Analysis (PCA) algorithm is used. The sample dimensionality reduction feature vector refers to the vector obtained after reducing the dimensionality of the sample comprehensive feature vector.

[0064] S205, based on the sample labels associated with the pre-labeled sample images and the sample dimensionality reduction feature vectors, the short-cut fiber orderliness evaluation model constructed based on the support vector machine algorithm is trained to obtain the trained short-cut fiber orderliness evaluation model.

[0065] The sample labels associated with the pre-labeled sample images are constructed as follows:

[0066] By collecting a large number of short-fiber CT image samples with known ordered labels (the labels can be derived from precise manual annotation, numerical simulation data, or gold standard measurement methods), and uniformly executing the above feature extraction process, the system constructs a training dataset for supervised learning. The dataset is in the form of standardized feature vectors and data pairs that match ordered labels, providing a high-quality and interpretable training foundation for subsequent machine learning models.

[0067] Among them, the short-cut fiber orderliness evaluation model is a neural network model built based on the Support Vector Machine (SVM) algorithm. For example, in the machine learning model training and orderliness recognition stages, the SVM supervised learning model is used to train the constructed feature vector dataset. This model can efficiently handle small sample, high-dimensional feature classification problems and has good generalization ability. Through the training process, SVM learns the accurate mapping relationship between the fixed-length feature vector composed of the SIFT directional statistical histogram and other auxiliary features and the final orderliness label (such as "ordered / unordered" classification or continuous regression value).

[0068] Compared to traditional measurement methods in existing technologies, this embodiment obtains sample images of meta-aramid insulating paper from the training samples of the dataset, then performs multi-scale feature extraction based on the scale-invariant feature transformation algorithm, and then fuses the sample features of each dimension. Principal component analysis is then used to filter the sample features. Finally, a short-cut fiber order evaluation model constructed based on the support vector machine algorithm is trained based on the pre-labeled sample images, associated sample labels, and sample dimensionality-reduced feature vectors. This design provides a safer measurement method with fewer operational steps, avoiding damage to the sample during sampling and observation analysis, simplifying the process, reducing labor costs, and preventing damage to the meta-aramid insulating paper during sampling. Combining machine learning and image technology, the order of short-cut fibers is determined based on computed tomography images, providing a macroscopic interpretation of the order of short-cut fibers, avoiding the randomness of traditional methods, reducing observation errors, and improving the efficiency and accuracy of detection.

[0069] Furthermore, due to the significant noise and partial beam hardening artifacts in the original CT images, the training method for the short-cut fiber orderliness evaluation model provided in this application can also include: preprocessing the sample images to obtain preprocessed sample images. Specifically, this includes using the MATLAB image processing toolbox to preprocess each slice image using a wavelet transform denoising method (e.g., using the "sym4" wavelet basis with a decomposition layer of 3), such as... Figure 3 As shown, Figure 3 The study demonstrates the effectiveness of preprocessing for shredded fiber CT images. Preprocessing effectively suppresses random noise while preserving the details of the fiber edges, significantly improving the signal-to-noise ratio (SNR by an average of approximately 5 dB).

[0070] In this embodiment, the advanced image processing algorithm wavelet transform denoising is used to preprocess the image, which suppresses noise, reduces artifacts, and enhances the image signal-to-noise ratio, providing a data foundation for subsequent feature extraction and orderliness analysis.

[0071] Optionally, in some embodiments of this application, based on the scale-invariant feature transformation algorithm, multi-scale feature extraction is performed on the sample image to obtain sample features of each dimension corresponding to the sample image, including: multi-scale detection of the sample image based on the scale-invariant feature transformation algorithm to obtain each image key point; feature representation of each image key point according to the spatial and dimensional characteristics of the image key points to obtain each represented feature; and quantization of the represented features to obtain sample features of each dimension corresponding to the sample image.

[0072] Among them, each image key point is a key point extracted based on multi-scale SIFT features. The spatial and dimensional characteristics of the image key points refer to the characteristics of each key point, such as position, scale, and orientation. The sample features are the features obtained after final quantization.

[0073] For example, in the multi-scale SIFT feature extraction and quantization stage, the multi-scale SIFT algorithm is first executed on the preprocessed image. This algorithm, with its inherent invariance to changes in light, rotation, and scale, can effectively resist common gray-scale unevenness and blur artifacts in CT images. Subsequently, each detected SIFT keypoint is characterized by features. Each keypoint includes position, scale, orientation (0-360°), and a 128-dimensional feature descriptor. Then, feature quantization transforms the fiber orderliness evaluation problem into a statistical modeling problem of keypoint orientation information.

