A smart method for evaluating the effect of curcumin composite fiber wound dressing

Abstract

The application discloses a kind of intelligent effect evaluation methods for curcumin composite fiber wound dressings, belong to medical evaluation technical field, specifically include: by constructing the dynamic wound simulation environment of multiple parameter coupling;Complete cycle is run and the fiber topography, environmental pH, curcumin concentration and viable bacteria number data are synchronously collected by placing the dressing sample in the environment;According to the change of environment and flora data, the continuous characteristic stage is divided;The geometric characteristic parameter of fiber topography in each stage is extracted, and the structure-release correlation parameter is obtained by time-domain alignment and correlation analysis of corresponding curcumin concentration data;Based on the parameter and environmental pH data, the drug release-antibacterial synergistic coefficient of each stage is calculated by staging model;Finally, according to the change relationship of synergistic coefficient between continuous characteristic stages, generate dynamic evaluation atlas.The application realizes the precise quantification and staging evaluation of dressing intelligent response behavior in the simulated real wound dynamic complex environment.

Description

Technical Field

[0001] This invention relates to the field of medical evaluation technology, specifically to an intelligent effect evaluation method for curcumin composite fiber wound dressings. Background Technology

[0002] With the interdisciplinary integration of materials science and biomedicine, functional wound dressings, especially composite fiber dressings with sustained drug release and antibacterial activity, have become an important research direction in the field of chronic wound management. Curcumin, due to its excellent anti-inflammatory, antioxidant, and antibacterial properties, is widely loaded into polymer fibers in the hope of developing smart dressings that combine physical barrier and active therapeutic functions. Accurate evaluation of the performance of such dressings is a key prerequisite for guiding material optimization and predicting clinical outcomes. The core of this approach lies in simulating the complex dynamic healing environment of real wounds and quantifying the "intelligent" response behavior of the dressings.

[0003] Currently, a certain foundation has been established for in vitro evaluation methods of such dressings. Conventional methods mainly include testing the dressing's drug release profile, inhibition zone size against specific planktonic bacteria, and scavenging ability against specific free radicals under static or single-factor controlled conditions. Some methods attempt to introduce light as an external trigger condition to assess photodynamic effects. More advanced studies are beginning to use standardized cytotoxicity assays and animal wound models to verify biosafety and preliminary efficacy. These methods constitute the main technical means of the current evaluation system.

[0004] However, existing assessment systems still exhibit significant limitations when faced with the complexity and dynamism of real wound environments. The wound healing process involves a complex scenario involving the coupling of multiple factors, such as bacterial biofilm formation and metabolism, the continuous degradation of the extracellular matrix by proteases, and dynamic fluctuations in exudate composition and pH. Existing methods often employ isolated, static test conditions, failing to recreate this dynamic environment of multi-parameter interactions. This results in a weak correlation between assessment results and the "intelligent" responsiveness of dressings in real clinical settings. Particularly for environmentally responsive materials like curcumin composite fibers, their drug release kinetics and bioactivity are highly dependent on changes in the surrounding microenvironment. Traditional assessment methods struggle to capture the synergistic relationship between real-time structural evolution and functional output under dynamic stress, thus creating bottlenecks in predicting actual therapeutic efficacy and guiding precise wound design. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent effect evaluation method for curcumin composite fiber wound dressings, solving the following technical problems:

[0006] Existing assessment methods are mainly limited by their static and isolated test conditions, which cannot reproduce the complex scenario of multi-factor dynamic coupling in real wound environments. This makes it difficult to capture the synergistic response relationship between the real-time structural evolution and functional release of curcumin composite fiber dressings under dynamic stress, resulting in a significant disconnect between the actual intelligent therapeutic effect in real clinical environments.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] A smart efficacy evaluation method for curcumin composite fiber wound dressings includes the following steps:

[0009] S1. Construct a dynamic wound simulation environment with multi-parameter coupling, wherein the dynamic wound simulation environment synchronously simulates bacterial proliferation, protein hydrolysis, pH fluctuations and periodic light switching.

[0010] S2. Place the wound dressing sample in the dynamic wound simulation environment and run it for a complete cycle, while simultaneously collecting fiber morphology data, environmental pH data, curcumin concentration data, and viable bacteria count data.

[0011] S3. Based on the changing characteristics of the environmental pH data and the live bacteria count data, the complete cycle is divided into multiple continuous characteristic stages;

[0012] S4. For each characteristic stage, extract the geometric feature parameters of the fiber morphology data, and perform time-domain alignment and correlation analysis with the curcumin concentration data of the corresponding stage to obtain the structure-release correlation parameters of each stage.

[0013] S5. Based on the structure-release correlation parameters and environmental pH data of each characteristic stage, the drug release-antibacterial synergy coefficient of each stage is calculated through a staged model.

[0014] S6. Based on the changing relationship between the drug release-antibacterial synergy coefficient and multiple consecutive characteristic stages, generate a dynamic evaluation map to characterize the effect of the wound dressing sample.

[0015] As a further aspect of the present invention: in S1, the process of simultaneously simulating bacterial proliferation, protein hydrolysis, pH fluctuations, and periodic light switching is as follows:

[0016] Read the environmental timing configuration, which includes multiple consecutive time windows and preset parameter groups for each window; the preset parameter groups include a microbial activity index, an enzymatic hydrolysis intensity index, and an acid-base balance index.

[0017] Within each time window, the bacterial community activity mapping table is queried based on the bacterial community activity index, and the bacterial community activity logical variable is updated; the enzyme hydrolysis intensity mapping table is queried based on the enzyme hydrolysis intensity index, and the protease concentration logical variable is updated; the acid-base balance mapping table is queried based on the acid-base balance index, and the hydrogen ion concentration logical value is updated; at the beginning of each time window, the illumination timing configuration is queried, and the illumination status logical variable is assigned to the on or off state.

[0018] The updated logical variables of microbial community activity, protease concentration, and hydrogen ion concentration are combined with the logical variable of light status to generate a comprehensive environmental state vector for the current time window. The comprehensive environmental state vectors are then output in the order of the time window to form a driving data stream.

[0019] As a further aspect of the present invention: the process of simultaneously collecting fiber morphology data, environmental pH data, curcumin concentration data, and viable bacteria count data in step S2 is as follows:

[0020] The midpoint of each time window in the driving data stream is identified as the synchronous acquisition time point; at each synchronous acquisition time point, a multi-scale texture analysis function is executed to process the sample surface image, and the fiber diameter distribution histogram and pore network connectivity are output and combined into fiber morphology data.

