Fiber recognition system in antibacterial polyester-cotton composite fiber based on graphic data analysis

Abstract

This invention discloses a fiber identification system for antibacterial polyester-cotton composite fibers based on image data analysis. The system includes a composite fiber sample terminal, a dual-channel microscopic imaging terminal, a fiber image preprocessing and segmentation terminal, and a fiber classification and identification terminal. The invention achieves multi-modal optical signal collaborative acquisition through the dual-channel microscopic imaging terminal, addressing the problem that a single imaging method cannot comprehensively acquire fiber morphology and texture features. It uses an adaptive distance regularization level set algorithm to dynamically adjust the weight of the regularization term, solving the problem of inaccurate segmentation of fiber adhesion regions in traditional segmentation methods. Simultaneously, it achieves accurate segmentation of fiber regions and background separation through a deep learning network model and the OTSU adaptive threshold algorithm. Finally, it uses multi-feature fusion and recombination, and supports a vector machine for classification and identification training output, solving the problem of insufficient identification detail in composite fiber identification and significantly improving the system's accuracy in identifying composite fibers.

Description

Technical Field

[0001] This invention relates to the field of textile fabric identification technology, specifically to a fiber identification system for antibacterial polyester-cotton composite fibers based on graphic data analysis. Background Technology

[0002] Fiber identification in antibacterial polyester-cotton composite fibers refers to the process of accurately distinguishing and identifying different fiber components in composite fiber materials composed of polyester (polyester fiber) and cotton fiber, which have undergone special treatment to acquire antibacterial functions. This process usually involves a comprehensive analysis and judgment of multiple aspects such as the chemical composition, physical structure, performance characteristics, and antibacterial effect of the fibers. Currently, image recognition technology is mostly used in textiles and textile products for the identification of fibers and fabric structures. By acquiring images of antibacterial polyester-cotton composite fibers, image processing algorithms and machine learning models are used to analyze and process the images, thereby achieving automatic fiber identification.

[0003] However, in composite fibers, it is often necessary to identify multiple fiber components simultaneously. These different fiber components have different characteristics and backgrounds, and most composite fabrics use adhesives during processing. These adhesives are difficult to completely remove during the quantitative analysis of fiber components, leading to unavoidable problems of multi-fiber crossing and adhesion during fiber identification. This results in the obscuring of some fiber features, often requiring fiber identification and classification even with missing features. Furthermore, each fiber component possesses unique physical, chemical, and optical properties, which not only demands high accuracy and resolution in image recognition technology but also the ability to distinguish and understand the subtle differences between these different fiber components and how they interact in the composite structure. Existing fiber identification systems do not have a high degree of detail when identifying image data acquired under complex conditions such as multi-fiber crossing and obscuring, resulting in poor accuracy in fiber identification among multiple fibers and ultimately, lower accuracy in correct classification. Summary of the Invention

[0004] The purpose of this invention is to provide a fiber identification system for antibacterial polyester-cotton composite fibers based on graphic data analysis, so as to solve the problems mentioned in the background art.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a fiber identification system for antibacterial polyester-cotton composite fibers based on image data analysis, comprising the following terminals: a composite fiber sample terminal, a dual-channel microscopic imaging terminal, a fiber image preprocessing and segmentation terminal, and a fiber classification and identification terminal; the composite fiber sample terminal is used to store the original composite fiber sample; the dual-channel microscopic imaging terminal is used to collaboratively acquire composite fiber image feature information through multimodal optical signals; the fiber image preprocessing and segmentation terminal is used to retrieve the target composite fiber image dataset from the composite fiber sample imaging database for region segmentation and corresponding label annotation; the fiber classification and identification terminal... The identification terminal is used to fuse and reconstruct the morphological and textural features of composite fibers, and to classify and identify fibers using a support vector machine. The composite fiber sample terminal is connected to the dual-channel microscopic imaging terminal network, the fiber image preprocessing and segmentation terminal is connected to the dual-channel microscopic imaging terminal network, and the fiber classification and identification terminal is connected to the fiber image preprocessing and segmentation terminal network. The dual-channel microscopic imaging terminal includes a first channel and a second channel. The first channel includes a confocal microscope, a polarization imaging unit, and a host computer module. The second channel includes a Raman spectroscopy point scanning unit, a multispectral fluorescence labeling unit, and a host computer module.

