Method and system for atopic dermatitis risk warning based on microplastic exposure monitoring

Abstract

This application discloses a method and system for early warning of atopic dermatitis risk based on microplastic exposure monitoring, relating to the field of data processing technology. In this method, skin contact signal sequences are acquired, and the dynamic distribution pattern of microplastics on the skin surface is determined using a convolutional neural network. This pattern is then fused with historical exposure data to calculate an accumulation effect index and generate a skin barrier interference risk score. If the score exceeds a threshold, real-time skin inflammation biomarker data is combined with a support vector machine to determine the potential disease induction probability. Based on this probability, a continuous tracking sequence is generated, and mutation signals are identified to correct calculation biases and obtain risk compensation weights. These weights are used to weighted map the induction probability to generate a target risk distribution map. Finally, accumulation features are extracted from the distribution map, instructions are generated, and an early warning is issued. Implementing the technical solution provided in this application can improve the accuracy of early warning for atopic dermatitis risk related to microplastic exposure.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, specifically a method and system for early warning of atopic dermatitis risk based on microplastic exposure monitoring. Background Technology

[0002] With the deepening development of environmental science and preventive medicine research, the potential impact of microplastic pollution on human health has received increasing attention. Microplastic particles are widely present in the daily environment and enter the human body through skin contact and inhalation. Studies have shown that microplastic exposure is associated with various skin health problems, among which atopic dermatitis, a common chronic inflammatory skin disease, is showing an increasing risk globally. The occurrence of atopic dermatitis involves multiple factors, including genetic susceptibility, immune dysfunction, and impaired skin barrier function, while long-term exposure to environmental factors, especially emerging pollutants such as microplastics, is considered an important risk factor. Therefore, establishing an effective microplastic exposure monitoring and atopic dermatitis risk early warning system is of great significance for the early prevention of the disease.

[0003] Currently, risk warning for atopic dermatitis caused by microplastic exposure mainly involves collecting skin contact signals, using data analysis algorithms to assess microplastic exposure levels, and combining this with monitoring of skin physiological parameters for risk assessment. This method determines the risk level by analyzing exposure characteristics and changes in physiological indicators during the monitoring process. It is relatively convenient to operate and has been applied to some extent in the field of health risk management.

[0004] However, with the increasing demand for precision prevention and personalized health management, the limitations of existing early warning methods are becoming increasingly apparent. In practical applications, due to the long-term cumulative nature of microplastic exposure, and the fact that the risk evolution of atopic dermatitis is the result of multiple dynamic factors, existing methods lack effective mechanisms for tracking the cumulative effects of exposure and the dynamic changes in risk, often leading to biased risk assessment results. This assessment bias results in insufficient accuracy in predicting the degree of risk, thereby reducing the accuracy of atopic dermatitis risk early warning based on microplastic exposure monitoring and failing to meet the early warning needs for precision prevention of chronic skin diseases. Summary of the Invention

[0005] To address the above problems, this invention provides a method and system for early warning of atopic dermatitis risk based on microplastic exposure monitoring. This method solves the problem that existing risk warning methods have limited ability to capture the long-term cumulative effects of microplastics and the dynamic fluctuations in individual physiological states, leading to biased assessment results. This invention can improve the accuracy of early warning of atopic dermatitis risk related to microplastic exposure.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A skin contact signal sequence was acquired, and a convolutional neural network was used to extract features from the skin contact signal sequence to determine the dynamic distribution pattern of microplastic particles on the skin surface. The dynamic distribution pattern is fused with historical exposure data records to calculate the cumulative effect index, and a skin barrier interference risk score is generated based on the cumulative effect index. When the skin barrier interference risk score is higher than a preset risk threshold, real-time skin inflammation marker data is acquired, and the real-time skin inflammation marker data is processed by a support vector machine to determine the probability of potential disease induction. A continuous tracking sequence is generated based on the potential disease induction probability. Risk compensation weights are obtained by identifying mutation signals in the continuous tracking sequence to correct risk measurement bias. The risk compensation weights are used to weight and map the probabilities of potential disease occurrence to generate a target risk distribution map. Accumulated features are extracted from the target risk distribution map, risk warning instructions are generated based on the accumulated features, and warnings are issued according to the risk warning instructions.

[0007] By employing the aforementioned technical solution, and using convolutional neural networks to extract features from skin contact signal sequences, the dynamic distribution pattern of microplastic particles on the skin surface can be determined. This accurately captures the spatial distribution characteristics and temporal evolution of microplastic particles, laying a precise data foundation for subsequent cumulative effect analysis. Based on this, the dynamic distribution pattern is fused with historical exposure data records to calculate the cumulative effect index, achieving a quantitative characterization of the long-term cumulative characteristics of microplastics. This overcomes the limitations of traditional methods that only focus on short-term exposure levels while neglecting cumulative effects, enabling the skin barrier interference risk score generated based on the cumulative effect index to truly reflect the impact of long-term exposure on skin barrier function. When the skin barrier interference risk score exceeds a preset risk threshold, the system promptly acquires real-time skin inflammation biomarker data and processes it using a support vector machine. This accurately identifies subtle changes in skin inflammation, resulting in a more accurate determination of the potential disease induction probability that closely reflects the individual's true health status. Furthermore, a continuous tracking sequence is generated based on the potential disease induction probability. By identifying mutation signals in the continuous tracking sequence, risk calculation biases are corrected, and risk compensation weights are obtained. This effectively captures the dynamic fluctuation characteristics of an individual's physiological state, compensating for the inability of static assessment methods to reflect the dynamic evolution of risk. A target risk distribution map is generated by weighting the probability of potential disease induction using risk-compensated weights. This integrates information from both cumulative exposure effects and dynamic physiological changes, resulting in a more comprehensive and accurate risk assessment. Finally, accumulated features are extracted from the target risk distribution map to generate risk alerts for early warning. This achieves a complete closed loop from exposure monitoring to risk assessment and early warning output, improving the accuracy of atopic dermatitis risk warning based on microplastic exposure monitoring and meeting the practical needs of precision prevention of chronic skin diseases.

[0008] Optionally, the skin contact signal sequence is subjected to joint analysis in the time and frequency domains to obtain a multidimensional spatiotemporal feature vector; the multidimensional spatiotemporal feature vector is input into the convolutional layer of a convolutional neural network for feature mapping to extract local concentration features; a concentration gradient evolution map is constructed based on the local concentration features; the topological structure difference matrix of the concentration gradient evolution map in adjacent time frames is calculated, and the dynamic distribution pattern of microplastic particles on the skin surface is determined according to the topological structure difference matrix.

[0009] Optionally, a multi-scale convolutional kernel is used to perform a sliding window scan on the multi-dimensional spatiotemporal feature vector to generate a multi-scale receptive field feature map; nonlinear activation processing is performed on the multi-scale receptive field feature map to obtain the activation response value in the feature space; spatial pooling dimensionality reduction is performed based on the activation response value to obtain the local concentration feature.

[0010] Optionally, the information entropy of the activation response value in different channel dimensions is calculated, and the channel attention weight is determined based on the information entropy; the activation response value is weighted using the channel attention weight to obtain a weighted response matrix; the weighted response matrix is ​​subjected to global average pooling to output the local concentration feature.

[0011] Optionally, the dynamic distribution pattern is aligned with the historical exposure data records on the time axis to extract the historical cumulative exposure amount; the historical cumulative exposure amount is weighted and integrated according to a preset time decay factor to calculate the cumulative effect index characterizing the degree of skin barrier damage; the deviation between the cumulative effect index and a preset inflammation trigger threshold is calculated; the deviation is mapped to a risk score mapping table to obtain the corresponding skin barrier interference risk score.

[0012] Optionally, real-time skin inflammation biomarker data, including skin temperature and transdermal water loss rate, are collected using physiological monitoring sensors; the real-time skin inflammation biomarker data are mapped to a multi-dimensional feature space to construct an inflammation feature vector; the inflammation feature vector is input into a support vector machine to calculate the classification boundary distance from the inflammation feature vector to the target classification hyperplane; the classification boundary distance is converted into an interval probability value through nonlinear exponential mapping to determine the potential disease induction probability.

[0013] Optionally, the skin temperature time series and the transdermal water loss rate time series are extracted from the real-time skin inflammation biomarker data, respectively; the covariance matrix of the skin temperature time series and the transdermal water loss rate time series is calculated, and the eigenvalues ​​of the covariance matrix are extracted to obtain cross-feature values ​​characterizing the imbalance state of skin hydrothermal coupling; Fourier transform is performed on the skin temperature time series to extract the frequency band energy spectrum characterizing the subcutaneous microcirculation metabolic state; the cross-feature values ​​and the frequency band energy spectrum are vector-concatenated to construct the inflammation feature vector.