[0021] Read the comprehensive environmental state vector of the current time window, extract the logical value of hydrogen ion concentration and perform linear scaling transformation, and output the environmental acidity and alkalinity data;

[0022] Input the current comprehensive environmental state vector and the curcumin concentration data at the previous synchronous acquisition time point into the curcumin dissolution kinetics model, perform calculations, and output new curcumin concentration data.

[0023] The current comprehensive environmental state vector and the viable bacteria count data from the previous synchronous acquisition time point are input into the microbial community growth inhibition model to perform calculations and output new viable bacteria count data.

[0024] As a further aspect of the present invention: In step S3, the process of dividing the characteristic stages based on the changing characteristics of environmental pH data and viable bacteria count data is as follows:

[0025] Piecewise linear fitting is performed on the environmental pH data sequence to identify the connection points where the slope of adjacent fitted line segments changes abruptly, which are recorded as pH inflection points.

[0026] The second difference of the natural logarithm sequence of viable bacterial count data is calculated, and the zero-crossing points where the sign changes in the second difference sequence are identified and recorded as the inflection points of the bacterial community dynamics.

[0027] The pH inflection point set and the microbial community dynamics inflection point set are merged, and the merged time points are arranged in ascending order to generate a total inflection point sequence. The starting time point of the complete cycle is used as the first segmentation point, and each time point in the total inflection point sequence is used as the subsequent segmentation point. A characteristic stage is defined by the closed time interval between adjacent segmentation points.

[0028] As a further aspect of the present invention: in step S4, the process of extracting geometric feature parameters and obtaining structure-release correlation parameters is as follows:

[0029] The variance of the fiber diameter distribution histogram at each time point in the fiber morphology data is calculated as the diameter variation parameter, and the reciprocal of the pore network connectivity at each time point is calculated as the pore barrier parameter. For the selected characteristic stage, the data segment of the corresponding time interval is extracted from the diameter variation parameter sequence and the pore barrier parameter sequence.

[0030] Linear regression was performed on the extracted diameter variation parameter data segment, and its slope was taken as the diameter trend parameter; linear regression was performed on the extracted pore barrier parameter data segment, and its slope was taken as the pore trend parameter; data segments with the same time interval were extracted from the curcumin concentration data sequence, and the average value of the absolute value of the concentration difference between adjacent sampling points in the data segment was calculated as the concentration dynamic parameter.

[0031] Input the diameter trend parameter, pore size trend parameter, and concentration dynamic parameter into the correlation calculation function, execute the calculation, and output the structure-release correlation parameter.

[0032] As a further aspect of the present invention: the process of the correlation calculation function performing the operation is as follows:

[0033] The three input parameters—diameter trend parameter, pore size trend parameter, and concentration dynamic parameter—are normalized to zero mean and unit variance, respectively, to obtain three normalized sequences.

[0034] Calculate the cross-correlation function between the normalized sequence corresponding to the concentration dynamics parameter and the normalized sequence corresponding to the diameter trend parameter, and search for the global maximum value of the cross-correlation function within a preset time delay range, which is denoted as the first peak value;

[0035] Calculate the cross-correlation function between the normalized sequence corresponding to the concentration dynamics parameter and the normalized sequence corresponding to the pore trend parameter, and search for the global maximum value of the cross-correlation function within the same preset time delay range, which is denoted as the second peak value;

[0036] Multiply the first peak and the second peak by the preset first weight coefficient and the second weight coefficient respectively, add the weighted first peak and the weighted second peak, and output the sum as the structure-release correlation parameter.

[0037] As a further aspect of the present invention: In step S5, the process of calculating the drug release-antibacterial synergy coefficient through a staged model is as follows:

[0038] A stage feature tensor is constructed for each feature stage. The tensor contains three input channels: a structure-release correlation parameter sequence, an environmental pH data sequence, and a viable bacteria count data sequence.

[0039] The stage feature tensor is input into the spatiotemporal feature extraction network. The network first performs one-dimensional convolution on each input channel to extract local temporal patterns, then concatenates the convolution outputs into channels, and performs multi-head self-attention computation on the concatenation results to capture cross-channel dependencies.

[0040] The feature matrix output by the multi-head self-attention is max-pooled along the time dimension and aggregated into a global feature vector. The global feature vector is then input into the co-coefficient prediction head, which contains two fully connected layers. The first layer is followed by a non-linear activation function, and the second layer outputs a scalar value as the drug release-antibacterial co-coefficient.

[0041] As a further aspect of the present invention: the network parameter training process of the spatiotemporal feature extraction network and the cooperative coefficient prediction head is as follows:

[0042] Construct a phased training sample library, where each sample contains a phase feature tensor of a historical feature phase and its corresponding standard covariance coefficient;

[0043] Initialize the parameters of the convolutional kernels, self-attention matrix, and fully connected layers in the network. Input the sample tensor into the network for forward propagation. Calculate the smooth L1 loss between the predicted co-coefficient and the standard co-coefficient. Calculate the loss gradient through backpropagation and update the network parameters using the adaptive moment estimation algorithm. Terminate training when the loss value no longer decreases for multiple consecutive rounds on the validation set, and fix the final network parameters.

[0044] As a further aspect of the present invention: in step S6, the process of generating the dynamic evaluation map is as follows:

[0045] Establish a two-dimensional coordinate system, with the horizontal axis labeled with the numbers of each characteristic stage in sequence, and the vertical axis labeled with the values ​​of the drug release-antibacterial synergy coefficient; in the coordinate system, plot the coordinate points determined by each characteristic stage number and its corresponding synergy coefficient value, identify the local maxima and local minima among all coordinate points, and label them using different shape markers;

[0046] Connect all coordinate points using a cubic spline interpolation function to generate a smooth trend curve. In the area below each stage number on the horizontal axis, fill in the name of the comprehensive environmental state vector type that appears most frequently in the driving data stream within the corresponding time interval of that stage. Render the two-dimensional coordinate system containing coordinate points, shape markers, smooth trend curves, and type names into an image file.