[0006] The specific operation process of the dual-channel microscopic imaging terminal includes the following steps:

[0007] Step S11: Input the original composite fiber sample to be tested into the confocal microscope in the first channel, scan the sample layer by layer with laser, construct the three-dimensional structure of the fiber based on the generated three-dimensional fiber volume data of the scan, and obtain the tomographic image of the sample to be tested.

[0008] Step S12: Based on the polarization imaging unit, rotate the polarizer group and sequentially acquire polarization angles of 0°, 45°, 90° and 135° in the tomographic image of the sample to be tested to obtain the fiber image of the sample to be tested with the maximum contrast at different angles.

[0009] Step S13: Simultaneously input the original composite fiber sample to be tested into the second channel, acquire the scattering spectrum image of the sample to be tested based on confocal microscopy, and use the multispectral fluorescent labeling unit to excite the fluorescent label in the sample to be tested, and acquire the fluorescence spectrum image of the sample to be tested at different wavelengths.

[0010] Step S14: Transmit the fiber image of the test sample with the maximum contrast obtained by the first channel to the host computer module one, and transmit the scattering spectrum image and fluorescence spectrum image of the test sample obtained by the second channel to the host computer module two.

[0011] Step S15: Integrate and summarize the image data of the test samples from the host computer module 1 and the host computer module 2 to obtain a composite fiber image dataset, and store it in the composite fiber test sample imaging database.

[0012] Preferably, the fiber image preprocessing segmentation terminal includes: a composite fiber target region segmentation module, an adaptive distance regularization level set algorithm module, a composite fiber morphological feature measurement and extraction module, and a texture feature extraction and calculation module; the composite fiber target region segmentation module is used to divide and mark multiple connected regions of the target composite fiber image; the adaptive distance regularization level set algorithm module is used to dynamically adjust the weight of the regularization term using image gradient information to obtain a binary image of the fiber contour of each target region; the composite fiber morphological feature measurement and extraction module is used to measure and extract the morphological features of the composite fiber, including direct and relative indicators; the texture feature extraction and calculation module is used to extract the gray-level co-occurrence matrix and calculate the feature parameters based on the gray-level co-occurrence matrix.

[0013] Preferably, the operation process of the fiber image preprocessing and segmentation terminal includes the following steps:

[0014] Step S21: Based on the composite fiber target region segmentation module, the target composite fiber image is divided into connected or disconnected fiber regions, and the constructed deep learning network model is used to segment isolated fiber regions and fiber adhesion regions.

[0015] Step S22: Initialize the level set function using the adaptive distance regularization level set algorithm module, set its zero level set as the initial contour, select the target image based on the labeled multi-region binary image, calculate the image gradient, calculate the weight of the adaptive distance regularization term based on the gradient information, iteratively update the level set function based on the gradient information and the adaptive distance regularization term until convergence, and the zero level set obtained after the level set function converges is the final segmentation result;

[0016] Step S23: The direct indicators in the composite fiber morphology feature measurement and extraction module include: fiber diameter, scale height, scale perimeter, and scale area; the relative indicators include: fiber diameter-to-height ratio, scale density, relative scale perimeter, and relative scale area. The direct indicators are directly measured and recorded, while the relative indicators are calculated using the linear and nonlinear relationships between the direct and relative indicators.

[0017] Step S24: The texture feature extraction and calculation module uses the obtained labeled multi-region binary image to delineate the region of interest for extracting texture features, calculates the gray-level co-occurrence matrix of the region of interest and the feature parameters based on the gray-level co-occurrence matrix, and quantizes the texture feature parameters.