[0014] Optionally, the potential disease induction probability is recorded according to a preset sampling frequency to generate a continuous tracking sequence reflecting the risk evolution trajectory; extreme value detection is performed on the continuous tracking sequence to identify mutation signals that deviate from the baseline fluctuation range in the sequence; the sampling density corresponding to the mutation signal is obtained, and when the sampling density exceeds the sampling density benchmark interval, it is determined that there is a risk calculation deviation, and a calculation correction coefficient is determined according to the deviation rate of the sampling density exceeding the sampling density benchmark interval; the initial risk assessment parameters are corrected using the calculation correction coefficient to obtain the risk compensation weight.

[0015] Optionally, a target monitoring area is determined based on the target risk distribution map, and a spatial heterogeneity gradient vector is extracted from the distribution data corresponding to the target monitoring area; the spatial heterogeneity gradient vector is normalized to construct the accumulated feature; the target similarity between the accumulated feature and each reference pattern in the preset skin reaction triggering pattern library is calculated; when the target similarity reaches a preset matching threshold, the accumulated feature is confirmed to be associated with the skin reaction, and a risk warning instruction containing the corresponding reference pattern is generated, and an early warning is issued according to the risk warning instruction.

[0016] Secondly, embodiments of this application provide an atopic dermatitis risk warning system based on microplastic exposure monitoring. The atopic dermatitis risk warning system based on microplastic exposure monitoring includes: one or more processors and a memory; the memory is coupled to the one or more processors and is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the atopic dermatitis risk warning system based on microplastic exposure monitoring to perform the method described in the first aspect and any possible implementation thereof.

[0017] In summary, one or more technical solutions provided in this application have at least the following technical effects or advantages: By employing the aforementioned technical solution, and using convolutional neural networks to extract features from skin contact signal sequences, the dynamic distribution pattern of microplastic particles on the skin surface can be determined. This accurately captures the spatial distribution characteristics and temporal evolution of microplastic particles, laying a precise data foundation for subsequent cumulative effect analysis. Based on this, the dynamic distribution pattern is fused with historical exposure data records to calculate the cumulative effect index, achieving a quantitative characterization of the long-term cumulative characteristics of microplastics. This overcomes the limitations of traditional methods that only focus on short-term exposure levels while neglecting cumulative effects, enabling the skin barrier interference risk score generated based on the cumulative effect index to truly reflect the impact of long-term exposure on skin barrier function. When the skin barrier interference risk score exceeds a preset risk threshold, the system promptly acquires real-time skin inflammation biomarker data and processes it using a support vector machine. This accurately identifies subtle changes in skin inflammation, resulting in a more accurate determination of the potential disease induction probability that closely reflects the individual's true health status. Furthermore, a continuous tracking sequence is generated based on the potential disease induction probability. By identifying mutation signals in the continuous tracking sequence, risk calculation biases are corrected, and risk compensation weights are obtained. This effectively captures the dynamic fluctuation characteristics of an individual's physiological state, compensating for the inability of static assessment methods to reflect the dynamic evolution of risk. A target risk distribution map is generated by weighting the probability of potential disease induction using risk-compensated weights. This integrates information from both cumulative exposure effects and dynamic physiological changes, resulting in a more comprehensive and accurate risk assessment. Finally, accumulated features are extracted from the target risk distribution map to generate risk alerts for early warning. This achieves a complete closed loop from exposure monitoring to risk assessment and early warning output, improving the accuracy of atopic dermatitis risk warning based on microplastic exposure monitoring and meeting the practical needs of precision prevention of chronic skin diseases. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating a method for early warning of atopic dermatitis risk based on microplastic exposure monitoring, as disclosed in an embodiment of this application. Figure 2 This is another schematic flowchart of an atopic dermatitis risk warning method based on microplastic exposure monitoring disclosed in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a system provided in an embodiment of this application.

[0019] In the diagram: 301, Central Processing Unit; 302, Read-Only Memory; 303, Random Access Memory; 304, Bus; 305, Input / Output Interface; 306, Input Section; 307, Output Section; 308, Storage Section; 309, Communication Section; 310, Driver; 311, Removable Media. Detailed Implementation

[0020] To enable those skilled in the art to better understand the technical solution, the present invention will be described in detail below with reference to embodiments. The description in this part is only exemplary and explanatory, and should not be used to limit the scope of protection of the present invention in any way.

[0021] It should be noted that, in this document, the terms "comprising," "including," and any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Specific examples have been used in this document to illustrate the principles and implementation methods of the present invention. These examples are merely for the purpose of helping to understand the method and core ideas of the present invention. The above are only preferred embodiments of the present invention. It should be pointed out that, due to the limitations of written expression and the objective existence of infinite specific structures, those skilled in the art can make several improvements, modifications, or variations without departing from the principles of the present invention, and can also combine the above technical features in an appropriate manner. These improvements, modifications, variations, or combinations, or the direct application of the concept and technical solution of the present invention to other situations without modification, should all be considered within the scope of protection of the present invention.

[0022] This application provides a method for early warning of atopic dermatitis risk based on microplastic exposure monitoring, referring to... Figure 1 , Figure 1 This is a flowchart illustrating a method for early warning of atopic dermatitis risk based on microplastic exposure monitoring, provided in an embodiment of this application. The method is applied to a system, which refers to a hardware and software integrated platform capable of executing an atopic dermatitis risk warning program based on microplastic exposure monitoring. The system can execute an atopic dermatitis risk warning program based on microplastic exposure monitoring. The method includes steps 101 to 106, as follows: Step 101: Obtain the skin contact signal sequence, and use a convolutional neural network to extract features from the skin contact signal sequence to determine the dynamic distribution pattern of microplastic particles on the skin surface.

[0023] Skin contact signal sequence refers to the collection of electrical or optical signals reflecting the adhesion state of microplastic particles, acquired over a continuous time period by a flexible sensor array attached to the body surface; convolutional neural network is a feedforward neural network with deep structure that includes convolutional computation, specifically designed to process data with grid structure to extract local and global features; dynamic distribution pattern refers to the regular characteristics of concentration changes, aggregation location shifts, and spatial diffusion of microplastic particles in a specific area of ​​the skin over time.

[0024] Specifically, the system receives raw multi-channel time-series data from a sensor array via a hardware interface, performs filtering, denoising, and normalization preprocessing to construct a two-dimensional or three-dimensional skin contact signal sequence matrix. This matrix is ​​then input into a pre-trained convolutional neural network. The network's front end uses multiple convolutional kernels of different sizes to perform a sliding window operation on the signal sequence matrix, executing local dot multiplication and summation calculations to extract low-level edge and texture features reflecting the concentration gradient of microplastic particles. Next, pooling layers downsample the extracted feature maps, preserving salient features and reducing data dimensionality. As the network deepens, deep convolutional layers nonlinearly combine low-level local features to extract global spatial features with high-order semantic information. Simultaneously, temporal convolutional operations in the network capture signal change trends between adjacent time frames. Finally, fully connected layers map the extracted spatiotemporal feature vectors to specific classification or regression results, outputting the concentration change rate and aggregation trend of microplastic particles at various spatial coordinate points, thereby accurately determining the dynamic distribution pattern of microplastic particles on the skin surface.

[0025] In one possible implementation, a convolutional neural network is used to extract features from the skin contact signal sequence to determine the dynamic distribution pattern of microplastic particles on the skin surface, specifically including steps 1011-1013, as follows: Step 1011: Perform joint analysis of the skin contact signal sequence in the time and frequency domains to obtain a multidimensional spatiotemporal feature vector.

[0026] Skin contact signal sequence refers to raw time-series data collected by an attached sensor array that reflects the adhesion state of microplastic particles on the skin surface; it records the continuous change of electrical or optical parameters over time. Multidimensional spatiotemporal feature vector refers to a high-dimensional numerical array formed by linearly concatenating extracted temporal features, frequency features, and sensor spatial location attributes; it is the standard input format for subsequent machine learning models to perform pattern recognition; for example, a 128-dimensional structured data column containing root mean square values, frequency band energy, and sensor two-dimensional coordinates.

