Fabric component intelligent analysis method
By using photothermal curing pretreatment and deep neural network models to identify the fiber types and areas of textile fabrics, the problems of fiber fragility and image blurring were solved, achieving efficient and accurate determination of fabric composition analysis.
Patent Information
- Application Number
- CN202410934816.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-12
- Publication Date
- 2024-11-01
AI Technical Summary
Existing technologies for textile fabric composition analysis suffer from problems such as fiber fragility, unclear images, and computational complexity caused by differences in fabric structure, making it difficult to quickly and efficiently identify the structure of new fabrics such as nonwovens and accurately determine the composition content.
Standard samples were prepared using a photothermal curing pretreatment method. The fabric was treated with a photosensitive resin solution and a light-transmitting agent, and combined with photothermal curing technology, the fibers were softened and the fiber spacing was increased. Cross-sectional images of the yarn were captured, fiber feature sets were extracted, and a deep neural network model was used to identify fiber types and areas, and the fabric composition content was directly calculated.
It improves the accuracy and efficiency of fabric composition analysis, solves the problems of fiber pulling and image blurring, and enables rapid and accurate composition determination of various fabrics.
Smart Images

Figure CN118883545A_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present application relate to but are not limited to the technical field of intelligent analysis of textile components, and in particular, to a method for intelligent analysis of fabric components. Background Art
[0002] In the field of textile fabric analysis and detection, traditional methods face many challenges and limitations. First, regarding the identification of fabric weaving structure, current technology mainly relies on a single method, which can only identify woven fabrics, weft knitted fabrics or warp knitted fabrics, and lacks effective identification methods for new fabric structures such as non-woven fabrics. Secondly, in terms of the determination of fabric component content, existing technologies generally require the fabric to be disassembled into yarns, and then the types and contents of each component fiber in the yarn are analyzed. However, this process is not only time-consuming and labor-intensive, but also has extremely low detection efficiency for fabrics with complex structures, especially warp knitted fabrics, woven fabrics, etc. that are difficult to disassemble, making it difficult to meet the market demand for fast and efficient detection. In addition, because the fibers in the yarn are tightly held together due to twisting, and the color is dark and the transmittance is low after dyeing, it is difficult for conventional observation methods to obtain clear fiber contour images, thereby increasing the error of the test results and even leading to false detection.
[0003] As a commonly used sample preparation technology, the resin embedding method also has significant defects in fiber detection. The resin material used in this method is hard and brittle, which can easily cause sample damage during the slicing process. At the same time, the toughness and rigidity of the fiber at room temperature make it difficult to keep the slicing operation flat, and it is easy to pull out, which in turn causes problems such as inconsistent depth of field and blurred images, seriously affecting the accuracy and reliability of the test results. The conventional calculation method for fabric composition analysis is to first calculate the mass percentage of each component fiber in the warp and weft directions respectively, and then convert the mass percentage of each component fiber in the entire fabric based on the proportion of the warp and weft directions. The calculation is complicated, and the calculation formula for fiber content in fabrics with different organizational structures is not unified, which greatly reduces the detection efficiency. Summary of the invention
[0004] The following is a summary of the subject matter described in detail herein. This summary is not intended to limit the scope of the claims.
[0005] The embodiment of the present application provides an intelligent analysis method for fabric composition, which solves the computational problems caused by fiber fragility, unclear images and fabric structure differences, thereby effectively improving the accuracy and efficiency of fabric composition analysis.
[0006] An intelligent fabric composition analysis method provided in an embodiment of the present application includes:
[0007] Obtaining a fabric sample to be tested, and identifying the fabric structure type of the fabric sample to be tested;
[0008] According to the fabric structure type, the fabric sample to be tested is subjected to photothermal curing pretreatment and standard sample rapid preparation to obtain a standard sample;
[0009] Taking a cross-sectional image of fibers in the short side yarn of the standard sample;
[0010] Extracting a feature set of each component fiber in the fiber cross-sectional image, and identifying the type and cross-sectional area of each fiber according to the feature set;
[0011] According to the type and cross-sectional area of each fiber, the density and total cross-sectional area of each component fiber are statistically obtained;
[0012] The mass percentage of each component fiber in the fabric sample to be tested is calculated based on the density and total cross-sectional area of each component fiber.
[0013] In one embodiment of the present application, the identifying and obtaining the fabric structure type of the fabric sample to be tested includes: inputting the fabric sample to be tested into a trained fabric component intelligent recognition model for fabric recognition to obtain the fabric structure type of the fabric sample to be tested;
[0014] The training steps of the trained fabric component intelligent recognition model include: obtaining a fabric recognition data set, denoising and normalizing the fabric recognition data set to obtain a standard fabric training set; constructing an initial fabric component intelligent recognition model; inputting the standard fabric training set into the initial fabric component intelligent recognition model to perform fabric recognition and obtain prediction information; calculating a first loss value based on the prediction information and a true value, adjusting the parameters of the initial fabric component intelligent recognition model by the first loss value until the first loss value meets a preset value, and obtaining a trained fabric component intelligent recognition model.
[0015] In one embodiment of the present application, obtaining a fabric identification data set includes: obtaining multiple samples of woven fabrics, warp knitted fabrics, weft knitted fabrics and non-woven fabrics respectively to obtain a first identification sample set; performing image acquisition on the first identification sample set to obtain a fabric identification data set.
[0016] In one embodiment of the present application, the fabric sample to be tested is subjected to photothermal curing pretreatment and standard sample rapid preparation according to the fabric tissue structure type to obtain a standard sample, including: preparing a sample to be tested according to the fabric tissue structure type, dripping a light-transmitting agent and a photosensitive resin solution onto the short side of the sample to be tested to obtain a pretreated sample; and subjecting the pretreated sample to multi-angle light irradiation and photothermal curing to obtain a standard sample.
[0017] In one embodiment of the present application, the test sample is prepared according to the type of fabric structure, including: selecting a first sample in the warp and weft directions of the woven fabric, selecting a second sample in the warp direction of the warp knitted fabric, selecting a third sample in the weft direction of the weft knitted fabric, selecting a fourth sample in the warp and weft directions of the non-woven fabric, and obtaining the test sample from the first sample, the second sample, the third sample or the fourth sample.
[0018] In one embodiment of the present application, the composition of the photosensitive resin includes at least one of acrylic, epoxy acrylate, epoxy acrylate, polyester and polyurethane; the translucent agent includes at least one of cedar oil, pine oil, Cargille oil and xylene.
[0019] In one embodiment of the present application, the pre-treated sample is subjected to multi-angle light irradiation and photothermal curing to obtain a standard sample, comprising: laying the pre-treated sample flat on a photothermal curing table, using a mixed light consisting of ultraviolet light and infrared light, and irradiating it from multiple angles in one direction or in both directions, to obtain a standard sample; wherein the angle formed by the multi-angle light irradiation direction and the corresponding plane is in the range of 30° to 150°, and the irradiation time T is in the range of 3 seconds to 60 seconds.
[0020] In one embodiment of the present application, the method of obtaining the type and cross-sectional area of each fiber according to the feature set includes: normalizing the feature value of the feature set to obtain a standardized feature set; inputting the standardized feature set into a fiber recognition model with the best indicators for classification and positioning identification to obtain the type and cross-sectional area of each fiber; the training process of the fiber recognition model with the best indicators includes: obtaining a fiber training data set, denoising and normalizing the fiber training data set to obtain a standard fiber training set; constructing an initial fiber recognition model, the initial fiber recognition model including an input layer, a hidden layer and an output layer; inputting the standard fiber training set into the input layer, and extracting features from the standard fiber training set using the hidden layer to obtain fiber feature information; using the output layer to classify and position the fiber feature information to obtain an output value; iteratively training the fiber recognition model until the error between the output value and the true value meets a preset error value, thereby obtaining a fiber recognition model with the best indicators.
[0021] In one embodiment of the present application, the method of obtaining a fiber training data set includes: obtaining multiple samples of fabrics of multiple tissue structures to obtain a second identification sample set; performing image acquisition on the second identification sample set to obtain a fiber cross-sectional image; performing image processing on the fiber cross-sectional image to extract the cross-sectional area characteristics, fiber diameter characteristics, fiber perimeter characteristics and fiber edge contour expansion line characteristics of each component fiber in the fiber cross-sectional image; obtaining the polarization color of each component fiber, converting the polarization color into a grayscale value, and obtaining a first grayscale characteristic; obtaining the fluorescent color of each component fiber, converting the fluorescent color into a grayscale value, and obtaining a second grayscale characteristic; obtaining the fiber melting point of each component fiber to obtain a fiber melting point characteristic; obtaining the solubility of each component fiber to obtain a solubility characteristic; performing feature value standardization processing on the cross-sectional area characteristics, fiber diameter characteristics, fiber perimeter characteristics, fiber edge contour expansion line characteristics, first grayscale characteristics, second grayscale characteristics, fiber melting point characteristics and solubility characteristic characteristics of each component fiber to obtain a fiber training data set.
