Dyeing effect prediction method, dyeing effect prediction model training method, apparatus, electronic device, storage medium, and program

Abstract

To provide a dyeing effect prediction method, a dyeing effect prediction model training method, an apparatus, an electronic device, a storage medium, and a program.SOLUTION: The present disclosure relates to the field of computers, and in particular to artificial intelligence techniques, neural network model techniques, and model training techniques. A method comprises: decomposing Raman spectrum data of a measurement target wound yarn package into a plurality of pieces of sub-signal data using wavelet basis functions; determining measurement target wound yarn package characteristics based on at least a portion of the plurality of pieces of sub-signal data; and predicting and obtaining a first dyeing label for indicating a grade of a dyeing effect based on the measurement target wound yarn package characteristics using a dyeing effect prediction model. According to the solution of the present disclosure, the dyeing grade of each lot of wound yarn package can be detected without dyeing, thereby improving the detection accuracy and detection efficiency.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to the field of computers, and in particular to artificial intelligence techniques, neural network model techniques, and model training techniques. [Background technology]

[0002] In the production of chemical fibers, it is difficult to ensure that the process parameters of multiple production devices are completely consistent, and there may be differences in the process parameters when the same device is used to produce different batches, so that when the chemical fiber packages are subsequently dyed, the coloring effect may change accordingly, which is difficult to accurately predict.Currently, in order to determine the dyeing effect of different batches of wound yarn packages, a method is adopted in which a part of the wound yarn packages of each batch is extracted as a sample, woven into socks and then dyed, and the dyeing effect is artificially identified and divided into several grades. Summary of the Invention [Problem to be solved by the invention]

[0003] The present disclosure provides a dyeing effect prediction method, a dyeing effect prediction model training method, an apparatus, an electronic device, a storage medium, and a program, which solve or alleviate one or more technical problems in the prior art. [Means for solving the problem]

[0004] In a first aspect, the present disclosure provides a method for predicting a staining effect, the method comprising: Decomposing Raman spectrum data of the measured wound yarn package into a plurality of sub-signal data using wavelet basis functions; determining a measured wound yarn package characteristic based on at least a portion of the target sub-signal data of the plurality of sub-signal data; and predicting and obtaining a first dyeing label representing a grade of dyeing effect based on the measured wound yarn package characteristics using a dyeing effect prediction model.

[0005] In a second aspect, the present disclosure provides a method for training a staining effect prediction model, the method comprising: Decomposing the Raman spectrum data of the sample wound yarn package into a plurality of sub-signal data using wavelet basis functions; determining a sample wound yarn package characteristic based on at least some of the sub-signal data of the plurality of sub-signal data; predicting a second dyed label based on the sample wound yarn package characteristics using a first predictive model; and Obtaining a loss function based on the actual dye label of the sample wound yarn package and the second dye label; and updating parameters of the first prediction model based on the loss function to obtain a trained dyeing effect prediction model.

[0006] In a third aspect, the present disclosure provides a dyeing effect prediction device, the device comprising: A first transformation module is used to decompose the Raman spectrum data of the measurement target wound yarn package, which is a crystalline polyester fiber product, by using a wavelet basis function to obtain a plurality of sub-signal data; a first determination module for determining a target wound yarn package characteristic based on at least a portion of the target sub-signal data of the plurality of sub-signal data; and a first prediction module for predicting and obtaining a first dyeing label for indicating a grade of dyeing effect based on the measured wound yarn package characteristics by using a dyeing effect prediction model.

[0007] In a fourth aspect, the present disclosure provides an apparatus for training a dyeing effect prediction model, the apparatus comprising: A second transformation module for decomposing the Raman spectrum data of the sample wound yarn package, which is a crystalline polyester fiber product, into a plurality of sub-signal data by using a wavelet basis function; a second determination module for determining a sample wound yarn package characteristic based on at least some of the sub-signal data of the plurality of sub-signal data; a second prediction module for predicting a second dyed label based on the sample wound yarn package characteristics using the first prediction model; a loss determination module for obtaining a loss function based on the actual dye label of the sample wound yarn package and the second dye label; and an update module for updating parameters of the first prediction model based on the loss function to obtain a trained dyeing effect prediction model.

[0008] In a fifth aspect, the present disclosure provides an electronic device, the device comprising: At least one processor; a memory in communication with the at least one processor; The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, cause the implementation of any one of the methods in the embodiments of the present disclosure.

[0009] In a sixth aspect, there is provided a non-transitory computer-readable storage medium having stored thereon computer instructions for causing a computer to perform any one of the methods in the embodiments of the present disclosure.

[0010] The beneficial effects of the technical solutions provided by the present disclosure include at least the following: That is, the dyeing grade of each lot of wound yarn package can be detected without dyeing, improving detection accuracy and detection efficiency.

[0011] It should be understood that the contents described herein are not intended to describe key or important features of the embodiments of the present disclosure, nor are they used to limit the scope of the present disclosure, other features of the present disclosure are facilitated by the following specification. [Brief description of the drawings]

[0012] In the accompanying drawings, the same numbers represent the same or similar parts or elements throughout the several views unless otherwise stated. The accompanying drawings are not necessarily drawn to scale. It should be understood that the drawings depict only some embodiments provided by the present disclosure and should not be considered as limiting the scope of the present disclosure. [Figure 1] 1 is a flowchart of a dyeing effect prediction method according to an embodiment of the present disclosure. [Diagram 2] 1 is a flowchart of a method for training a dyeing effect prediction model according to an embodiment of the present disclosure. [Diagram 3] FIG. 1 is a schematic diagram showing the configuration of a dyeing effect prediction model according to an embodiment of the present disclosure. [Figure 4] 1 is a schematic diagram showing a configuration of a dyeing effect prediction device according to an embodiment of the present disclosure. [Diagram 5] FIG. 1 is a schematic diagram showing a configuration of a training device for a dyeing effect prediction model according to an embodiment of the present disclosure. [Figure 6] FIG. 1 is a block diagram of an electronic device according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] The present disclosure will now be described in more detail with reference to the accompanying drawings, in which like reference numbers represent functionally identical or similar elements, and in which various aspects of embodiments are illustrated, and which, unless otherwise noted, are not necessarily drawn to scale.

