Methods for testing the antioxidant properties of jute fiber

Abstract

This invention relates to the field of textile material testing technology and discloses a method for testing the antioxidant properties of jute fiber. The method includes acquiring the initial response signal of a jute fiber sample under specific wavelength light illumination and extracting spectral response fragments by comparing it with a standard database. Based on these fragments, a dynamic response trajectory is constructed, and a hierarchical feature stripping engine is used to separate a set of feature markers related to the oxidation process layer by layer. These feature markers are reorganized into a multidimensional feature matrix and matched with a pre-built jute fiber aging evolution knowledge graph to locate key evolution paths affecting antioxidant properties. Based on these paths, the performance degradation curves of the sample under different environmental parameters are simulated and calculated. Within an integrated performance extrapolation framework, the curves, paths, and spectral fragments are fused for iterative calculations, outputting an antioxidant performance stability index under specified conditions. This invention achieves analytical and predictive evaluation of the dynamic evolution process of antioxidant properties.

Description

Technical Field

[0001] This invention relates to the field of textile material testing technology, specifically to a method for testing the antioxidant properties of jute fiber. Background Technology

[0002] Currently, the evaluation of the antioxidant properties of jute fiber commonly employs endpoint titration based on chemical reactions or free radical scavenging rate determination based on absorbance measurements at specific wavelengths. These methods constitute the standard technical means in the industry. They indirectly estimate the total antioxidant capacity of the fiber sample by measuring the amount of reagent consumed or the color change of the solution after the reaction, essentially obtaining a static value characterizing the instantaneous state.

[0003] These conventional technical solutions have inherent limitations. Their detection processes cannot record and reflect the continuous dynamic response of antioxidant components under external light or oxidants. The results are endpoint values ​​that are a mixture of multiple components, failing to distinguish the order and interaction mechanisms of different antioxidants in the oxidation process. Existing methods can only answer how much antioxidant capacity is currently available, but cannot reveal how it decays or why it decays in this way. This results in a superficial assessment of material durability, lacking insight into the performance degradation trajectory, and failing to effectively predict the long-term stability of fibers under different usage environments. Developing a detection method capable of analyzing dynamic oxidation processes and predicting performance evolution has become a key requirement for improving the level of material evaluation. Summary of the Invention

[0004] The purpose of this invention is to provide a method for testing the antioxidant properties of jute fiber, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for testing the antioxidant properties of jute fiber, the method comprising:

[0006] The initial response signal of jute fiber sample under light of a specific wavelength is collected, and the initial response signal is synchronously compared with a preset standard response database to extract the spectral response fragments characterizing antioxidant components from the initial response signal.

[0007] Based on the spectral response fragment, a dynamic response trajectory is constructed, and the dynamic response trajectory is input into a hierarchical feature stripping engine to separate the set of feature markers related to the oxidation process layer by layer.

[0008] The feature tag set is reorganized and mapped in time series to generate a multidimensional feature matrix, and the multidimensional feature matrix is ​​matched with a pre-established jute fiber aging evolution knowledge graph.

[0009] Based on the matching results between the multidimensional feature matrix and the jute fiber aging evolution knowledge graph, the key evolution path affecting antioxidant performance is located. Based on the key evolution path, the performance degradation curves of jute fiber samples under different combinations of environmental parameters are simulated and calculated.

[0010] By integrating the performance degradation curve, the key evolution path, and the spectral response fragment, iterative calculations are performed within an integrated performance extrapolation framework to ultimately output the antioxidant stability index of jute fiber samples under specified detection conditions.

[0011] Preferably, based on the spectral response fragment, a dynamic response trajectory is constructed, and the dynamic response trajectory is input into a hierarchical feature stripping engine to separate a set of feature markers related to the oxidation process layer by layer, including:

[0012] Based on the time series data of the spectral response segment, a continuously changing dynamic response trajectory is generated using curve fitting technology;

[0013] The dynamic response trajectory is input into the first feature stripping layer of the hierarchical feature stripping engine. The first feature stripping layer filters out the basic waveform components in the dynamic response trajectory based on a predefined physicochemical feature filter.

[0014] The basic waveform components are passed to the second feature stripping layer of the hierarchical feature stripping engine. The second feature stripping layer uses a convolution kernel to slide and scan the basic waveform components to extract the local abrupt change features and periodic oscillation features hidden in the basic waveform components.

[0015] The local mutation features and the periodic oscillation features are jointly input into the feature integration layer of the hierarchical feature stripping engine. The feature integration layer performs cross-validation and association binding on the local mutation features and the periodic oscillation features to form a set of feature tags related to the oxidation process.

[0016] Preferably, the local mutation features and the periodic oscillation features are jointly input into the feature integration layer of the hierarchical feature stripping engine. The feature integration layer performs cross-validation and association binding on the local mutation features and the periodic oscillation features to form a set of feature tags related to the oxidation process, including:

[0017] In the feature integration layer, a shared feature space is established, and the local mutation features and the periodic oscillation features are projected into the shared feature space;

[0018] Within the shared feature space, the covariance tensor between the local mutation feature and the periodic oscillation feature is calculated, and the local mutation feature and the periodic oscillation feature are re-encoded based on the covariance tensor;

[0019] A bidirectional attention mechanism is applied to the re-encoded local mutation features and periodic oscillation features to calculate the attention weight of the local mutation features on the periodic oscillation features, and at the same time, the attention weight of the periodic oscillation features on the local mutation features is calculated.

[0020] Based on the attention weights calculated by the bidirectional attention mechanism, the re-encoded local mutation features and periodic oscillation features are weighted and fused to generate a fused feature vector.

[0021] Cluster analysis is performed on the fused feature vectors, and each feature cluster obtained after clustering is labeled as a set of feature labels related to the oxidation process.

[0022] Preferably, the step of reorganizing and mapping the feature tag set in a time series to generate a multidimensional feature matrix, and matching the multidimensional feature matrix with a pre-established jute fiber aging evolution knowledge graph, includes:

[0023] For each feature marker in the feature marker set, a sliding window sampling is performed along the time axis to obtain feature marker subsets under multiple time slices;

[0024] Each subset of feature labels under each time slice is classified and mapped onto the corresponding coordinate axis of a high-dimensional vector space according to its physicochemical properties.

[0025] All time slices are combined and mapped to the coordinates in the high-dimensional vector space to form a multidimensional feature matrix that expands in the time and attribute dimensions;

[0026] Extract all graph nodes representing different aging stages from the knowledge graph of jute fiber aging evolution, and calculate the feature vector of each graph node;

[0027] The similarity between the multidimensional feature matrix and the feature vectors of all graph nodes extracted from the jute fiber aging evolution knowledge graph is calculated. Rows or columns in the multidimensional feature matrix with similarity exceeding a preset threshold are matched to the corresponding graph nodes in the jute fiber aging evolution knowledge graph.