[0022] In one embodiment of the present application, the fabric tissue structure types of the fabric sample to be tested include woven fabrics, warp knitted fabrics, weft knitted fabrics and non-woven fabrics, and the mass percentage of each component fiber of the fabric sample to be tested is calculated according to the total cross-sectional area and density of each component fiber, including: if the fabric sample to be tested is a woven fabric, the warp and weft fibers of the woven fabric are combined and counted; if the fabric sample to be tested is a weft knitted fabric, the weft fibers of the weft knitted fabric are counted; if the fabric sample to be tested is a warp knitted fabric, the warp fibers of the warp knitted fabric are counted; if the fabric sample to be tested is a non-woven fabric, the warp and weft fibers of the non-woven fabric are combined and counted; wherein, the mass percentage calculation formula of the i-th component fiber of the fabric sample to be tested is:
[0023]
[0024] In the formula, ρ i represents the density of the i-th component fiber, i∈[1,I], I is a positive integer, s i It represents the total cross-sectional area of the i-th component fiber, and M represents the total mass of each component fiber of the fabric sample to be tested.
[0025] The fabric composition intelligent analysis method provided by the embodiment of the present application first obtains and identifies the fabric structure type of the fabric sample to be tested, and then performs photothermal curing pretreatment and rapid preparation of standard samples on the fabric sample to be tested according to the fabric structure to obtain the standard sample. By performing photothermal curing pretreatment on the fabric sample to be tested, the sample fiber can be effectively softened and the fiber spacing can be increased, thereby avoiding the phenomenon of fiber lodging and image blurring due to pulling during the cutting process, and significantly improving the accuracy and credibility of the subsequent analysis results. After obtaining the standard sample, the fiber cross-sectional image in the yarn of the short side of the standard sample is photographed, and then the feature set of each component fiber in the fiber cross-sectional image is extracted. According to these feature sets, the type and cross-sectional area of each fiber in the fiber cross-sectional image can be identified. Since the type and content of each component fiber in the standard sample are determined by directly observing the fiber cross-sectional image, when determining the fabric component content of the standard sample, it is not necessary to remove the yarn from the fabric and then determine the content of each group of fiber components, and then convert and calculate according to the fabric characteristics, thereby significantly improving the efficiency of detection. According to the type and cross-sectional area of each fiber, the density and total cross-sectional area of each component fiber can be statistically obtained. According to the density and total cross-sectional area of each component fiber, the mass percentage of each component fiber in the fabric sample to be tested can be calculated. This application directly prepares fabric cross-sectional slices and directly performs intelligent recognition and measurement on fabric cross-sectional images, effectively solving the problems of difficulty in yarn unravelling, fiber pulling, image blurring, and difficulty in unifying the calculation formula for fiber content in fabrics with different tissue structures, thereby improving the accuracy and efficiency of fabric composition analysis. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] Figure 1 is a flow chart of the intelligent analysis method of fabric composition provided in an embodiment of the present application;
[0027] Figure 2 It is a flow chart of the training steps of the fabric composition intelligent recognition model provided in the embodiment of the present application;
[0028] Figure 3 The embodiment of this application provides Figure 1 A specific flow chart of step 120;
[0029] Figure 4 is a flow chart of the training steps of the fiber recognition model provided in the embodiment of the present application;
[0030] Figure 5 is a flow chart of obtaining a fiber training data set provided in an embodiment of the present application;
[0031] Figure 6 It is a schematic diagram of a twill fabric provided in a specific embodiment of the present application;
[0032] Figure 7is a cross-sectional view of a polyester / cotton blended twill fabric sample provided in a specific embodiment of the present application;
[0033] Figure 8 This is a cross-sectional view of a polyester / viscose / spandex blended mesh warp knitted fabric provided in a specific embodiment of the present application;
[0034] Fig. 9 This is a cross-sectional view of a wool / acrylic / spandex blended weft knitted sample provided in a specific embodiment of the present application.
[0035] Fig.10 This is a cross-sectional view of a viscose / cotton mixed spunlace nonwoven fabric provided in a specific embodiment of the present application. DETAILED DESCRIPTION
[0036] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0037] It should be noted that, although the logical order is shown in the flowchart, in some cases, the steps shown or described can be performed in a different order from that in the flowchart. The terms "first", "second", etc. in the specification, claims and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be noted that the structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification for people familiar with this technology to understand and read, and are not used to limit the limiting conditions that the present invention can be implemented, so they have no technical substantive significance. Any modification of the structure, change of the proportional relationship or adjustment of the size, without affecting the effects that the present invention can produce and the purposes that can be achieved, should still fall within the scope of the technical content disclosed by the present invention. At the same time, the terms such as "upper", "lower", "left", "right", "middle" and "one" quoted in this specification are only for the convenience of narration, and are not used to limit the scope of the implementation of the present invention. The change or adjustment of the relative relationship should also be regarded as the scope of the implementation of the present invention without substantially changing the technical content.
[0038] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.
[0039] As a commonly used sample preparation technology, the traditional resin embedding method has significant defects in fiber testing. The resin material used in this method is hard and brittle, which can easily lead to sample damage during the slicing process. At the same time, the fiber's toughness and rigidity at room temperature make it difficult to keep the slices flat and prone to pulling, which in turn causes problems such as inconsistent depth of field and blurred images, seriously affecting the accuracy and reliability of the test results.
[0040] In view of this, the fabric composition intelligent analysis method provided in the embodiment of the present application first obtains and identifies the fabric structure type of the fabric sample to be tested, and then performs photothermal curing pretreatment and rapid preparation of standard samples on the fabric sample to be tested according to the fabric structure to obtain a standard sample. By performing photothermal curing pretreatment on the fabric sample to be tested, the sample fiber can be effectively softened and the fiber spacing can be increased, thereby avoiding the phenomenon of fiber lodging and image blurring due to pulling during the cutting process, and significantly improving the accuracy and credibility of subsequent analysis results. After obtaining the standard sample, the fiber cross-sectional image in the yarn of the short side of the standard sample is photographed, and then the feature set of each component fiber in the fiber cross-sectional image is extracted. According to these feature sets, the type and cross-sectional area of each fiber in the fiber cross-sectional image can be identified. Since the type and content of each component fiber in the standard sample are determined by directly observing the fiber cross-sectional image, when determining the fabric component content of the standard sample, it is not necessary to remove the yarn from the fabric and then determine the content of each group of fiber components, and then convert and calculate according to the fabric characteristics, thereby significantly improving the efficiency of detection. According to the type and cross-sectional area of each fiber, the density and total cross-sectional area of each component fiber can be statistically obtained. According to the density and total cross-sectional area of each component fiber, the mass percentage of each component fiber in the fabric sample to be tested can be calculated. The present application adopts a method for directly preparing fabric cross-sectional slices, and directly performs intelligent identification and measurement on fabric cross-sectional images, which effectively solves the problem of directly measuring the content of components suitable for various fabrics, solves the problem of uniformly identifying the fabric weaving structure when directly preparing various fabric slices, solves the problem of inconsistent depth of field and blurred images caused by easy pulling of fibers in slices, and solves the problem of directly measuring the content of components suitable for various fabrics, thereby improving the accuracy and efficiency of fabric component analysis.
[0041] The embodiments of the present application are further described below in conjunction with the accompanying drawings.
[0042] See also Figure 1 , Figure 1 Flowchart of the intelligent analysis method of fabric composition provided by the embodiment of the present application. Figure 1 As shown, the calculation method includes but is not limited to steps 110 to 160.
[0043] Step 110: obtaining a fabric sample to be tested, and identifying the fabric structure type of the fabric sample to be tested;
[0044] Step 120: Perform photothermal curing pretreatment and rapid preparation of standard samples on the fabric sample to be tested according to the fabric structure type to obtain a standard sample;
[0045] Step 130: photographing a fiber cross-section image in the short side yarn of the standard sample;
[0046] Step 140: extracting a feature set of each component fiber in the fiber cross-sectional image, and identifying the type and cross-sectional area of each fiber according to the feature set;
[0047] Step 150: Statistically obtain the density and total cross-sectional area of each component fiber according to the type and cross-sectional area of each fiber;
[0048] Step 160: Calculate the mass percentage of each component fiber in the fabric sample to be tested based on the density and total cross-sectional area of each component fiber.
[0049] Steps 110 to 160 are described in detail below.