[0047] The beneficial effects of this invention are:

[0048] This invention addresses the fundamental problem of existing assessment systems failing to reflect the complex dynamic scenarios of real wounds due to static and isolated testing. It constructs a multi-parameter coupled dynamic wound simulation environment and collects time-series data on dressing fiber morphology, environmental pH, curcumin concentration, and viable bacterial count within this environment. Based on actual changes in environmental and microbial data, it automatically divides healing characteristic stages and extracts geometric feature parameters of the fiber structure and drug release data for temporal alignment and correlation analysis. This allows for precise capture of the dynamic synergistic relationship between the microstructural evolution and drug release behavior of the dressing at different environmental stress stages. Furthermore, this invention designs a spatiotemporal feature extraction network incorporating a multi-head self-attention mechanism to perform deep analysis of multi-channel time-series data. This network automatically learns and quantifies the complex nonlinear mapping and cross-channel dependence between environmental factors, fiber structure changes, and antibacterial effects, ultimately outputting a dynamic assessment map that intuitively reflects the intelligent response intensity of the dressing at different stages. This invention realizes a closed loop from simulation environment, data acquisition, stage identification to intelligent analysis, which makes the evaluation results highly correlated with the dynamic performance of the dressing in the real clinical environment, providing a reliable technical basis for the precise design, performance optimization and efficacy prediction of curcumin composite fiber dressing. Attached Figure Description

[0049] The invention will now be further described with reference to the accompanying drawings.

[0050] Figure 1 This is a flowchart illustrating the present invention. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Please see Figure 1 As shown, this invention provides an intelligent effect evaluation method for curcumin composite fiber wound dressings, comprising the following steps:

[0053] S1. Construct a dynamic wound simulation environment with multi-parameter coupling, wherein the dynamic wound simulation environment synchronously simulates bacterial proliferation, protein hydrolysis, pH fluctuations and periodic light switching.

[0054] S2. Place the wound dressing sample in the dynamic wound simulation environment and run it for a complete cycle, while simultaneously collecting fiber morphology data, environmental pH data, curcumin concentration data, and viable bacteria count data.

[0055] S3. Based on the changing characteristics of the environmental pH data and the live bacteria count data, the complete cycle is divided into multiple continuous characteristic stages;

[0056] S4. For each characteristic stage, extract the geometric feature parameters of the fiber morphology data, and perform time-domain alignment and correlation analysis with the curcumin concentration data of the corresponding stage to obtain the structure-release correlation parameters of each stage.

[0057] S5. Based on the structure-release correlation parameters and environmental pH data of each characteristic stage, the drug release-antibacterial synergy coefficient of each stage is calculated through a staged model.

[0058] S6. Based on the changing relationship between the drug release-antibacterial synergy coefficient and multiple consecutive characteristic stages, generate a dynamic evaluation map to characterize the effect of the wound dressing sample.

[0059] In a preferred embodiment of the present invention, the process of simultaneously simulating bacterial proliferation, protein hydrolysis, pH fluctuations, and periodic light switching in S1 is as follows:

[0060] The construction of the dynamic wound simulation environment begins with reading an environmental time-series configuration file. This file defines a complete evaluation cycle structure containing multiple consecutive time windows and their corresponding preset parameter sets. Each time window represents a basic environmental state maintenance unit, and its duration can be set according to the speed of the simulated physiological process, for example, 30 to 120 seconds for simulating rapid metabolic changes, or 300 to 600 seconds for simulating slow tissue fluid changes. The preset parameter set for each window includes three core indices: a bacterial community activity index, an enzymatic hydrolysis intensity index, and an acid-base balance index. The bacterial community activity index represents the relative level of bacterial metabolism and proliferation rate in a numerical range of 0 to 100. The enzymatic hydrolysis intensity index is associated with the combined hydrolytic activity of simulated proteases such as matrix metalloproteinases MMP-2 and MMP-9. The acid-base balance index is mapped to a specific pH range of 5.5 to 8.0.

[0061] Within each time window, the bacterial community activity mapping table is consulted based on the bacterial community activity index to obtain the corresponding bacterial growth rate parameter, and the logical variable of bacterial community activity is updated accordingly. Simultaneously, the enzymatic hydrolysis intensity mapping table is consulted based on the enzymatic hydrolysis intensity index to obtain the protease catalytic efficiency parameter, and the logical variable of protease concentration is updated accordingly. The acid-base balance mapping table is consulted based on the acid-base balance index to obtain the specific hydrogen ion concentration parameter, and the logical value of hydrogen ion concentration is updated accordingly. The bacterial community activity mapping table, enzymatic hydrolysis intensity mapping table, and acid-base balance mapping table are all lookup tables established through calibration using previous experimental data, mapping index values ​​to biologically meaningful kinetic or concentration parameters.

[0062] At the start of each time window, the independently defined illumination timing configuration is queried. If the illumination on-time conditions defined in the configuration are met at the current time, the illumination status logical variable is assigned a status flag representing on; otherwise, it is assigned a status flag representing off. The illumination timing configuration can simulate various illumination modes, such as simulating the circadian rhythm switching every 12 hours, or simulating a 20-minute irradiation followed by a 10-minute off cycle in clinical phototherapy. The switching of illumination status is recorded as a key environmental trigger event.

[0063] Subsequently, the updated logical variables for microbial community activity, protease concentration, hydrogen ion concentration, and illumination status are combined and encapsulated to generate a comprehensive environmental state vector representing the complete environmental state of the current time window. This vector is a fixed-dimensional data structure. Finally, the comprehensive environmental state vector generated for each time window is output sequentially, forming a continuous and strictly time-stamped driving data stream. This driving data stream provides a unified time reference and core environmental driving force for the entire evaluation process.

[0064] In another preferred embodiment of the present invention, the process of simultaneously collecting fiber morphology data, environmental pH data, curcumin concentration data, and viable bacteria count data in step S2 is as follows:

[0065] The synchronous acquisition of multi-source data is based on the driving data stream generated in the aforementioned embodiments. First, the start and end timestamps of each time window in the driving data stream are parsed, and the midpoint of each window is calculated and set as the synchronous acquisition time point. For example, for a time window lasting 120 seconds and starting at 0 seconds, its synchronous acquisition time point is set at the 60th second. This setting aims to avoid fluctuations during environmental state transitions and to acquire data that better represents the stable state within the window.

[0066] At each synchronous acquisition time point, four data generation processes are executed in parallel to acquire fiber morphology data, environmental pH data, curcumin concentration data, and live bacteria count data, respectively.