[0027] Specifically, the process begins by receiving the original skin contact signal sequence. A sliding time window technique is used to divide the continuous long sequence into multiple overlapping short data frames to ensure signal stability. For each short data frame, the maximum, minimum, root mean square (RMS) value, and waveform factor are calculated in the time domain. These time-domain statistics are obtained by traversing all sampling points within the data frame and performing basic algebraic operations, and are used to quantify the instantaneous physical friction amplitude generated when microplastic particles contact the skin. Subsequently, a Fast Fourier Transform (FFT) is performed on the data frame in the frequency domain to convert the time-domain waveform into a frequency-domain spectrum. In the generated spectrum data, the band energy integral, centroid frequency, and peak power spectral density within a specific high-frequency band are calculated to accurately reflect the unique high-frequency micro-vibration response of the microplastic particles. After completing the independent calculations in the time and frequency domains, the extracted time-domain statistical features and frequency-domain spectral features are normalized to eliminate numerical differences between different physical dimensions. Finally, according to the preset feature arrangement rules, the normalized time-domain features and frequency-domain features, together with the sensor physical space coordinates corresponding to the signal, are matrix-stitched to construct a multi-dimensional spatiotemporal feature vector containing complete spatiotemporal and frequency information, providing a standardized data foundation for subsequent deep feature extraction.

[0028] Step 1012: Input the multidimensional spatiotemporal feature vector into the convolutional layer of the convolutional neural network for feature mapping and extract local concentration features; construct a concentration gradient evolution map based on the local concentration features.

[0029] The convolutional layer of a convolutional neural network is the core mathematical computation unit in a deep learning model that performs local perception and weight sharing. It extracts the spatial local topological patterns of the input data through pre-defined filters; for example, it uses a 3x3 convolutional kernel to slide across the input matrix to extract edge contour information. Local concentration features are quantitative representations of the aggregation density and distribution of microplastic particles in a specific micro-region of skin; they reflect the degree of particle retention at the microscopic spatial scale; for example, the quantification score of high-density aggregation of microplastic particles in a pore monitoring area on the back of a hand. A concentration gradient evolution map is a multidimensional matrix map generated by arranging local concentration features at different spatial locations in temporal order and calculating spatial differences; it visually shows the spatial direction of particle concentration change from high to low and the rate of evolution over time.

[0030] Specifically, the constructed multidimensional spatiotemporal feature vector is first reshaped into a two-dimensional or three-dimensional tensor format that meets the network input requirements, and then input into the convolutional layer of a pre-trained convolutional neural network. The convolutional layer contains multiple convolutional kernels with different receptive field sizes, which slide across the spatial dimension of the input tensor with a fixed stride. At each sliding window position, the weight matrix of the convolutional kernel is multiplied element-wise with the local receptive field data corresponding to the input tensor. All product results are summed and a bias term is added, followed by a nonlinear activation value output through a rectified linear function (ReLU). This series of matrix multiplication and addition calculations effectively filters out background physiological noise and accurately extracts the local concentration features reflecting the microplastic adhesion density in a specific region. Next, the local concentration features of all spatial locations within the same time frame are extracted and mapped to a two-dimensional plane coordinate system consistent with the physical topology of the actual skin sensor array. Within this coordinate system, the first-order spatial partial derivatives of the local concentration feature values ​​between adjacent spatial nodes are calculated. The calculation results of these partial derivatives contain information about the magnitude and direction of concentration changes, thus forming the concentration gradient vector. Finally, the concentration gradient vector matrices of multiple consecutive time frames are stacked and correlated in chronological order to construct a concentration gradient evolution diagram that includes a time axis, spatial coordinate axis, and concentration gradient vector values, thereby completely recording the spatiotemporal evolution trajectory of microplastic particle concentration.

[0031] In one possible implementation, the multidimensional spatiotemporal feature vector is input into the convolutional layer of the convolutional neural network for feature mapping to extract local concentration features, specifically including steps 10111-10112, as follows: Step 10111: Use multi-scale convolution kernels to perform sliding window scanning on multi-dimensional spatiotemporal feature vectors to generate multi-scale receptive field feature maps.

[0032] Specifically, the system first receives the multi-dimensional spatiotemporal feature vector output from the previous step and uses it as the input tensor. A set of multi-scale convolutional kernels with different receptive field sizes are pre-configured in the convolutional layer to capture the distribution details of microplastic particles at different spatial scales. For each kernel size, the system performs a sliding window scan along the spatial dimension of the multi-dimensional spatiotemporal feature vector. During the scan, the kernel moves row-by-row and column-by-column across the input tensor with a preset fixed stride. At each position where the sliding window stops, the core convolution calculation is performed: the weight values ​​inside the kernel are multiplied element-wise with the feature values ​​within the corresponding local receptive field of the input tensor. All the product results are then summed and a fixed bias term is added to obtain a scalar feature value for that local region. To maintain the integrity of the feature map edge information, zero-padding is applied to the edges of the input tensor before the sliding scan. Through the above matrix multiplication and addition calculations, each specific-sized convolutional kernel traverses the entire input tensor, generating an independent single-scale feature map. The feature maps generated by small-sized convolutional kernels retain the fine-grained high-frequency edge information of microplastic particles, while the feature maps generated by large-sized convolutional kernels extract the coarse-grained low-frequency contour information of particle distribution. Finally, all single-scale feature maps are concatenated and combined along the channel dimension to construct a high-dimensional multi-scale receptive field feature map, achieving comprehensive capture of local spatial features.

[0033] Step 10112: Perform nonlinear activation processing on the multi-scale receptive field feature map to obtain the activation response value in the feature space; perform spatial pooling dimensionality reduction based on the activation response value to obtain local concentration features.

[0034] Specifically, the generated multi-scale receptive field feature map is first received, and nonlinear activation processing is performed on each element within it. This is achieved by using the Rectified Linear Unified Array (ReLU) as the activation function, iterating through all values ​​in the feature map matrix for conditional judgment and replacement calculations. The calculation process is as follows: extract the value at the current position; if the value is greater than zero, it remains unchanged; if the value is less than or equal to zero, it is forcibly replaced with zero. Through this element-wise nonlinear calculation, negative responses caused by background noise are effectively filtered out, preserving and highlighting the positive signals representing the presence of microplastic particles, thus obtaining a pure activation response value matrix in the feature space. Subsequently, spatial pooling dimensionality reduction is performed based on the obtained activation response value matrix. A fixed-size pooling window (e.g., 2x2) and a corresponding sliding step size are set, allowing the pooling window to slide along the spatial dimension of the activation response value matrix. Within each local region covered by the pooling window, max pooling is performed: comparing all activation response values ​​within the window, the element with the largest value is selected as the sole output representative for that local region, discarding the remaining smaller values. This method of extracting local maxima not only significantly reduces the spatial resolution and data dimensionality of the feature map, thus lowering the resource consumption of subsequent calculations, but also strictly preserves the most significant microplastic concentration signal within the region, giving the feature tolerance to minute spatial displacements. After complete spatial pooling dimensionality reduction calculation, the output low-dimensional feature matrix is ​​the local concentration feature, accurately quantifying the particle aggregation state in each local region.

[0035] In one possible implementation, spatial pooling dimensionality reduction is performed based on the activation response value to obtain local concentration features, specifically including steps 10112-1 to 10112-2, as follows: Step 10112-1: Calculate the information entropy of the activation response value in different channel dimensions, and determine the channel attention weights based on the information entropy.

[0036] Specifically, the system first receives the activation response value tensor containing multiple channel dimensions from the previous step. To evaluate the effective information contained in each channel, the system iterates through all channel dimensions one by one, independently calculating the information entropy of each channel. The specific calculation process is as follows: For the currently selected single channel, the system extracts the activation response values ​​at all spatial locations within that channel; these continuous response values ​​are divided into multiple discrete numerical intervals, and the frequency of the response value within each numerical interval is counted. This frequency is then divided by the total number of response values ​​in that channel to obtain the probability distribution value corresponding to each numerical interval. Subsequently, the product of each probability distribution value and its own logarithmic value is calculated. The product results of all intervals within that channel are summed, and the final sum is negative. This negative value is the information entropy of that channel dimension. The larger the information entropy value, the wider the feature distribution within that channel and the richer the detailed information about the microplastic distribution. After completing the information entropy calculation for all channels, these information entropy values ​​are extracted and constructed into a one-dimensional vector. To convert the information entropy into standard weight coefficients, a normalization calculation is performed on this one-dimensional vector. Specifically, the Softmax normalization function is used to calculate the exponential value of the information entropy of each channel, with the natural constant as the base. Then, the exponential value of a channel is divided by the sum of the exponential values ​​of all channels. Through this exponential normalization calculation, a set of numerical sequences with a sum of 1 is output. This sequence is the channel attention weight, which accurately quantifies the relative importance of each channel feature in subsequent calculations.