[0018] Preferably, the step of using the constructed deep learning network model to segment isolated fiber regions and fiber-adhesive regions includes: constructing a deep learning network model based on the segmentation of the target composite fiber image; inputting the target composite fiber image to be segmented and the label dataset corresponding to each initially defined fiber category in the target fiber image; the label dataset includes a set of label data for multiple fiber samples in the same composite fiber image with multiple labels added; performing image enhancement and denoising preprocessing on the target composite fiber image to be segmented; and using the OTSU adaptive thresholding algorithm to generate a binary mask to initially separate the image background region and the fiber region; further, labeling each independent fiber cluster including isolated fiber regions and fiber-adhesive regions through 8-neighborhood connected component analysis to form a label matrix; and outputting the geometric center coordinates, area, and bounding box data values ​​of each labeled region based on the formed label matrix to obtain a labeled multi-region binary image.

[0019] Preferably, the process of defining the region of interest for extracting texture features includes: traversing the marked region binary image, extracting the circumscribed moment of the target fiber region in the grayscale image, using the circumscribed moment as the region of interest, extracting the grayscale co-occurrence matrix in four directions (0°, 45°, 90°, and 135°), summing the four grayscale co-occurrence matrices to obtain a new grayscale co-occurrence matrix, and normalizing it. The new co-occurrence matrix extracted by summing the four grayscale co-occurrence matrices reflects the texture features in each direction. The texture feature parameters of energy, entropy, contrast, and correlation of the four normalized co-occurrence matrices are calculated, and the average value is calculated for the same feature parameter. The texture features of the fiber are described based on the average value.

[0020] Preferably, the fiber classification and recognition terminal includes: a multi-feature fusion module, a feature array recombination module, and a support vector machine training model; the multi-feature fusion module is used to fuse the morphological and texture features of synthetic fibers; the feature array recombination module is used to perform higher-dimensional combination of the fused feature array; the support vector machine training model is used for classification and recognition based on the constructed support vector machine training model, and outputs the final classification and recognition result.

[0021] Preferably, the operation process of the fiber classification and identification terminal includes the following steps:

[0022] Step S31: The multi-feature fusion module normalizes the extracted morphological and texture features of the composite fiber, defining the morphological and texture features within a unified range. The feature array recombination module then merges them into a higher-dimensional feature array. That is, by normalizing the morphological and texture features of the composite fiber to between 0 and 1, they are recombined into a new dimensionless five-dimensional feature array.

[0023] Step S32: After the morphological and texture features of the composite fiber are normalized, the n new five-dimensional feature arrays composed of the normalized morphological and texture features are processed by the five-fold cross-validation method.

[0024] Step S33: In the support vector machine training model, the weight of each feature vector is automatically determined using the Gaussian kernel radial basis function. The feature array processed by the five-fold cross-validation method is sequentially input into the support vector machine for training and recognition, and the final classification and recognition result is output.

[0025] Preferably, the specific processing method of the five-fold cross-validation method includes: processing each of the n new five-dimensional feature arrays composed of normalized morphological features and texture features using the five-fold cross-validation method, that is, performing five-fold cross-validation experiments on the new five-dimensional feature arrays. In each five-fold cross-validation experiment, all feature data samples in the five-dimensional feature array are randomly divided into five parts, each containing the same number of morphological feature arrays and texture feature arrays. Four parts are used as training datasets and one part as test datasets in turn. The recognition rate of the five experiments is output, the average of the five experimental results is calculated, and the calculated average is used as the recognition accuracy of this five-fold cross-validation. The average of the recognition accuracy of the five five-fold cross-validation experiments is calculated, and the average of the recognition accuracy is used as the final recognition accuracy of the support vector machine training model.

[0026] Compared with existing technologies, the beneficial effects achieved by this invention are as follows: This invention achieves multimodal optical signal collaborative acquisition through a dual-channel microscopic imaging terminal, solving the problem that a single imaging method cannot fully acquire fiber morphology and texture features. At the same time, by combining three-dimensional structure reconstruction, polarized light imaging, and fluorescence spectroscopy imaging, the resolution and information richness of composite fiber images are improved. By calculating texture feature parameters through gray-level co-occurrence matrix, the problem of insufficient quantization of texture features in traditional methods is solved. By normalizing and recombining morphological and texture features into a high-dimensional feature array through a multi-feature fusion module, the problem of low classification accuracy of single features is solved. Finally, support vector machines combined with Gaussian kernel radial basis functions are used to automatically determine feature weights, and a five-fold cross-validation method is used to optimize model training, ensuring the reliability and generalization ability of classification results, and improving the accuracy and robustness of classification and recognition. Attached Figure Description