[0050] In one embodiment, in step 110, the fabric structure types of the fabric sample to be tested include but are not limited to woven fabrics, warp knitted fabrics (also known as warp knitted fabrics), weft knitted fabrics (also known as weft knitted fabrics) and non-woven fabrics, wherein the components of these fabrics include but are not limited to pure fabrics such as cotton, wool, linen, silk, polyester, nylon, polypropylene, acrylic, Tencel, modal, viscose and their different combinations and blended fabrics. Specifically, woven fabrics include plain weave, twill, satin and jacquard fabrics, etc., warp knitted fabrics include mesh fabrics, terry fabrics, pleated fabrics, plush fabrics, weft-inserted fabrics, etc., weft knitted fabrics include plain weft stitch, rib stitch and double reverse stitch, and non-woven fabrics include meltblown fabrics, spunbond fabrics, spunlace fabrics, needle-punched fabrics, etc.
[0051] In one embodiment, when the fabric structure type of the fabric sample to be tested is identified, the fabric sample to be tested can be input into a trained fabric component intelligent recognition model for fabric recognition to obtain the fabric structure type of the fabric sample to be tested. The fabric component intelligent recognition model can be composed of a conventional deep neural network model or a convolutional neural network model, etc. For the relevant model structure and model principle of the deep neural network model or the convolutional neural network model, reference can be made to the introduction in the relevant technology, which will not be elaborated here.
[0052] Reference Figure 2 In one embodiment, the training steps of the fabric composition intelligent recognition model may include but are not limited to the following steps 210 to 230.
[0053] Step 210: obtaining a fabric recognition data set, performing denoising and normalization processing on the fabric recognition data set, and obtaining a standard fabric training set;
[0054] Step 220: construct an initial fabric component intelligent recognition model, input the standard fabric training set into the initial fabric component intelligent recognition model to perform fabric recognition, and obtain prediction information;
[0055] Step 230: Calculate a first loss value based on the predicted information and the true value, and adjust the parameters of the initial fabric component intelligent recognition model by the first loss value until the first loss value meets the preset value, thereby obtaining a trained fabric component intelligent recognition model.
[0056] Steps 210 to 230 are described in detail below.
[0057] In one embodiment, in step 210, the process of obtaining a fabric identification data set includes: obtaining multiple samples of woven fabrics, warp knitted fabrics, weft knitted fabrics, and non-woven fabrics, respectively, to obtain a first identification sample set, and then performing image acquisition on the first identification sample set to obtain a fabric identification data set. Specifically, samples of four fabrics (woven fabrics, warp knitted fabrics, weft knitted fabrics, and non-woven fabrics) can be collected respectively, and N representative samples of each fabric are collected, where N is a positive integer; then all the collected samples are placed under a microscope and magnified 50 times, and then collected, and clear images of all the samples after being placed under a microscope and magnified 50 times are obtained as the fabric identification data set.
[0058] In one embodiment, denoising means removing noise from an image (fabric recognition dataset), such as impurities, blur, distortion, etc. in the image, to reduce interference with model training. Normalization means adjusting the pixel values of the image data to the same scale (such as 0-1 or between -1 and 1), which helps the model converge faster and achieve better performance. By denoising and normalizing the fabric recognition dataset, the data quality of the input fabric composition intelligent recognition model can be ensured, thereby improving the efficiency and accuracy of model training.
[0059] In one embodiment, in step 220, when constructing the initial fabric component intelligent recognition model, a suitable neural network (such as a YOLO V9 model, a convolutional neural network, etc.) can be selected as the initial model, and this application does not limit this. It should be noted that the standard fabric training set includes a large number of fabric images marked with fabric types and position information (such as bounding boxes). In the process of training the initial fabric component intelligent recognition model using the standard fabric training set, the initial fabric component intelligent recognition model can learn the fabric type and bounding box coordinates in the fabric image, and perform preliminary prediction to obtain prediction information. The prediction information includes the prediction results of the fabric structure type, that is, the probability distribution of the fabric category and the position of the fabric in the image.
[0060] In one embodiment, after the fabric identification task processing in step 220, prediction information including fabric category and fabric position can be obtained. In order to improve the accuracy of the prediction of the initial fabric composition intelligent identification model, these prediction information can be calculated with the true value of the image (that is, the fabric type and position information of the fabric image in the standard fabric training set) to obtain the difference between the predicted value and the true value, that is, the first loss value. Specifically, for each image in the training set, the model will output a prediction result. By comparing this prediction result with the true label of the image, the loss value corresponding to the image can be calculated. Then, the loss values of all images in the entire training set are averaged or summed to obtain the first loss value (that is, the loss value of the entire training set).
[0061] In one embodiment, the weights, biases and other parameters of the initial fabric composition intelligent recognition model can also be adjusted according to the first loss value using back propagation and other algorithms until the first loss value meets a preset value (such as reaching below a certain threshold or the first loss value no longer decreases significantly after multiple rounds of training), thereby obtaining a trained fabric composition intelligent recognition model that can more accurately identify the input fabric image and output the corresponding fabric tissue structure type.
[0062] In one embodiment, in step 120, the photothermal curing pretreatment is to improve certain properties of the fabric (the fabric sample to be tested), such as hardness, wear resistance, weather resistance, etc., or to form a specific pattern, texture, etc. on the fabric surface by photothermal curing technology. Standard sample rapid preparation is to prepare the fabric sample to be tested that has undergone photothermal curing pretreatment into a sample that meets the test or analysis standard, so as to carry out subsequent performance testing or analysis.
[0063] Reference Figure 3 In one embodiment, in step 120, the fabric sample to be tested is subjected to photothermal curing pretreatment and standard sample rapid preparation according to the fabric tissue structure type to obtain a processing flow of the standard sample, which may include but is not limited to the following steps 310 to 320.
[0064] Step 310: Prepare a test sample according to the fabric structure type, and drop a light-transmitting agent and a photosensitive resin solution onto the short side of the test sample to obtain a pretreated sample;
[0065] Step 320: Perform multi-angle light irradiation and photothermal curing on the pre-treated sample to obtain a standard sample.
[0066] In one embodiment, in step 310, it is understood that, since different fabrics may require different pretreatment methods and standard sample rapid preparation processes, the trained fabric component intelligent recognition model can be used to identify the fabric sample to be tested to obtain the fabric tissue structure type, and then the corresponding pretreatment method and standard sample rapid preparation process are selected according to the fabric tissue structure type. Exemplarily, when the fabric is cotton fabric, since cotton fabric generally has good water absorption, it may be necessary to perform a water washing treatment before testing to remove impurities and slurry on the surface and reach a certain humidity balance state. According to the test standards (such as ASTM, ISO, etc.), the cotton fabric sample may need to be cut into a specific size and shape, such as a rectangle or dumbbell shape, to facilitate tensile testing, etc.
[0067] In one embodiment, in the process of preparing the sample to be tested according to the type of fabric structure, a first sample is selected in the warp and weft directions of the woven fabric, a second sample is selected in the warp direction of the warp knitted fabric, a third sample is selected in the weft direction of the weft fabric, and a fourth sample is selected in the warp and weft directions of the non-woven fabric, and the sample to be tested is obtained from the first sample, the second sample, the third sample, and the fourth sample. Specifically, if the fabric sample to be tested is a woven fabric, a laser cutting method is used to take one sample in the warp and weft directions of the sample to be tested; if the fabric sample to be tested is a warp knitted fabric, a laser cutting method is used to take one warp sample in the warp direction of the sample to be tested; if the fabric sample to be tested is a weft fabric, a laser cutting method is used to take one weft sample in the weft direction of the sample to be tested; if the fabric sample to be tested is a non-woven fabric, a laser cutting method is used to take one sample in the warp and weft directions of the sample to be tested.
[0068] In one embodiment, the shape of the sample to be tested can be a rectangle. In the process of preparing the sample to be tested, a laser can be used to cut a sample to be tested of a certain width and length on the fabric sample to be tested. For example, the length of the sample to be tested ranges from 1 mm to 500 mm, and the width ranges from 1 mm to 80 mm. It should be noted that the width and length of the sample to be tested can be adjusted according to actual conditions and are not limited here. Preferably, the width ranges from 10 mm to 30 mm, and the length ranges from 40 mm to 50 mm.
[0069] In one embodiment, a translucent agent can be dripped onto the short side end (short side) of the sample to be tested, so that the translucent agent can cover the surface of all fibers in the sample to be tested, thereby improving the clarity of the cross-sectional image of the fibers in the sample during observation; then a photosensitive resin solution is dripped onto the short side end of the sample to be tested, so that the liquid photosensitive resin solution penetrates into the tightly hugged fibers in the yarn in the fabric sample (sample to be tested), thereby expanding the distance between the fibers and obtaining a pretreated sample. It should be noted that the short side end generally refers to an edge portion with a relatively small length or size. For example, in a rectangular sample to be tested, the short side end refers to the edge portion where the short side of the rectangle is located, and there are generally two short side ends. For a polygonal or irregularly shaped sample to be tested, the short side end may refer to the edge portion where the relatively short line segments are located.