[0014] Furthermore, in order to better explain the present disclosure, many specific details are described in the following specific embodiments. It should be understood by those skilled in the art that the present disclosure can be similarly implemented without specific details. In some embodiments, methods, means, components, circuits, etc. that are well known to those skilled in the art are not described in detail so as not to obscure the gist of the present disclosure.

[0015] The detection method of manually judging dyeing grade described in the background art requires sampling, weaving and dyeing of chemical fiber packages, which is time-consuming and inefficient, and has problems such as missing products with dyeing abnormalities. In addition, the accuracy of the manual color judgment results is influenced by the inspector's experience and has low accuracy. At the same time, the manual color judgment results are influenced by the inspector's subjectivity and have low objectivity.

[0016] In order to at least partially address one or more of the above problems and other potential problems, an embodiment of the present disclosure provides a method for predicting dyeing effect, and by utilizing the technical solution according to an embodiment of the present disclosure, the dyeing grade of each lot of wound yarn package can be objectively predicted without dyeing, thereby improving detection accuracy and detection efficiency.

[0017] 1 is a method for predicting dyeing effect provided by one embodiment of the present disclosure. As shown in FIG. 1, the method includes at least the following steps:

[0018] In S101, the Raman spectrum data of the wound yarn package to be measured is decomposed into a plurality of sub-signal data using wavelet basis functions.

[0019] In the embodiment of the present disclosure, one or more measurement points on the wound yarn package to be measured can be detected using a Raman spectrometer to obtain Raman spectrum data for each measurement point. The measurement points can be selected on the circular side of the wound yarn package to be measured. For example, 3 to 6 measurement points can be selected on the upward circular side of the wound yarn package to be measured while the wound yarn package to be measured moves on the production line. Alternatively, the wound yarn package to be measured can be lifted by a lifting device to select 2 to 5 measurement points on each of the upper and lower circular sides to improve the accuracy of the measurement results. The number of measurement points is not limited and can be adjusted according to the actual situation.

[0020] The measured yarn package may be a polyester fiber product having a crystalline structure, or may be a polyester fiber product having an amorphous structure. The polyester fiber product may be a polyester filament obtained by spinning and post-treatment. The crystalline structure refers to a state in which polymers are arranged in an orderly manner in the fiber. In the crystalline structure, the polymer chains are densely arranged with each other to form an orderly lattice structure. The fiber having a crystalline structure generally has a high degree of crystallinity, and is shown to have a sharp diffraction peak in an experiment such as X-ray diffraction. Due to the orderly arrangement of the molecules, the fiber having a crystalline structure is generally relatively stable, and also has relatively high strength and hardness. The amorphous structure refers to the polymer being in a disordered state in the fiber and not forming a regular lattice structure. In the amorphous structure, the polymer chains are relatively loose and do not have a clear periodic arrangement. The fiber having an amorphous structure generally has a relatively low degree of crystallinity. Usually, the lattice structure is not colored during dyeing, so the degree of crystallinity, crystal distribution, and crystal orientation all have a certain correlation with the dyeing effect.

[0021] Wavelet transform is a signal processing technique that can decompose a signal into sub-signals of different frequency ranges, thereby obtaining the local features of the signal. In staining effect prediction, the wavelet transform can be used to efficiently capture the frequency domain features in Raman spectrum data.

[0022] In one example, the process of wavelet transformation includes: Pre-processing the collected Raman spectral data, including operations such as peak correction, baseline correction, spectral smoothing, and noise reduction, to reduce the influence of noise on the wavelet transformation results; Applying wavelet transformation to the pre-processed Raman spectral data; Selecting appropriate wavelet basis functions to decompose the spectral data into sub-signals of different scales and frequency ranges. Any of the following wavelet basis functions can be utilized:

[0023] Haar Wavelet: The simplest wavelet basis function, used in the discrete wavelet transform, is a simple square wave function with equalized frequency and time domain localization.

[0024] Morlet wavelet: used in the continuous wavelet transform, it is a complex wavelet basis function used to analyze signals in the time-frequency domain.

[0025] Daubechies wavelet: commonly used in discrete wavelet transform, it is a family of orthogonal wavelet basis functions, different Daubechies wavelet basis functions have different scale and frequency characteristics.

[0026] Approximate Symmetric Symlet Wavelet: A symmetric version of the Daubechies wavelet, also commonly used in discrete wavelet transforms.

[0027] In S102, a measurement target wound yarn package characteristic is determined based on at least a portion of the target sub-signal data of the plurality of sub-signal data.

[0028] A representative target sub-signal data is obtained by screening the multiple sub-signal data obtained by the wavelet transform. A feature extraction is performed on the target sub-signal data to obtain a measurement target wound yarn package feature. The feature extraction may use statistical features (e.g., mean, variance, kurtosis, etc.), frequency domain features (e.g., energy distribution, spectral morphology, etc.), or other related features.

[0029] In S103, a first dyeing label is predicted based on the measured yarn package characteristics using the dyeing effect prediction model, where the first dyeing label represents a grade of dyeing effect.

[0030] The dyeing effect prediction model may be a pre-trained deep learning model, and the deep learning model may use a convolutional neural network (CNN) or a recurrent neural network (RNN). The model is trained using labeled sample Raman spectrum data. The deep learning model automatically learns a high-level feature representation of the sample data to form a feature learner suitable for dyeing grade prediction.

[0031] According to the solutions of the embodiments of the present disclosure, Raman spectrum data can be used to realize online detection in the production scenario, and the dyeing effect of the yarn package can be detected in real time, so that the parameters of the production equipment and the production process can be timely fed back and adjusted according to the detection results; the dyeing effect prediction model can learn the characteristic information of polymer fibers with crystalline structure and amorphous structure, so that the model can predict the dyeing effects of two kinds of polyester fibers; and the feature extraction method based on wavelet transform of Raman spectrum data can capture signal changes of different scales and frequency ranges, so as to efficiently extract more abundant and accurate feature information.