[0028] Preferably, the step of calculating the similarity between the multidimensional feature matrix and the feature vectors of all graph nodes extracted from the jute fiber aging evolution knowledge graph, and matching rows or columns in the multidimensional feature matrix with similarity exceeding a preset threshold to the corresponding graph nodes in the jute fiber aging evolution knowledge graph, includes:

[0029] For each graph node in the jute fiber aging evolution knowledge graph, generate a graph node embedding vector based on its connected relation edges and attribute values;

[0030] Each row of the multidimensional feature matrix is ​​regarded as a feature vector, and the cosine similarity between each feature vector and the embedding vector of each graph node is calculated.

[0031] A similarity matrix is ​​established, wherein the rows of the similarity matrix correspond to the rows of the multidimensional feature matrix, the columns correspond to the graph nodes of the jute fiber aging evolution knowledge graph, and the matrix elements are the calculated cosine similarity values.

[0032] Traverse the similarity matrix and find the graph node with the largest cosine similarity value in each row. If the largest cosine similarity value exceeds the preset matching threshold, mark the current row of the multidimensional feature matrix as a successful match with the graph node in the jute fiber aging evolution knowledge graph.

[0033] All successfully matched rows and their corresponding graph node information in the jute fiber aging evolution knowledge graph are recorded as a matching result mapping table.

[0034] Preferably, the step of locating the key evolutionary path affecting antioxidant performance based on the matching result between the multidimensional feature matrix and the jute fiber aging evolution knowledge graph, and simulating and calculating the performance degradation curve of jute fiber samples under different combinations of environmental parameters based on the key evolutionary path, includes:

[0035] Parse the matching result mapping table to locate all graph nodes that successfully match the multidimensional feature matrix in the jute fiber aging evolution knowledge graph;

[0036] In the knowledge graph of jute fiber aging evolution, the shortest connected subgraph that connects all successfully matched graph nodes is found, and the directed edge sequence in the shortest connected subgraph constitutes the initial evolution path.

[0037] The initial evolution path is pruned and optimized by removing edges whose weights are lower than the average weight of the path, resulting in a simplified critical evolution path.

[0038] The key evolution path is transformed into a state transition model, where the states in the state transition model correspond to graph nodes in the key evolution path, and the transition probability is calculated based on the edge weights between graph nodes and the matching strength in the matching result mapping table.

[0039] Multiple sets of different environmental parameter combinations are set, and each set of environmental parameter combinations is used as an external input to drive the state transition model. The trajectory of jute fiber sample evolving from the initial state along the key evolution path is iteratively calculated. The vertical axis of the trajectory represents the performance index, thereby generating a performance degradation curve corresponding to each set of environmental parameter combinations.

[0040] Preferably, the initial response signal of the jute fiber sample under specific wavelength light illumination is collected, and the initial response signal is synchronously compared with a preset standard response database to extract the spectral response fragments characterizing antioxidant components from the initial response signal, including:

[0041] Using a spectral detection device, jute fiber samples are irradiated with excitation light of a specific wavelength, and light signals reflected or transmitted from the jute fiber samples are continuously collected. The light signals are converted into an electrical signal sequence as an initial response signal. Standard response signal templates of the same type of jute fiber under known good antioxidant properties are retrieved from a preset standard response database.

[0042] The initial response signal is time-aligned with the standard response signal template, and the signal intensity difference sequence between the initial response signal and the standard response signal template at each sampling time point is calculated. A sliding window peak detection is applied to the signal intensity difference sequence to identify continuous time intervals where the difference exceeds a dynamic threshold.

[0043] From the initial response signal, the original signal segment corresponding to the continuous time interval is extracted, and the original signal segment is marked as a spectral response segment characterizing the antioxidant component.

[0044] Preferably, the re-encoded local mutation features and periodic oscillation features are subjected to a bidirectional attention mechanism to calculate the attention weight of the local mutation features on the periodic oscillation features, and simultaneously calculate the attention weight of the periodic oscillation features on the local mutation features, including:

[0045] The re-encoded local mutation feature matrix and the re-encoded periodic oscillation feature matrix are respectively input into the attention calculation module;

[0046] In the attention calculation module, the local mutation feature matrix is ​​used as the query vector, and the periodic oscillation feature matrix is ​​used as the key vector and value vector. The similarity score between the query vector and the key vector is calculated by the dot product operation. The softmax function is applied to normalize the similarity score to obtain the attention weight distribution of the local mutation feature to the periodic oscillation feature.

[0047] Meanwhile, using the periodic oscillation feature matrix as the query vector and the local mutation feature matrix as the key vector and value vector, the attention weight distribution of the periodic oscillation feature to the local mutation feature is calculated through the same dot product operation and softmax normalization.

[0048] The attention weight distributions are weighted and summed, with the weight coefficients adaptively adjusted according to the feature dimensions, to generate the fused attention weight matrix.

[0049] Preferably, the step of transforming the critical evolution path into a state transition model, wherein the states in the state transition model correspond to graph nodes in the critical evolution path, and the transition probability is calculated based on the edge weights between graph nodes and the matching strength in the matching result mapping table, includes:

[0050] The weight values ​​of the connecting edges between each graph node on the key evolution path are extracted from the jute fiber aging evolution knowledge graph. The weight values ​​represent the ease or difficulty of transitioning between aging stages, and the higher the weight value, the easier the transition.

[0051] Extract the matching strength corresponding to each successfully matched graph node from the matching result mapping table. The matching strength is quantified by the cosine similarity value in the similarity calculation.

[0052] For each pair of adjacent states in the state transition model, the transition probability is calculated as the product of the edge weight and the matching strength. Then, the product of all possible transitions is normalized so that the sum of the out-transition probabilities of each state is one.

[0053] The calculated transition probabilities are filled into the state transition matrix, where the rows and columns of the state transition matrix correspond to the graph nodes on the key evolution path, thus completing the parameterization construction of the state transition model.

[0054] Preferably, the step of applying sliding window peak detection to the signal intensity difference sequence to identify continuous time intervals where the difference exceeds a dynamic threshold includes:

[0055] A sliding window of fixed length is set, the length of which is preset according to the sampling rate of the signal strength difference sequence;

[0056] The sliding window is gradually slid along the time axis of the signal strength difference sequence, and for each window position, the average value and standard deviation of the signal strength difference within the window are calculated;

[0057] A dynamic threshold is set based on the mean and standard deviation. The dynamic threshold is equal to the mean plus a multiple of the standard deviation, wherein the multiple factor is calibrated experimentally. Within each window, points where the signal strength difference exceeds the dynamic threshold are identified, and the location index of the points is recorded.

[0058] Points exceeding the dynamic threshold identified in adjacent windows are merged to form continuous time intervals, while continuous time intervals with a length less than the minimum duration threshold are filtered out to ensure that only significant spectral response segments are retained.