[0037] Step 10112-2: Use channel attention weights to weight the activation response values ​​to obtain a weighted response matrix; perform global average pooling on the weighted response matrix to output local concentration features.

[0038] Specifically, the system first synchronously receives the original multi-channel activation response value tensor and the calculated channel attention weight vectors. During weighted allocation calculation, the system performs alignment matching along the channel dimension. For each specific feature channel, its corresponding single-channel attention weight scalar value is extracted. Subsequently, the activation response values ​​at all spatial coordinate positions within that channel are traversed, and each activation response value is multiplied element-wise with its corresponding weight scalar value. Through this multiplicative modulation, the response values ​​of key channels with high information entropy and rich microplastic features are proportionally amplified, while the response values ​​of low-weight channels containing more background noise or redundant information are significantly suppressed. After completing the multiplication operations for all channels, a three-dimensional tensor with unchanged spatial dimensions but recalibrated values ​​is output—the weighted response matrix. Next, global average pooling is performed on this weighted response matrix to achieve dimensionality reduction. The calculation process is as follows: For each channel in the weighted response matrix independently, traverse all rows and columns in the two-dimensional spatial plane of that channel, and sum the weighted response values ​​at all positions. Then, divide this sum by the total number of spatial nodes within that channel (i.e., the product of spatial height and width) to calculate the arithmetic mean of that channel in the global space. This mean replaces the original two-dimensional spatial matrix and becomes the unique scalar representing the overall activation intensity of that channel. The global mean values ​​calculated for all channels are concatenated in channel order to finally output a one-dimensional numerical vector, which is the local concentration feature. This calculation process completely eliminates the sensitivity of the feature to small spatial translations and extracts highly condensed microplastic concentration characterization data.

[0039] Step 1013: Calculate the topological difference matrix of the concentration gradient evolution map in adjacent time frames, and determine the dynamic distribution pattern of microplastic particles on the skin surface based on the topological difference matrix.

[0040] A concentration gradient evolution map refers to a multi-dimensional spatiotemporal data structure containing information on the spatial rate of change and direction of microplastic concentration; it provides basic quantitative data on particle movement trends; for example, a three-dimensional tensor recording the concentration vector changes of various skin regions over the past ten minutes. Adjacent time frames refer to data slices of two closely connected time nodes in a continuous time series, with a fixed sampling interval; they are used to compare state abrupt changes within a very short time window; for example, a data matrix of two consecutive monitoring periods with timestamps T1 and T2. A topological structure difference matrix is ​​a two-dimensional numerical array that quantifies the changes in spatial distribution and connectivity between two time frames through mathematical operations; it accurately identifies active regions where significant displacement or aggregation occurs; for example, a difference matrix recording the absolute values ​​of concentration gradient changes at each pixel. Dynamic distribution patterns refer to the macroscopic laws governing the overall migration, diffusion, or enrichment of microplastic particles on the skin surface over time; it is the final motion state determination result output by the system; for example, a kinematic classification label showing particles spreading in a fan shape from the wrist to the back of the hand.

[0041] Specifically, the concentration gradient matrices of the two adjacent timeframes (the current and previous timeframes) are first extracted from the concentration gradient evolution map. To calculate the topological difference matrix, element-wise spatial difference calculations are performed on the concentration gradient matrices of these two adjacent timeframes. Specifically, for each identical spatial coordinate point in the matrix, the Euclidean distance between the concentration gradient vector of the current timeframe and the concentration gradient vector of the previous timeframe is calculated. The calculated distance difference values ​​are then filled into the corresponding coordinate positions of a newly constructed blank matrix, thereby generating the topological difference matrix. High-value regions in this matrix directly correspond to skin locations where microplastic particles undergo drastic movement or rapid aggregation. Subsequently, connected component analysis and clustering calculations are performed on the topological difference matrix. A significance threshold is set, and active nodes with values ​​exceeding this threshold are extracted from the matrix. Density-based spatial clustering (DBSCAN) is used to divide spatially adjacent active nodes into different evolutionary clusters. The centroid coordinates of each evolutionary cluster are calculated, and the centroid displacement vector is obtained by comparing it with historical centroid positions. Simultaneously, the area expansion rate of the evolutionary cluster is calculated. Finally, these centroid displacement vectors and area expansion rate features are input into a preset decision tree classifier. By comparing them with preset kinematic feature boundaries, specific kinematic categories, such as directional migration, local enrichment, or random shedding, are matched to accurately determine the dynamic distribution pattern of microplastic particles on the skin surface.

[0042] Step 102: Integrate the dynamic distribution pattern with historical exposure data records, calculate the cumulative effect index, and generate a skin barrier interference risk score based on the cumulative effect index.

[0043] Historical exposure data records refer to quantitative data and environmental parameters stored in a database that document microplastic exposure suffered by a target individual over a specific time period. The accumulation effect index is a comprehensive quantitative value used to characterize the physical accumulation caused by long-term retention of microplastic particles on the skin surface and its residual effects over time if not removed. The skin barrier disturbance risk score transforms the accumulation effect index into an intuitive numerical assessment result reflecting the tendency of the skin's protective structures, such as the stratum corneum, to be damaged. In conceptual applications, the dynamic distribution pattern reflects the current state, while historical exposure data records provide the temporal context; only by combining both can the cumulative characteristics of microplastics be truly reflected.

[0044] Specifically, firstly, using timestamp alignment technology, the currently acquired dynamic distribution pattern data and historical exposure data extracted from the database are spliced ​​and resampled on a timeline to construct a complete exposure timeline including both historical and current states. When calculating the cumulative effect index, an integral calculation method with a time decay factor is used. Specifically, each time point in the historical exposure data record is assigned a decay weight, the magnitude of which is inversely proportional to the interval between the data occurrence time and the current time; that is, the older the data, the smaller its weight, thus simulating the natural shedding and metabolic process of microplastics on the skin surface. The current dynamic distribution pattern exposure amount is weighted and summed with the historical exposure amount adjusted for decay weights to obtain the cumulative effect index reflecting the true retention amount. Subsequently, a mapping model between the cumulative effect index and the degree of skin barrier damage is established. The calculated cumulative effect index is input into this mapping model, and the degree of deviation of the current cumulative amount from the baseline is calculated by comparing it with preset health baseline data. Using a linear or nonlinear proportional conversion algorithm, this deviation degree is converted into a percentage or standardized interval value, ultimately generating a skin barrier interference risk score for intuitively characterizing the degree of barrier damage.

[0045] In one possible implementation, the dynamic distribution pattern is fused with historical exposure data records to calculate a cumulative effect index, and a skin barrier disturbance risk score is generated based on the cumulative effect index. Specifically, steps 1021-1023 are included, as follows: Step 1021: Align the dynamic distribution pattern with the historical exposure data records along the timeline and extract the historical cumulative exposure amount.

[0046] Historical exposure data records refer to a set of data stored in a database that records the concentration and duration of microplastic particle exposure of a target object over a specific period of time in the past; it serves as the data basis for assessing long-term cumulative effects; for example, the average daily microplastic exposure concentration curve over the past thirty days. Historical cumulative exposure refers to the sum or integral value of microplastic particle exposure concentration within a specific historical time window; it quantifies the total microplastic load borne by the target object over the long term.

[0047] Specifically, the system first extracts the dynamic distribution pattern obtained during the current monitoring period and historical exposure data records retrieved from the database. To eliminate differences in sampling frequency and recording time points between different data sources, the system performs strict timeline alignment calculations. This is implemented by establishing a unified standard time baseline and extracting all timestamps from the dynamic distribution pattern and historical exposure data records. For the two data sequences, a linear interpolation algorithm is used to fill in missing time nodes, resampling all data into a standard time sequence with fixed time intervals (e.g., hourly or daily). After timeline alignment, the two data sequences have corresponding feature values ​​at the same time nodes. Subsequently, a fixed historical backtracking time window is set on the aligned timeline. Within this time window, all concentration feature values ​​from the historical exposure data records are extracted, and cumulative summation calculations are performed at discrete time points. The calculation process involves iterating through each standard time node within the time window, extracting the corresponding microplastic exposure concentration value, and continuously adding these concentration values ​​arithmetically. By performing this traversal and cumulative calculation along the aligned time axis, discrete single-point exposure data is transformed into a macroscopic scalar representing the total long-term exposure. The final output value of this scalar is the historical cumulative exposure, providing accurate data support for subsequent skin barrier damage assessment.