[0027] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0028] Figure 1 This is a schematic diagram of the system module composition provided in an embodiment of the present invention;

[0029] Figure 2 This is a schematic diagram of the operation process of the dual-channel microscopic imaging terminal provided in an embodiment of the present invention;

[0030] Figure 3 This is a schematic diagram of the operation process of the fiber image preprocessing and segmentation terminal provided in an embodiment of the present invention;

[0031] Figure 4 This is a schematic diagram of the operation process of the fiber classification and identification terminal provided in an embodiment of the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] The embodiments of the present invention are combined with Figure 1 As shown, the following technical solution is provided: a fiber identification system for antibacterial polyester-cotton composite fibers based on graphic data analysis, including the following terminals: a composite fiber sample terminal, a dual-channel microscopic imaging terminal, a fiber image preprocessing and segmentation terminal, and a fiber classification and identification terminal.

[0034] In this embodiment, the composite fiber test sample terminal is used to store the original composite fiber test sample; the dual-channel microscopic imaging terminal is used to collaboratively acquire composite fiber image feature information through multimodal optical signals: that is, the core function of the dual-channel microscopic imaging terminal is to simultaneously capture the physical structure and chemical and optical properties of each layer of fiber in the composite fiber, providing differentiated feature input for subsequent algorithms, and further eliminating background noise in the single fiber imaging mode through complementary design at the optical hardware level, solving the problems of feature information loss and material distinction ambiguity in composite fiber cross-occlusion scenarios.

[0035] For example, the dual-channel microscopic imaging terminal includes a first channel and a second channel. The first channel includes a confocal microscope, a polarization imaging unit, and a host computer module. The confocal microscope is used to reconstruct the three-dimensional structure of fibers by using laser layer-by-layer scanning and filtering out stray light from the defocus plane using a pinhole to obtain tomographic images. The polarization imaging unit is used to insert rotatable linear polarizers into the confocal optical path and obtain fiber images with maximum contrast by rotating the polarizer group. Specifically, it triggers fiber image acquisition sequentially based on four configured polarization angles including 0°, 45°, 90°, and 135°. The host computer module is used to transmit the data from the first channel via a data transmission device. The fiber imaging data is displayed in real time on the computer. The second channel includes a Raman spectroscopy point scanning unit, a multispectral fluorescence labeling unit, and a host computer module. The Raman spectroscopy point scanning unit is used to eliminate stray light in the defocused area of ​​the original fiber sample through a confocal microscope. Based on the generated scattering spectrum, it obtains the structure, composition, and state information of the original fiber sample and provides a high-resolution spectrum of the original fiber sample. The multispectral fluorescence labeling unit is used to simultaneously obtain the fluorescence spectral imaging information of the original fiber sample at different wavelengths. The host computer module is used to display the fiber imaging data of the second channel in real time on the computer through a data transmission device.

[0036] For example, in combination Figure 2 As shown, the specific operation process of the dual-channel microscopic imaging terminal includes the following steps:

[0037] Step S11: Input the original composite fiber sample to be tested into the confocal microscope in the first channel, scan the sample layer by layer with laser, construct the three-dimensional structure of the fiber based on the three-dimensional data of the fiber generated by the scan, and obtain the tomographic image of the sample to be tested.

[0038] Step S12: Based on the polarization imaging unit, rotate the polarizer group and sequentially acquire polarization angles of 0°, 45°, 90° and 135° in the tomographic image of the sample to be tested to obtain the fiber image of the sample to be tested with the maximum contrast at different angles.

[0039] Step S13: Simultaneously input the original composite fiber sample to be tested into the second channel, acquire the scattering spectrum image of the sample to be tested based on the confocal microscope, and use the multispectral fluorescent labeling unit to excite the fluorescent label in the sample to be tested, and acquire the fluorescence spectrum image of the sample to be tested at different wavelengths.

[0040] Step S14: Transmit the fiber image of the test sample with the maximum contrast obtained by the first channel to the host computer module one, and transmit the scattering spectrum image and fluorescence spectrum image of the test sample obtained by the second channel to the host computer module two.