[0074] More specifically, the preprocessed images are batch-imported into a pre-written Matlab processing flow. The SIFT interface is used to extract multi-scale features from each image, with a contrast threshold of 0.03 and an edge threshold of 10. On average, approximately 500 to 800 stable SIFT keypoints can be extracted from a typical slice image. Figure 4 As shown, each key point is characterized, and its orientation angle is used for subsequent statistics.

[0075] In this embodiment, by relying on multi-scale detection to accurately capture key points of the image, combining spatial and dimensional characteristics to characterize features, and then enhancing the accuracy of features through quantization, effective information of the image can be extracted comprehensively, greatly improving the accuracy and reliability of feature extraction, and providing high-quality feature data for subsequent processes.

[0076] Optionally, in some embodiments of this application, the characterized features are quantized to obtain sample features of each dimension corresponding to the sample image, including: determining the orientation angles of each key point of each image corresponding to each two-dimensional image to construct an orientation histogram; determining the three-dimensional orientation vectors of each key point of each image corresponding to each three-dimensional volume data image to obtain the data distribution in the spherical coordinate system; and quantizing each characterized feature based on the orientation histogram and the data distribution in the spherical coordinate system to obtain sample features of each dimension corresponding to the sample image.

[0077] The sample images include two-dimensional images and three-dimensional volume data images, namely two-dimensional slices or three-dimensional volume data of the original shredded fiber CT scan images. Specifically, for two-dimensional images, the orientation angles of all key points are directly counted to construct an orientation histogram, while for three-dimensional volume data, the three-dimensional orientation vectors of the key points are calculated and their distribution in the spherical coordinate system is statistically analyzed.

[0078] For example, Figure 5 The orientation histogram of the detected key points is shown, such as... Figure 5 As shown, in this embodiment, for two-dimensional slice images, the orientation angles (0-360°) of all key points are statistically analyzed at 10° intervals to generate a 36-dimensional orientation histogram.

[0079] In this embodiment, an appropriate quantization method can be selected based on the different characteristics of two-dimensional and three-dimensional images, accurately capturing key information of images in different dimensions, making feature quantization more targeted and accurate, and providing more reliable data support for subsequent operations.

[0080] Optionally, in some embodiments of this application, the sample features of each dimension are fused to obtain a fused sample comprehensive feature vector with a preset length, including: extracting the key point density corresponding to the image key points and the distribution of the image key points at different scales; using the key point density and the distribution at different scales as auxiliary features to construct the sample comprehensive feature vector.

[0081] It is understood that during the feature engineering process, this application can also extract key point density and its distribution at different scales as auxiliary features to jointly form a multi-dimensional feature vector expressing the ordered state of fibers. For example, the orientation statistical histogram obtained by multi-scale SIFT analysis of each CT image and other auxiliary features (such as key point density and scale distribution) are fused and spliced ​​into a fixed-length comprehensive feature vector. This vector efficiently encodes the overall orientation pattern and spatial distribution characteristics of fibers in the image.

[0082] In this embodiment, by introducing key point auxiliary features, the constructed feature vector can more comprehensively express the fiber ordering status, greatly improving the information carrying capacity of the feature vector and providing richer and more reliable data support for subsequent processes.

[0083] Optionally, in some embodiments of this application, based on the principal component analysis algorithm, the sample features in the sample comprehensive feature vector are screened to obtain the sample dimensionality-reduced feature vector corresponding to the sample comprehensive feature vector. This includes: calculating the variance, correlation, and contribution to the classification target of each sample feature in the sample comprehensive feature vector; removing redundant and irrelevant features based on the variance, correlation, and contribution to the classification target to obtain the retained core features; and mapping the core features to a low-dimensional space through linear or nonlinear transformation to obtain the sample dimensionality-reduced feature vector.

[0084] For example, the high-dimensional feature vector is first processed by the feature selection and dimensionality reduction algorithm Principal Component Analysis (PCA): by calculating the variance, correlation and contribution of the features to the classification target, redundant and irrelevant feature information is removed, and the core features are mapped to a low-dimensional space through linear or nonlinear transformations.

[0085] More specifically, the initially constructed feature vectors contain some redundancy. Therefore, this embodiment uses principal component analysis to reduce the dimensionality of the 64-dimensional features of the training set. Analysis reveals that the first 43 principal components can cumulatively explain 92.18% of the variance of the original data. Figure 6 The visualization of the principal component analysis results is shown, such as... Figure 6 As shown. Therefore, we choose to map the original 64-dimensional features onto these 43 principal components to form a new 43-dimensional low-dimensional feature vector.