[0059] Compared with the prior art, the beneficial effects of the present invention are:

[0060] By constructing dynamic response trajectories and applying a hierarchical feature stripping engine, the composite spectral signals generated under light excitation can be decomposed step-by-step according to the time dimension and signal characteristics. This allows for the stripping of feature marker sets corresponding to different oxidation stages, such as rapid initial consumption, synergistic effects in the middle stage, and slow decay in the later stage. This transforms the analysis of the antioxidant properties of jute fibers from a simple measurement of total capacity to a detailed deconstruction of its complex time-varying behavior, enabling direct observation and feature extraction of oxidation reaction kinetics.

[0061] By matching the multidimensional feature matrix constructed from experimental data with a knowledge graph of jute fiber aging evolution, the specific chemical reaction pathways or structural change pathways that dominate the current sample's performance degradation can be identified. Simulations based on these pathways can deduce the curves of fiber antioxidant properties changing over time under different temperature, humidity, or light intensity conditions. Deeply integrating single-test experiments with material failure mechanism models allows the evaluation conclusions to include predictive information about future performance evolution, outputting a performance stability index correlated with specific environmental stresses, rather than an isolated capability value. Attached Figure Description

[0062] Figure 1 This is a schematic diagram illustrating the working principle of the method for detecting the antioxidant properties of jute fiber described in this invention.

[0063] Figure 2 A flowchart for constructing dynamic response trajectories and feature stripping;

[0064] Figure 3 A flowchart for generating feature integration and feature label sets;

[0065] Figure 4 Heatmap of the oxidation state transition probability matrix of jute fiber;

[0066] Figure 5 A comparison of local mutations and periodic oscillations in the peeling engine for detecting antioxidant characteristics of jute fiber. Detailed Implementation

[0067] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0068] Please see Figure 1 This invention provides a method for detecting the antioxidant properties of jute fiber. The method includes: First, acquiring the initial response signal of a jute fiber sample under specific wavelength light illumination using a spectral detection device, and synchronously comparing the initial response signal with a preset standard response database to extract spectral response segments that characterize antioxidant components from the initial response signal. Then, constructing a dynamic response trajectory based on the extracted spectral response segments, and inputting the dynamic response trajectory into a hierarchical feature stripping engine, which separates a set of feature markers related to the oxidation process layer by layer. Next, reorganizing and mapping the obtained feature marker set in a time series to generate a multidimensional feature matrix, and matching this multidimensional feature matrix with a pre-established jute fiber aging evolution knowledge graph. Then, based on the matching result between the multidimensional feature matrix and the jute fiber aging evolution knowledge graph, locating the key evolution path affecting antioxidant properties in the knowledge graph, and simulating the performance degradation curve of the jute fiber sample under different combinations of environmental parameters based on this key evolution path. Finally, by integrating performance degradation curves, key evolution paths, and spectral response fragments, iterative calculations are performed within an integrated performance extrapolation framework to output the antioxidant stability index of jute fiber samples under specified detection conditions.

[0069] Example 1: See Figure 2 After acquiring the spectral response fragment, a continuously changing dynamic response trajectory is generated based on the time-series data of this spectral response fragment using curve fitting technology. The generated dynamic response trajectory is input to the first feature stripping layer of a hierarchical feature stripping engine. This first feature stripping layer filters out the basic waveform components in the dynamic response trajectory according to predefined physicochemical feature filters. Next, the obtained basic waveform components are passed to the second feature stripping layer of the hierarchical feature stripping engine. This second feature stripping layer uses a convolutional kernel to slide and scan the basic waveform components, extracting the implicit local abrupt change features and periodic oscillation features. Then, the extracted local abrupt change features and periodic oscillation features are jointly input to the feature integration layer of the hierarchical feature stripping engine. This feature integration layer performs cross-validation and association binding on the local abrupt change features and periodic oscillation features, forming a set of feature labels related to the oxidation process.

[0070] In practice, the acquired spectral response fragments are presented as a set of time-series data. For example, under 650 nm wavelength illumination, the spectral response fragments include data from the time stamp. arrive The light intensity response value sequence of a total of n+1 sampling points constitutes the original time series data. Based on this time series data, a continuously changing dynamic response trajectory is generated through curve fitting technology. One implementation method is to use a polynomial fitting algorithm to fit the discrete sampling points into a smooth continuous curve. The formula for polynomial fitting can be expressed as:

[0071] ;

[0072] in: This represents the fitted response value at time t. Denotes the coefficient of the i-th term in a polynomial. Represents the highest degree of a polynomial, with coefficients... The continuous curve determined by the least squares method is the dynamic response trajectory. In some embodiments, spline interpolation or Gaussian process regression can also be used to generate the dynamic response trajectory, with the goal of obtaining a trajectory that can characterize the continuous change of the response signal.

[0073] It is understood that the generated dynamic response trajectory will be input into a hierarchical feature stripping engine. The first feature stripping layer of the hierarchical feature stripping engine processes the dynamic response trajectory according to a predefined physicochemical feature filter. The physicochemical feature filter can be a set of bandpass filters, whose passband range is set according to the characteristic absorption or reflection frequency bands of known antioxidant components in jute fiber. In specific implementation, the first feature stripping layer applies this set of filters to filter the dynamic response trajectory, screening and separating waveform components that conform to the preset passband range. These screened waveform components are defined as basic waveform components, which correspond to the responses generated by the vibration or electronic transitions of certain types of specific chemical bonds.

[0074] Next, the basic waveform components are passed to the second feature stripping layer of the hierarchical feature stripping engine. This second layer uses a convolution kernel to slide and scan the basic waveform components. Optionally, the convolution kernel can be a one-dimensional sequence of difference operators, such as [-1, 0, 1]. The kernel slides across the data points of the basic waveform components with a fixed stride, calculating the dot product between the kernel and the local data window. This operation highlights rapidly changing regions in the waveform. During the sliding scan, local locations where the absolute value of the convolution result exceeds a preset threshold are recorded; the features corresponding to these locations are extracted as local abrupt change features. Simultaneously, Fourier transform or wavelet transform is performed on the basic waveform components to analyze their frequency domain components, extracting oscillation modes with energy concentrated in specific frequency ranges as periodic oscillation features. In some embodiments, the extraction of periodic oscillation features can be accomplished through autocorrelation function analysis, identifying periods with significant peaks in the autocorrelation function.

[0075] Subsequently, the extracted local mutation features and periodic oscillation features are jointly input into the feature integration layer of the hierarchical feature stripping engine. In the feature integration layer, the local mutation features and periodic oscillation features are cross-validated and correlated. Cross-validation can be achieved by calculating the statistical correlation between the occurrence time of the local mutation feature and the phase of the periodic oscillation feature; for example, calculating the probability of a local mutation feature occurring near the peak or trough of a periodic oscillation. Correlation binding pairs local mutation features with significant statistical correlations with periodic oscillation features to form a combined feature unit. Finally, all combined feature units formed through cross-validation and correlation binding, along with unbound independent salient features, constitute a feature tag set related to the oxidation process. Each tag in the feature tag set corresponds to an identifier describing a specific morphology or pattern of the signal.