[0048] Step 1022: Calculate the cumulative effect index characterizing the degree of skin barrier damage by weighting the historical cumulative exposure amount according to the preset time decay factor.

[0049] The preset time decay factor is a numerical weighting coefficient that decreases over time; it is used to simulate the biological forgetting mechanism by which the impact of early microplastic exposure on the current skin condition gradually weakens; for example, a decimal that decays exponentially to near zero with increasing time intervals. The cumulative effect index is a characteristic value that quantifies the degree of substantial damage to the skin barrier caused by long-term microplastic exposure after weighted integral calculation; it is the core basis for triggering subsequent risk assessment; for example, a comprehensive quantitative score that characterizes the risk of stratum corneum damage after time decay correction.

[0050] Specifically, the system first receives time-series data corresponding to historical cumulative exposure extracted in the previous steps and retrieves the preset time decay factor stored internally. To accurately reflect the true impact of exposure at different time points on the current skin barrier, the system introduces a time decay mechanism for weighted integral calculation. The specific calculation process is as follows: taking the current assessment time as the zero point, the system iterates backward through each discrete time node within the time window. For each time node, the time interval between that node and the current time is calculated. Using this time interval as the independent variable, it is substituted into the preset exponential decay function to calculate the time decay factor specific to that time node. The closer the time node is to the current time, the closer its calculated decay factor is to 1; the farther the time node is from the current time, the closer its decay factor is to 0. Subsequently, the historical exposure concentration value at that time node is extracted, and it is multiplied element-wise with the calculated specific time decay factor to obtain the effective exposure equivalent at the current time. After completing the multiplicative weighted calculation for all historical time nodes, the effective exposure equivalents at all time nodes are summed. This multiplication and addition process, which combines continuous decay over time with discrete exposure data, is called weighted integration. After a complete weighted integration calculation, a single scalar value is output, which is the cumulative effect index. It objectively and accurately quantifies the true cumulative effect of historical microplastic exposure on current skin barrier damage.

[0051] Step 1023: Calculate the degree of deviation between the cumulative effect index and the preset inflammation trigger threshold; map the degree of deviation to the risk score mapping table to obtain the corresponding skin barrier interference risk score.

[0052] The preset inflammation trigger threshold refers to a critical concentration or cumulative amount of microplastics that the system pre-sets to characterize the initiation of a pathological inflammatory response in skin tissue; it serves as a baseline reference for determining risk levels; for example, the minimum cumulative amount of microplastics required to induce erythema in the skin, as determined by clinical trial data. Deviation refers to the relative numerical difference between the current cumulative effect indicator and the preset inflammation trigger threshold; it quantifies the distance between the current skin state and the initiation of substantial inflammation; for example, the percentage difference obtained by subtracting the two values ​​and dividing by the threshold. The risk score mapping table is a pre-constructed key-value pair database containing deviation ranges and corresponding risk score values; it is used to convert continuous deviations into standardized risk levels; for example, a two-dimensional data table specifying a deviation range corresponding to a score of 80. The skin barrier interference risk score is a standardized score output after mapping queries, intuitively reflecting the severity of microplastic interference to the skin barrier; it is the final assessment result presented to the user or downstream system; for example, an integer score between 0 and 100, with higher values ​​representing greater risk.

[0053] Specifically, the system first obtains the calculated cumulative effect index and reads the preset inflammation trigger threshold from the system configuration library. To quantify the actual threat posed by the current cumulative exposure to the skin barrier, the system calculates the degree of deviation. The specific calculation process is as follows: the cumulative effect index value is extracted, and an arithmetic subtraction operation is performed between it and the preset inflammation trigger threshold to calculate the absolute difference between the two. Then, this absolute difference is divided by the preset inflammation trigger threshold to obtain a dimensionless ratio. This ratio accurately reflects the proportion by which the current cumulative effect index overflows or falls below the safety threshold; this ratio is the degree of deviation. After obtaining the degree of deviation, the system calls the internally stored risk score mapping table. This mapping table is pre-divided into multiple continuous and non-overlapping deviation value intervals, each strictly bound to a specific risk score value. The system uses the calculated deviation value as a search keyword to perform interval matching traversal in the risk score mapping table. It determines which preset value interval the current deviation value falls into. Once a match is successful, the system directly extracts the score value bound to that interval as the output result. This table lookup mapping process transforms the complex continuous calculation results into intuitive discrete scores. The final extracted score is the skin barrier disturbance risk score, which standardizes and quantifies the potential disturbance risk to the skin barrier caused by microplastic exposure, providing a clear decision-making basis for subsequent protective intervention measures.

[0054] Step 103: When the skin barrier interference risk score is higher than the preset risk threshold, real-time skin inflammation biomarker data is obtained, and support vector machine is used to process the real-time skin inflammation biomarker data to determine the probability of potential disease induction.

[0055] The preset risk threshold is a pre-defined numerical limit used to trigger the next stage of in-depth physiological monitoring, serving as a data filter and system resource scheduling mechanism. Real-time skin inflammation biomarker data refers to physiological parameters that are collected in real time through biosensors and can reflect the skin microcirculation status and inflammatory response, such as abnormal increases in skin surface temperature or transdermal water loss rate. Support vector machine is a supervised learning algorithm based on statistical learning theory, aiming to find an optimal hyperplane to achieve classification or regression prediction of data. Potential disease induction probability refers to the quantitative likelihood that a target individual will develop skin diseases such as atopic dermatitis in the future under the current microplastic exposure and physiological state.

[0056] Specifically, the system continuously monitors the skin barrier interference risk score. Once the score is determined to be significantly higher than a preset risk threshold, a control command is immediately sent to activate the corresponding biosensor module. The biosensor module collects multi-dimensional physiological signals, which, after analog-to-digital conversion and baseline calibration, form a structured real-time skin inflammation biomarker data vector. This data vector is then input into a support vector machine (SVM) model. During the SVM computation, a kernel function (such as the radial basis function) is first used to nonlinearly map the low-dimensional real-time skin inflammation biomarker data vector to a high-dimensional feature space to address the linear inseparability of the original data. In the high-dimensional space, the geometric distance from the input data points to the pre-trained optimal classification hyperplane is calculated. This optimal hyperplane is a boundary derived from training with a large amount of atopic dermatitis diagnosis and healthy sample data. Based on the relative position of the input data points on the hyperplane and their distance from the hyperplane, the geometric distance is converted into a continuous probability value between 0 and 1 using the Sigmoid function or Platt scaling method. This probability value directly reflects the similarity between the current physiological biomarker features and typical inflammatory pathogenesis features, thereby accurately determining the probability of potential disease induction.

[0057] Step 104: Generate a continuous tracking sequence based on the potential disease induction probability, and correct the risk calculation bias by identifying mutation signals in the continuous tracking sequence to obtain the risk compensation weight.

[0058] A continuous tracking sequence refers to a one-dimensional time series data formed by arranging multiple potential disease induction probability values ​​calculated sequentially over time according to timestamps; a mutation signal refers to a sudden and abrupt change in the value within adjacent time points or short time windows in the sequence that does not conform to the normal physiological evolution pattern; risk measurement bias refers to the error between the probability calculation result and the actual physiological risk caused by non-exposure factors such as sensor physical friction, sudden changes in the external environment, or electromagnetic interference; risk compensation weight is a multiplicative or additive coefficient used to reverse the original probability value, aiming to eliminate the impact of measurement bias.

[0059] Specifically, the system establishes a fixed-length First-In-First-Out (FIFO) data buffer, storing the potential disease induction probability output from each calculation cycle along with its timestamp sequentially into the buffer to construct a dynamically updated continuous tracking sequence. To identify mutation signals, an algorithm combining extreme value detection and rate of change analysis is employed. The first and second differences between adjacent data points in the continuous tracking sequence are calculated. When the absolute value of the difference between a data point exceeds a preset physiological change limit threshold, it is marked as a candidate mutation point. Further judgment is made based on sampling density, calculating the data fluctuation variance within the time window of the candidate mutation point. If the variance is extremely large and the duration is extremely short, it is confirmed as a mutation signal. After confirming the mutation signal, the risk measurement deviation is quantified. The difference between the peak value of the mutation signal and the mean value of the data in the stable period before the mutation occurs is calculated; this difference represents the deviation magnitude. Subsequently, a risk compensation weight is calculated, the calculation logic of which is inversely proportional to the deviation magnitude. Specifically, the deviation ratio is obtained by dividing the deviation amplitude by the normal fluctuation range of the baseline. This deviation ratio is then converted into a value between 0 and 1 using a monotonically decreasing exponential or inverse proportional function. The larger the deviation, the smaller the risk compensation weight generated, thereby ensuring that false high-risk assessments caused by abrupt signals can be effectively weakened in subsequent processing.