[0041] Step S15: Integrate and summarize the image data of the test samples from the host computer module 1 and host computer module 2 to obtain the composite fiber image dataset, and store it in the composite fiber test sample imaging database.

[0042] In this embodiment, the fiber image preprocessing and segmentation terminal is network-connected to the dual-channel microscopic imaging terminal. The fiber image preprocessing and segmentation terminal is used to retrieve the target composite fiber image dataset from the composite fiber sample imaging database for region segmentation and corresponding label annotation.

[0043] For example, the fiber image preprocessing segmentation terminal includes: a composite fiber target region segmentation module, an adaptive distance regularization level set algorithm module, a composite fiber morphological feature measurement and extraction module, and a texture feature extraction and calculation module, wherein: the composite fiber target region segmentation module is used to divide and mark multiple connected regions of the target composite fiber image; the adaptive distance regularization level set algorithm module is used to dynamically adjust the weight of the regularization term using image gradient information to obtain a binary image of the fiber contour of each target region; the composite fiber morphological feature measurement and extraction module is used to measure and extract the morphological features of the composite fiber, including direct and relative indicators; and the texture feature extraction and calculation module is used to extract the gray-level co-occurrence matrix and calculate the feature parameters based on the gray-level co-occurrence matrix.

[0044] For example, in combination Figure 3 As shown, the operation process of the fiber image preprocessing and segmentation terminal includes the following steps:

[0045] Step S21: Based on the composite fiber target region segmentation module, the target composite fiber image is divided into connected or disconnected fiber regions, and the constructed deep learning network model is further used to segment isolated fiber regions and fiber adhesion regions.

[0046] For example, by initially defining all adjacent pixels within a region as having the same or similar attributes, if a pixel cannot reach another pixel through pixels within that region, then these two pixels are not in the same connected region. If a pixel can reach another pixel through pixels within that region, then these two pixels are in the same connected region. Further, the pixel attributes within the same or different connected regions are determined. If the pixel attributes within the same connected region are determined to be the same or similar, then the region is an isolated fiber region within the connected region. Conversely, if the pixel attributes outside the same connected region are determined to be the same or similar, then the region is an isolated fiber region outside the connected region. If the pixel attributes within the same connected region are determined to be different or dissimilar, then the region is a fiber-bonded region within the connected region. If the pixel attributes outside the same connected region are determined to be different or dissimilar, then the region is a fiber-bonded region outside the connected region.

[0047] Specifically, a deep learning network model based on target composite fiber image segmentation is constructed. The input is the target composite fiber image to be segmented and the label dataset corresponding to each fiber category initially defined in the target fiber image. The label dataset includes a set of label data with multiple labels added to multiple fiber samples in the same composite fiber image. The target composite fiber image to be segmented is preprocessed with image enhancement and denoising, and a binary mask is generated using the OTSU adaptive thresholding algorithm to initially separate the image background region and fiber region. Furthermore, each independent fiber cluster, including isolated fiber regions and adhered fiber regions, is labeled through 8-neighbor connected component analysis to form a label matrix. Based on the formed label matrix, the geometric center coordinates, area, and bounding box data values ​​of each labeled region are output to obtain a labeled multi-region binary image.

[0048] Step S22: Initialize the level set function using the adaptive distance regularization level set algorithm module, set its zero level set as the initial contour, select the target image based on the labeled multi-region binary image, calculate the image gradient, and calculate the weight of the adaptive distance regularization term based on the gradient information. Iteratively update the level set function based on the gradient information and the adaptive distance regularization term until convergence. The zero level set obtained after the level set function converges is the final segmentation result.

[0049] Step S23: The direct indicators in the composite fiber morphology feature measurement and extraction module include: fiber diameter, scale height, scale perimeter, and scale area. The relative indicators include: fiber diameter-to-height ratio, scale density, relative scale perimeter, and relative scale area. The direct indicators are directly measured and recorded, while the relative indicators are calculated using the linear and nonlinear relationships between the direct and relative indicators.