[0067] Fiber morphology data is acquired by executing a multi-scale texture analysis function. This function receives digital images of the dressing sample surface acquired at the current acquisition time. First, the images are subjected to multi-scale filtering and segmentation to distinguish fiber structures from background pores. Then, the diameters of all identified fibers in the image are statistically analyzed to generate a fiber diameter distribution histogram, for example, counting the number of fibers in the range of 0.5 to 3.0 micrometers with a bin width of 0.1 micrometers. Simultaneously, image morphology algorithms are applied to analyze the topological connectivity of the pore network, calculating pore network connectivity indices, such as the ratio of the area of ​​the largest connected pore cluster to the total pore area using eight-neighborhood analysis. Finally, the statistical characteristics of the diameter distribution histogram are combined with the pore network connectivity values ​​to form the fiber morphology data for the current time point.

[0068] Environmental pH data is acquired by parsing the driving data stream. The comprehensive environmental state vector corresponding to the current time window is read, and the logical value of hydrogen ion concentration is extracted. Then, using a preset linear scaling transformation relationship, this logical value is converted into a pH value with a clear physical meaning. The transformation relationship is determined through calibration, for example, using the formula pH=aX+b, where X is the logical value, and a and b are calibration coefficients, mapping the output value to a pH range of 5.0 to 8.5. The converted pH value is recorded as the environmental pH data for the current time point.

[0069] Curcumin concentration data is generated through iterative calls to the curcumin dissolution kinetics model. At the first data collection time point, the model begins calculations with an initial concentration value, e.g., 0 μg / mL. At each subsequent data collection time point, the comprehensive environmental state vector of the current time window is input into the model along with the curcumin concentration data from the previous data collection time point. The model integrates kinetic rules describing drug diffusion, dissolution, and changes in chemical stability under specific pH, enzyme activity, and light conditions. After the model performs the calculations, it outputs a new predicted curcumin concentration value, which serves as the curcumin concentration data for the current time point.

[0070] The generation of viable bacterial count data is achieved through iterative calls to a bacterial community growth inhibition model. At the first data collection time point, the model begins calculations with an initial viable bacterial count. At each subsequent data collection time point, the comprehensive environmental state vector of the current time window is input into the model along with the viable bacterial count data from the previous data collection time point. The model integrates kinetic rules describing the intrinsic bacterial growth, the inhibitory effect of curcumin concentration, and the photodynamic antibacterial effect that may be enhanced by light. After the model performs the calculations, it outputs a new predicted viable bacterial count value, for example, expressed as the logarithm of colony-forming units per milliliter (CFU / mL), which serves as the viable bacterial count data for the current time point. All four data generation processes are triggered at a unified, synchronous data collection time point and bound to the same timestamp for storage, forming an internally correlated multidimensional time series.

[0071] In another preferred embodiment of the present invention, the process of dividing the characteristic stages based on the changing characteristics of environmental pH data and viable bacteria count data in step S3 is as follows:

[0072] First, piecewise linear fitting is performed on the environmental pH data sequence. This fitting process divides the entire sequence into several continuous straight line segments to minimize the overall fitting error. Then, the slope difference at the connection points of adjacent line segments is calculated. When the absolute value of the slope difference at a connection point exceeds a preset sensitivity threshold, such as 0.1 pH units per minute, a significant change in pH trend is determined, and this point is marked as a pH inflection point. All identified inflection points constitute the pH inflection point set.

[0073] Secondly, the viable bacterial count data sequence is processed to capture changes in bacterial community dynamics. First, the natural logarithm of the original sequence is taken to linearize the underlying exponential relationship. Then, the second difference sequence of this logarithmic sequence is calculated, reflecting the rate of change in bacterial growth. This second difference sequence is scanned to identify all points where the numerical sign changes, i.e., zero-crossing points. For example, a zero-crossing point changing from a positive to a negative value may indicate that bacterial growth has shifted from acceleration to deceleration. All identified zero-crossing points are marked as inflection points in bacterial community dynamics, forming a set of inflection points.

[0074] Subsequently, the pH inflection point set and the microbial community dynamics inflection point set are merged. Duplicate points with extremely close timestamps in the merged set are deduplicated; for example, adjacent points with time intervals within 10 seconds are considered the same event point, and only one is retained. Then, the timestamps of all inflection points are sorted in ascending order to generate an ordered sequence of total inflection points. Each point in this sequence represents a potential wound state transition event indicated by data.

[0075] Finally, using the start time of the complete evaluation cycle as the first segmentation point, each time point in the overall inflection point sequence is used as a subsequent segmentation point. The closed time interval between two adjacent segmentation points is defined as a characteristic stage. For example, the interval from the start point to the first inflection point is the first stage, the interval from the first inflection point to the second inflection point is the second stage, and so on, until the interval from the last inflection point to the end of the evaluation cycle is the final stage. Each characteristic stage thus corresponds to a time segment with relatively consistent environmental and microbial dynamics or a specific evolutionary pattern, providing a refined time frame for subsequent analysis of the dressing's response in different physiological contexts.

[0076] In another preferred embodiment of the present invention, the process of extracting geometric feature parameters and obtaining structure-release correlation parameters in step S4 is as follows:

[0077] Fiber morphology data itself contains a histogram of fiber diameter distribution over time and information on pore network connectivity. First, the variance of the fiber diameter distribution histogram at each sampling time point is calculated. This variance characterizes the dispersion or uniformity of fiber diameter within the sample field of view at that moment. For example, a distribution with a variance of 0.05 square micrometers indicates relatively uniform fiber diameters, while a distribution with a variance of 0.25 square micrometers indicates uneven fiber diameters and high structural heterogeneity. This series of variance values ​​calculated in chronological order constitutes the diameter variation parameter sequence. Simultaneously, the reciprocal of the pore network connectivity value at each sampling time point is calculated. Pore network connectivity typically describes the degree of interconnection between pores, and its reciprocal can be understood as an indicator of the "barrier" or "separation" of the pore structure. For example, a higher connectivity value, such as 0.8, has a reciprocal of 1.25; a lower connectivity value, such as 0.2, has a reciprocal of 5.0. The larger the reciprocal, the more isolated and disconnected the pore structure tends to be, potentially creating a stronger physical barrier to material diffusion. This series of reciprocals constitutes the sequence of pore barrier parameters. The diameter variation parameter and the pore barrier parameter together form two core quantitative indicators describing the dynamic evolution of the fiber scaffold geometry.

[0078] Next, for a selected characteristic phase defined in Example 3 above, data extraction and trend analysis are performed. Assume the time boundary of this characteristic phase is from minute 180 to minute 320. Based on this time interval, all data points with timestamps between 180 and 320 minutes are extracted from the diameter variation parameter sequence to form a diameter variation data segment. Similarly, data points within the same time interval are extracted from the pore barrier parameter sequence to form a pore barrier data segment.