[0060] In one possible implementation, a continuous tracking sequence is generated based on the potential disease induction probability. Risk assessment bias is corrected by identifying mutation signals in the continuous tracking sequence to obtain a risk compensation weight. Specifically, this includes steps 1041-1043, as follows: Step 1041: Record the probability of potential disease induction according to the preset sampling frequency, and generate a continuous tracking sequence that reflects the trajectory of risk evolution.

[0061] The preset sampling frequency refers to a fixed rhythm parameter that is pre-set to control the time interval of data collection; it is used to standardize the time span and density of data acquisition; for example, setting it to trigger data collection once every ten minutes. The potential disease induction probability refers to a numerical value that quantifies the degree of risk of developing a related disease under specific exposure conditions; it reflects the immediate health threat level at the current monitoring node; for example, a pathogenic risk index with a value of 0.15 obtained through pre-calculation. The risk evolution trajectory refers to the dynamic change path of the risk probability value over time; it is used to visually demonstrate the development trend of disease induction risk; for example, showing a risk evolution path that gradually increases from low to high on a timeline.

[0062] Specifically, the system first initializes its internal hardware timer or software clock module, strictly configuring its trigger period to a preset sampling frequency. Once the monitoring period begins, the timer module generates a trigger signal at each fixed time interval. Upon receiving this signal, the system immediately extracts the instantaneous value of the potential disease-inducing probability output by the pre-calculation module. Simultaneously, the system acquires the current absolute timestamp and binds it to the extracted probability value, forming a time-stamped independent data node. This newly generated data node is then appended sequentially to the end of a pre-allocated linear storage space or data buffer. With the continuous periodic triggering of the timer module, multiple time-stamped data nodes accumulate in the storage space in strict chronological order. This fixed-frequency iterative recording process fully captures the dynamic changes in the probability value throughout the monitoring time window. Finally, the system outputs all data nodes sorted by time in the storage space; this one-dimensional array is the continuous tracking sequence. This sequence, through discrete time nodes and corresponding probability values, accurately and completely depicts the risk evolution trajectory of the potential disease-inducing probability over time.

[0063] Step 1042: Perform extreme value detection on the continuously tracked sequence to identify abrupt change signals in the sequence that deviate from the baseline fluctuation range.

[0064] The baseline fluctuation range refers to the upper and lower threshold range representing the normal fluctuation state of data; it serves as a reference benchmark for judging whether a data point has undergone an abnormal change; for example, the normal value range consisting of a fluctuation of 10% above or below the historical average. A mutation signal refers to an abnormal data point identified in the sequence whose value significantly exceeds the normal fluctuation range; it characterizes a sharp deterioration or abnormal change in the risk state at a specific moment; for example, a probability peak node that suddenly spikes to twice the upper limit of normal fluctuation.

[0065] Specifically, the system first receives the continuous tracking sequence output from the previous step. To establish an objective standard for judging whether the data is abnormal, the system needs to calculate the baseline fluctuation range of the sequence. The specific calculation method is as follows: a sliding window algorithm is used to traverse the entire continuous tracking sequence, calculating the arithmetic mean and standard deviation of the sequence within a global or local time window. Using this arithmetic mean as the center, a set multiple of the standard deviation is added upwards as an upper limit threshold, and a set multiple of the standard deviation is subtracted downwards as a lower limit threshold, thus constructing a baseline fluctuation range containing both upper and lower limits. Subsequently, the system performs extreme value detection on the continuous tracking sequence. Each data node in the sequence is traversed, the probability value of the current node is extracted, and it is compared with the probability values ​​of the adjacent preceding and following nodes. If the value of the current node is strictly greater than the values ​​of its two adjacent nodes, it is marked as a local maximum; if it is strictly less than the values ​​of its adjacent nodes, it is marked as a local minimum. After extracting all local extreme values, the system enters the anomaly identification stage. Each extracted local extreme value is numerically compared with the pre-calculated baseline fluctuation range. If a local maximum value is significantly greater than the upper limit of the baseline fluctuation range, or a local minimum value is significantly less than the lower limit, it indicates that the data point has broken the normal statistical distribution pattern and experienced an abnormal jump. The system extracts these specific data nodes that exceed the boundaries of the baseline fluctuation range. These extracted nodes are the mutation signals, which precisely pinpoint the specific time and magnitude of the abnormal fluctuations during the risk evolution process.

[0066] Step 1043: Obtain the sampling density corresponding to the mutation signal. When the sampling density exceeds the sampling density benchmark range, it is determined that there is a risk calculation deviation. The calculation correction coefficient is determined according to the deviation rate of the sampling density exceeding the sampling density benchmark range. The initial risk assessment parameters are corrected using the calculation correction coefficient to obtain the risk compensation weight.

[0067] A mutation signal refers to an abnormal data point identified in the preliminary steps that deviates from normal fluctuations; it serves as the time anchor point for triggering sampling density verification; for example, a probability peak node that suddenly increases in a time series. Sampling density refers to the number of data collections per unit time within the local time period of the mutation signal; it reflects the density of data acquisition during that critical period; for example, sixty data points were collected within one minute before and after the mutation. The sampling density baseline range refers to a pre-set reasonable sampling frequency range to ensure the accuracy of the calculation results; it serves as a standard for judging whether the current sampling is too dense or too sparse; for example, a density range of ten to twenty collections per minute. The calculation correction coefficient refers to a multiplication factor calculated based on the deviation rate, used to adjust the evaluation parameters; it is used to offset errors caused by abnormal sampling density; for example, a weighting coefficient set to 0.8. The initial risk assessment parameter refers to the uncorrected original risk quantification indicator; it serves as the basic input for correction calculations.

[0068] Specifically, the system first extracts the timestamp corresponding to the mutation signal identified in the previous step. A local time window is constructed by extending a predetermined length forward and backward from this timestamp. The total number of data nodes in the continuously tracked sequence within this local time window is counted, and this total is divided by the time span of the time window to calculate the actual sampling density at the time of the mutation signal. Subsequently, a preset sampling density benchmark interval is retrieved, and its upper and lower limits are extracted. The calculated actual sampling density is compared with this benchmark interval. If the actual sampling density is strictly greater than the upper limit or strictly less than the lower limit, the system determines that the current data acquisition is either too dense or too sparse, thus confirming a potential measurement deviation. After confirming the deviation, the system calculates the deviation rate: the absolute difference between the actual sampling density and the nearest benchmark interval boundary value (upper or lower limit) is calculated, and this absolute difference is divided by the boundary value; the resulting ratio is the deviation rate. After obtaining the deviation rate, the system substitutes it into a preset mapping function (such as a linear inverse proportional function or an exponential decay function) to calculate the corresponding measurement correction coefficient. The greater the deviation rate, the more the calculated correction coefficient deviates from the value of one. Finally, the system obtains the initial risk assessment parameters and performs element-wise multiplication on these parameters with the calculated correction coefficient. Through this multiplication modulation, the initial assessment parameters are proportionally amplified or reduced, thereby offsetting the numerical distortion caused by abnormal sampling density. The final output value of this multiplication operation is the risk compensation weight, which provides rigorously calibrated and accurate parameters for subsequent overall risk quantification.

[0069] Step 105: Use risk compensation weights to weight the probability of potential disease induction and generate a target risk distribution map.

[0070] A target risk distribution map is a multidimensional data visualization or matrix representation that binds the corrected risk values ​​to the spatial coordinates and time dimension of the skin surface, intuitively presenting the risk situation in different areas.