[0070] In one embodiment, the components of the photosensitive resin may include at least one of acrylic, epoxy acrylate, epoxy acrylate, polyester and polyurethane, or it may be a compound of multiple of these components; the components of the translucent agent may include at least one of cedar oil, pine oil, Cargil oil and xylene, or it may be a compound of multiple of these components.
[0071] In one embodiment, in the process of multi-angle light irradiation and photothermal curing of the pretreated sample to obtain a standard sample, the pretreated sample can be laid flat on a photothermal curing table, and a mixed light composed of ultraviolet light and infrared light is used to irradiate from multiple angles in one direction or in both directions to obtain a standard sample. Specifically, the short side of the pretreated sample is placed on the photothermal curing table, and the irradiation time is T, and multi-angle ultraviolet light and infrared light are irradiated to obtain a standard sample with stable fiber spacing in the yarn. Among them, the angle formed by the multi-angle irradiation direction and the corresponding plane ranges from 30° to 150°, and the irradiation time T ranges from 3 seconds to 60 seconds, so that the curing agent that penetrates into the fibers is quickly cured, and the division and fixation of the tightly hugged fibers in the yarn in the fabric sample (pretreated sample) are achieved. It should be pointed out that the preferred range of the angle formed by the irradiation direction and the corresponding plane is 45° to 90°, and the preferred range of the irradiation time T is 10 seconds to 20 seconds.
[0072] In one embodiment, the processing flow from step 310 to step 320 is to embed and fix the fibers with photosensitive resin, and then use photothermal irradiation, especially infrared irradiation, to increase the temperature of the cured resin and the fixed fibers, causing them to soften, making it easier to cut the fiber cross section flat, thereby solving the problem of the fibers being pulled and falling over, resulting in inconsistent depth of field and blurred images during image observation, which seriously affects the accuracy of the test results.
[0073] In one embodiment, in step 130, when taking a cross-sectional image of the fibers in the yarn on the short side of the standard sample, the short side of the standard sample in step 220 can be first clamped with a fabric slicer; then the short side of the standard sample is cut flat to make the upper and lower surfaces flush; then the cut sample is moved to the observation platform of the transflective microscope and the focus is adjusted to obtain a clear image, and all cross-sectional images of the fibers in the yarn on the short side of the cut sample are collected from left to right or from right to left.
[0074] In one embodiment, in step 140, the types of fibers may include at least natural fibers, regenerated fibers and chemical fibers, and the subdivided types of fibers may include at least cotton, wool, linen, silk, polyester, nylon, polypropylene, acrylic, Tencel, modal and viscose.
[0075] In one embodiment, when extracting the feature set of each component fiber in the fiber cross-sectional image, the obtained fiber cross-sectional image can be first input into Matlab software for image processing to obtain the cross-sectional area feature, fiber diameter feature, fiber perimeter feature, and fiber edge contour expansion line feature of each component fiber in the fiber cross-sectional image; then the polarization color of each component fiber is obtained through the microscope polarization component, and the polarization color is converted into a grayscale value to obtain the first grayscale feature; the fluorescence color of each component fiber is obtained through the microscope fluorescence component, and the fluorescence color is converted into a grayscale value to obtain the second grayscale feature; the fiber melting point of each component fiber is obtained through the melting point determination component of the microscope to obtain the fiber melting point feature; the dissolution performance of each component fiber is obtained through the in-situ dissolution component of the microscope to obtain the dissolution performance feature. At this time, a feature set consisting of the cross-sectional area feature, the fiber diameter feature, the fiber perimeter feature, the fiber edge contour expansion line feature, the first grayscale feature, the second grayscale feature, the fiber melting point feature, and the dissolution performance feature is obtained. It should be pointed out that these eight characteristics refer to the collection of eight characteristics for each fiber under the microscope, and generally refer to the acquisition of fibers of various components.
[0076] In one embodiment, after obtaining the feature set of each component fiber, the type and cross-sectional area of each fiber can be obtained according to the feature set identification. In this process, the feature set can be first subjected to feature value standardization to obtain a standardized feature set, and then the standardized feature set is input into the fiber identification model with the best indicators for classification and identification to obtain the type and cross-sectional area of each fiber. It should be noted that feature value standardization is a commonly used method in machine learning and data processing. Its main purpose is to eliminate the impact of different dimensions or inconsistent dimension ranges, so that all features can be compared and analyzed on the same scale.
[0077] In one embodiment, the feature set is subjected to feature value standardization to obtain a standardized feature set, that is, the eight features of each component fiber are subjected to feature value standardization to obtain a standardized feature set of each component fiber. Specifically, the process of standardizing a certain feature value includes: first, for each feature, the maximum value of the feature in all fibers is calculated. This maximum value represents the highest performance of all fibers on this feature; then, 1.5 times this maximum value is used as a reference value. It should be noted that this reference value can ensure that even the most extreme feature values can be effectively scaled, and there is no restriction on the multiple here, which can be set according to actual needs; then, for each feature value of each component, it is divided by this reference value. In this way, all feature values will be scaled to the range of 0 to 1.
[0078] Reference Figure 4 In one embodiment, the training steps of the fiber identification model with the best indicators may include but are not limited to the following steps 410 to 450.
[0079] Step 410: obtaining a fiber training data set, performing denoising and normalization processing on the fiber training data set, and obtaining a standard fiber training set;
[0080] Step 420: constructing an initial fiber recognition model, the initial fiber recognition model comprising an input layer, a hidden layer and an output layer;
[0081] Step 430: input the standard fiber training set into the input layer, and use the hidden layer to extract features of the standard fiber training set to obtain fiber feature information;
[0082] Step 440: using the output layer to classify and locate the fiber feature information to obtain an output value;
[0083] Step 450: Iteratively train the fiber recognition model until the error between the output value and the true value meets the preset error value, thereby obtaining the fiber recognition model with the best indicators.
[0084] Steps 410 to 450 are described in detail below.
[0085] In one embodiment, in step 410, the fiber training data set includes at least cotton, wool, linen, silk, polyester, nylon, polypropylene, acrylic, Tencel, modal and viscose, and these fiber training data sets include fiber images or feature data of different types and states. Since the actual collected data may contain noise, such as impurities in the image, light changes, etc., these noises will affect the training effect of the model, so denoising is required to improve data quality. Normalization refers to adjusting the numerical range of the data to the same scale, usually between 0 and 1 or -1 to 1, which helps to speed up the convergence of the model and improve training efficiency.
[0086] In one embodiment, in step 420, when constructing the initial fiber recognition model, a suitable neural network (such as a back propagation neural network (Mat l ab BP neural network) etc.) can be selected as the initial model, and the present application does not limit this. These models can include at least an input layer, two hidden layers and one output layer, wherein the input layer can receive preprocessed data, the hidden layer is responsible for feature extraction and nonlinear transformation, and the output layer is responsible for the final classification or recognition task.
[0087] In one embodiment, in step 430, the preprocessed standard fiber training set is input into the input layer of the model, and fiber feature information useful for classification and location recognition can be extracted through nonlinear transformation of the hidden layer. Such feature information may include the shape, texture, color, etc. of the fiber.
[0088] In one embodiment, in step 440, the output layer can classify or identify the fiber according to the extracted feature information to obtain an output value. The fiber feature information extracted by the hidden layer is passed to the output layer, and the fiber is classified using a classifier (such as a softmax classifier) in the output layer, and the fiber is located and identified in combination with a positioning algorithm (such as a target detection algorithm). The output value obtained is the prediction result of the model for the input fiber sample.
[0089] In one embodiment, in step 450, the iterative training aims to make the predicted output value of the model for the training data gradually approach the true value by repeatedly adjusting the parameters of the model (such as weights and biases). Specifically, by comparing the difference (i.e., error) between the output value and the true value, the model parameters are adjusted using an optimization algorithm such as a back propagation algorithm to reduce the error until the error meets a preset error value (a preset threshold value or reaches a preset training round), and a fiber recognition model with the best indicators is obtained. It should be pointed out that the fiber recognition model with the best indicators refers to a model that performs well in multiple evaluation indicators. It can not only accurately and comprehensively identify the type and cross-sectional area of the fiber, but also has high computational efficiency and low resource consumption.
[0090] In one embodiment, when the initial fiber recognition model is a Matlab BP neural network toolbox, the initial fiber recognition model includes 1 input layer, 2 hidden layers, and 1 output layer, wherein the input layer includes 8 neurons, representing 8 input feature values, and the activation function is a Sigmoid function; the hidden layer has 2 layers, each hidden layer includes 8 neurons, and the activation function of the hidden layer is a Sigmoid function; the output of the output layer is a value between 0 and 10, and the activation function of the output layer is a linear function. The standardized feature set of each component fiber is input into the input layer, and after processing by the hidden layer and the output layer, the corresponding output value (i.e., the type of each component fiber) can be obtained. By iteratively training the fiber recognition model, the training is terminated when the output error (i.e., the difference between the output value and the true value) is less than or equal to 0.001. After the training is completed, a set of weight matrix tables can be obtained, which is the fiber recognition model with the best indicators. It should be pointed out that different output values can represent different types of fibers. For example, an output value of 0.0 represents cotton, an output value of 0.5 represents wool, an output value of 1.0 represents linen, etc. This application does not impose any restrictions on this, and the fiber type represented by the output value can be set according to actual needs.