[0032] In one possible implementation form, in step S102, determining the measured wound yarn package characteristic based on at least a portion of the target sub-signal data of the plurality of sub-signal data further includes the following steps.

[0033] In S1021, at least some target sub-signal data are selected from the plurality of sub-signal data according to a preset scale or a preset frequency to obtain first Raman data, where the plurality of sub-signal data correspond to wavelet coefficients of different scales or spectral components of different frequencies.

[0034] In S1022, a wound yarn package characteristic to be measured is determined based on the first Raman data.

[0035] In the embodiment of the present disclosure, a plurality of sub-signal data are screened to obtain more representative target sub-signal data, and a plurality of sets of sub-signal data are obtained by performing a wavelet transform multiple times using a plurality of wavelet basis functions on the Raman spectrum data of the winding package to be measured. Then, a plurality of sets of sub-signal data are screened to obtain more representative target sub-signal data, and the corresponding transform scale or transform frequency is a preset scale or preset frequency. Different wavelet basis functions can be used for crystalline and amorphous polymer fibers.

[0036] Screening the plurality of sub-signal data may include the following steps. a. Select the scale range of interest. Select the wavelet scale range of interest based on the characteristics of the spectral data. Wavelet transforms of different scales will highlight the features of different frequency ranges, so select the appropriate scale range according to your research needs. b. Extract the target frequency band. In the image of wavelet coefficients, extract the sub-signals of the target frequency band according to the selected scale range. By setting a threshold on the image of wavelet coefficients, the low amplitude wavelet coefficients can be set to 0 and the high amplitude sub-signals can be retained. c) Inverse wavelet transform: Perform inverse wavelet transform on the extracted target sub-signal to convert it into the original spectral data representation. d. Visualization and comparison. The target sub-signal data after inverse transformation is visualized, and its similarity and difference with the original spectrum data are compared. In order to intuitively evaluate the representativeness of the screened sub-signal data, a comparison diagram between the target sub-signal data and the original spectrum data can be drawn. e. Select representative sub-signals. Based on the comparison result, select a more representative target sub-signal as the final result. A more representative sub-signal should be able to retain the main features and information of the original spectrum data, and remove more noise and irrelevant information.

[0037] The above screening process can be adjusted and optimized according to the specific production scenarios of different synthetic fiber products. Different wavelet scale ranges or threshold settings can be tried to achieve better screening effects. At the same time, if prior knowledge of the characteristics of the spectral data is available, it can be added to the screening process, which further improves the representativeness of the sub-signals. Through the screening process, more representative and information-rich target sub-signal data can be selected from the multiple sub-signals obtained by wavelet transformation.

[0038] In one example, a wavelet transform is performed on the Raman spectrum data of the wound yarn package to be measured using a first wavelet basis function to obtain first sub-signal data A, first sub-signal data B, and first sub-signal data C. A wavelet transform is performed on the Raman spectrum data of the wound yarn package to be measured using a second wavelet basis function to obtain second sub-signal data A, second sub-signal data B, and second sub-signal data C. The above multiple sub-signal data correspond to multiple wavelet coefficients of different scales or multiple spectral components of different frequencies. Next, screening is performed on the multiple sub-signal data to obtain target sub-signal data (e.g., second sub-signal data A, first sub-signal data B).

[0039] According to the method of the embodiments of the present disclosure, signal changes of different scales and frequency ranges can be captured, thereby extracting richer and more accurate feature information, which helps the dyeing effect prediction to obtain more accurate and reliable results.

[0040] In one possible implementation, determining the wound yarn package characteristic to be measured based on the first Raman data in step S1022 further includes:

[0041] In S110, a characteristic of the wound yarn package to be measured is determined based on the micro image data of the wound yarn package to be measured and the first Raman data.

[0042] In the embodiment of the present disclosure, the first Raman data can read some structural features of the measured wound yarn package, and by introducing micro image data of the measured wound yarn package, more feature information that may affect the dyeing effect can be further obtained. The multi-modal data can fully utilize the complementary information between the Raman spectrum data and the image data, and can improve the accuracy and robustness of the dyeing grade prediction.

[0043] In one possible implementation, determining the measured wound yarn package characteristic based on the micro image data of the measured wound yarn package and the first Raman data in step S110 further includes:

[0044] In S1101, a multimodal autoencoder included in the dyeing effect prediction model is used to determine target image features based on micro image data of the wound yarn package to be measured.

[0045] In S1102, a multimodal autoencoder is utilized to determine target spectral features based on the first Raman data.

[0046] In S1103, a feature fusion layer included in the dyeing effect prediction model is used to determine the measurement target wound yarn package feature based on the target image feature and the target spectrum feature.

[0047] In the embodiment of the present disclosure, as shown in FIG. 3, an autoencoder is used as a feature extractor during feature extraction. An autoencoder is an unsupervised learning method that can learn a compact representation from input data and reconstruct the original data. To accommodate multi-modal data, a multi-modal autoencoder can be constructed to perform feature extraction on Raman spectrum data and micro-image data, respectively. By using an autoencoder, the model can automatically learn useful feature representations that can be applied to different data without the need to manually design a feature extraction method.

[0048] The network structure of the autoencoder can be composed of multiple encoding and decoding layers. Different encoding and decoding modules, such as convolutional encoders and deconvolutional decoders, are utilized to better capture the spatial features and contextual information of the spectral and image data. Furthermore, cross-modal connections can be added between the encoding layers to enable the model to exchange information between the spectral and image data, thereby further enhancing the feature learning ability.

[0049] The feature fusion layer fuses the features from the two feature extraction networks. This can be accomplished by a simple combining operation or a more complex fusion layer. The fusion layer may be a fully connected layer, a weighted fusion, or other fusion strategies to combine features from the spectral and image data.

[0050] According to the solution of the embodiments of the present disclosure, the information of different data sources can be comprehensively utilized through multi-modal data fusion, so that the characteristics related to the dyeing effect prediction task can be more comprehensively described, and the complementarity of multiple data sources can be fully utilized, so as to provide more accurate and robust dyeing effect prediction results.