[0079] A univariate linear regression analysis was performed on the extracted diameter variation data segment. The regression analysis used time as the independent variable and the diameter variation parameter as the dependent variable to fit a straight line that best represents the trend of diameter variation within that period. The slope of this fitted line was extracted and is called the diameter trend parameter. The value and sign of this parameter have clear physical meanings: a positive slope, for example, an increase of 0.005 square micrometers per minute, indicates that the uniformity of fiber diameter is decreasing within this characteristic period, and the structure may become more uneven due to swelling or degradation; a negative slope indicates that the fiber structure is evolving towards a more uniform direction. Similarly, a univariate linear regression analysis was performed on the pore barrier data segment to obtain the slope of the fitted line, called the pore trend parameter. A positive pore trend parameter, for example, an increase of 0.02 per minute, indicates that the "barrier effect" of the pore structure increases over time; a negative value indicates that pore connectivity is improving.

[0080] To characterize the drug release kinetics during this characteristic phase, data points within the same time interval, from 180 minutes to 320 minutes, were extracted from the curcumin concentration data sequence to form a curcumin concentration data segment. The absolute value of the difference in curcumin concentration between any two adjacent sampling points within this data segment was calculated, and then the average of all these absolute differences was taken. This average value was defined as the concentration kinetic parameter. For example, if the sampling interval is 10 minutes, there are 15 intervals between 180 and 320 minutes. The average absolute value of the concentration change within these 15 intervals was calculated to be 0.15 micrograms per milliliter per second. The concentration kinetic parameter intuitively reflects the average drasticness of drug concentration changes or the intensity of release rate fluctuations within this time period. It integrates the processes of release acceleration and deceleration and is an effective indicator for describing the dynamic characteristics of release behavior.

[0081] After obtaining the diameter trend parameter, porosity trend parameter, and concentration dynamic parameter representing the structural change trend, respectively, a correlation calculation function is used to comprehensively evaluate the coupling strength between them. The function is designed to capture the possible time-delay correlation and overall synergy between structural change and release behavior.

[0082] The correlation calculation function involves several steps. The first step is data normalization. Since the diameter trend parameter, porosity trend parameter, and concentration dynamic parameter may have different dimensions and numerical ranges, direct comparison or calculation may lead to bias. Therefore, these three input parameters are first converted into standard scores. Specifically, the arithmetic mean and standard deviation of each parameter are calculated separately; then, for each parameter, its original value is subtracted from its mean, and then divided by its standard deviation. After this processing, each parameter is transformed into a normalized sequence with a mean of 0 and a standard deviation of 1. This step eliminates the influence of dimensions, allowing parameters with different properties to be analyzed for correlation under the same standard.

[0083] The second step is to calculate the cross-correlation function and find the peak value. The cross-correlation function is a tool used to measure the similarity between two time series at different time shifts. First, the cross-correlation function is calculated between the normalized series corresponding to the concentration dynamics parameter and the normalized series corresponding to the diameter trend parameter. The calculation is performed within a preset time delay range, for example, from -30 minutes to +30 minutes. This range is set based on prior knowledge of the physiological timescale of how changes in fiber structure affect drug release, aiming to capture situations where structural changes may lead or lag behind changes in release behavior. The correlation coefficient between the two series is calculated for each time shift value within this time delay range. Among all calculated correlation coefficients, the value with the largest absolute value is found; this value is called the first peak value. It represents the strongest linear correlation between the concentration dynamics and the fiber diameter trend at the optimal matching time delay. Simultaneously, the time shift value corresponding to reaching this peak value is recorded. This value implies potential causal time difference information, but in this embodiment, only the peak value itself is used for subsequent calculations. The exact same cross-correlation calculation process was repeated between the normalized sequence of concentration dynamic parameters and the normalized sequence of pore trend parameters, and the search was conducted within the same preset time delay range. The absolute value of the maximum correlation coefficient obtained was recorded as the second peak value.

[0084] The third step is weighting and synthesis. Considering that the influence of changes in fiber diameter and pore structure on drug release behavior may have different weights, the first and second peak values ​​need to be weighted before synthesizing the final correlation index. The first peak value is multiplied by a preset first weighting coefficient, for example, 0.6; the second peak value is multiplied by a preset second weighting coefficient, for example, 0.4. The preset weighting coefficients can be based on theoretical understanding from materials science, such as the assumption that changes in fiber diameter uniformity may have a more direct impact on the diffusion path than on pore connectivity, thus assigning a higher weight. The weighted first peak value and the weighted second peak value are added together, and the sum is the final structure-release correlation parameter.

[0085] The structure-release correlation parameter is a dimensionless scalar value. Its magnitude comprehensively reflects the overall statistical correlation between the evolution trend of the dressing fiber structure (including changes in diameter uniformity and pore barrier properties) and the dynamic intensity of curcumin release within a specific characteristic stage, considering possible time lags. A high positive value, such as close to 1.5, indicates a strong positive correlation between structural changes and release dynamics; that is, specific evolution patterns of the structure (such as diameter becoming less uniform and pores becoming more barrier-like) are closely accompanied by dramatic changes in drug release. A value close to zero indicates a weaker correlation between the two within that stage. By calculating the unique structure-release correlation parameter for each characteristic stage, this method achieves a phased and quantitative characterization of the dressing's intelligent response behavior (i.e., adjusting release behavior according to its own structural changes), providing key input features for the core evaluation model.

[0086] In another preferred embodiment of the present invention, the process of calculating the drug release-antibacterial synergy coefficient through a staged model in step S5 is as follows:

[0087] First, for each independent feature stage defined by the aforementioned method, a dedicated stage feature tensor is constructed as the model input. This feature tensor is a three-dimensional data structure, with its dimensions corresponding to the time step, the number of feature channels, and the batch size, respectively. Specifically, for a feature stage that may last 140 minutes, if the data sampling interval is 10 minutes, then this stage contains 15 consecutive time steps. The feature channels are fixed at three: the first channel inputs the sequence of "structure-release correlation parameters" arranged chronologically within this feature stage; the second channel inputs the sequence of "environmental pH data" (i.e., pH value sequence) within the same time period; and the third channel inputs the sequence of "viable bacteria count data." The sequence data for each channel is preprocessed, such as normalized, to ensure that the numerical range is suitable for neural network processing. If multiple samples are processed in a batch, the batch size is increased accordingly. Thus, each feature stage is represented as a tensor with a shape of time step multiplied by 3 multiplied by 1, which encapsulates the complete dynamic information of this stage in the three dimensions of structural response, environmental stress, and microbial response.