[0091] In one embodiment, the processing flow from step 410 to step 450 is intended to obtain a fiber recognition model with good performance through continuous iteration and optimization for actual fiber classification or recognition tasks. It should be noted that in actual applications, the parameters and structure of the model can be adjusted according to specific circumstances to obtain the best recognition effect.
[0092] See also Figure 5 In one embodiment, the process of obtaining the fiber training data set may include but is not limited to the following steps 510 to 570.
[0093] Step 510: obtaining a plurality of samples of fabrics of various tissue structures respectively to obtain a second identification sample set, and performing image acquisition on the second identification sample set to obtain a fiber cross-sectional image;
[0094] Step 520: performing image processing on the fiber cross-sectional image to extract the cross-sectional area characteristics, fiber diameter characteristics, fiber perimeter characteristics and fiber edge contour unfolding line characteristics of each component fiber in the fiber cross-sectional image;
[0095] Step 530: obtaining the polarization color of each component fiber, converting the polarization color into a gray value, and obtaining a first gray feature;
[0096] Step 540: Obtain the fluorescence color of each component fiber, convert the fluorescence color into a gray value, and obtain a second gray feature;
[0097] Step 550: Obtain the fiber melting point of each component fiber to obtain the fiber melting point characteristics;
[0098] Step 560: Obtaining the solubility properties of each component fiber to obtain solubility properties characteristics;
[0099] Step 570: Perform feature value standardization processing on the cross-sectional area characteristics, fiber diameter characteristics, fiber perimeter characteristics, fiber edge contour unfolding line characteristics, first grayscale characteristics, second grayscale characteristics, fiber melting point characteristics and solubility performance characteristics of each component fiber to obtain a fiber training data set.
[0100] Steps 510 to 570 are described in detail below.
[0101] In one embodiment, in step 510, samples of fabrics with various tissue structures are collected to ensure the diversity and representativeness of the samples. The fiber cross-sectional images obtained by image acquisition of these samples contain rich morphological and structural information, and by establishing a diverse data set, subsequent fiber identification and feature extraction are facilitated.
[0102] In one embodiment, in step 520, necessary image processing is performed on the fiber cross-sectional image, such as denoising, contrast enhancement, etc., to improve the image quality. The cross-sectional area, diameter, perimeter, and edge contour expansion line of each component fiber are extracted. These features reflect the morphology and size information of the fiber, which are crucial for identifying different types of fibers.
[0103] In one embodiment, in step 530 and step 540, the polarized color is the color of the fiber under specific lighting conditions, and the fluorescent color is the light emitted by the fiber when excited by ultraviolet light. By converting the optical properties of the fiber (polarized color and fluorescent color) into grayscale values, the feature set can be further enriched and the accuracy and robustness of recognition can be improved.
[0104] In one embodiment, in step 550 and step 560, the solubility of each component fiber is obtained, that is, the dissolution of the fiber in different solvents, which reflects the chemical properties of the fiber and its interaction with the solvent. By measuring the physical and chemical properties of the fiber (such as melting point and solubility), the dimension of the feature set can be further increased, and the comprehensiveness of fiber identification can be improved.
[0105] In one embodiment, in step 570, all characteristic values of each component fiber, such as the cross-sectional area, diameter, circumference, edge contour expansion line, two grayscale features, melting point, and solubility properties, are standardized so that all characteristic values are converted to the same scale to ensure smooth subsequent model training.
[0106] In one embodiment, illustratively, the specific process of obtaining the fiber training data set includes: first, collecting identification samples of 11 kinds of fibers respectively, collecting M representative samples of each kind of fiber, where M is a positive integer; the 11 kinds of fibers include: cotton, wool, linen, silk, polyester, nylon, polypropylene, acrylic, Tencel, modal and viscose; then, placing all the identification samples under a microscope and magnifying them by 500 to 1000 times, collecting fiber cross-sectional images of all the identification samples; then, inputting the obtained fiber cross-sectional images of all the identification samples into Mat l ab software is used for image processing to extract the cross-sectional area characteristics, fiber diameter characteristics, fiber circumference characteristics and fiber edge contour unfolding line characteristics of each component fiber in the fiber cross-sectional image; the polarization color of each component fiber is obtained through the polarization component of the microscope, and then the polarization color is converted into a grayscale value to obtain the first grayscale characteristic; the fluorescence color of each component fiber is obtained through the fluorescence component of the microscope, and then the fluorescence color is converted into a grayscale value to obtain the second grayscale characteristic; the fiber melting point of each component fiber is obtained through the melting point determination component of the microscope to obtain the fiber melting point characteristic; the solubility performance of each component fiber is obtained through the in-situ dissolution component of the microscope to obtain the solubility performance characteristic; the eight characteristics of each fiber are subjected to characteristic value standardization processing to obtain eight standardized characteristics of each fiber, and these eight characteristics can be used as a fiber training data set.
[0107] In one embodiment, the processing flow from step 510 to step 570 constructs an identification training set containing a variety of fiber feature information through systematic sample collection, image collection, image processing, feature extraction and feature value standardization steps, providing a solid foundation for subsequent fiber recognition model training.
[0108] In one embodiment, in step 150, since the density of each fiber is usually public or conventional, once the type and cross-sectional area of each fiber are identified, the density of the fiber can be directly corresponded and confirmed. In other words, after the type and cross-sectional area of each fiber are identified with the help of the fiber recognition model with the best indicators in advance, the density of each component fiber can be derived based on these recognition results, combined with the known correspondence between fiber type and density, and cross-sectional area data, and further statistics are calculated for the density value and total cross-sectional area of all component fibers, providing a solid foundation for subsequent analysis.
[0109] In one embodiment, in step 160, it should be pointed out that when analyzing the mass percentage of each component fiber in the fabric sample to be tested (such as woven fabric, knitted fabric or nonwoven fabric) (i.e., the distribution and composition of the fiber), the directionality of the fiber can be determined according to the structural characteristics of the sample. Specifically, since the woven fabric is formed by the yarns in the warp and weft directions interwoven perpendicularly to each other, it has a clear warp and weft structure, so by combining the warp and weft fibers of the woven fabric, the distribution and composition of the fibers in the sample can be fully understood. In addition, since the weft knitted fabric is a fabric formed by one or more yarns fed into the working needle bed of the loom along the transverse direction (weft direction) of the fabric, knitted into loops in a certain order and interlaced with each other, for the weft knitted fabric, the fibers in the weft direction are mainly counted. Since the warp knitted fabric is a knitted fabric formed by one or more groups of parallel yarns, which are fed into all the working needles of the knitting machine in the warp direction and looped at the same time, for the warp knitted fabric, the fibers in the warp direction are mainly counted. Since nonwoven fabrics (also known as non-woven fabrics) are thin sheets or felt-like products made by reinforcing fiber nets by physical or chemical methods, in some cases, the warp and weft fibers of nonwoven fabrics can be combined and counted. That is to say, when determining the directionality of the fibers to be counted based on the structural characteristics of the sample, for the embodiments of the present application, if the fabric sample to be tested is a woven fabric, the warp and weft fibers of the sample to be tested are combined and counted; if the fabric sample to be tested is a weft knitted fabric, the weft fibers of the sample to be tested are counted; if the fabric sample to be tested is a warp knitted fabric, the warp fibers of the sample to be tested are counted; if the fabric sample to be tested is a nonwoven fabric, the warp and weft fibers of the sample to be tested are combined and counted;
[0110] Among them, the calculation formula for the mass percentage of the i-th component fiber of the fabric sample to be tested is:
[0111]
[0112] In the formula, ρ i represents the density of the i-th component fiber, i∈[1,I], I is a positive integer, s i It represents the total cross-sectional area of the i-th component fiber, and M represents the total mass of each component fiber of the fabric sample to be tested.
[0113] The present application is further explained below in conjunction with specific examples, but the specific examples do not limit the present application in any way. Unless otherwise specified, the reagents and methods involved in the examples are commonly used reagents and methods in the art. Specific embodiment one:
[0115] See also Figure 6 and Figure 7 , Figure 6 It is a schematic diagram of a twill fabric provided in a specific embodiment of the present application; Figure 7This is a cross-sectional view of a polyester / cotton blended twill fabric sample provided by a specific embodiment of the present application; Figure 6 and Figure 7 This example is described.