[0051] In one possible implementation form, determining target image features based on micro image data of the measured wound yarn package using a multimodal autoencoder included in the dyeing effect prediction model in step S1101 further includes:

[0052] Based on the micro image data of the measured wound yarn package, structural parameters of the crystal structure are determined, where the structural parameters include the degree of crystallinity and / or the degree of orientation.

[0053] A multimodal autoencoder included in the dyeing effect prediction model is used to determine target image features based on the micro image data and structure parameters.

[0054] In the embodiment of the present disclosure, a high-resolution microscope such as a transmission electron microscope or a scanning electron microscope can be used to image the measurement point of the winding package to be measured. The microscope can provide a high-resolution image to observe the microstructure of the fiber. The micro image data can intuitively observe the distribution and orientation of the crystals, and the crystallinity is determined by calculating the area occupied by the crystals (the ratio of the number of pixels in the crystalline region to the total number of pixels). In polyester fibers, polyester molecules can simultaneously exist in both crystalline and amorphous structures. A higher crystallinity indicates that a crystalline region in which more polymer molecules are arranged in an orderly manner is formed in the fiber. The image may also be divided into multiple regions, and the crystallinity may be calculated for each region. By analyzing the crystallinity distribution in different regions, a frequency distribution diagram or histogram of the crystallinity can be obtained, thereby obtaining crystal distribution information.

[0055] The degree of orientation refers to the degree of order in the direction of the polymer chains in the fiber. The degree of orientation is obtained by performing statistics on the alignment direction of the polymer chains. For polyester fibers, the degree of orientation is related to the fiber's stretchability, mechanical properties, thermal properties, transparency, etc., and has a certain impact on the coloring effect of the fiber.

[0056] Furthermore, the degree of crystallinity and orientation are affected by processes such as spinning, drawing and solidification. During spinning, the degree of orientation of the fiber can be affected by factors such as spinning speed, spinning tension, spinning temperature, etc. The temperature and speed during the drawing and solidification steps can also affect the ordered arrangement of the polymer chains within the fiber.

[0057] According to the method of the embodiment of the present disclosure, crystallinity and orientation information can be extracted from micro image data, and then used in conjunction with spectral data to improve the accuracy of prediction results.

[0058] In one possible implementation, determining target spectral features based on the first Raman data using a multimodal autoencoder in step S1102 further includes:

[0059] A multimodal autoencoder of the staining effect prediction model is used to determine target spectral features based on at least one of Raman peak intensity information, peak position information, peak area information, or peak shape information included in the first Raman data.

[0060] In the embodiments of the present disclosure, Raman spectroscopy can be used to analyze the degree of crystallinity in fiber materials. Molecular vibrational modes in crystalline and non-crystalline regions generate different Raman scattering signals. By measuring the Raman spectrum of a sample and analyzing the intensity or area of ​​certain Raman peaks, an index of crystallinity, such as the percentage of crystalline regions or the crystallinity index, can be calculated.

[0061] Raman spectroscopy can also be used to analyze the degree of orientation in fiber materials. The degree of orientation of fiber molecular chains can affect their Raman spectrum characteristics. By measuring the Raman spectrum of a sample and analyzing information such as specific Raman peak intensity, peak position, peak shape, etc., the orientation information of fiber molecular chains can be obtained.

[0062] According to the method of the embodiment of the present disclosure, crystallinity and orientation information can be extracted from Raman spectrum data, which can be used in combination with micro-image data to improve the accuracy of prediction results.

[0063] In one possible implementation form, in step S1103, determining the measured wound yarn package features based on the target image features and the target spectral features using a feature fusion layer included in the dyeing effect prediction model further includes: The target image features are input to the first branch of a feature fusion layer included in the dyeing effect prediction model. The target spectral features are input into the second branch of the feature fusion layer. An attention module of the feature fusion layer is utilized to determine a first weight corresponding to the first branch and a second weight corresponding to the second branch. A measured wound yarn package feature is determined based on the target image feature, the first weight, the target spectral feature and the second weight.

[0064] In the embodiment of the present disclosure, as shown in Figure 3, an attention mechanism is used in the feature fusion layer to realize cross-modal feature fusion. The attention mechanism can automatically learn the feature importance of each modality and perform weighted fusion on features of different modalities according to the importance. By introducing the attention mechanism, the model can more flexibly combine features of different spectral data, improving the effect of feature fusion.

[0065] The feature fusion layer may be composed of multiple branches, each branch is used to process features of different modalities. In one example, the first branch is used to process image data, and the second branch is used to process spectrum data. In each branch, the attention module performs importance evaluation and weighting on the features. Then, the weighted features are fused to form a final multi-modal feature representation. The design of the feature fusion layer can be flexibly adjusted based on the actual problem and the features of the dataset to obtain a better fusion effect.

[0066] The subsequent network structure in the dyeing effect prediction model receives the fusion feature output to the feature fusion layer, and the network may be a fully connected neural network (DNN), a recurrent neural network (RNN) or other network structure that can be applied to prediction tasks. The output of the prediction model is a dyeing label corresponding to the dyeing grade.

[0067] As described above, according to the solution of the embodiment of the present disclosure, through the feature extraction network and feature fusion layer, the multi-modal data fusion dyeing effect prediction model can more efficiently and accurately utilize the information of Raman spectrum data and image data, thereby obtaining stronger prediction ability.

[0068] 2 is a flowchart of a method for training a dyeing effect prediction model provided by an embodiment of the present disclosure. As shown in FIG. 2, the method includes at least:

[0069] In S201, the Raman spectrum data of the sample wound yarn package is decomposed into a plurality of sub-signal data using wavelet basis functions.

[0070] In S202, a sample wound yarn package characteristic is determined based on at least a portion of the plurality of sub-signal data.

[0071] In S203, the first prediction model is utilized to predict a second dyed label based on the sample wound yarn package characteristics.

[0072] In S204, a loss function is obtained based on the actual dyed label and the second dyed label of the sample wound yarn package.

[0073] In S205, the parameters of the first prediction model are updated based on the loss function to obtain a trained dyeing effect prediction model.