[0076] Example 2: See Figure 3In the feature integration layer, a shared feature space is first established, and local mutation features and periodic oscillation features are projected into this shared feature space. Within the shared feature space, the covariance tensor between the local mutation features and the periodic oscillation features is calculated, and the local mutation features and periodic oscillation features are re-encoded based on this covariance tensor. A bidirectional attention mechanism is applied to the re-encoded local mutation feature matrix and periodic oscillation feature matrix. Specifically, the re-encoded local mutation feature matrix and the re-encoded periodic oscillation feature matrix are respectively input into the attention calculation module. In the attention calculation module, using the local mutation feature matrix as the query vector and the periodic oscillation feature matrix as the key vector and value vector, the similarity score between the query vector and the key vector is calculated through a dot product operation, and the similarity score is normalized using a softmax function to obtain the attention weight distribution of the local mutation features on the periodic oscillation features. Simultaneously, using the periodic oscillation feature matrix as the query vector and the local mutation feature matrix as the key vector and value vector, the attention weight distribution of the periodic oscillation features on the local mutation features is calculated through the same dot product operation and softmax normalization. The attention weight distributions are weighted and summed, with the weight coefficients adaptively adjusted according to the feature dimensions, to generate a fused attention weight matrix. Based on the attention weights calculated using the bidirectional attention mechanism, the re-encoded local mutation features and periodic oscillation features are weighted and fused to generate a fused feature vector. Finally, cluster analysis is performed on this fused feature vector, and each feature cluster is labeled as a set of feature labels related to the oxidation process.

[0077] In practical implementation, the feature integration layer establishes a shared feature space, projecting local mutation features and periodic oscillation features into this shared feature space. The projection operation can be achieved through a linear transformation: the local mutation feature matrix is ​​multiplied by a projection matrix to obtain its representation in the shared space, and the periodic oscillation feature matrix is ​​multiplied by another projection matrix to obtain its representation in the shared space. Within the shared feature space, the covariance tensor between the local mutation features and the periodic oscillation features is calculated. The covariance tensor reflects the collaborative change relationship between the local mutation features and the periodic oscillation features across different dimensions. Based on the covariance tensor, the local mutation features and the periodic oscillation features are re-encoded. This re-encoding can be accomplished by aligning the feature vectors with the principal component directions of the covariance tensor.

[0078] In some embodiments, a bidirectional attention mechanism is applied to the re-encoded local mutation features and periodic oscillation features to calculate the attention weights of the local mutation features on the periodic oscillation features, and simultaneously calculate the attention weights of the periodic oscillation features on the local mutation features. The re-encoded local mutation feature matrix and the re-encoded periodic oscillation feature matrix are input into the attention calculation module. In the attention calculation module, using the local mutation feature matrix as the query vector and the periodic oscillation feature matrix as the key and value vectors, a similarity score between the query vector and the key vector is calculated through a dot product operation. The softmax function is then applied to normalize the similarity score, yielding the attention weight distribution of the local mutation features on the periodic oscillation features. Simultaneously, using the periodic oscillation feature matrix as the query vector and the local mutation feature matrix as the key and value vectors, the same dot product operation and softmax normalization are used to calculate the attention weight distribution of the periodic oscillation features on the local mutation features. The formula for calculating the attention weight distribution can be expressed as:

[0079] ;

[0080] in: The expression represents the attention weight of the u-th query vector to the v-th key vector, with the symbol... This represents the natural exponential function. It is the vector in the u-th row of the query vector matrix. It is the v-th row vector of the key vector matrix. is the dimension of the vector, and the dot product operation represents the vector inner product. The attention weight distribution is weighted and summed, where the weight coefficients are adaptively adjusted according to the feature dimension, to generate the fused attention weight matrix.

[0081] It is understandable that the attention weights calculated based on the bidirectional attention mechanism are used to weight and fuse the re-encoded local mutation features and periodic oscillation features to generate a fused feature vector. Weighted fusion can be achieved by multiplying the attention weight matrix with the value vector. For the attention weight distribution of local mutation features on periodic oscillation features, multiplying the weight matrix with the periodic oscillation feature matrix yields the fused representation from the perspective of local mutation features. For the attention weight distribution of periodic oscillation features on local mutation features, multiplying the weight matrix with the local mutation feature matrix yields the fused representation from the perspective of periodic oscillation features. Then, the fused representations from the two perspectives are concatenated or averaged to generate the final fused feature vector. Optionally, cluster analysis is performed on the fused feature vector, and each feature cluster obtained after clustering is labeled as a set of feature labels related to the oxidation process. The cluster analysis can use the K-means clustering algorithm, dividing the fused feature vector into K clusters. The center of each cluster represents a feature pattern, each feature vector is assigned to the nearest cluster, and each cluster is assigned a unique identifier as a feature label.

[0082] Example 3: When generating the multidimensional feature matrix and performing node matching, for each feature label in the feature label set, a sliding window sampling is performed along the time axis to obtain feature label subsets under multiple time slices. Each feature label subset under each time slice is classified and mapped onto the corresponding coordinate axis of a high-dimensional vector space according to its physicochemical properties. The coordinate points of all time slices mapped to the high-dimensional vector space are combined to form a multidimensional feature matrix expanded in the time and attribute dimensions. All graph nodes representing different aging stages are extracted from the jute fiber aging evolution knowledge graph, and the feature vector of each graph node is calculated. For matching, for each graph node in the jute fiber aging evolution knowledge graph, a graph node embedding vector is generated based on its connected relation edges and attribute values. Each row of the multidimensional feature matrix is ​​considered as a feature vector, and the cosine similarity between each feature vector and each graph node embedding vector is calculated. A similarity matrix is ​​established, where the rows of the similarity matrix correspond to the rows of the multidimensional feature matrix, the columns correspond to the graph nodes of the jute fiber aging evolution knowledge graph, and the matrix elements are the calculated cosine similarity values. Traverse this similarity matrix and find the graph node with the highest cosine similarity value in each row. If the highest cosine similarity value exceeds a preset matching threshold, mark the current row of the multidimensional feature matrix as a successful match with that graph node in the jute fiber aging evolution knowledge graph. Record all successfully matched rows and their corresponding graph node information in the jute fiber aging evolution knowledge graph as a matching result mapping table.