[0116] Step 1: Get a picture of the fabric sample to be tested, input the picture of the fabric sample to be tested into the trained fabric component intelligent recognition model, and obtain the fabric structure type of the fabric sample to be tested; the specific process is:
[0117] 100 representative samples of woven fabrics, warp knitted fabrics, weft knitted fabrics and non-woven fabrics were collected respectively, covering pure fabrics such as cotton, wool, linen, silk, polyester, nylon, polypropylene, acrylic, Tencel, modal, viscose and their different combinations and blended fabrics; in terms of structure, woven fabrics include plain, twill, satin and jacquard fabrics, warp knitted fabrics include mesh fabrics, terry fabrics, pleated fabrics, plush fabrics, weft-inserted fabrics, etc., weft knitted fabrics include weft plain stitch, rib stitch and double-sided stitch, non-woven fabrics include meltblown fabrics, spunbond fabrics, spunlace fabrics, needle-punched fabrics, etc.
[0118] Magnify 50 times under a microscope, collect clear images of each standard sample, input the collected clear images of each standard sample into the trained fabric component intelligent recognition model to identify the weaving structure type of the fabric sample, and obtain the weaving structure type of the sample to be tested, that is, the fabric tissue structure type of the fabric sample to be tested.
[0119] Step 2: Prepare the test sample according to the fabric structure type of the fabric sample to be tested. The test sample is rectangular. The specific process is as follows:
[0120] According to the characteristics of twill woven fabrics, laser cutting is used, controlled by software, with or without the assistance of a mold, and one sample is taken in the warp and weft directions respectively, and laser cutting is used. The width of each sample is 10 mm and the length is 40 mm.
[0121] Step 3: Add cedar oil translucent agent to the short side of the sample so that the translucent agent covers the surface of all fibers in the fabric sample.
[0122] Step 4: Add the epoxy acrylate and polyurethane composite photosensitive resin solution to the short side of the sample, so that the liquid photosensitive resin solution can penetrate into the tightly hugged fibers in the yarn of the fabric sample to expand the distance between the fibers.
[0123] Step 5: Spread the small fabric sample with the photosensitive resin solution added on the transparent quartz glass, and irradiate it with a mixed light composed of ultraviolet light and infrared light from a unidirectional 30° angle for 3 seconds, so that the curing agent that penetrates into the fibers can be quickly cured, thereby achieving the segmentation and fixation of the tightly hugged fibers in the yarn in the fabric sample.
[0124] Step 6: Use a fabric slicer to clamp the short side of the pre-treated sample, and cut the clamped short side of the pre-treated sample flat so that the upper and lower surfaces are flush.
[0125] Step 7: Move the clamped sample to the observation platform of the transflective microscope, automatically adjust the focus to obtain a clear image, and automatically collect the cross-sectional images of all the fibers in the yarn on the short side of the fabric sample from left to right, such as Figure 7 As shown. The obtained fiber cross-sectional image is input into Matlab software for image processing, and the cross-sectional area characteristics, fiber diameter characteristics, fiber perimeter characteristics, and fiber edge contour unfolding line characteristics of each component fiber in the fiber cross-sectional image are extracted; the polarization color of each component fiber is obtained by the polarization component of the microscope, and then the polarization color is converted into a gray value to obtain the first grayscale characteristic; the fluorescence color of each component fiber is obtained by the fluorescence component of the microscope, and then the fluorescence color is converted into a gray value to obtain the second grayscale characteristic; the fiber melting point of each component fiber is obtained by the melting point determination component of the microscope, and the fiber melting point characteristics are obtained; the solubility performance of each component fiber is obtained by the in-situ dissolution component of the microscope, and the solubility performance characteristics are obtained;
[0126] The eight characteristics of each component fiber are subjected to characteristic value standardization processing to obtain eight standardized characteristics of each component fiber; the eight characteristics include: first grayscale characteristic, second grayscale characteristic, fiber melting point characteristic, solubility characteristic, cross-sectional area characteristic, diameter characteristic, perimeter characteristic and fiber edge contour unfolding line characteristic;
[0127] Step 8: Use the fiber recognition model with the best indicators to determine the type and quantity of fibers. The specific process is to input the eight standardized features of each fiber to be tested into the fiber recognition model with the best indicators, and output the type of each fiber; then, based on the type of each fiber and the corresponding cross-sectional area of each fiber, the total cross-sectional area of each component fiber is calculated.
[0128] Step 9: According to the density and total cross-sectional area of each component fiber, all fibers of the warp and weft samples are combined and analyzed to directly calculate the mass percentage of each component fiber in the woven fabric sample. The mass percentage calculation formula of a component fiber is: (density of the component fiber × total cross-sectional area of the component fiber) / total fiber mass, and the polyester is 30.2% and the cotton is 68.8%. Compared with the test results of the chemical dissolution method, the error is less than 1%. Specific embodiment 2:
[0130] See also Figure 8 , Figure 8 This is a cross-sectional view of a polyester / viscose / spandex blended mesh warp knitted fabric provided in a specific embodiment of the present application. Figure 8 This example is described.
[0131] Step A: Obtain a picture of the fabric sample to be tested, input the picture of the fabric sample to be tested into the trained fabric component intelligent recognition model, and obtain the fabric structure type of the fabric sample to be tested; the specific process is:
[0132] 100 representative samples of woven fabrics, warp knitted fabrics, weft knitted fabrics and non-woven fabrics were collected respectively, covering pure fabrics such as cotton, wool, linen, silk, polyester, nylon, polypropylene, acrylic, Tencel, modal, viscose and their different combinations and blended fabrics; in terms of structure, woven fabrics include plain, twill, satin and jacquard fabrics, warp knitted fabrics include mesh fabrics, loop fabrics, pleated fabrics, plush fabrics, weft-inserted fabrics, etc., weft knitted fabrics include weft plain needle fabrics, rib fabrics and double-sided fabrics, non-woven fabrics include meltblown fabrics, spunbond fabrics, spunlace fabrics, needle-punched fabrics, etc., magnified 50 times under a microscope, and collected a clear image of each standard sample, magnified 50 times under a microscope, and collected a clear image of each standard sample, and input the collected clear image of each standard sample into the trained fabric component intelligent recognition model to identify the weaving structure type of the fabric sample, and obtain the weaving structure type of the sample to be tested, that is, the fabric structure type of the fabric sample to be tested.
[0133] Step B: Prepare a test sample according to the fabric structure type of the fabric sample to be tested, and the test sample is rectangular; the specific process is:
[0134] According to the characteristics of mesh warp knitted fabrics, a laser cutting method is used, through software control, with or without the assistance of a mold, a sample is taken in the warp direction and cut by laser, and the width of each sample is 20 mm and the length is 50 mm.
[0135] Step C: Add the composite translucent agent of pine oil and Cargille oil dropwise to the short side of the sample so that the translucent agent covers the surface of all fibers in the fabric sample.
[0136] Step D: Add epoxy acrylate photosensitive resin solution to the short side of the sample, so that the liquid photosensitive resin solution penetrates into the tightly hugged fibers in the yarn of the fabric sample to expand the distance between the fibers.
[0137] Step E: Spread a small fabric sample with a drop of photosensitive resin solution on a transparent quartz glass, and irradiate it with a mixed light consisting of ultraviolet light and infrared light from a single direction at 45° for 10 seconds, so that the curing agent that penetrates between the fibers can be quickly cured, thereby achieving the segmentation and fixation of the tightly hugged fibers in the yarn in the fabric sample.
[0138] Step F: Use a fabric slicer to clamp the short side of the pre-treated sample, and cut the clamped short side of the pre-treated sample flat so that the upper and lower surfaces are flush.
[0139] Step G: Move the clamped sample to the observation platform of the transflective microscope, automatically adjust the focus to obtain a clear image, and automatically collect the cross-sectional images of all the fibers in the yarn on the short side of the fabric sample from right to left, such as Figure 8 The obtained fiber cross-sectional image is input into Matlab software for image processing, and the cross-sectional area characteristics, fiber diameter characteristics, fiber perimeter characteristics, and fiber edge contour unfolding line characteristics of each component fiber in the fiber cross-sectional image are extracted;
[0140] The polarization color of each component fiber is obtained by using a polarization component of a microscope, and then the polarization color is converted into a gray value to obtain a first gray feature;
[0141] The fluorescence color of each component fiber is obtained by using a microscope fluorescence component, and then the fluorescence color is converted into a gray value to obtain a second gray feature;
[0142] The fiber melting point of each component fiber is obtained by using the melting point determination component equipped with the microscope to obtain the fiber melting point characteristics;
[0143] The dissolution properties of each component fiber are obtained through the in-situ dissolution component equipped with the microscope, and the dissolution performance characteristics are obtained;
[0144] The eight characteristics of each component fiber are subjected to characteristic value standardization processing to obtain eight standardized characteristics of each component fiber; the eight characteristics include: first grayscale characteristic, second grayscale characteristic, fiber melting point characteristic, solubility characteristic, cross-sectional area characteristic, diameter characteristic, perimeter characteristic and fiber edge contour unfolding line characteristic;
[0145] Step H: Input the eight standardized features of each fiber to be tested into the fiber recognition model with the best indicators, and output the type of each fiber; then, according to the type of each fiber and the corresponding cross-sectional area of each fiber, calculate the total cross-sectional area of each component fiber.