[0074] In an embodiment of the present disclosure, the first prediction model may be an initial prediction model, or may be a prediction model after the initial prediction model has been trained and updated several times (i.e., S201 to S205 have been performed several times).

[0075] In one example, a convolutional neural network (CNN) can be used as the initial prediction model, which may include three sets of convolutional layers and pooling layers to perform feature extraction on the input spectrum data. The convolutional layers can capture local features of the data, and the pooling layers are used to retain important features and reduce the dimensionality and computational complexity of the data. A fully connected layer is added after the convolutional layers to further extract and learn global features. Finally, the output encoded feature representation is used for subsequent feature fusion and prediction tasks.

[0076] During model training, labeled sample data must be used. Based on samples with known staining grades, each sample is labeled with its corresponding actual staining label, which becomes the target variable of the model. The labeled sample data is used to train the model, and the hyperparameters of the model are adjusted to obtain optimal performance.

[0077] The second dyeing label is used to represent the grade of the dyeing effect, and the grade of the dyeing effect corresponding to the second dyeing label may adopt a standard that divides it finer than the actual dyeing label. In the related art, when judging the color with the human naked eye, it can only be divided into five grades, and among the five grades, one stage (five grades, nine stages) is divided into each half grade. Due to the limited discrimination ability of the naked eye, it is difficult to further divide the actual label into finer grades. Therefore, when calculating the loss function, a loss function such as Focal Loss, which is suitable for fine-grained classification, can be used, and the discrimination ability for fine-grained grades can be improved. At the same time, through the multimodal data fusion of Raman spectrum data and image data, the multimodal data fusion can improve the understanding and discrimination ability of the model for fine-grained grades. As a result, the second dyeing label can correspond to more grades of dyeing effect, for example, 15 grades, thereby improving the prediction accuracy of the dyeing effect prediction model.

[0078] In the model training stage, the established model can be evaluated using methods such as cross-validation to measure prediction accuracy and generalization ability. The performance of the model can be evaluated using evaluation metrics such as precision, accuracy, recall rate, F1 score, etc.

[0079] For the method and examples of using wavelet basis functions to perform wavelet transformation, please refer to the relevant description of the corresponding steps in the embodiment of the dyeing effect prediction method above, and they will not be repeated here.

[0080] In one possible implementation, in step S202, determining the sample wound yarn package characteristic based on at least some of the sub-signal data of the plurality of sub-signal data further includes: At least some of the sub-signal data are selected from the plurality of sub-signal data based on the predetermined scale or the predetermined frequency to obtain second Raman data, where the plurality of sub-signal data correspond to wavelet coefficients of the plurality of different scales or spectral components of the plurality of different frequencies. Based on the second Raman data, a sample wound yarn package characteristic is determined.

[0081] In addition, when performing staining prediction, the prediction model may need to be slightly modified due to differences in the characteristics of crystalline and amorphous structures, as follows:

[0082] Regarding the selection of the feature extraction method, since crystalline and amorphous fibers have different molecular arrangement manners and characteristics, the prediction model may need to select an appropriate feature extraction method according to different structure types. For example, for crystalline fibers, a feature extraction method that emphasizes the periodic arrangement of molecules can be selected, and for amorphous fibers, a feature extraction method that emphasizes the disorder of molecules can be selected.

[0083] Regarding data pre-processing, different data pre-processing steps may need to be performed for fibers with different structures, for example for crystalline fibers it may be necessary to perform crystallinity calculations, whereas for amorphous fibers consideration may need to be given to how to handle the disorder information.

[0084] Regarding tuning of model structure, the structure of the predictive model may need to be adjusted according to fibers with different structures, for example, different sub-models can be designed for crystalline and amorphous fibers and then merged, or an attention mechanism can be used to dynamically adjust the attentional degree of the model to different structures.

[0085] Regarding data enhancement, in order to improve the robustness and generalization ability of the model, data can be enhanced in the training stage, so that the model can better adapt to fibers with different structures. For example, the data can be transformed, such as rotated, flipped, scaled, etc., so that the model can better learn the features under different structures.

[0086] In one possible implementation, in step S2021, determining the sample wound yarn package characteristics based on the second Raman data includes: A sample wound yarn package characteristic is determined based on the micro-image data of the sample wound yarn package and the second Raman data.

[0087] In one possible implementation, in step S210, determining the sample wound yarn package characteristics based on the micro-image data and the second Raman data of the sample wound yarn package further includes: In S2101, a multimodal autoencoder included in a first prediction model is used to determine sample image features based on micro image data of a sample wound yarn package. In S2102, a multimodal autoencoder is used to determine sample spectral features based on the second Raman data. In S2103, a feature fusion layer included in the first predictive model is utilized to determine sample wound yarn package features based on the sample image features and the sample spectral features.

[0088] In one possible implementation form, in step S2101, determining sample image features based on micro image data of the sample wound yarn package using a multimodal autoencoder included in the first predictive model further includes: Based on the micro-image data of the sample wound yarn package, structural parameters of the crystalline structure are determined, where the structural parameters include the degree of crystallinity and / or the degree of orientation. A multimodal autoencoder included in the first prediction model is used to determine sample image features based on the micro-image data and the structure parameters.

[0089] In the embodiment of the present disclosure, when the sample wound yarn package is a polymer fiber with a crystalline structure, the characteristic information can be based on the crystallinity, distribution of crystals, degree of orientation, etc. in the micro image data. On the other hand, when the sample wound yarn package is a polymer fiber with an amorphous structure, the characteristic information can be based on the proportion of amorphous regions, distribution of amorphous regions, and degree of disorder of crystals in the micro image data.

[0090] In one possible implementation, determining sample spectral characteristics based on the second Raman data using a multimodal autoencoder in step S2102 further includes: A multimodal autoencoder is utilized to determine sample wound yarn package characteristics based on at least one of Raman peak intensity information, peak position information, peak area information, or peak shape information contained in the second Raman data.