[0088] The constructed stage feature tensor is then fed into a specially designed spatiotemporal feature extraction network. The forward propagation process of this network involves multiple layers of computation.

[0089] The first layer is a one-dimensional convolutional layer. This layer performs independent convolution operations on the three channels of the input tensor. The convolution kernel slides along the time dimension, and its size defines the receptive field; for example, a kernel of length 3 or 5 is used. The convolution operation for each channel aims to extract local temporal patterns and short-term dependencies in the data sequence of that channel. For example, in the structure-release correlation parameter channel, the convolution operation might identify a pulse pattern where the correlation strength rises and falls rapidly in a short period; in the environmental pH channel, it might identify a local trend of slowly decreasing pH. After each channel undergoes convolution, a non-linear activation function, such as the ReLU function, is typically used to introduce non-linear transformation capabilities. Assuming 16 different convolution kernels are used for each channel, the input sequence for each channel will be transformed into 16 new feature maps, generating a total of 48 feature maps across the three channels. These feature maps maintain the same length as the input sequence in the time dimension.

[0090] The second layer is a channel concatenation and multi-head self-attention layer. First, the 48 feature maps output from the previous convolutional layer, corresponding to the three original channels respectively, are concatenated along the channel dimension to form a composite feature tensor that integrates multi-source information. Then, this composite feature tensor is input into a multi-head self-attention module. This module is a key part of the model, used to capture complex cross-channel dependencies and long-range temporal dependencies between different feature channels. Its working principle is as follows: the input features are linearly projected onto multiple subspaces, and query, key, and value vectors are calculated in each subspace. Attention weights are obtained by calculating the similarity (scaling after dot product) between the query vector and all key vectors. These weights are then used to perform a weighted summation of the value vectors to generate the output of that subspace. For example, one head might focus on learning "how drug release dynamics affect the microbial response when porosity is enhanced and the environment is alkaline," while another head might focus on "the temporal relationship between pH mutations and structural association changes under light triggering." The outputs of the multiple heads are finally concatenated and linearly projected again to form the final output feature matrix of the self-attention layer. Each row of the matrix corresponds to a time step, while each column integrates high-level abstract features extracted from the original three channels through deep interaction.

[0091] The third layer is a temporal max-pooling layer. The feature matrix output by the multi-head self-attention layer retains the length of the original sequence in the temporal dimension. To aggregate this sequence information into a fixed-length vector representation for input into the subsequent fully connected network, pooling is required along the temporal dimension. Max-pooling is used here, taking the maximum value across all time steps for each column of the feature matrix. For example, if the feature matrix has 64 columns, max-pooling results in a global feature vector with 64 elements. Each element in this vector represents the most significant intensity of its corresponding feature across the entire feature stage time span.

[0092] Finally, the global feature vector obtained from pooling is input into a small fully connected neural network called the "synergy coefficient prediction head." This prediction head typically consists of two fully connected layers. The first fully connected layer receives a 64-dimensional input, which may be mapped to a 32-dimensional hidden space, and is immediately followed by a non-linear activation function, such as ReLU or Tanh, to further enhance the model's non-linear expressive power. The second fully connected layer maps the 32-dimensional hidden features to the final output dimension, i.e., 1-dimensional. The output of this layer does not use a non-linear activation function and directly produces a continuous scalar value. This scalar value is the "drug release-antibacterial synergy coefficient" calculated by the model, corresponding to the current input feature stage. This coefficient is a comprehensive evaluation index, and its value theoretically reflects the strength of the synergistic effect between the drug release behavior of the dressing and its achieved antibacterial effect under a specific environmental dynamic and structural evolution mode.

[0093] All learnable parameters in the aforementioned spatiotemporal feature extraction network and co-occurrence coefficient prediction head, including the weights and biases of convolutional kernels, the query, key, and value projection matrices in the self-attention module, and the weights and biases of fully connected layers, need to be trained through supervised learning.

[0094] The first step in training is to construct a high-quality, phased training sample library. This library is built upon a large number of historical wound dressing experimental cases that have undergone comprehensive evaluation. For each historical case, based on its complete evaluation cycle data, multiple characteristic phases are defined using the same method as the evaluation process of this invention. For each historical characteristic phase, a corresponding "phase feature tensor" is constructed as the input feature X. Simultaneously, a "standard co-correlation coefficient" needs to be determined for each historical phase as the training target y. This standard co-correlation coefficient needs to be obtained independently of this model through more reliable but potentially more costly methods. For example, it can be obtained by performing expert quantitative scoring on high-resolution wound healing images of the same phase, or by combining detailed molecular biological detection results after the end of that phase (such as a comprehensive index of the decrease rate of specific inflammatory factors and bacterial clearance rate), after standardization. Thus, each historical characteristic phase constitutes a training sample, i.e., a pair of X and y.

[0095] Before training begins, all parameters in the network need to be initialized. A random initialization strategy is typically used; for example, the weights of convolutional kernels and fully connected layers are randomly sampled from a normal distribution with a mean of 0 and a standard deviation of 0.01, and the biases are initialized to 0. The projection matrix in the self-attention module is also initialized similarly using random initialization.

[0096] The training process employs an iterative optimization approach. In each training round, a small batch of samples is selected from the sample library, for example, a batch size of 32. The feature tensors of this batch of samples are input into the network, and the aforementioned complete forward propagation process is executed to obtain the network's "predicted covariance coefficients" for this batch of samples. Next, the loss between the predicted values ​​and the corresponding "standard covariance coefficients" is calculated. Here, a smoothed L1 loss function is used, which imposes a linear penalty for larger errors and a squared penalty for smaller errors. Compared to mean squared error, it is less sensitive to outliers, which helps improve training stability.

[0097] After the loss function is calculated, the gradient of the loss value with respect to each trainable parameter in the network is calculated using the backpropagation algorithm. The gradient indicates the direction and magnitude by which each parameter should be adjusted to reduce the loss. Subsequently, the parameters are updated using the adaptive moment estimation algorithm. This optimizer maintains an adaptive learning rate for each parameter, taking into account historical gradient information, and can converge more smoothly and efficiently.

[0098] To monitor the training process and prevent overfitting, the sample library is typically divided into a training set and a validation set. During training, the loss is periodically calculated on the validation set, which is not involved in parameter updates. When the loss value on the validation set decreases by less than a very small threshold over multiple consecutive training epochs, such as 20 epochs, the model is considered to have been sufficiently trained and reached convergence. At this point, the training process is terminated, and all the current parameter values ​​of the network are fixed as the parameters of the final trained model. This model can then be used to calculate the drug release-antibacterial synergistic coefficient at various characteristic stages of newly evaluated curcumin composite fiber wound dressing samples.