[0083] In practice, the feature marker set is reorganized and mapped over time. For each feature marker in the set, a sliding window sampling is performed along the time axis. For example, if the feature marker set contains markers A, B, and C, and the total time series length is 100 sampling points, a sliding window with a window length of 10 and a step size of 5 is used for sampling, resulting in feature marker subsets under multiple time slices. The first time slice contains feature markers A and B appearing from time point 1 to 10, and the second time slice contains feature markers A and C appearing from time point 6 to 15. Each feature marker subset under each time slice constitutes a local observation sample. The feature marker subset under each time slice is then classified and mapped to a corresponding coordinate axis in a high-dimensional vector space according to its physicochemical properties. For example, feature marker A represents carbonyl absorption characteristics and is mapped to the first dimension of the vector space, while feature marker B represents phenolic hydroxyl vibration characteristics and is mapped to the second dimension. The mapped values ​​can be the frequency or average intensity of the feature marker appearing in that time slice. Through this mapping, each time slice is represented as a coordinate point in a high-dimensional vector space. All time slices are mapped to coordinates in a high-dimensional vector space to form a multidimensional feature matrix that expands in the time and attribute dimensions. The rows of the multidimensional feature matrix correspond to different time slice numbers, the columns correspond to different physicochemical attribute dimensions, and the values ​​of the matrix elements are the mapped values.

[0084] All graph nodes representing different aging stages are extracted from the jute fiber aging evolution knowledge graph. The jute fiber aging evolution knowledge graph includes graph nodes for "initial state," "mild oxidation," "moderate oxidation," and "severe degradation." The feature vector of each graph node is calculated, and the calculation process can be vectorized based on the type and weight of the relational edges directly connected to the graph node. For each graph node in the jute fiber aging evolution knowledge graph, a graph node embedding vector is generated based on its connected relational edges and attribute values. In some embodiments, a graph neural network algorithm is used to generate the graph node embedding vector, taking the graph node and its adjacency relationships as input and outputting a fixed-length numerical vector as the graph node embedding vector. Each row of the multidimensional feature matrix is ​​considered as a feature vector, and the cosine similarity between each feature vector and each graph node embedding vector is calculated. The cosine similarity calculation formula is:

[0085] ;

[0086] in: Let represent the cosine similarity value between the eigenvector of the p-th row of the multidimensional feature matrix and the embedding vector of the q-th graph node. This represents the p-th row vector of the multidimensional feature matrix. This represents the embedding vector of the q-th graph node. The dot sign indicates the vector dot product, and the double vertical lines indicate the Euclidean norm of the vector. Representing vectors The Euclidean norm, Representing vectors The Euclidean norm of the matrix is ​​used. A similarity matrix is ​​constructed where rows correspond to rows of the multidimensional feature matrix, columns correspond to nodes in the jute fiber aging evolution knowledge graph, and matrix elements are calculated cosine similarity values. The similarity matrix is ​​traversed, and the node with the highest cosine similarity value in each row is found. If the highest cosine similarity value exceeds a preset matching threshold, the current row of the multidimensional feature matrix is ​​marked as a successful match with a node in the jute fiber aging evolution knowledge graph. For example, if the matching threshold is set to 0.75, a row with a cosine similarity of 0.82 to the "moderate oxidation" node is considered a successful match. All successfully matched rows and their corresponding node information in the jute fiber aging evolution knowledge graph are recorded in a matching result mapping table. This matching result mapping table is a data structure where each record contains at least the row index of the multidimensional feature matrix, the identifier of the successfully matched node, and the corresponding cosine similarity value. It is understandable that the matching result mapping table establishes a correspondence between time slices of observed data and theoretical aging stages in the knowledge graph of jute fiber aging evolution.

[0087] Example 4: Based on the matching results between the multidimensional feature matrix and the jute fiber aging evolution knowledge graph, the matching result mapping table is analyzed to locate all graph nodes that successfully match the multidimensional feature matrix in the jute fiber aging evolution knowledge graph. In the jute fiber aging evolution knowledge graph, the shortest connected subgraph connecting all successfully matched graph nodes is found. The directed edge sequence in this shortest connected subgraph constitutes the initial evolution path. This initial evolution path is pruned and optimized, removing edges with weights lower than the average path weight to obtain a simplified key evolution path. The key evolution path is transformed into a state transition model, where the states correspond to graph nodes in the key evolution path. The transition probability is calculated based on the edge weights between graph nodes and the matching strength in the matching result mapping table. Specifically, the weight values ​​of the connecting edges between each graph node on the key evolution path are extracted from the jute fiber aging evolution knowledge graph. These weight values ​​represent the ease of transition between aging stages. The matching strength corresponding to each successfully matched graph node is extracted from the matching result mapping table, and this matching strength is quantified using the cosine similarity value in the similarity calculation. For each pair of adjacent states in the state transition model, the transition probability is calculated as the product of the edge weight and the matching strength. Then, the product of all transitions is normalized so that the sum of the transition probabilities for each state is one. The calculated transition probabilities are filled into the state transition matrix, where the rows and columns correspond to the graph nodes on the critical evolution path, thus completing the parameterization of the state transition model. Multiple different combinations of environmental parameters are set, and each combination is used as an external input to drive this state transition model. The trajectory of the jute fiber sample evolving from its initial state along the critical evolution path is iteratively calculated. The ordinate of this trajectory represents the performance index, thereby generating a performance degradation curve corresponding to each combination of environmental parameters.

[0088] In practice, the parsing is performed based on the matching result mapping table. This table contains row indices of the multidimensional feature matrix, identifiers of successfully matched graph nodes, and cosine similarity values. The parsing operation involves reading each record in the matching result mapping table and locating all graph nodes in the jute fiber aging evolution knowledge graph that successfully match the multidimensional feature matrix based on the graph node identifiers in the records. For example, if the matching result mapping table indicates that rows 10, 25, and 40 of the multidimensional feature matrix match the graph nodes "mild oxidation," "moderate oxidation," and "severe degradation," respectively, then these three corresponding graph nodes are found in the jute fiber aging evolution knowledge graph. Within the jute fiber aging evolution knowledge graph, all successfully matched graph nodes are taken as a subset. The shortest connected subgraph connecting all nodes in this subset is searched. This search can be performed using the shortest path algorithm in graph theory to calculate the shortest path between the subset nodes. These paths are then merged to form a connected subgraph containing all target nodes. The directed edge sequence in this shortest connected subgraph constitutes the initial evolution path, for example, "mild oxidation -> moderate oxidation -> severe degradation."

[0089] The initial evolution path is pruned and optimized by removing edges with weights lower than the path's average weight. This pruning optimization requires calculating the average weight of all edges in the initial evolution path, comparing the weight of each edge to this average, and deleting edges with weights lower than the average to obtain a simplified critical evolution path. In practice, the critical evolution path is transformed into a state transition model, where states correspond to graph nodes in the critical evolution path. For example, state S1 corresponds to "mild oxidation," state S2 to "moderate oxidation," and state S3 to "severe degradation." In practice, each state directly corresponds to a graph node in the path; for example, state S1 corresponds to the "mild oxidation" node, state S2 to the "moderate oxidation" node, and state S3 to the "severe degradation" node, thus establishing a one-to-one mapping between states and aging stages. Next, the weight values ​​of the connecting edges between adjacent graph nodes on the critical evolution path are extracted from the jute fiber aging evolution knowledge graph. These weight values ​​reflect the ease of transition between aging stages. Simultaneously, the matching strength corresponding to each successfully matched graph node is obtained from the matching result mapping table, and this strength is quantified by the cosine similarity value. For each pair of adjacent states in the state transition model, the transition probability is calculated as the product of the edge weight and the matching strength. Then, the product of all transition probabilities originating from the current state is normalized to ensure that the sum of all outgoing transition probabilities for each state is one. Finally, the calculated transition probabilities are filled into the state transition matrix, where the rows and columns correspond to graph nodes on the key evolution path, thus completing the parameterized construction of the state transition model. The transition probability is calculated based on the edge weights between graph nodes and the matching strength in the matching result mapping table. The weight values ​​of the connecting edges between each graph node on the key evolution path are extracted from the jute fiber aging evolution knowledge graph; these edge weight values ​​represent the ease of transition between aging stages. The matching strength corresponding to each successfully matched graph node is extracted from the matching result mapping table, and the matching strength is quantified using the cosine similarity value in the similarity calculation.