[0146] Step I: According to the density and total cross-sectional area of each component fiber, all fibers in the warp sample are analyzed and counted, and the mass percentage of each component fiber in the warp knitted fabric sample is directly calculated. The mass percentage calculation formula of a component fiber is: (density of the component fiber × total cross-sectional area of the component fiber) / total fiber mass, and the results are 10.3% for polyester, 85.0% for viscose, and 4.7% for spandex. Compared with the test results of the chemical dissolution method, the error is less than 1%. Specific embodiment three:
[0148] See also Fig. 9 , Fig. 9 This is a cross-sectional view of a wool / acrylic / spandex blended weft knitted sample provided in a specific embodiment of the present application. Fig. 9 This example is described.
[0149] Step (1): obtain a picture of the fabric sample to be tested, input the picture of the fabric sample to be tested into the trained fabric component intelligent recognition model, and obtain the fabric structure type of the fabric sample to be tested; the specific process is:
[0150] 100 representative samples of woven fabrics, warp knitted fabrics, weft knitted fabrics and non-woven fabrics were collected respectively, covering pure fabrics such as cotton, wool, linen, silk, polyester, nylon, polypropylene, acrylic, Tencel, modal, viscose and their different combinations of blended fabrics; in terms of structure, woven fabrics include plain, twill, satin and jacquard fabrics, warp knitted fabrics include mesh fabrics, terry fabrics, pleated fabrics, plush fabrics, weft-inserted fabrics, etc., weft knitted fabrics include weft plain needle fabrics, rib fabrics and double reverse fabrics, non-woven fabrics include meltblown fabrics, spunbond fabrics, spunlace fabrics, needle-punched fabrics, etc. The fabric samples to be tested were magnified 50 times under a microscope, and clear images of each standard sample were collected. The clear images of each standard sample were input into the trained fabric component intelligent recognition model to identify the type of weaving structure of the fabric sample, and the type of weaving structure of the sample to be tested, that is, the type of fabric structure of the fabric sample to be tested, was obtained.
[0151] Step (2): Prepare a test sample according to the fabric structure type of the test fabric sample, and the test sample is rectangular; the specific process is:
[0152] According to the characteristics of weft knitted fabrics with plain weft needles, a sample is taken in the weft direction and cut by laser. The width of the sample is 30 mm and the length is 100 mm.
[0153] Step (3): Add pine tar oil translucent agent to the short side of the sample so that the translucent agent covers the surface of all fibers in the fabric sample.
[0154] Step (4): drip the epoxy acrylate photosensitive resin solution onto the short side of the sample, so that the liquid photosensitive resin solution penetrates into the spaces between the tightly packed fibers in the yarn of the fabric sample, thereby expanding the distance between the fibers.
[0155] Step (5): Spread a small fabric sample with the photosensitive resin solution added on a transparent quartz glass, and irradiate it with a mixed light consisting of ultraviolet light and infrared light from a unidirectional angle of 60° for 20 seconds to quickly solidify the curing agent that penetrates between the fibers, thereby achieving the segmentation and fixation of the tightly hugged fibers in the yarn in the fabric sample.
[0156] Step (6): Use a fabric slicer to clamp the short side of the pre-treated sample, and cut the clamped short side of the pre-treated sample flat so that the upper and lower surfaces are flush.
[0157] Step (7): Move the clamped sample to the observation platform of the transflective microscope, automatically adjust the focus to obtain a clear image, and automatically collect the cross-sectional images of all the fibers in the yarn on the short side of the fabric sample from right to left, such as Fig. 9 As shown. The obtained fiber cross-sectional image is input into Matlab software for image processing, and the cross-sectional area characteristics, fiber diameter characteristics, fiber perimeter characteristics, and fiber edge contour unfolding line characteristics of each component fiber in the fiber cross-sectional image are extracted; the polarization color of each component fiber is obtained by the polarization component of the microscope, and then the polarization color is converted into a gray value to obtain the first grayscale characteristic; the fluorescence color of each component fiber is obtained by the fluorescence component of the microscope, and then the fluorescence color is converted into a gray value to obtain the second grayscale characteristic; the fiber melting point of each component fiber is obtained by the melting point determination component of the microscope, and the fiber melting point characteristics are obtained; the solubility performance of each component fiber is obtained by the in-situ dissolution component of the microscope, and the solubility performance characteristics are obtained;
[0158] The eight characteristics of each component fiber are subjected to characteristic value standardization processing to obtain eight standardized characteristics of each component fiber; the eight characteristics include: first grayscale characteristic, second grayscale characteristic, fiber melting point characteristic, solubility characteristic, cross-sectional area characteristic, diameter characteristic, perimeter characteristic and fiber edge contour unfolding line characteristic;
[0159] Step (8): Use the fiber recognition model with the best indicators to determine the type and quantity of fibers. Specifically, the eight standardized features of each fiber to be tested are input into the fiber recognition model with the best indicators, and the type of each fiber is output; then, based on the type of each fiber and the corresponding cross-sectional area of each fiber, the total cross-sectional area of each component fiber is calculated.
[0160] Step (9): According to the density and total cross-sectional area of each component fiber, all fibers in the weft sample are analyzed and counted, and the mass percentage of each component fiber in the weft knitted fabric sample is directly calculated. The mass percentage calculation formula of a component fiber is: (density of the component fiber × total cross-sectional area of the component fiber) / total fiber mass, and the results are 70.0% for wool, 25.0% for acrylic, and 5.0% for spandex. Compared with the test results of the chemical dissolution method, the error is less than 1%. Specific embodiment four:
[0162] See also Fig.10 , Fig.10 This is a cross-sectional view of a viscose / cotton mixed spunlace nonwoven fabric provided in a specific embodiment of the present application. Fig.10 This example is described.
[0163] Step I: Obtain a picture of the fabric sample to be tested, input the picture of the fabric sample to be tested into the trained fabric component intelligent recognition model, and obtain the fabric structure type of the fabric sample to be tested; the specific process is:
[0164] 100 representative samples of woven fabrics, warp knitted fabrics, weft knitted fabrics and non-woven fabrics were collected respectively, covering pure fabrics such as cotton, wool, linen, silk, polyester, nylon, polypropylene, acrylic, Tencel, modal, viscose and their different combinations of blended fabrics; in terms of structure, woven fabrics include plain, twill, satin and jacquard fabrics, warp knitted fabrics include mesh fabrics, terry fabrics, pleated fabrics, plush fabrics, weft-inserted fabrics, etc., weft knitted fabrics include weft plain needle fabrics, rib fabrics and double reverse fabrics, non-woven fabrics include meltblown fabrics, spunbond fabrics, spunlace fabrics, needle-punched fabrics, etc. The samples to be tested were magnified 50 times under a microscope, and clear images of each standard sample were collected. The clear images of each standard sample were input into the trained fabric component intelligent recognition model to identify the weaving structure type of the fabric sample, and the weaving structure type of the sample to be tested was obtained, that is, the fabric structure type of the fabric sample to be tested.
[0165] Step II: Prepare the test sample according to the fabric structure type of the fabric sample to be tested. The test sample is rectangular. The specific process is as follows:
[0166] According to the characteristics of spunlace nonwovens, laser cutting is used, controlled by software, with or without the assistance of a mold, and one sample is taken in the warp and weft directions, and laser cutting is used. The width of each sample is 20 mm and the length is 500 mm.
[0167] Step III: Add pine tar oil translucent agent to the short side of the sample so that the translucent agent covers the surface of all fibers in the fabric sample.
[0168] Step IV: Add the polyester photosensitive resin solution dropwise to the short side of the sample, so that the liquid photosensitive resin solution penetrates into the tightly hugged fibers in the yarn of the fabric sample to expand the distance between the fibers.
[0169] Step V: Spread a small fabric sample with a drop of photosensitive resin solution on a transparent quartz glass, and irradiate it with a mixed light consisting of ultraviolet light and infrared light from a single direction at 90° for 30 seconds, so that the curing agent that penetrates between the fibers can be quickly cured, thereby achieving the segmentation and fixation of the tightly hugged fibers in the yarn in the fabric sample.
[0170] Step VI: Use a fabric slicer to clamp the short side of the pre-treated sample, and cut the clamped short side of the pre-treated sample flat so that the upper and lower surfaces are flush.