[0086] In this embodiment, not only is the data dimensionality significantly reduced, avoiding the curse of dimensionality and eliminating noisy features, but the representativeness and separability of the feature set are also effectively improved, and the computational efficiency is increased. This provides cleaner and more efficient input data for subsequent model training, ultimately greatly improving the training speed, prediction accuracy and generalization ability of the model. In addition, it can also provide quantitative indicators of feature importance, providing data support for the physical mechanism analysis of fiber order.

[0087] In addition, this application also includes validating the trained model using the validation set in the dataset, specifically including:

[0088] The dimensionality-reduced 43-dimensional feature dataset was randomly divided into an 80% training set and a 20% validation set. The Support Vector Machine (SVM) algorithm in Matlab was used for model training, with a Radial Basis Function (RBF) selected as the kernel function. Grid search and 5-fold cross-validation were used to optimize the penalty parameter C and the kernel coefficient gamma. After training, the optimal model achieved an R² = 93.21% determination coefficient on the validation set, demonstrating excellent prediction accuracy and generalization ability. Figure 7 The model's classification results are shown, such as... Figure 7 As shown, the closer the SIFT orientation concentration is to 10, the more concentrated the fiber orientation is and the higher the degree of order.

[0089] In another embodiment, such as Figure 8 As shown, a training method for an evaluation model of the orderliness of short-cut fibers in meta-aramid insulating paper is provided, including the following steps:

[0090] S801, Obtain sample images of meta-aramid insulating paper from the training samples of the dataset;

[0091] S802, based on the scale-invariant feature transformation algorithm, performs multi-scale detection on sample images to obtain key points of each image, performs feature representation on each key point of the image, and obtains the represented features.

[0092] S803, determine the orientation angles of each key point in each two-dimensional image to construct an orientation histogram, and determine the three-dimensional orientation vectors of each key point in each three-dimensional volume data image to obtain the data distribution in the spherical coordinate system;

[0093] S804 quantizes each represented feature to obtain the sample features of each dimension corresponding to the sample image.

[0094] S805 extracts the keypoint density and the distribution of keypoints at different scales in the image, and uses them as auxiliary features to construct a sample comprehensive feature vector.

[0095] S806 calculates the variance, correlation, and contribution to the classification target of each sample feature in the sample comprehensive feature vector, and removes redundant and irrelevant features to obtain the core features to be retained.

[0096] S807 maps core features to a low-dimensional space through linear or nonlinear transformations to obtain sample dimensionality-reduced feature vectors.

[0097] S808: Based on the sample labels associated with the pre-labeled sample images and the sample dimensionality reduction feature vectors, the short-cut fiber orderliness evaluation model constructed based on the support vector machine algorithm is trained to obtain the trained short-cut fiber orderliness evaluation model.

[0098] It should be noted that the specific limitations of the above steps can be found in the specific limitations of the training method for a meta-aramid insulating paper short fiber order evaluation model mentioned above, and will not be repeated here.

[0099] In addition, from the perspective of model application, such as Figure 9 As shown in the embodiments of this application, a method for evaluating the orderliness of short-cut fibers in meta-aramid insulating paper is provided. The method includes:

[0100] S901, Obtain the image of the meta-aramid insulating paper to be evaluated.

[0101] Among them, the image of the meta-aramid insulating paper to be evaluated is the image of the product to be tested or evaluated during the model application stage.

[0102] S902, based on the scale-invariant feature transformation algorithm, performs multi-scale feature extraction on the image to be evaluated, and obtains the features to be evaluated in each dimension of the image to be evaluated.

[0103] S903 performs feature fusion on the features to be evaluated in each dimension to obtain a fused comprehensive feature vector to be evaluated with a preset length.

[0104] S904, based on the principal component analysis algorithm, filters each feature to be evaluated in the comprehensive feature vector to be evaluated, and obtains the dimension-reduced feature vector to be evaluated corresponding to the comprehensive feature vector to be evaluated.

[0105] It should be noted that the processes of image acquisition or preprocessing, feature extraction and fusion, and filtering in steps S901 to S904 are based on the same technical principles as... Figure 2 The technical principles in the embodiments correspond to the data processing parts of the model application stage and the model training stage, so they will not be elaborated here.