[0090] For each pair of adjacent states in the state transition model, the transition probability is calculated as the product of the edge weight and the matching strength. Then, the product of all transitions is normalized so that the sum of the exit transition probabilities for each state is one. The formula for calculating the state transition probability can be expressed as:

[0091] ;

[0092] in: This represents the probability of transitioning from state i to state j. This represents the weight of the edge from graph node i to graph node j. The summation in the denominator represents the matching strength of state i, and it is performed on all target states k transitioning from state i. The calculated transition probabilities are then filled into the state transition matrix, where the rows and columns correspond to the graph nodes on the critical evolution path, thus completing the parameterization of the state transition model. See Table 1, which illustrates the calculation of state transition probabilities for a model with three graph nodes.

[0093] Table 1: Edge Weights and Matching Strengths Between Graph Nodes

[0094]

[0095] In some embodiments, the combination of environmental parameters may include multiple dimensions such as temperature, humidity, and light intensity. For example, a set of environmental parameters could be "temperature 30°C, humidity 60%RH, light intensity 5000 lux". Multiple different combinations of environmental parameters are set, and each combination is used as an external input to drive the state transition model. Environmental parameters can intervene in the model by influencing the probability values ​​in the state transition matrix. One approach is to define a scaling factor for each environmental parameter, which acts on the corresponding transition probability. The trajectory of the jute fiber sample evolving from its initial state along a key evolution path is iteratively calculated. The iterative process simulates the evolution of the state sequence through continuous multiplication of the state transition matrix. The vertical axis of the trajectory represents the performance index; for example, the performance index for the state "mildly oxidized" is defined as 0.9, "moderately oxidized" as 0.6, and "severely degraded" as 0.3. By simulating the state transition process, the changes in performance index over time or the number of transition steps are recorded, thereby generating a performance degradation curve corresponding to each combination of environmental parameters. It can be understood that different combinations of environmental parameters will lead to different state evolution speeds and paths by modifying the transition probabilities, thus producing different forms of performance degradation curves. Optionally, the iterative computation of the state transition model can be simulated using the Markov chain Monte Carlo method to evaluate the statistical characteristics of performance degradation.

[0096] See Figure 4In the construction of the state transition model for detecting the antioxidant properties of jute fiber, a heatmap visually presents the probability distribution of transitions between different oxidation states. Specifically, the "current state" and "target state" in the heatmap correspond to the "mild oxidation," "moderate oxidation," and "severe degradation" nodes in the jute fiber aging evolution knowledge graph, respectively, and the matrix element values ​​represent the transition probabilities between states. The heatmap shows that when the current state is "mild oxidation," the probability of transitioning to "moderate oxidation" is 0.80, and the probability of transitioning to "severe degradation" is 0.20, with no possibility of self-transition; when the current state is "moderate oxidation," the probability of transitioning only to "severe degradation" is 1.00; and when the current state is "severe degradation," there is no transition path. This probability distribution is obtained by multiplying and normalizing the edge weights between nodes in the jute fiber aging evolution knowledge graph with the matching strength in the matching result mapping table. Its numerical value reflects the ease or difficulty of state transitions between different oxidation stages, providing a quantitative basis for simulating performance degradation curves under different environmental parameters.

[0097] Example 5: When acquiring the initial response signal and extracting the spectral response fragment, a spectral detection device is used to irradiate the jute fiber sample with excitation light of a specific wavelength, continuously acquiring the light signals reflected or transmitted from the jute fiber sample, and converting this light signal into an electrical signal sequence as the initial response signal. Standard response signal templates of the same type of jute fiber under known good antioxidant properties are retrieved from a preset standard response database. The initial response signal and the standard response signal template are time-aligned, and the signal intensity difference sequence between the initial response signal and the standard response signal template at each sampling time point is calculated. A sliding window peak detection is applied to this signal intensity difference sequence to identify continuous time intervals where the difference exceeds a dynamic threshold. Specifically, a fixed-length sliding window is set, the length of which is preset according to the sampling rate of the signal intensity difference sequence. This sliding window is gradually slid along the time axis of the signal intensity difference sequence, and for each window position, the average and standard deviation of the signal intensity difference within the window are calculated. A dynamic threshold is set based on the calculated average and standard deviation, which is equal to the average plus a multiple of the standard deviation, where the multiple factor is calibrated experimentally. Within each window, points whose signal intensity difference exceeds a dynamic threshold are identified, and their location indices are recorded. Points exceeding the dynamic threshold identified in adjacent windows are merged to form continuous time intervals, while continuous time intervals shorter than the minimum duration threshold are filtered out. From the initial response signal, the original signal segment corresponding to the final continuous time interval is extracted, and this original signal segment is marked as a spectral response fragment characterizing the antioxidant component.

[0098] In practice, a spectroscopic detection device is used to collect the initial response signal of jute fiber samples under illumination of a specific wavelength. This device can be a spectrometer equipped with a monochromator or a laser source of a specific wavelength, such as ultraviolet light at 365 nm, to excite the fluorescence or reflection characteristics of antioxidant components in the jute fiber. The reflected or transmitted light signals from the jute fiber sample are continuously collected. A photodetector converts the light signals into a time-varying electrical signal sequence, with a sampling rate of 1000 points per second. This electrical signal sequence serves as the initial response signal. From a pre-set standard response database, standard response signal templates of the same type of jute fiber under known good antioxidant properties are retrieved. The standard response database stores multiple sets of response signal curves of jute fiber samples measured under standard experimental conditions with calibrated antioxidant properties. By querying the sample number or attribute matching, a standard response signal template identical to the jute fiber type being tested is retrieved. The template is a data sequence with the same time length and sampling rate as the initial response signal.