[0171] Step VII: Move the clamped sample to the observation platform of the transflective microscope, automatically adjust the focus to obtain a clear image, and automatically collect the cross-sectional images of all the fibers in the yarn on the short side of the fabric sample from right to left, such as Fig.10As shown. The obtained fiber cross-sectional image is input into Matlab software for image processing, and the cross-sectional area characteristics, fiber diameter characteristics, fiber perimeter characteristics, and fiber edge contour unfolding line characteristics of each component fiber in the fiber cross-sectional image are extracted; the polarization color of each component fiber is obtained by the polarization component of the microscope, and then the polarization color is converted into a gray value to obtain the first grayscale characteristic; the fluorescence color of each component fiber is obtained by the fluorescence component of the microscope, and then the fluorescence color is converted into a gray value to obtain the second grayscale characteristic; the fiber melting point of each component fiber is obtained by the melting point determination component of the microscope, and the fiber melting point characteristics are obtained; the solubility performance of each component fiber is obtained by the in-situ dissolution component of the microscope, and the solubility performance characteristics are obtained;
[0172] The eight characteristics of each component fiber are subjected to characteristic value standardization processing to obtain eight standardized characteristics of each component fiber; the eight characteristics include: first grayscale characteristic, second grayscale characteristic, fiber melting point characteristic, solubility characteristic, cross-sectional area characteristic, diameter characteristic, perimeter characteristic and fiber edge contour unfolding line characteristic;
[0173] Step VIII: Use the fiber recognition model with the best indicators to determine the type and quantity of fibers. Specifically, the eight standardized features of each fiber to be tested are input into the fiber recognition model with the best indicators, and the type of each fiber is output; then, based on the type of each fiber and the corresponding cross-sectional area of each fiber, the total cross-sectional area of each component fiber is calculated.
[0174] Step IX: According to the density and total cross-sectional area of each component fiber, all fibers in the warp sample are analyzed and counted, and the mass percentage of each component fiber in the spunlace nonwoven fabric sample is directly calculated. The mass percentage calculation formula of a component fiber is: (density of the component fiber × total cross-sectional area of the component fiber) / total fiber mass, and the viscose is 55.0% and the cotton is 45.0%. Compared with the test results of the chemical dissolution method, the error is less than 1%.
[0175] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A fabric composition intelligent analysis method, characterized in that: include: Obtaining a fabric sample to be tested, and identifying the fabric structure type of the fabric sample to be tested; According to the fabric structure type, the fabric sample to be tested is subjected to photothermal curing pretreatment and standard sample rapid preparation to obtain a standard sample; Taking a cross-sectional image of fibers in the short side yarn of the standard sample; Extracting a feature set of each component fiber in the fiber cross-sectional image, and identifying the type and cross-sectional area of each fiber according to the feature set; According to the type and cross-sectional area of each fiber, the density and total cross-sectional area of each component fiber are statistically obtained; The mass percentage of each component fiber in the fabric sample to be tested is calculated based on the density and total cross-sectional area of each component fiber.
2. The intelligent analysis method of fabric composition according to claim 1, characterized in that: The identifying and obtaining the fabric structure type of the fabric sample to be tested includes: inputting the fabric sample to be tested into a trained fabric component intelligent recognition model for fabric recognition to obtain the fabric structure type of the fabric sample to be tested; The training steps of the trained fabric component intelligent recognition model include: obtaining a fabric recognition data set, denoising and normalizing the fabric recognition data set to obtain a standard fabric training set; constructing an initial fabric component intelligent recognition model; inputting the standard fabric training set into the initial fabric component intelligent recognition model to perform fabric recognition and obtain prediction information; calculating a first loss value based on the prediction information and a true value, adjusting the parameters of the initial fabric component intelligent recognition model by the first loss value until the first loss value meets a preset value, and obtaining a trained fabric component intelligent recognition model.
3. A fabric composition intelligent analysis method according to claim 2, characterized in that: The method of obtaining a fabric recognition data set comprises: respectively obtaining a plurality of samples of woven fabrics, warp knitted fabrics, weft knitted fabrics and non-woven fabrics to obtain a first recognition sample set; and performing image acquisition on the first recognition sample set to obtain a fabric recognition data set.
4. A fabric composition intelligent analysis method according to claim 3, characterized in that: The method of performing photothermal curing pretreatment on the fabric sample to be tested and rapidly preparing a standard sample according to the fabric tissue structure type to obtain a standard sample includes: preparing a sample to be tested according to the fabric tissue structure type, dripping a light-transmitting agent and a photosensitive resin solution onto the short side of the sample to be tested to obtain a pretreated sample; and performing multi-angle light irradiation and photothermal curing on the pretreated sample to obtain a standard sample.
5. A fabric composition intelligent analysis method according to claim 4, characterized in that: The test sample is prepared according to the type of fabric structure, including: selecting a first sample in the warp and weft directions of the woven fabric, selecting a second sample in the warp direction of the warp knitted fabric, selecting a third sample in the weft direction of the weft knitted fabric, selecting a fourth sample in the warp and weft directions of the non-woven fabric, and obtaining the test sample from the first sample, the second sample, the third sample or the fourth sample.
6. According to the intelligent analysis method of fabric composition according to claim 4, the composition of the photosensitive resin includes at least one of acrylic acid, epoxy acrylate, epoxy acrylate, polyester and polyurethane; the translucent agent includes at least one of cedar oil, pine oil, Cargille oil and xylene.
7. The method for intelligent analysis of fabric composition according to claim 4, characterized in that: The method of subjecting the pre-treated sample to multi-angle light irradiation and photothermal curing to obtain a standard sample comprises: laying the pre-treated sample flat on a photothermal curing table, using mixed light consisting of ultraviolet light and infrared light, irradiating the sample from multiple angles in one direction or in both directions to obtain a standard sample; The angle formed by the multi-angle light irradiation direction and the corresponding plane is in the range of 30° to 150°, and the irradiation time T is in the range of 3 seconds to 60 seconds.
8. The method for intelligent analysis of fabric composition according to claim 1, characterized in that: The method of obtaining the type and cross-sectional area of each fiber according to the feature set includes: performing feature value standardization processing on the feature set to obtain a standardized feature set; inputting the standardized feature set into a fiber recognition model with the best indicators for classification and positioning recognition to obtain the type and cross-sectional area of each fiber; The training process of the fiber recognition model with the best indicators includes: obtaining a fiber training data set, denoising and normalizing the fiber training data set to obtain a standard fiber training set; constructing an initial fiber recognition model, wherein the initial fiber recognition model includes an input layer, a hidden layer, and an output layer; inputting the standard fiber training set into the input layer, extracting features of the standard fiber training set using the hidden layer to obtain fiber feature information; classifying and locating the fiber feature information using the output layer to obtain an output value; iteratively training the fiber recognition model until the error between the output value and the true value meets a preset error value, thereby obtaining a fiber recognition model with the best indicators.
9. A fabric composition intelligent analysis method according to claim 8, characterized in that: The method for obtaining a fiber training data set includes: obtaining multiple samples of fabrics of multiple tissue structures respectively to obtain a second identification sample set; performing image acquisition on the second identification sample set to obtain a fiber cross-sectional image; performing image processing on the fiber cross-sectional image to extract and obtain the cross-sectional area feature, fiber diameter feature, fiber perimeter feature and fiber edge contour expansion line feature of each component fiber in the fiber cross-sectional image; obtaining the polarization color of each component fiber, converting the polarization color into a grayscale value, and obtaining a first grayscale feature; obtaining the fluorescent color of each component fiber, converting the fluorescent color into a grayscale value, and obtaining a second grayscale feature; obtaining the fiber melting point of each component fiber to obtain a fiber melting point feature; obtaining the solubility property of each component fiber to obtain a solubility property feature; performing feature value standardization processing on the cross-sectional area feature, fiber diameter feature, fiber perimeter feature, fiber edge contour expansion line feature, first grayscale feature, second grayscale feature, fiber melting point feature and solubility property feature of each component fiber to obtain a fiber training data set.
10. The intelligent analysis method of fabric composition according to claim 2, characterized in that: The fabric structure types of the fabric sample to be tested include woven fabrics, warp knitted fabrics, weft knitted fabrics and non-woven fabrics. The mass percentage of each component fiber of the fabric sample to be tested is calculated according to the total cross-sectional area and density of each component fiber, including: If the fabric sample to be tested is a woven fabric, the fibers in the warp and weft directions of the woven fabric are combined and counted; if the fabric sample to be tested is a weft knitted fabric, the fibers in the weft direction of the weft knitted fabric are counted; if the fabric sample to be tested is a warp knitted fabric, the fibers in the warp direction of the warp knitted fabric are counted; if the fabric sample to be tested is a non-woven fabric, the fibers in the warp and weft directions of the non-woven fabric are combined and counted; The calculation formula for the mass percentage of the i-th component fiber of the fabric sample to be tested is: In the formula, ρ i represents the density of the i-th component fiber, i∈[1,I], I is a positive integer, s i It represents the total cross-sectional area of the i-th component fiber, and M represents the total mass of each component fiber of the fabric sample to be tested.