[0091] In one possible implementation form, in step S2103, determining a sample wound yarn package characteristic based on the sample image characteristic and the sample spectral characteristic using a feature fusion layer included in the first predictive model further includes: The sample image features are input to a first branch of a feature fusion layer included in a first prediction model. The sample spectral features are input into the second branch of the feature fusion layer. An attention module of the feature fusion layer is utilized to determine a first weight corresponding to the first branch and a second weight corresponding to the second branch. A sample wound yarn package characteristic is determined based on the sample image characteristic, the first weight, the sample spectral characteristic and the second weight.

[0092] In addition, in the training stage, the dyeing effect prediction model uses multi-modal data to pre-train the model, and saves the weight values ​​for the second Raman data and the micro image data in the feature extraction network, and the weight values ​​for the multiple sub-signal data. Then, in the inference stage, the Raman spectrum data is simply input into the feature extraction network to obtain feature representations, and then these features are used for dyeing effect prediction. In this way, the feature extraction network can be pre-trained using multi-modal data, and by extracting rich feature representations, the model can better learn useful information in the Raman spectrum data with the assistance of micro image data, so as to obtain better prediction results in the inference stage. In this way, only a small amount of micro image data can be used in the model training stage, and in the inference prediction stage, only spectrum data is used without using micro image data, thereby realizing online detection, greatly reducing utilization costs, and improving industrial application capabilities.

[0093] 4 is a dyeing effect prediction device provided by an embodiment of the present disclosure. As shown in FIG. 4, the device includes: The first transformation module 401 is used to decompose the Raman spectrum data of the measured wound yarn package into a number of sub-signal data by using wavelet basis functions. The first determination module 402 is used to determine a measured wound yarn package characteristic based on at least a portion of the target sub-signal data of the plurality of sub-signal data. The first prediction module 403 is used to predict and obtain a first dyeing label based on the measured yarn package characteristics using the dyeing effect prediction model, where the first dyeing label is used to represent a grade of the dyeing effect.

[0094] In one possible implementation, the first determination module 402 comprises: The selection sub-module is used for selecting at least some target sub-signal data from the plurality of sub-signal data according to a preset scale or a preset frequency to obtain first Raman data, where the plurality of sub-signal data correspond to wavelet coefficients of different scales or spectral components of different frequencies. The characteristic determination sub-module is used to determine a characteristic of the wound yarn package to be measured based on the first Raman data. In one possible implementation, the characteristic determination submodule is used to determine a characteristic of the wound yarn package to be measured based on the micro image data of the wound yarn package to be measured and the first Raman data.

[0095] In one possible implementation, the feature determination sub-module: determining a target image feature based on the micro image data of the measured wound yarn package using a multimodal autoencoder included in a dyeing effect prediction model; determining target spectral features based on the first Raman data using the multimodal autoencoder; and and determining a measurement target wound yarn package feature based on the target image feature and the target spectral feature by using a feature fusion layer included in the dyeing effect prediction model.

[0096] In one possible implementation, the feature determination sub-module: determining structural parameters of a crystal structure based on micro image data of the measured wound yarn package, the structural parameters including a degree of crystallinity and / or a degree of orientation; A multimodal autoencoder included in the dyeing effect prediction model is used to determine target image features based on the micro image data and the structure parameters.

[0097] In one possible implementation, the feature determination sub-module: A multimodal autoencoder of the dyeing effect prediction model is used to determine target spectral features based on at least one of Raman peak intensity information, peak position information, peak area information, or peak shape information contained in the first Raman data.

[0098] In one possible implementation, the feature determination sub-module: Inputting the target image features into a first branch of a feature fusion layer included in the dyeing effect prediction model; inputting the target spectral features into a second branch of the feature fusion layer; determining a first weight corresponding to the first branch and a second weight corresponding to the second branch using an attention module of the feature fusion layer; determining a measured wound yarn package feature based on the target image feature, the first weight, the target spectral feature, and the second weight.

[0099] The specific functions and exemplary descriptions of each module and sub-module of the apparatus in the embodiments of the present disclosure may refer to the relevant descriptions of the corresponding steps in the embodiments of the above method, and will not be repeated here.

[0100] 5 is a training device for a dyeing effect prediction model provided by an embodiment of the present disclosure. As shown in FIG. 5, the device includes: The second transformation module 501 is used to decompose the Raman spectrum data of the sample wound yarn package into multiple sub-signal data using wavelet basis functions. The second determination module 502 is used to determine a sample wound yarn package characteristic based on at least some of the sub-signal data of the plurality of sub-signal data. The second prediction module 503 is used to predict and obtain a second dyed label based on the sample wound yarn package characteristics using the first prediction model. The loss determination module 504 is used to obtain a loss function based on the actual dye label and the second dye label of the sample wound yarn package. The update module 505 is used to update the parameters of the first prediction model based on the loss function to obtain a trained dyeing effect prediction model.

[0101] In one possible implementation, the second determination module 502 comprises: The second selecting sub-module is used for selecting at least some sub-signal data from the plurality of sub-signal data according to a preset scale or a preset frequency to obtain second Raman data, where the plurality of sub-signal data correspond to wavelet coefficients of different scales or spectral components of different frequencies. A second characteristic determination sub-module is used to determine a sample wound yarn package characteristic based on the second Raman data.

[0102] In one possible implementation, the second feature determination submodule: Based on the micro-image data of the sample wound yarn package and the second Raman data, a sample wound yarn package characteristic is determined.

[0103] In one possible implementation, the second feature determination submodule: determining sample image features based on micro image data of the sample wound yarn package using a multimodal autoencoder included in a first predictive model; determining sample spectral characteristics based on the second Raman data using the multimodal autoencoder; and and determining a sample wound yarn package feature based on the sample image feature and the sample spectral feature using a feature fusion layer included in the first predictive model.

[0104] In one possible implementation, the second feature determination submodule: determining structural parameters of a crystalline structure based on micro image data of the sample wound yarn package, the structural parameters including a degree of crystallinity and / or a degree of orientation; and determining sample image features based on the micro-image data and the structure parameters using a multi-modal autoencoder included in the first prediction model.

[0105] In one possible implementation, the second feature determination submodule: The multi-modal autoencoder is used to determine sample wound yarn package characteristics based on at least one of Raman peak intensity information, peak position information, peak area information, or peak shape information contained in the second Raman data.