[0099] In practical implementation, the initial response signal and the standard response signal template are time-aligned. Time alignment can be achieved by finding the time shift corresponding to the maximum correlation between the two signals through cross-correlation calculation. After alignment, the signal intensity difference sequence between the initial response signal and the standard response signal template at each sampling time point is calculated. The signal intensity difference sequence is a new sequence, where the value at each point is the initial response signal intensity minus the standard response signal template intensity. A sliding window peak detection is applied to this signal intensity difference sequence to identify continuous time intervals where the difference exceeds the dynamic threshold. A fixed-length sliding window is set, the length of which is preset according to the sampling rate of the signal intensity difference sequence. For example, when the sampling rate is 1000 Hz, the sliding window length can be set to 100 data points, corresponding to a time span of 0.1 seconds. The sliding window is gradually slid along the time axis of the signal intensity difference sequence. For each window position, the average and standard deviation of the signal intensity difference within the window are calculated. The formula for calculating the dynamic threshold θ is:

[0100] ;

[0101] in: Indicates dynamic threshold. This represents the average value of the signal strength differences within the current sliding window. This represents the standard deviation of the signal strength difference within the current sliding window. Indicates the factor. Factor Through experimental calibration, it was determined in some embodiments that the multiple factor... This can be obtained through statistical analysis of training samples with known response ranges of antioxidant components, for example, by... The value is set to 2.5. Within each window, the signal strength difference exceeding the dynamic threshold is identified. The points are identified and their position indices are recorded, which are the sequence numbers of these points in the signal strength difference sequence.

[0102] It is understandable that points exceeding the dynamic threshold identified within adjacent windows are merged to form continuous time intervals. A continuous time interval consists of a series of points with consecutive position indices. Simultaneously, continuous time intervals shorter than the minimum duration threshold are filtered out. The minimum duration threshold is used to eliminate transient noise interference; for example, setting the minimum duration threshold to 10 data points means that intervals with a duration shorter than 0.01 seconds will be filtered out, ensuring that only significant spectral response fragments are retained. From the initial response signal, the original signal segment corresponding to the final continuous time interval is extracted. The original signal segment is the portion of the electrical signal sequence directly extracted from the initial response signal that corresponds to the continuous time interval. The original signal segment is marked as a spectral response fragment characterizing the antioxidant component. In some embodiments, the standard response database contains multiple templates. When calculating the signal intensity difference sequence, the template closest to the current experimental conditions can be selected, or the average sequence of differences from multiple templates can be calculated. Optionally, the calculation of the dynamic threshold θ can also use the absolute median difference within the window instead of the standard deviation to enhance robustness to outliers.

[0103] See Figure 5 In the feature extraction stage of jute fiber antioxidant performance testing, the distribution trajectories of two core features (local mutation features and periodic oscillation features) output by the hierarchical feature stripping engine in the 100-dimensional feature space are shown. The red curve corresponds to the local mutation feature, which exhibits high-frequency and dramatic fluctuations in the feature dimension, reflecting sudden signal variations in the spectral response of anti-oxidative components in jute fiber (related to rapid changes in local components during the oxidation process). The blue curve corresponds to the periodic oscillation feature, whose fluctuations exhibit relatively regular periodicity, reflecting the rhythmic changes in the antioxidant component response signal (related to the dynamic equilibrium process of fiber oxidation). The differentiated distribution of these two types of features in the feature dimension provides a visual basis for feature differentiation for subsequent cross-validation and association binding in the feature integration layer, supporting the construction of a set of feature labels related to the oxidation process.

Claims

1. A method for testing the antioxidant properties of jute fiber, characterized in that, The method includes: The initial response signal of jute fiber sample under light of a specific wavelength is collected, and the initial response signal is synchronously compared with a preset standard response database to extract the spectral response fragments characterizing antioxidant components from the initial response signal. Based on the spectral response fragment, a dynamic response trajectory is constructed, and the dynamic response trajectory is input into a hierarchical feature stripping engine to separate the feature label set related to the oxidation process layer by layer. This includes generating a continuously changing dynamic response trajectory based on the time series data of the spectral response fragment through curve fitting technology. The feature tag set is reorganized and mapped in time series to generate a multidimensional feature matrix, and the multidimensional feature matrix is ​​matched with a pre-established jute fiber aging evolution knowledge graph. Based on the matching results between the multidimensional feature matrix and the jute fiber aging evolution knowledge graph, the key evolution paths affecting antioxidant performance are identified. Based on these key evolution paths, the performance degradation curves of jute fiber samples under different combinations of environmental parameters are simulated and calculated, including: The matching result mapping table is parsed, and all graph nodes that successfully match the multidimensional feature matrix are located in the jute fiber aging evolution knowledge graph. In the knowledge graph of jute fiber aging evolution, the shortest connected subgraph that connects all successfully matched graph nodes is found, and the directed edge sequence in the shortest connected subgraph constitutes the initial evolution path. The initial evolution path is pruned and optimized by removing edges whose weights are lower than the average weight of the path, resulting in a simplified critical evolution path. The critical evolution path is transformed into a state transition model, where the states in the state transition model correspond to graph nodes in the critical evolution path. The transition probabilities are calculated based on the edge weights between graph nodes and the matching strength in the matching result mapping table, including: The weight values ​​of the connecting edges between each graph node on the key evolution path are extracted from the jute fiber aging evolution knowledge graph. The weight values ​​represent the ease or difficulty of transitioning between aging stages, and the higher the weight value, the easier the transition. The matching strength corresponding to each successfully matched graph node is extracted from the matching result mapping table, and the matching strength is quantified by the cosine similarity value in the similarity calculation. For each pair of adjacent states in the state transition model, the transition probability is calculated as the product of the edge weight and the matching strength. Then, the product of all possible transitions is normalized so that the sum of the out-transition probabilities of each state is one. The calculated transition probabilities are filled into the state transition matrix, where the rows and columns of the state transition matrix correspond to the graph nodes on the key evolution path, thus completing the parameterization construction of the state transition model. Multiple sets of different environmental parameter combinations are set, and each set of environmental parameter combinations is used as an external input to drive the state transition model. The trajectory of the jute fiber sample evolving from the initial state along the key evolution path is iteratively calculated. The vertical axis of the trajectory represents the performance index, thereby generating a performance degradation curve corresponding to each set of environmental parameter combinations. By integrating the performance degradation curve, the key evolution path, and the spectral response fragment, iterative calculations are performed within an integrated performance extrapolation framework to ultimately output the antioxidant stability index of jute fiber samples under specified detection conditions.

2. The method for detecting the antioxidant properties of jute fiber according to claim 1, characterized in that, Based on the spectral response fragment, a dynamic response trajectory is constructed, and the dynamic response trajectory is input into a hierarchical feature stripping engine to separate a set of feature markers related to the oxidation process layer by layer, including: The dynamic response trajectory is input into the first feature stripping layer of the hierarchical feature stripping engine. The first feature stripping layer filters out the basic waveform components in the dynamic response trajectory based on a predefined physicochemical feature filter. The basic waveform components are passed to the second feature stripping layer of the hierarchical feature stripping engine. The second feature stripping layer uses a convolution kernel to slide and scan the basic waveform components to extract the local abrupt change features and periodic oscillation features hidden in the basic waveform components. The local mutation features and the periodic oscillation features are jointly input into the feature integration layer of the hierarchical feature stripping engine. The feature integration layer performs cross-validation and association binding on the local mutation features and the periodic oscillation features to form a set of feature tags related to the oxidation process.