[0050] Step S24: The texture feature extraction and calculation module uses the obtained labeled multi-region binary image to delineate the region of interest for extracting texture features, calculates the gray-level co-occurrence matrix of the region of interest and the feature parameters based on the gray-level co-occurrence matrix, and quantizes the texture feature parameters;

[0051] For example, defining the region of interest for extracting texture features involves: traversing the marked binary image of the region, extracting the circumscribed moment of the target fiber region in the grayscale image, and using this as the region of interest; extracting the grayscale co-occurrence matrix in four directions (0°, 45°, 90°, and 135°); summing the four grayscale co-occurrence matrices to obtain a new grayscale co-occurrence matrix, and normalizing it; the new co-occurrence matrix extracted by summing the grayscale co-occurrence matrices reflects the texture features in each direction; calculating the texture feature parameters of energy, entropy, contrast, and correlation of the four normalized co-occurrence matrices; calculating the average value for the same feature parameter; and describing the texture features of the fiber based on this average value.

[0052] In this embodiment, the fiber classification and recognition terminal is network-connected with the fiber image preprocessing and segmentation terminal. The fiber classification and recognition terminal is used to define the morphological and texture features of the composite fiber within a unified range, fuse the morphological and texture features to form a higher-dimensional feature array, and identify the fiber category in the composite fiber through a support vector machine with the same configuration.

[0053] For example, the fiber classification and recognition terminal includes: a multi-feature fusion module, a feature array recombination module, and a support vector machine training model; the multi-feature fusion module is used to fuse the morphological and texture features of synthetic fibers; the feature array recombination module is used to perform higher-dimensional combination of the fused feature array; the support vector machine training model is used for classification and recognition based on the constructed support vector machine training model, and outputs the final classification and recognition result.

[0054] For example, in combination Figure 4 As shown, the operation process of the fiber classification and identification terminal includes the following steps:

[0055] Step S31: The multi-feature fusion module normalizes the extracted morphological and texture features of the composite fiber, defining the morphological and texture features within a unified range. The feature array recombination module then merges them into a higher-dimensional feature array. That is, by normalizing the morphological and texture features of the composite fiber to between 0 and 1, they are recombined into a new dimensionless five-dimensional feature array.

[0056] Step S32: After normalizing the morphological and texture features of the composite fiber, the n new five-dimensional feature arrays composed of the normalized morphological and texture features are processed by the five-fold cross-validation method.

[0057] For example, the specific processing method of the five-fold cross-validation method includes: processing each of the n new five-dimensional feature arrays composed of normalized morphological features and texture features using the five-fold cross-validation method, that is, conducting five-fold cross-validation experiments on the new five-dimensional feature arrays. In each five-fold cross-validation experiment, all feature data samples in the five-dimensional feature array are randomly divided into five parts, each containing the same number of morphological feature arrays and texture feature arrays. Four parts are used as training datasets and one part is used as test datasets in turn. The recognition rate of the five experiments is output, the average of the five experimental results is calculated, and the calculated average is used as the recognition accuracy of this five-fold cross-validation. The average of the recognition accuracy of the five five-fold cross-validation experiments is calculated, and the average of the recognition accuracy is used as the final recognition accuracy of the support vector machine training model.

[0058] Step S33: In the support vector machine training model, the weight of each feature vector is automatically determined using the Gaussian kernel radial basis function. The feature array processed by the five-fold cross-validation method is then sequentially input into the support vector machine for training and recognition, and the final classification and recognition result is output.