[0099] In another preferred embodiment of the present invention, the process of generating the dynamic evaluation map in step S6 is as follows:

[0100] A two-dimensional coordinate system is established, where the horizontal axis represents an abstract sequence of time progression, specifically labeled with the sequential numbering of each characteristic stage. For example, if a complete evaluation period is divided into five consecutive characteristic stages, these are labeled from left to right on the horizontal axis as Stage 1, Stage 2, Stage 3, Stage 4, and Stage 5. The vertical axis represents the numerical range of the drug release-antimicrobial synergy coefficient. The upper and lower limits of this range are typically dynamically set based on the synergy coefficient distribution of all samples to be evaluated or historical data; for example, the minimum value is set to 0, and the maximum value to 2.5, to ensure that all data points are clearly contained within the plotting area.

[0101] After establishing the coordinate system, the next step is to transform the calculation results of each feature stage into visualized data points. For each feature stage, its sequential number is used as the x-axis value, and the drug release-antimicrobial synergy coefficient calculated through the staged model for that stage is used as the y-axis value. For example, if the synergy coefficient for stage two is 1.2, then a coordinate point is determined at the position of x-axis 2 and y-axis 1.2. This operation is repeated for all feature stages to obtain a series of discrete coordinate points. These points are plotted in the coordinate system, usually represented by solid circles, and the point color can be set to dark blue to enhance visibility.

[0102] To highlight critical turning points in the dressing's response performance, extreme value analysis is required for all coordinate points. By comparing the ordinate value of each coordinate point with its preceding and following points, local maxima and minima can be identified. A local maximum is a point where the synergistic coefficient value is higher than that of its left and right adjacent points, marking a relative peak in the dressing's intelligent synergistic effect at that stage. Conversely, a local minimum indicates a relative trough in response intensity. These identified extreme points will be highlighted using markers of different geometric shapes. For example, a local maximum might be marked with a red pentagram overlaid on the original dot; a local minimum might be marked with a green triangle. This visual distinction allows assessors to readily identify the key stages of performance fluctuations.

[0103] To more clearly observe the overall trend of the synergy coefficient over the entire time series, a curve fitting method is used to connect the discrete coordinate points. Here, a cubic spline interpolation function is employed to generate a smooth trend curve. This interpolation method ensures that the curve passes through every original data point while creating a smooth transition between adjacent points, avoiding the abruptness of a broken line connection. The calculation process solves for a set of piecewise cubic polynomial coefficients based on the positions of all coordinate points, thus defining a continuous, smooth curve with continuous first and second derivatives. Finally, this curve is plotted on a coordinate system, typically represented by a solid line with a width of 2 pixels, using black or dark gray as the color to clearly present it as a background trend line.

[0104] The area below the horizontal axis of the graph provides crucial environmental context information. For each stage number on the horizontal axis, a short text label needs to be entered in the area directly below it, describing the dominant environmental condition type for that stage. This label is derived from a statistical analysis of all comprehensive environmental state vectors driving the data stream within the time interval corresponding to that characteristic stage. Specifically, it counts the frequency of occurrence of different environmental state vector types within that time period. The type is a predefined category based on typical combinations of logical variables such as bacterial activity, protease concentration, pH, and light conditions in the vector, such as "high bacterial acidic dark period" or "low bacterial alkaline light period." The most frequently occurring type name is selected and entered below the corresponding stage. For example, stage three might be labeled "medium bacterial neutral light period."

[0105] Finally, all the aforementioned visual elements—a scaled two-dimensional coordinate system, coordinate points representing data at each stage, shape markers highlighting extreme points, a smooth trend curve, and environment type labels below the horizontal axis—are combined and rendered to generate a high-resolution digital image file. The rendering process sets appropriate image dimensions, such as a width of 1600 pixels and a height of 900 pixels, and configures clear fonts and a harmonious color scheme. The generated image file is typically saved in PNG or SVG format; this file serves as the final dynamic evaluation map. This map integrates four core pieces of information—time stage, synergy coefficient, performance extreme points, and environmental background—in a compact and intuitive format, providing a powerful visualization tool for comprehensively interpreting the intelligent effects of the dressing.

[0106] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for intelligent efficacy assessment of curcumin composite fiber wound dressing, characterized by, The method comprises the following steps: S1, constructing a multi-parameter coupled dynamic wound simulation environment, which synchronously simulates bacterial proliferation, proteolysis, pH fluctuation and periodic light switching; S2, placing a wound dressing sample in the dynamic wound simulation environment for a complete cycle, and synchronously collecting fiber morphology data, environmental pH data, curcumin concentration data and viable bacteria count data; S3, according to the change characteristics of the environmental pH data and the viable bacteria count data, dividing the complete cycle into a plurality of continuous characteristic stages; S4, for each characteristic stage, extracting the geometric feature parameters of the fiber morphology data, and performing time domain alignment and correlation analysis with the curcumin concentration data of the corresponding stage to obtain structure-release correlation parameters of each stage; S5, based on the structure-release correlation parameters of each characteristic stage and the environmental pH data, calculating the drug release-antibacterial synergistic coefficient of each stage through a staged model; S6, according to the change relationship of the drug release-antibacterial synergistic coefficient between the plurality of continuous characteristic stages, generating a dynamic evaluation atlas for characterizing the effect of the wound dressing sample; In S5, the process of calculating the drug release-antibacterial synergistic coefficient through the staged model is: For each characteristic stage, a stage feature tensor is constructed, which includes a structure-release correlation parameter sequence, an environmental pH data sequence and a viable bacteria count data sequence as three input channels; Input the stage feature tensor into the spatio-temporal feature extraction network; the network first performs one-dimensional convolution on each input channel to extract local time patterns, then concatenates the convolution outputs, and performs multi-head self-attention calculation on the concatenated results to capture cross-channel dependency; The feature matrix output by the multi-head self-attention is maximally pooled along the time dimension to aggregate into a global feature vector, which is input into the synergistic coefficient prediction head. The prediction head includes two fully connected layers, the first layer is followed by a nonlinear activation function, and the second layer outputs a scalar value as the drug release-antibacterial synergistic coefficient; The network parameter training process of the spatio-temporal feature extraction network and the synergistic coefficient prediction head is: Construct a staged training sample library, each sample includes a stage feature tensor of a historical characteristic stage and its corresponding standard synergistic coefficient; Initialize the parameters of the convolution kernel, self-attention matrix and fully connected layer in the network, input the sample tensor into the network for forward propagation, calculate the smooth L1 loss between the predicted synergistic coefficient and the standard synergistic coefficient, calculate the loss gradient through back propagation, and update the network parameters using the adaptive moment estimation algorithm. When the loss value does not decrease continuously on the validation set for multiple rounds, terminate the training and fix the final network parameters.