3. The method for detecting the antioxidant properties of jute fiber according to claim 2, characterized in that, The local mutation features and the periodic oscillation features are jointly input into the feature integration layer of the hierarchical feature stripping engine. The feature integration layer performs cross-validation and association binding on the local mutation features and the periodic oscillation features to form a set of feature tags related to the oxidation process, including: In the feature integration layer, a shared feature space is established, and the local mutation features and the periodic oscillation features are projected into the shared feature space; Within the shared feature space, the covariance tensor between the local mutation feature and the periodic oscillation feature is calculated, and the local mutation feature and the periodic oscillation feature are re-encoded based on the covariance tensor; A bidirectional attention mechanism is applied to the re-encoded local mutation features and periodic oscillation features to calculate the attention weight of the local mutation features on the periodic oscillation features, and at the same time, the attention weight of the periodic oscillation features on the local mutation features is calculated. Based on the attention weights calculated by the bidirectional attention mechanism, the re-encoded local mutation features and periodic oscillation features are weighted and fused to generate a fused feature vector. Cluster analysis is performed on the fused feature vectors, and each feature cluster obtained after clustering is labeled as a set of feature labels related to the oxidation process.

4. The method for detecting the antioxidant properties of jute fiber according to claim 1, characterized in that, The step of reorganizing and mapping the feature tag set over time to generate a multidimensional feature matrix, and then matching the multidimensional feature matrix with a pre-established jute fiber aging evolution knowledge graph, includes: For each feature marker in the feature marker set, a sliding window sampling is performed along the time axis to obtain feature marker subsets under multiple time slices; Each subset of feature labels under each time slice is classified and mapped onto the corresponding coordinate axis of a high-dimensional vector space according to its physicochemical properties. All time slices are combined and mapped to the coordinates in the high-dimensional vector space to form a multidimensional feature matrix that expands in the time and attribute dimensions; Extract all graph nodes representing different aging stages from the knowledge graph of jute fiber aging evolution, and calculate the feature vector of each graph node; The similarity between the multidimensional feature matrix and the feature vectors of all graph nodes extracted from the jute fiber aging evolution knowledge graph is calculated. Rows or columns in the multidimensional feature matrix with similarity exceeding a preset threshold are matched to the corresponding graph nodes in the jute fiber aging evolution knowledge graph.

5. The method for detecting the antioxidant properties of jute fiber according to claim 4, characterized in that, The step of calculating the similarity between the multidimensional feature matrix and the feature vectors of all graph nodes extracted from the jute fiber aging evolution knowledge graph, and matching rows or columns in the multidimensional feature matrix with similarity exceeding a preset threshold to the corresponding graph nodes in the jute fiber aging evolution knowledge graph, includes: For each graph node in the jute fiber aging evolution knowledge graph, generate a graph node embedding vector based on its connected relation edges and attribute values; Each row of the multidimensional feature matrix is ​​regarded as a feature vector, and the cosine similarity between each feature vector and the embedding vector of each graph node is calculated. A similarity matrix is ​​established, wherein the rows of the similarity matrix correspond to the rows of the multidimensional feature matrix, the columns correspond to the graph nodes of the jute fiber aging evolution knowledge graph, and the matrix elements are the calculated cosine similarity values. Traverse the similarity matrix and find the graph node with the largest cosine similarity value in each row. If the largest cosine similarity value exceeds the preset matching threshold, mark the current row of the multidimensional feature matrix as a successful match with the graph node in the jute fiber aging evolution knowledge graph. All successfully matched rows and their corresponding graph node information in the jute fiber aging evolution knowledge graph are recorded as a matching result mapping table.

6. The method for testing the antioxidant properties of jute fiber according to claim 1, characterized in that, The initial response signal of the jute fiber sample under specific wavelength light is collected, and the initial response signal is synchronously compared with a preset standard response database to extract the spectral response fragments characterizing antioxidant components from the initial response signal, including: Using a spectral detection device, jute fiber samples are irradiated with excitation light of a specific wavelength, and light signals reflected or transmitted from the jute fiber samples are continuously collected. The light signals are converted into an electrical signal sequence as an initial response signal. Standard response signal templates of the same type of jute fiber under known good antioxidant properties are retrieved from a preset standard response database. The initial response signal is time-aligned with the standard response signal template, and the signal intensity difference sequence between the initial response signal and the standard response signal template at each sampling time point is calculated. A sliding window peak detection is applied to the signal intensity difference sequence to identify continuous time intervals where the difference exceeds a dynamic threshold. From the initial response signal, the original signal segment corresponding to the continuous time interval is extracted, and the original signal segment is marked as a spectral response segment characterizing the antioxidant component.

7. The method for detecting the antioxidant properties of jute fiber according to claim 3, characterized in that, The re-encoded local mutation features and periodic oscillation features are then subjected to a bidirectional attention mechanism to calculate the attention weight of the local mutation features on the periodic oscillation features, and simultaneously calculate the attention weight of the periodic oscillation features on the local mutation features, including: The re-encoded local mutation feature matrix and the re-encoded periodic oscillation feature matrix are respectively input into the attention calculation module; In the attention calculation module, the local mutation feature matrix is ​​used as the query vector, and the periodic oscillation feature matrix is ​​used as the key vector and value vector. The similarity score between the query vector and the key vector is calculated by the dot product operation. The softmax function is applied to normalize the similarity score to obtain the attention weight distribution of the local mutation feature to the periodic oscillation feature. Meanwhile, using the periodic oscillation feature matrix as the query vector and the local mutation feature matrix as the key vector and value vector, the attention weight distribution of the periodic oscillation feature to the local mutation feature is calculated through the same dot product operation and softmax normalization. The attention weight distributions are weighted and summed, with the weight coefficients adaptively adjusted according to the feature dimensions, to generate the fused attention weight matrix.

8. The method for testing the antioxidant properties of jute fiber according to claim 6, characterized in that, The step of applying sliding window peak detection to the signal intensity difference sequence to identify continuous time intervals where the difference exceeds a dynamic threshold includes: A sliding window of fixed length is set, the length of which is preset according to the sampling rate of the signal strength difference sequence; The sliding window is gradually slid along the time axis of the signal strength difference sequence, and for each window position, the average value and standard deviation of the signal strength difference within the window are calculated; A dynamic threshold is set based on the mean and standard deviation. The dynamic threshold is equal to the mean plus a multiple of the standard deviation, wherein the multiple factor is calibrated experimentally. Within each window, points where the signal strength difference exceeds the dynamic threshold are identified, and the location index of the points is recorded. Points exceeding the dynamic threshold identified in adjacent windows are merged to form continuous time intervals, while continuous time intervals with a length less than the minimum duration threshold are filtered out to ensure that only significant spectral response segments are retained.

Images