[0071] Specifically, firstly, a weighted mapping calculation is performed, multiplying the potential disease induction probability output in step 103 with the risk compensation weight obtained in step 104 point-to-point. This multiplication operation directly pulls the artificially high probability value caused by noise back to the true level, obtaining the corrected true risk probability value. Subsequently, the target risk distribution map generation process is initiated. The two-dimensional or three-dimensional spatial physical coordinates of each acquisition node in the sensor array are extracted, and the corrected true risk probability value is mapped one-to-one with the corresponding spatial coordinates to construct a discrete risk space matrix. To eliminate monitoring blind spots between sensor nodes, a spatial interpolation algorithm (such as bilinear interpolation or Kriging interpolation) is used to smooth the discrete matrix, calculating the estimated risk value for areas without deployed sensors. After interpolation, the entire continuous spatial matrix is ​​mapped to a preset color space (such as a pseudo-color mapping table), with low-risk values ​​corresponding to cool colors and high-risk values ​​corresponding to warm colors. Finally, a target risk distribution map containing a complete spatial topology and accurate risk gradient is output. This map is stored in the form of a digital matrix and can be directly used for subsequent feature extraction and early warning decision-making.

[0072] Step 106: Extract accumulated features from the target risk distribution map, generate risk warning instructions based on the accumulated features, and issue warnings according to the risk warning instructions.

[0073] Accumulated features refer to the quantitative attributes extracted from the target risk distribution map that reflect the spatial clustering degree, area expansion rate, or overall risk intensity distribution of high-risk areas; risk warning instructions are machine-executable code or structured text automatically generated by the system based on the extracted features, containing specific intervention levels and operational suggestions; early warning refers to the process by which the system transforms risk warning instructions into sound, light, and electrical signals or data pushes that can be perceived by the user through a human-computer interaction interface or external communication interface.

[0074] Specifically, the generated target risk distribution map matrix is ​​first processed and statistically analyzed to extract accumulated features. A risk threshold is set, and a threshold segmentation algorithm is used to binarize the distribution map matrix, separating high-risk areas (foreground) and low-risk areas (background). The connected component area, edge perimeter, and integral sum of risk values ​​within the high-risk areas are calculated. Simultaneously, the expansion rate of the high-risk area is calculated by comparing it with the distribution map from the previous time period. These areas, integral sums, and expansion rates together constitute a multi-dimensional accumulated feature vector. Next, this accumulated feature vector is input into a pre-configured decision rule engine. The rule engine contains multi-level judgment conditions. By comparing the values ​​in the accumulated feature vector with a preset intervention level judgment tree, the current risk level is determined, and a risk warning instruction containing specific control codes and text information is generated accordingly. Finally, the warning action is executed. The system parses the control codes in the risk warning instruction and calls the corresponding hardware driver. If the instruction is a low-level warning, the status icon will only be updated and the log will be recorded in the accompanying application interface; if the instruction is a high-level warning, the mobile terminal's speaker will be triggered to play an alarm audio, a prominent warning pop-up window will be activated on the screen, and an emergency data packet containing accumulated feature data will be sent to the bound medical and health management platform through the wireless communication module, thus completing the closed-loop warning process for the risk of atopic dermatitis.

[0075] In one possible implementation, accumulated features are extracted based on the target risk distribution map, risk warning instructions are generated based on the accumulated features, and warnings are issued according to the risk warning instructions. Specifically, this includes steps 1061-1063, as follows: Step 1061: Determine the target monitoring area based on the target risk distribution map, and extract the spatial heterogeneity gradient vector from the distribution data corresponding to the target monitoring area.

[0076] Specifically, the system first receives the input target risk distribution map. It extracts the risk values ​​of all spatial coordinate nodes in the distribution map and compares them one by one with a pre-set risk benchmark threshold. Spatial coordinate nodes with risk values ​​strictly greater than the benchmark threshold are extracted. A connected component labeling algorithm is used to merge these spatially adjacent nodes that meet the threshold condition, forming one or more closed spatial polygon boundaries. The area inside this boundary is determined as the target monitoring area. Subsequently, the system extracts the risk values ​​corresponding to all coordinate nodes within this target monitoring area, forming distribution data. To extract the spatial heterogeneity gradient vector, the system traverses each central coordinate node in the distribution data, performing difference calculations in the horizontal (X-axis) and vertical (Y-axis) directions. Specifically, the system obtains the risk value of the node adjacent to the right of the central coordinate node, subtracts the risk value of the node adjacent to the left, and uses the result as the horizontal gradient component; similarly, it obtains the risk value of the node adjacent to the top of the central coordinate node, subtracts the risk value of the node adjacent to the bottom, and uses the result as the vertical gradient component. The horizontal and vertical gradient components corresponding to the central coordinate node are then combined and encapsulated in a fixed order. After traversing all nodes within the target monitoring area, the gradient components corresponding to all nodes are combined and arranged in spatial coordinate order into a multidimensional data array. This array is the extracted spatial heterogeneity gradient vector, which accurately quantifies the degree of spatial mutation and the direction of spread of risk within the monitoring area.

[0077] Step 1062: Normalize the spatial heterogeneity gradient vector to construct accumulated features; calculate the target similarity between the accumulated features and each reference pattern in the preset skin reaction triggering pattern library.

[0078] First, the system receives the spatially heterogeneous gradient vector output from the previous step. It then iterates through all numerical elements in the vector, finding and extracting the global maximum and minimum values. Next, it performs a linear scaling calculation on each numerical element: subtracting the global minimum from the current element's value yields the difference; simultaneously, it calculates the range between the global maximum and minimum; dividing the difference by this range gives the normalized standard value of the element. In this way, all gradient components are strictly confined to the range of zero to one, completing the normalization process. The resulting normalized vector is then used to construct the accumulated feature. Next, the system establishes a connection with a pre-defined skin reaction triggering pattern library and sequentially reads each reference pattern stored in the library. For each extracted reference pattern, the system calculates its target similarity with the accumulated feature. The specific calculation uses the cosine of the angle between the vectors: the accumulated feature vector is multiplied element-wise with the corresponding elements in the reference pattern vector, and all product results are summed to obtain the dot product of the two vectors. Subsequently, the square roots of the sum of squares of each element in the accumulated feature vector and the sum of squares of each element in the reference pattern vector are calculated to obtain the magnitudes of the two vectors. Finally, the calculated dot product is divided by the product of the magnitudes of the two vectors, and the resulting ratio is the target similarity. The closer this ratio is to one, the higher the consistency between the accumulated features and the reference pattern in terms of spatial distribution. The system iteratively executes the above calculations until it obtains the target similarity set between the accumulated features and all reference patterns in the pattern library.

[0079] Step 1063: When the target similarity reaches the preset matching threshold, confirm the association between the accumulated features and the skin reaction, and generate a risk warning instruction containing the corresponding reference pattern at the intervention level, and issue an early warning according to the risk warning instruction.

[0080] Specifically, the system first receives the target similarity set calculated in the previous steps and extracts a preset matching threshold from the system configuration register. The system iterates through the target similarity set, strictly comparing each target similarity value with the preset matching threshold. If all target similarity values ​​in the set are strictly less than the preset matching threshold, it is determined that there is no match, and the warning process terminates. If one or more target similarity values ​​in the set are greater than or equal to the preset matching threshold, the system extracts the target similarity value with the highest value. The reference pattern corresponding to this highest similarity value is locked as the corresponding reference pattern. At this point, the system logically confirms that the accumulated features and skin reaction are associated. After confirming the match, the system performs a mapping query in the attribute table of the pattern library based on the unique identifier of the corresponding reference pattern, extracting the intervention level parameter strictly bound to that pattern. Subsequently, the system encapsulates the identifier information of the corresponding reference pattern, the extracted intervention level parameter, and the absolute timestamp of the current system into a standardized data frame according to a preset communication protocol format. This data frame is the generated risk warning instruction. Finally, the system sends the risk warning instruction to the warning execution module. The early warning execution module parses the intervention level parameter in the instruction and calls the corresponding hardware interface based on the parameter. For example, when a high intervention level is detected, the system sends a control level to the display driver module, triggering the screen to display a bright red warning indicator. Simultaneously, it sends a pulse width modulation signal to the audio driver module, driving the speaker to play a high-frequency alarm sound, thus strictly following the risk warning instruction to complete the automated early warning action.

[0081] In the above embodiments, basic skin reaction risk warning functions were achieved through feature accumulation construction and similarity matching. To further improve the accuracy of disease risk assessment and establish a quantitative relationship between physiological markers and disease risk, this application also provides a method for atopic dermatitis risk warning based on microplastic exposure monitoring. This method analyzes the time-series characteristics of markers such as skin temperature and transdermal water loss rate, constructs a mapping relationship between multidimensional inflammatory features and a target classification hyperplane, and performs nonlinear exponential probability transformation, enabling the monitoring system to more accurately assess the probability of potential disease induction under complex physiological conditions. The following section combines... Figure 2 Another method for early warning of atopic dermatitis risk based on microplastic exposure monitoring in the embodiments of this application is described below: Step 201: Collect real-time skin inflammation biomarker data including skin temperature and transdermal water loss rate through physiological monitoring sensors; map the real-time skin inflammation biomarker data to a multi-dimensional feature space to construct an inflammation feature vector.