2. A method for intelligent evaluation of the effect of curcumin composite fiber wound dressing according to claim 1, characterized by, In S1, the process of synchronously simulating bacterial proliferation, proteolysis, pH fluctuation and periodic light switching is: Read the environmental time sequence configuration containing a plurality of continuous time windows and preset parameter groups of each window; the preset parameter group includes a bacterial activity index, an enzyme decomposition intensity index and an acid-base balance index; In each time window, the bacterial activity index is used to query the bacterial activity mapping table to update the bacterial activity logical variable; the enzyme hydrolysis intensity index is used to query the enzyme hydrolysis intensity mapping table to update the protease concentration logical variable; the acid-base balance index is used to query the acid-base balance mapping table to update the hydrogen ion concentration logical value; at the start of each time window, the light timing configuration is queried to assign the light state logical variable to an open state or a closed state; The updated bacterial activity logical variable, protease concentration logical variable, hydrogen ion concentration logical value and light state logical variable are combined to generate a comprehensive environmental state vector of the current time window, and each comprehensive environmental state vector is output in time window order to form a driving data stream.

3. A method for intelligent evaluation of the effect of curcumin composite fiber wound dressing according to claim 2, characterized by, In S2, the process of synchronously collecting fiber morphology data, environmental pH data, curcumin concentration data and viable bacteria count data is as follows: An intermediate time point of each time window in the driving data stream is identified as a synchronous collection time point; at each synchronous collection time point, a multi-scale texture analysis function is used to process the sample surface image to output a fiber diameter distribution histogram and a pore network connectivity, which are combined into fiber morphology data; The comprehensive environmental state vector of the current time window is read, the hydrogen ion concentration logical value is extracted and linear scaling is performed to output the environmental pH data; The current comprehensive environmental state vector and the curcumin concentration data at the last synchronous collection time point are input into the curcumin dissolution kinetics model to perform calculation and output new curcumin concentration data; The current comprehensive environmental state vector and the viable bacteria count data at the last synchronous collection time point are input into the bacterial growth inhibition model to perform calculation and output new viable bacteria count data.

4. A method for intelligent evaluation of the effect of curcumin composite fiber wound dressing according to claim 1, characterized by, In S3, the process of dividing the feature stages according to the change characteristics of the environmental pH data and the viable bacteria count data is as follows: Segmented linear fitting is performed on the environmental pH data sequence, and the connection points between adjacent fitting line segments where the slope changes abruptly are identified as pH turning points; The second-order difference of the natural logarithm sequence of the viable bacteria count data sequence is calculated, and the zero-crossing points where the sign changes in the second-order difference sequence are identified as bacterial dynamic turning points; The pH turning point set and the bacterial dynamic turning point set are combined, and the combined time points are arranged in ascending order to generate a total turning point sequence. The first division point is the starting time point of a complete cycle, and each time point in the total turning point sequence is sequentially used as a subsequent division point. A feature stage is defined by the closed time interval between adjacent division points.

5. A method for intelligent evaluation of the effect of curcumin composite fiber wound dressing according to claim 1, characterized by, In S4, the process of extracting geometric feature parameters and obtaining structure-release correlation parameters is as follows: The variance of the fiber diameter distribution histogram at each time point in the fiber morphology data is calculated as a diameter variation parameter, and the reciprocal of the pore network connectivity at each time point is calculated as a pore barrier parameter. For selected feature stages, data segments corresponding to the time intervals are extracted from the diameter variation parameter sequence and the pore barrier parameter sequence; Linear regression is performed on the intercepted diameter variation parameter data segment, and the slope is taken as the diameter trend parameter; linear regression is performed on the intercepted pore barrier parameter data segment, and the slope is taken as the pore trend parameter; the data segment of the same time interval is intercepted from the data sequence of the curcumin concentration, and the average value of the absolute value of the concentration difference between adjacent sampling points in the data segment is calculated as the concentration dynamic parameter; The diameter trend parameter, the pore trend parameter and the concentration dynamic parameter are input into the correlation calculation function, and the structure-release correlation parameter is output after operation.

6. A method for intelligent evaluation of the effect of curcumin composite fiber wound dressing according to claim 5, characterized by, The operation process of the correlation calculation function is as follows: The diameter trend parameter, the pore trend parameter and the concentration dynamic parameter are normalized respectively to obtain three normalized sequences; The cross-correlation function between the normalized sequence corresponding to the concentration dynamic parameter and the normalized sequence corresponding to the diameter trend parameter is calculated, and the global maximum value of the cross-correlation function is searched within the preset time delay range, which is recorded as the first peak value; The cross-correlation function between the normalized sequence corresponding to the concentration dynamic parameter and the normalized sequence corresponding to the pore trend parameter is calculated, and the global maximum value of the cross-correlation function is searched within the same preset time delay range, which is recorded as the second peak value; The first peak value and the second peak value are multiplied by the first weight coefficient and the second weight coefficient respectively, and the weighted first peak value and the weighted second peak value are added to output the addition result as the structure-release correlation parameter.

7. A method for intelligent evaluation of the effect of curcumin composite fiber wound dressing according to claim 1, characterized by, In S6, the process of generating a dynamic evaluation atlas is as follows: A two-dimensional coordinate system is established, the horizontal axis is labeled with the numbers of each feature stage in order, and the vertical axis is labeled with the values of the drug release-antibacterial synergy coefficient; in the coordinate system, the coordinate points determined by each feature stage number and its corresponding synergy coefficient value are plotted, and the local maximum points and the local minimum points in all coordinate points are identified and labeled with different shapes respectively; All coordinate points are connected using a cubic spline interpolation function to generate a smooth trend curve, and the most frequently occurring comprehensive environmental state vector type name in the driving data stream in the corresponding time interval of each stage number on the horizontal axis is filled in the area below the stage number; The two-dimensional coordinate system containing coordinate points, shape markers, smooth trend curves and type names is rendered into an image file.

Images