[0106] In one possible implementation, the second feature determination submodule: inputting the sample image features into a first branch of a feature fusion layer included in the first prediction model; inputting the sample spectral features into a second branch of the feature fusion layer; determining a first weight corresponding to the first branch and a second weight corresponding to the second branch using an attention module of the feature fusion layer; determining a sample wound yarn package characteristic based on the sample image feature, the first weight, the sample spectral feature, and the second weight.

[0107] The specific functions and exemplary descriptions of each module and sub-module of the apparatus in the embodiments of the present disclosure may refer to the relevant descriptions of the corresponding steps in the embodiments of the above method, and will not be repeated here.

[0108] FIG. 6 is a structural block diagram of an electronic device according to an embodiment of the present disclosure. As shown in FIG. 6, the electronic device includes a memory 610 and a processor 620, and the memory 610 stores a computer program executable by the processor 620. The number of the memory 610 and the processor 620 may be one or more. The memory 610 may store one or more computer programs, which, when executed by the electronic device, cause the electronic device to perform the method provided by the above method embodiment. The electronic device may further include: The communication interface 630 is used for communicating with an external device and performing data interaction and transmission.

[0109] When the memory 610, the processor 620, and the communication interface 630 are implemented independently, the memory 610, the processor 620, and the communication interface 630 are connected to each other via a bus and can communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The bus can be classified into an address bus, a data bus, a control bus, and the like. For ease of explanation, only one thick line is shown in FIG. 6, but this does not indicate only one bus or only one type of bus.

[0110] Optionally, in a specific implementation, when the memory 610, the processor 620, and the communication interface 630 are integrated on one chip, the memory 610, the processor 620, and the communication interface 630 can communicate with each other via an internal interface.

[0111] It should be understood that the processor may be a Central Processing Unit (CPU) or may be other general purpose processors, Digital Signal Processing (DSP), Application Specific Integrated Circuits (ASIC), Field Programmable Gate Arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general purpose processor may be a microprocessor or any conventional processor, etc. Additionally, the processor may be a processor supporting the Advanced RISC Machines (ARM) architecture.

[0112] In addition, optionally, the memory may include read-only memory and random access memory, and may further include non-volatile random access memory. The memory may be either volatile or non-volatile memory, or may include both volatile and non-volatile memory. Here, non-volatile memory may include Read-Only Memory (ROM), Programmable ROM (PROM), Erasable PROM (EPROM), Electrically EPROM (EEPROM), or flash memory. Volatile memory may include Random Access Memory (RAM) that acts as an external cache. By way of example and not limitation, many forms of RAM are available. For example, static random access memory (Static RAM, SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (Enhanced SDRAM, ESDRAM), synchlink DRAM (SLDRAM), and direct RAMBUS RAM (DR RAM).

[0113] The above-mentioned embodiments may be implemented in whole or in part in software, hardware, firmware, or any combination thereof. When implemented in software, the whole or in part may be implemented in the form of a computer program product. A computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, a process or function according to the embodiments of the present disclosure is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another, for example, the computer instructions may be transmitted from one website site, computer, server, or data center to another website site, computer, server, or data center via wire (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, Bluetooth, microwave, etc.). The computer-readable storage medium may be any available medium that can be accessed by a computer, or a data storage device including a server, data center, etc. integrated with one or more available media. The usable medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a magnetic tape), an optical medium (e.g., a digital versatile disk (DVD)), or a semiconductor medium (e.g., a solid-state disk (SSD)). Note that the computer-readable storage medium referred to in this disclosure may be a non-volatile storage medium, in other words, a non-transitory storage medium.

[0114] Those skilled in the art can understand that all or part of the steps for realizing the above embodiments may be implemented by hardware, or may be implemented by instructing relevant hardware by a program, and the program may be stored in a computer-readable storage medium, and the storage medium may be a read-only memory, a magnetic disk, an optical disk, etc.

[0115] In describing the embodiments of the present disclosure, the reference words "one embodiment", "some embodiments", "example", "particular example", or "some examples" mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present disclosure. In addition, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, a person skilled in the art may combine different embodiments or examples described herein and features of different embodiments or examples to the extent that they are not mutually inconsistent.

[0116] In the description of the embodiments of the present disclosure, unless otherwise stated, " / " means "or", for example, A / B can mean either A or B. In this specification, "and / or" is merely for describing the related relationship of related objects, and indicates that three kinds of relationships may exist, for example, A and / or B can indicate the following three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0117] In the description of the embodiments of the present disclosure, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying a relative importance, nor should they be construed as implying the number of technical features depicted. Thus, features defined as "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the embodiments of the present disclosure, "plurality" means two or more, unless otherwise specified.

[0118] The above are merely illustrative examples of the present disclosure, and are not intended to limit the present disclosure. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present disclosure should be included within the scope of protection of the present disclosure.

Claims

1. Decomposing Raman spectrum data of the measured wound yarn package into a plurality of sub-signal data using wavelet basis functions; determining a measured wound yarn package characteristic based on at least a portion of the target sub-signal data of the plurality of sub-signal data; and predicting a first dyeing label representing a grade of dyeing effect based on the measured wound yarn package characteristics using a dyeing effect prediction model. A method for predicting dyeing results.

2. Determining a measured wound yarn package characteristic based on at least a portion of the target sub-signal data of the plurality of sub-signal data includes: Selecting at least some target sub-signal data from the plurality of sub-signal data corresponding to wavelet coefficients of a plurality of different scales or spectral components of a plurality of different frequencies based on a preset scale or a preset frequency to obtain first Raman data; and determining a measured wound yarn package characteristic based on the first Raman data. The method for predicting dyeing effect according to claim 1 .

3. determining a target wound yarn package characteristic based on the first Raman data, determining a characteristic of the measured wound yarn package based on the micro-image data of the measured wound yarn package and the first Raman data; The method for predicting dyeing effect according to claim 2.