[0106] S905 inputs the dimensionality-reduced feature vector to be evaluated into any trained short-cut fiber orderliness evaluation model to output the orderliness evaluation result associated with the image to be evaluated.

[0107] For example, in practical recognition applications, for new, unknown shredded fiber CT images, the feature vector generated after preprocessing, feature extraction and quantization is input into the trained SVM model, which can quickly and automatically output the quantitative evaluation result of its fiber orderliness.

[0108] Tests showed that after processing unknown test samples as described above, inputting them into a trained SVM model allowed the system to output a quantitative order prediction value within an average of 0.5 seconds, achieving rapid and automated recognition.

[0109] Optionally, in some embodiments of this application, the method for evaluating the orderliness of chopped fibers in meta-aramid insulating paper further includes: obtaining comprehensive orientation reports corresponding to meta-aramid insulating paper with different processes and formulations, constructing a standard database of chopped fiber orderliness based on data aggregation; obtaining comprehensive quantitative indicators from the orderliness evaluation results associated with the image to be evaluated, comparing and fitting the comprehensive quantitative indicators with the standard database, and generating a target evaluation report of chopped fiber orderliness.

[0110] The comprehensive orientation report is the result of each assessment, while the target assessment report refers to the final complete assessment result obtained after comparison with the database.

[0111] For example, in this embodiment, a standard database can be established, which specifically includes: collecting a large number of samples with different processes and formulations, evaluating them using a trained model, obtaining their comprehensive orientation reports, and summarizing these orderliness data to construct a standard database of chopped fiber orderliness.

[0112] Then, the orientation of short-cut fibers in unknown samples can be evaluated by image analysis. Specifically, for a new unknown sample, its comprehensive quantitative index is obtained through model evaluation. Then, these indexes are compared and fitted with the established standard database to achieve a rapid and non-destructive evaluation of the orderliness of its short-cut fibers.

[0113] Furthermore, the training method for the short-cut fiber order evaluation model in meta-aramid insulating paper and the short-cut fiber order evaluation method provided in this application both belong to the process of analyzing the order of short-cut fibers in meta-aramid insulating paper based on multi-scale SIFT features and machine learning. Figure 10 This demonstrates the technical approach for analyzing the orderliness of chopped fibers. The following section combines... Figure 10 This provides an overall explanation of the orderliness analysis process.

[0114] In this embodiment, computed tomography (CT) images of chopped fibers are acquired, and a well-trained machine learning model is used to quickly and accurately identify their orderliness. In practical applications, chopped fiber images acquired by CT scans may produce some unavoidable artifacts due to lighting or operational errors, so image downsampling is necessary. By combining handcrafted SIFT features with machine learning, effective information from the images is extracted for the orderliness analysis of chopped fibers.

[0115] The overall approach of this embodiment is as follows: instead of directly performing difficult segmentation or recognition on the original image that may contain artifacts, it extracts SIFT feature points that can resist slight image degradation and have strong physical significance. The statistical information of these feature points is used as input to a machine learning model to indirectly and robustly evaluate the overall orderliness. This allows for the orderliness analysis of the orientation of short-cut fibers in a certain type of meta-aramid insulating paper. Based on machine learning, object recognition is performed. Combining image processing and machine learning to identify short-cut fibers with different properties makes the evaluation results more scientifically interpretable.

[0116] Figures 1 to 10 Any technical feature in the embodiments corresponding to any of the above items is also applicable to the embodiments of this application. Figures 11 to 14 The corresponding implementation examples will not be repeated hereafter.

[0117] The model training and evaluation methods in the embodiments of this application have been described above. The apparatus for performing the above methods is described below.

[0118] Reference Figure 11 The training device for evaluating the orderliness of short-cut fibers in meta-aramid insulating paper is described. The device includes:

[0119] Image acquisition module 1101 is used to acquire sample images of meta-aramid insulating paper from training samples of the dataset;

[0120] The feature extraction module 1102 is used to perform multi-scale feature extraction on the sample image based on the scale-invariant feature transformation algorithm to obtain the sample features of each dimension corresponding to the sample image.

[0121] The feature fusion module 1103 is used to fuse the sample features of each dimension to obtain a fused comprehensive feature vector of the sample with a preset length.

[0122] The feature filtering module 1104 is used to filter the features of each sample in the sample comprehensive feature vector based on the principal component analysis algorithm, so as to obtain the sample dimension-reduced feature vector corresponding to the sample comprehensive feature vector.