[0059] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0060] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A fiber recognition system for antibacterial polyester-cotton composite fiber based on pattern data analysis, characterized by, The system includes the following terminals: a composite fiber sample testing terminal, a dual-channel microscopic imaging terminal, a fiber image preprocessing and segmentation terminal, and a fiber classification and recognition terminal. The composite fiber sample testing terminal stores the original composite fiber sample. The dual-channel microscopic imaging terminal collaboratively acquires composite fiber image feature information through multimodal optical signals. The fiber image preprocessing and segmentation terminal retrieves the target composite fiber image dataset from the composite fiber sample imaging database for region segmentation and corresponding label annotation. The fiber classification and recognition terminal fuses and recombines the morphological and texture features of the composite fibers, and uses a support vector machine for fiber classification and recognition. The composite fiber sample testing terminal is network-connected to the dual-channel microscopic imaging terminal, the fiber image preprocessing and segmentation terminal is network-connected to the dual-channel microscopic imaging terminal, and the fiber classification and recognition terminal is network-connected to the fiber image preprocessing and segmentation terminal. The dual-channel microscopic imaging terminal includes a first channel and a second channel. The first channel includes a confocal microscope, a polarization imaging unit, and a host computer module. The second channel includes a Raman spectroscopy point scanning unit, a multispectral fluorescence labeling unit, and a host computer module. The specific operation process of the dual-channel microscopic imaging terminal includes the following steps: Step S11: Input the original composite fiber sample to be tested into the confocal microscope in the first channel, scan the sample layer by layer with laser, construct the three-dimensional structure of the fiber based on the generated three-dimensional fiber volume data of the scan, and obtain the tomographic image of the sample to be tested. Step S12: Based on the polarization imaging unit, rotate the polarizer group and sequentially acquire polarization angles of 0°, 45°, 90° and 135° in the tomographic image of the sample to be tested to obtain the fiber image of the sample to be tested with the maximum contrast at different angles. Step S13: Simultaneously input the original composite fiber sample to be tested into the second channel, acquire the scattering spectrum image of the sample to be tested based on confocal microscopy, and use the multispectral fluorescent labeling unit to excite the fluorescent label in the sample to be tested, and acquire the fluorescence spectrum image of the sample to be tested at different wavelengths. Step S14: Transmit the fiber image of the test sample with the maximum contrast obtained by the first channel to the host computer module one, and transmit the scattering spectrum image and fluorescence spectrum image of the test sample obtained by the second channel to the host computer module two. Step S15: Integrate and summarize the image data of the test samples from the host computer module 1 and the host computer module 2 to obtain a composite fiber image dataset, and store it in the composite fiber test sample imaging database.

2. The fiber identification system for antibacterial polyester-cotton composite fibers based on graphic data analysis according to claim 1, characterized in that: The fiber image preprocessing and segmentation terminal includes: a composite fiber target region segmentation module, an adaptive distance regularization level set algorithm module, a composite fiber morphological feature measurement and extraction module, and a texture feature extraction and calculation module. The composite fiber target region segmentation module is used to divide and mark multiple connected regions of the target composite fiber image; the adaptive distance regularization level set algorithm module is used to dynamically adjust the weight of the regularization term using image gradient information to obtain a binary image of the fiber contour of each target region; the composite fiber morphological feature measurement and extraction module is used to measure and extract the morphological features of the composite fiber, including direct and relative indicators; the texture feature extraction and calculation module is used to extract the gray-level co-occurrence matrix and calculate the feature parameters based on the gray-level co-occurrence matrix.

3. The fiber identification system for antibacterial polyester-cotton composite fibers based on graphic data analysis according to claim 2, characterized in that: The operation process of the fiber image preprocessing and segmentation terminal includes the following steps: Step S21: Based on the composite fiber target region segmentation module, the target composite fiber image is divided into connected or disconnected fiber regions, and the constructed deep learning network model is used to segment isolated fiber regions and fiber adhesion regions. Step S22: Initialize the level set function using the adaptive distance regularization level set algorithm module, set its zero level set as the initial contour, select the target image based on the labeled multi-region binary image, calculate the image gradient, calculate the weight of the adaptive distance regularization term based on the gradient information, iteratively update the level set function based on the gradient information and the adaptive distance regularization term until convergence, and the zero level set obtained after the level set function converges is the final segmentation result; Step S23: The direct indicators in the composite fiber morphology feature measurement and extraction module include: fiber diameter, scale height, scale perimeter, and scale area; the relative indicators include: fiber diameter-to-height ratio, scale density, relative scale perimeter, and relative scale area. The direct indicators are directly measured and recorded, while the relative indicators are calculated using the linear and nonlinear relationships between the direct and relative indicators. Step S24: The texture feature extraction and calculation module uses the obtained labeled multi-region binary image to delineate the region of interest for extracting texture features, calculates the gray-level co-occurrence matrix of the region of interest and the feature parameters based on the gray-level co-occurrence matrix, and quantizes the texture feature parameters.