4. determining a characteristic of the wound yarn package to be measured based on the micro image data of the wound yarn package to be measured and the first Raman data, determining a target image feature based on the micro image data of the measured wound yarn package using a multimodal autoencoder included in a dyeing effect prediction model; determining target spectral features based on the first Raman data using the multimodal autoencoder; and determining a measured wound yarn package feature based on the target image feature and the target spectral feature by using a feature fusion layer included in the dyeing effect prediction model; The method for predicting dyeing effect according to claim 3.

5. determining a target image feature based on micro image data of the measured wound yarn package using a multimodal autoencoder included in a dyeing effect prediction model; determining structural parameters of a crystal structure based on micro image data of the measured wound yarn package, the structural parameters including a degree of crystallinity and / or a degree of orientation; determining a target image feature based on the micro image data and the structure parameters by using a multimodal autoencoder included in a dyeing effect prediction model; The method for predicting dyeing effect according to claim 4.

6. Determining target spectral features based on the first Raman data utilizing the multimodal autoencoder includes: determining a target spectral feature based on at least one of Raman peak intensity information, peak position information, peak area information, or peak shape information included in the first Raman data using a multimodal autoencoder of a staining effect prediction model; The method for predicting dyeing effect according to claim 4.

7. determining a measurement target wound yarn package feature based on the target image feature and the target spectrum feature by utilizing a feature fusion layer included in the dyeing effect prediction model; Inputting the target image features into a first branch of a feature fusion layer included in the dyeing effect prediction model; inputting the target spectral features into a second branch of the feature fusion layer; determining a first weight corresponding to the first branch and a second weight corresponding to the second branch using an attention module of the feature fusion layer; determining a measured wound yarn package feature based on the target image feature, the first weight, the target spectral feature, and the second weight. The method for predicting dyeing effect according to claim 4.

8. Decomposing the Raman spectrum data of the sample wound yarn package into a plurality of sub-signal data using wavelet basis functions; determining a sample wound yarn package characteristic based on at least some of the sub-signal data of the plurality of sub-signal data; predicting a second dyed label based on the sample wound yarn package characteristics using a first predictive model; and Obtaining a loss function based on the actual dye label of the sample wound yarn package and the second dye label; and updating parameters of the first prediction model based on the loss function to obtain a trained dyeing effect prediction model. Method for training dyeing effect prediction model.

9. Determining a sample wound yarn package characteristic based on at least a portion of the sub-signal data of the plurality of sub-signal data includes: selecting at least some sub-signal data from the plurality of sub-signal data based on a preset scale or a preset frequency to obtain second Raman data, the plurality of sub-signal data corresponding to wavelet coefficients of a plurality of different scales or spectral components of a plurality of different frequencies; and determining the sample wound yarn package characteristics based on the second Raman data. The method for training a dyeing effect prediction model according to claim 8.

10. determining a sample wound yarn package characteristic based on the second Raman data; determining a sample wound yarn package characteristic based on the micro-image data of the sample wound yarn package and the second Raman data. The method for training a dyeing effect prediction model according to claim 9.

11. determining a sample wound yarn package characteristic based on the micro-image data of the sample wound yarn package and the second Raman data; determining sample image features based on micro image data of the sample wound yarn package using a multimodal autoencoder included in a first predictive model; determining sample spectral characteristics based on the second Raman data using the multimodal autoencoder; and and determining sample wound yarn package features based on the sample image features and the sample spectral features using a feature fusion layer included in the first predictive model. The method for training a dyeing effect prediction model according to claim 10.

12. determining sample image features based on micro image data of the sample wound yarn package using a multimodal autoencoder included in the first predictive model, determining structural parameters of a crystalline structure based on micro image data of the sample wound yarn package, the structural parameters including a degree of crystallinity and / or a degree of orientation; determining sample image features based on the micro-image data and the structure parameters using a multi-modal autoencoder included in a first predictive model; The method for training a dyeing effect prediction model according to claim 11.

13. Determining sample spectral characteristics based on the second Raman data utilizing the multimodal autoencoder includes: utilizing the multi-modal autoencoder to determine sample wound yarn package characteristics based on at least one of Raman peak intensity information, peak position information, peak area information, or peak shape information included in the second Raman data. The method for training a dyeing effect prediction model according to claim 11.

14. determining a sample wound yarn package feature based on the sample image feature and the sample spectral feature using a feature fusion layer included in the first predictive model, inputting the sample image features into a first branch of a feature fusion layer included in the first prediction model; inputting the sample spectral features into a second branch of the feature fusion layer; determining a first weight corresponding to the first branch and a second weight corresponding to the second branch using an attention module of the feature fusion layer; determining a sample wound yarn package characteristic based on the sample image characteristics, the first weights, the sample spectral characteristics, and the second weights. The method for training a dyeing effect prediction model according to claim 11.

15. a first transformation module for decomposing the Raman spectrum data of the measured wound yarn package using a wavelet basis function to obtain a plurality of sub-signal data; a first determination module for determining a target wound yarn package characteristic based on at least a portion of the target sub-signal data of the plurality of sub-signal data; a first prediction module for predicting and obtaining a first dyeing label for indicating a grade of dyeing effect based on the measured wound yarn package characteristics by using a dyeing effect prediction model; Dyeing effect prediction device.

16. a second transformation module for decomposing the Raman spectrum data of the sample wound yarn package into a plurality of sub-signal data using wavelet basis functions; a second determination module for determining a sample wound yarn package characteristic based on at least a portion of the plurality of sub-signal data; a second prediction module for predicting a second dyed label based on the sample wound yarn package characteristics using the first prediction model; and a loss determination module for obtaining a loss function based on the actual dye label of the sample wound yarn package and the second dye label; and an update module for updating parameters of the first prediction model based on the loss function to obtain a trained dyeing effect prediction model. Training device for dyeing effect prediction model.

17. At least one processor; a memory in communication with the at least one processor; 15. An electronic device, comprising: a memory configured to store instructions executable by the at least one processor, the instructions, when executed by the at least one processor, causing the at least one processor to perform a method according to any one of claims 1 to 14.

18. A non-transitory computer readable storage medium having stored thereon computer instructions for causing a computer to perform the method of any one of claims 1 to 14.

19. A program for implementing the dyeing effect prediction method according to any one of claims 1 to 14 when executed by a processor in a computer.

Images