[0123] The model training module 1105 is used to train the short-cut fiber orderliness evaluation model constructed based on the support vector machine algorithm according to the sample labels associated with the pre-labeled sample images and the sample dimensionality reduction feature vectors, so as to obtain the trained short-cut fiber orderliness evaluation model.

[0124] Reference Figure 12 A device for evaluating the orderliness of short-cut fibers in meta-aramid insulating paper is described, the device comprising:

[0125] The image acquisition module 1201 is used to acquire the image of the meta-aramid insulating paper to be evaluated.

[0126] The feature extraction module 1202 is used to extract features from the image to be evaluated at multiple scales based on the scale-invariant feature transformation algorithm, so as to obtain the features to be evaluated in each dimension of the image to be evaluated.

[0127] The feature fusion module 1203 is used to fuse the features to be evaluated in various dimensions to obtain a fused comprehensive feature vector to be evaluated with a preset length.

[0128] The feature selection module 1204 is used to select each feature to be evaluated in the comprehensive feature vector to be evaluated based on the principal component analysis algorithm, so as to obtain the dimension reduction feature vector to be evaluated corresponding to the comprehensive feature vector to be evaluated.

[0129] The orderliness evaluation result output module 1205 is used to input the dimensionality reduction feature vector to be evaluated into any trained short-cut fiber orderliness evaluation model, so as to output the orderliness evaluation result associated with the image to be evaluated.

[0130] In this embodiment, through the cooperation of the above modules, a safe measurement method with fewer operation steps is designed, avoiding damage to the sample to be tested during sampling, observation and analysis, simplifying the steps, reducing labor costs, and avoiding damage to the meta-aramid insulating paper during sampling; combining machine learning and image technology, the degree of order of chopped fibers is determined based on computed tomography images, and a macroscopic interpretation of the degree of order of chopped fibers is made, avoiding the randomness of traditional methods, reducing observation errors, and improving the efficiency and accuracy of detection.

[0131] In another embodiment, a device for evaluating the orderliness of chopped fibers in meta-aramid insulating paper and for training a model is provided. This device can be a computer device, such as a server, and its internal structure diagram can be as follows: Figure 13As shown, the device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The device's database stores relevant data. The I / O interfaces are used for exchanging information between the processor and external devices. The device's communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements the methods described in the above embodiments.

[0132] In yet another embodiment, a device for evaluating the orderliness of short-cut fibers in meta-aramid insulating paper and for training a model is provided. This device can be a computer device, such as a terminal, and its internal structure diagram can be as follows: Figure 14 As shown, the device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements the methods described in the above embodiments. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the computer equipment casing, or external keyboards, touchpads, or mice, etc.

[0133] Those skilled in the art will understand that Figure 13 and Figure 14 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the device to which the present application is applied. The specific device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements, in order to realize the functions of computer devices such as terminals or servers.

[0134] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0135] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0136] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, apparatuses, or modules, and may be electrical, mechanical, or other forms.

[0137] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0138] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0139] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0140] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state disk (SSD)).

[0141] The technical solutions provided in the embodiments of this application have been described in detail above. Specific examples have been used in the embodiments of this application to illustrate the principles and implementation methods of the embodiments of this application. The description of the above embodiments is only for the purpose of helping to understand the methods and core ideas of the embodiments of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the embodiments of this application. Therefore, the content of this specification should not be construed as a limitation on the embodiments of this application.

Claims

1. A training method for an evaluation model of the orderliness of short-cut fibers in meta-aramid insulating paper, characterized in that, The training method for the short-cut fiber order evaluation model in the meta-aramid insulating paper includes: Sample images of meta-aramid insulating paper were obtained from the training samples of the dataset; Based on the scale-invariant feature transformation algorithm, multi-scale feature extraction is performed on the sample image to obtain the sample features of each dimension corresponding to the sample image; The sample features of each dimension are fused to obtain a fused comprehensive feature vector of sample with a preset length; Based on the principal component analysis algorithm, the sample features in the sample comprehensive feature vector are filtered to obtain the sample dimensionality reduction feature vector corresponding to the sample comprehensive feature vector; Based on the pre-labeled sample images associated with the sample labels and the sample dimensionality reduction feature vectors, the short-cut fiber orderliness evaluation model constructed based on the support vector machine algorithm is trained to obtain the trained short-cut fiber orderliness evaluation model.