4. The fiber identification system for antibacterial polyester-cotton composite fibers based on graphic data analysis according to claim 3, characterized in that: The process of segmenting isolated fiber regions and fiber-adhesive regions using a constructed deep learning network model includes: constructing a deep learning network model based on the segmentation of the target composite fiber image; inputting the target composite fiber image to be segmented and the label dataset corresponding to each initially defined fiber category in the target fiber image; the label dataset includes a set of label data with multiple labels added to multiple fiber samples within the same composite fiber image; performing image enhancement and denoising preprocessing on the target composite fiber image to be segmented; and using the OTSU adaptive thresholding algorithm to generate a binary mask to initially separate the image background region from the fiber region; further, labeling each independent fiber cluster including isolated fiber regions and adherent fiber regions through 8-neighbor connected component analysis to form a label matrix; and outputting the geometric center coordinates, area, and bounding box data values ​​of each labeled region based on the formed label matrix to obtain a labeled multi-region binary image.

5. The fiber identification system for antibacterial polyester-cotton composite fibers based on graphic data analysis according to claim 4, characterized in that: The process of defining the region of interest (ROI) for extracting texture features includes: traversing the marked binary image of the region, extracting the circumscribed moments of the target fiber region from the grayscale image, using the circumscribed moments as the ROI, extracting the grayscale co-occurrence matrices in four directions (0°, 45°, 90°, and 135°), summing the four grayscale co-occurrence matrices to obtain a new grayscale co-occurrence matrix, and normalizing it. The new co-occurrence matrix extracted by summing the four grayscale co-occurrence matrices reflects the texture features in each direction. The texture feature parameters of energy, entropy, contrast, and correlation of the four normalized co-occurrence matrices are calculated, and the average value is calculated for the same feature parameter. The texture features of the fiber are described based on the average value.

6. The fiber identification system for antibacterial polyester-cotton composite fibers based on graphic data analysis according to claim 5, characterized in that: The fiber classification and recognition terminal includes: a multi-feature fusion module, a feature array recombination module, and a support vector machine training model; the multi-feature fusion module is used to fuse the morphological and texture features of synthetic fibers; the feature array recombination module is used to perform higher-dimensional combination of the fused feature array; the support vector machine training model is used for classification and recognition based on the constructed support vector machine training model, and outputs the final classification and recognition result.

7. The fiber identification system for antibacterial polyester-cotton composite fibers based on graphic data analysis according to claim 6, characterized in that: The operation process of the fiber classification and identification terminal includes the following steps: Step S31: The multi-feature fusion module normalizes the extracted morphological and texture features of the composite fiber, defining the morphological and texture features within a unified range. The feature array recombination module then merges them into a higher-dimensional feature array. That is, by normalizing the morphological and texture features of the composite fiber to between 0 and 1, they are recombined into a new dimensionless five-dimensional feature array. Step S32: After the morphological and texture features of the composite fiber are normalized, the n new five-dimensional feature arrays composed of the normalized morphological and texture features are processed by the five-fold cross-validation method. Step S33: In the support vector machine training model, the weight of each feature vector is automatically determined using the Gaussian kernel radial basis function. The feature array processed by the five-fold cross-validation method is sequentially input into the support vector machine for training and recognition, and the final classification and recognition result is output.

8. The fiber identification system for antibacterial polyester-cotton composite fibers based on graphic data analysis according to claim 7, characterized in that: The specific processing method of the five-fold cross-validation method includes: processing each of the n new five-dimensional feature arrays composed of normalized morphological features and texture features using the five-fold cross-validation method, that is, performing five-fold cross-validation experiments on the new five-dimensional feature arrays. In each five-fold cross-validation experiment, all feature data samples in the five-dimensional feature array are randomly divided into five parts, each containing the same number of morphological feature arrays and texture feature arrays. Four parts are used as training datasets and one part is used as test datasets in turn. The recognition rate of the five experiments is output, the average of the five experimental results is calculated, and the calculated average is used as the recognition accuracy of this five-fold cross-validation. The average of the recognition accuracy of the five five-fold cross-validation experiments is calculated, and the average of the recognition accuracy is used as the final recognition accuracy of the support vector machine training model.

Images