2. The method according to claim 1, characterized in that, The scale-invariant feature transformation algorithm performs multi-scale feature extraction on the sample image to obtain sample features of each dimension corresponding to the sample image, including: Based on the scale-invariant feature transformation algorithm, multi-scale detection is performed on the sample images to obtain key points for each image; Based on the spatial and dimensional characteristics of the image key points, feature representation is performed on each of the image key points to obtain the represented features. The represented features are quantized to obtain the sample features of each dimension corresponding to the sample image.

3. The method according to claim 2, characterized in that, The sample image includes two-dimensional images and three-dimensional volumetric data images. The quantization of the represented features to obtain sample features corresponding to each dimension of the sample image includes: Determine the orientation angles of each key point in each of the two-dimensional images to construct an orientation histogram; Determine the three-dimensional direction vectors of each key point in each of the three-dimensional volume data images to obtain the data distribution in spherical coordinates; Based on the orientation histogram and the data distribution in the spherical coordinate system, the represented features are quantized to obtain the sample features of each dimension corresponding to the sample image.

4. The method according to claim 1, characterized in that, The step of fusing the sample features of each dimension to obtain a fused comprehensive feature vector of a preset length includes: The keypoint density and the distribution of keypoints at different scales in the image are extracted. The sample comprehensive feature vector is constructed by using the keypoint density and the distribution at different scales as auxiliary features.

5. The method according to claim 1, characterized in that, The principal component analysis algorithm is used to filter the sample features in the comprehensive feature vector of the samples to obtain the sample dimensionality-reduced feature vector corresponding to the comprehensive feature vector of the samples, including: Calculate the variance, correlation, and contribution to the classification target of each sample feature in the comprehensive feature vector of the samples; Redundant and irrelevant features are removed based on the variance, correlation, and contribution to the classification target to obtain the retained core features; The core features are mapped to a low-dimensional space through linear or nonlinear transformations to obtain the sample's dimensionality-reduced feature vector.

6. A method for evaluating the degree of order of short-cut fibers in meta-aramid insulating paper, characterized in that, The method for evaluating the orderliness of short-cut fibers in the meta-aramid insulating paper includes: Obtain the image of the meta-aramid insulating paper to be evaluated; Based on the scale-invariant feature transformation algorithm, multi-scale feature extraction is performed on the image to be evaluated to obtain the features to be evaluated in each dimension of the image to be evaluated. The features to be evaluated in each dimension are fused to obtain a fused comprehensive feature vector to be evaluated with a preset length. Based on the principal component analysis algorithm, each feature to be evaluated in the comprehensive feature vector to be evaluated is screened to obtain the dimension-reduced feature vector to be evaluated corresponding to the comprehensive feature vector to be evaluated. The dimensionality-reduced feature vector to be evaluated is input into the short-cut fiber orderliness evaluation model trained as described in any one of claims 1 to 5 above, so as to output the orderliness evaluation result associated with the image to be evaluated.

7. The method according to claim 6, characterized in that, The method for evaluating the orderliness of short-cut fibers in the meta-aramid insulating paper also includes: Obtain comprehensive orientation reports for meta-aramid insulating paper with different processes and formulations, and construct a standard database of chopped fiber order based on the data summary; Obtain the comprehensive quantitative index from the orderliness evaluation results associated with the image to be evaluated, compare and fit the comprehensive quantitative index with the standard database, and generate a target evaluation report on the orderliness of chopped fibers.

8. A training device for evaluating the orderliness of short-cut fibers in meta-aramid insulating paper, characterized in that, The device includes: The image acquisition module is used to acquire sample images of meta-aramid insulating paper from the training samples of the dataset; The feature extraction module is used to perform multi-scale feature extraction on the sample image based on the scale-invariant feature transformation algorithm to obtain the sample features of each dimension corresponding to the sample image. The feature fusion module is used to fuse the sample features of each dimension to obtain a fused sample comprehensive feature vector with a preset length. The feature filtering module is used to filter the sample features in the sample comprehensive feature vector based on the principal component analysis algorithm, so as to obtain the sample dimensionality reduction feature vector corresponding to the sample comprehensive feature vector. The model training module is used to train the short-cut fiber orderliness evaluation model constructed based on the support vector machine algorithm according to the pre-labeled sample labels associated with the sample images and the sample dimensionality reduction feature vectors, so as to obtain the trained short-cut fiber orderliness evaluation model.

9. A device for evaluating the orderliness of short-cut fibers in meta-aramid insulating paper and for training a model, characterized in that, The device includes: At least one processor and memory; The memory is used to store program code, and the processor is used to call the program code stored in the memory to execute the method as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, It includes instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1 to 7.

Images