[0082] Physiological monitoring sensors are hardware detection devices used to attach to or approach the human body surface to acquire physiological physical quantities; they are responsible for converting biological characteristics into electrical signals; examples include patch thermistors and humidity probes. Transdermal water loss rate refers to the mass of water evaporated through a unit area of ​​the stratum corneum per unit time; it is used to assess the integrity of the skin barrier function; for example, a water loss of 15 grams per square meter per hour. Real-time skin inflammation biomarker data refers to a set of physiological indicators continuously collected at the current time point that reflects the degree of skin inflammation; it constitutes the original data source for subsequent feature extraction; for example, a matrix containing temperature and water loss rate values ​​for ten consecutive seconds. Multidimensional feature space refers to a mathematical expression environment composed of multiple independent data feature dimensions; it is used to expand one-dimensional scalar data into a structured form containing richer information; for example, a three-dimensional coordinate system containing time-domain and frequency-domain features. Inflammation feature vectors are oriented numerical sequences representing the state of skin inflammation in multidimensional feature space; they serve as the standard input format for machine learning models; for example, a one-dimensional array containing temperature gradients and water loss variance.

[0083] In this step, the physiological monitoring sensor is first controlled via a hardware interface to continuously detect at a fixed sampling frequency. The sensor's internal analog-to-digital converter converts the detected analog electrical signals into digital signals, extracting skin temperature and transdermal water loss rates at each sampling time point. These two values ​​are then aligned according to timestamps and combined to form real-time skin inflammation biomarker data. Subsequently, a data mapping operation is performed. A multidimensional feature space is established to extract the attributes of the real-time skin inflammation biomarker data across different dimensions. Specifically, the collected raw one-dimensional time-series data undergoes mathematical transformation to extract quantitative indicators in statistical, frequency, and cross-correlation dimensions. These extracted quantitative indicators are arranged and combined in a pre-defined fixed order to form a structured multidimensional data array. This array is the constructed inflammation feature vector. Through this mapping and construction method, the raw, single-dimensional physical measurements are transformed into a comprehensive mathematical expression that fully characterizes the skin pathology state, providing a standardized data foundation for subsequent classification calculations.

[0084] In one possible implementation, real-time skin inflammation biomarker data is mapped to a multidimensional feature space to construct an inflammation feature vector, specifically including steps 2011-2013, as follows: Step 2011: Extract the skin temperature time series and transdermal water loss rate time series from the real-time skin inflammation marker data respectively.

[0085] Specifically, upon receiving real-time skin inflammation biomarker data, the system initiates a data parsing and separation process. It iterates through each record in the data packet, identifying the data identifier in each record. Based on the identifier definition, the data field containing the skin temperature label is extracted, and its corresponding timestamp information is read. The extracted skin temperature values ​​are sorted in ascending order according to their timestamps, and invalid null values ​​or abnormal noise exceeding physical limits are removed. After smoothing and filtering, a continuous skin temperature time series is generated. Simultaneously, based on the water loss rate data identifier, the transdermal water loss rate field is extracted from the same data packet, and its timestamp information is also read. These water loss rate values ​​are arranged in the same timestamp order, and the same denoising and smoothing operations as the temperature data are performed to generate a transdermal water loss rate time series. During this process, strict time alignment verification is performed, comparing the timestamps of corresponding data points in the two series to ensure that every value in the skin temperature time series corresponds one-to-one with every value in the transdermal water loss rate time series in the time dimension, without any time offset or data misalignment. This separation and alignment operation transforms the mixed raw data into two independent and synchronized one-dimensional time-series vectors, providing standard data input for subsequent cross-correlation analysis.

[0086] Step 2012: Calculate the covariance matrix of the skin temperature time series and the transdermal water loss rate time series, and extract the eigenvalues ​​of the covariance matrix to obtain the cross eigenvalues ​​characterizing the imbalance state of skin hydrothermal coupling.

[0087] Specifically, firstly, the arithmetic mean of the skin temperature time series and the arithmetic mean of the transdermal water loss rate time series are calculated separately. Next, the average value is subtracted from each value in the skin temperature time series to obtain the temperature deviation series; the average value is subtracted from each value in the transdermal water loss rate time series to obtain the water loss deviation series. Then, the inner product of these two deviation series is calculated, that is, the deviation values ​​at corresponding positions are multiplied, summed, and then divided by the series length minus one, to obtain the covariance between skin temperature and transdermal water loss rate. Simultaneously, the variances of the temperature deviation series and the water loss deviation series are calculated separately. The two calculated variances are placed on the main diagonal of a two-row, two-column matrix, and the calculated covariance is placed on the secondary diagonal of this matrix, thus constructing the covariance matrix. After construction, eigenvalue decomposition is performed on the covariance matrix. Specifically, the roots of the equation are obtained by solving the characteristic equation of the matrix, i.e., setting the determinant of the matrix minus the product of the unknown scalar and the identity matrix to zero. These roots are the eigenvalues ​​of the covariance matrix. The eigenvalue with the largest value was extracted. This eigenvalue represents the energy in the direction of the greatest variation in the joint change of skin temperature and transepidermal water loss rate. This eigenvalue was used as the cross-feature value, which accurately quantifies the degree of synchronous disorder of skin hydrothermal metabolism under inflammatory conditions, i.e., the state of imbalance in skin hydrothermal coupling.

[0088] Step 2013: Perform Fourier transform on the skin temperature time series to extract the frequency band energy spectrum that characterizes the subcutaneous microcirculation metabolic state; concatenate the cross-feature values ​​with the frequency band energy spectrum to construct the inflammation feature vector.

[0089] Specifically, after acquiring the skin temperature time series, a Discrete Fourier Transform (DFT) is applied. In the calculation, each discrete data point in the time series is multiplied and accumulated with a complex exponential basis function, mapping the temperature fluctuation data in the time domain to the frequency domain, generating a complex spectral sequence containing amplitude and phase information. Then, the square of the modulus of the complex number corresponding to each frequency point in the complex spectral sequence is calculated to obtain the power spectral density. A target frequency range characterizing the subcutaneous microcirculation metabolic state is defined, and the power spectral density values ​​within this target frequency range are extracted. These values ​​are then discretely integrated or directly summed to calculate the total energy within this specific frequency band. The target frequency range is then divided into multiple sub-bands, and the energy values ​​of each sub-band are calculated and arranged sequentially to form a frequency band energy spectrum array. After extracting the frequency band energy spectrum, the cross-feature values ​​calculated in the previous steps are retrieved. A vector concatenation operation is performed to create a new blank one-dimensional array. The cross-feature value is written as the first element of this new array, and then each energy value in the frequency band energy spectrum array is appended to the subsequent positions of the new array in ascending order of frequency. Through this continuous allocation of memory addresses and data filling, the fusion of time-domain cross features and frequency-domain energy features is completed. The final output of this complete one-dimensional array is the constructed inflammatory feature vector, which accurately quantifies the inflammatory pathological state of the skin from multiple dimensions.

[0090] Step 202: Input the inflammation feature vector into the support vector machine and calculate the classification boundary distance from the inflammation feature vector to the target classification hyperplane.

[0091] Support Vector Machines (SVMs) are binary or multi-class machine learning models based on statistical learning theory. They achieve optimal data partitioning by finding the maximum margin boundary; for example, they are algorithm engines that use kernel functions to handle nonlinear classification problems. The target classification hyperplane is the optimal decision boundary in the high-dimensional feature space of an SVM that completely or approximately separates data samples of different classes; it is a fixed mathematical surface or plane determined after model training; for example, a two-dimensional plane that divides two sets of points in three-dimensional space. The classification boundary distance is the shortest geometric distance from the inflammatory feature vector to the target classification hyperplane in multidimensional space; it quantifies the confidence level or deviation of a sample belonging to a certain class; for example, it is the calculated absolute length of the perpendicular segment from a point to the plane.

[0092] Specifically, the system first loads pre-trained support vector machine (SVM) model parameters, including support vectors, corresponding Lagrange multipliers, bias terms, and the selected kernel function type. Upon receiving the constructed inflammation feature vector, it is input as the test sample vector into the SVM's computation engine. The computation engine then initiates the calculation of the classification boundary distance. Specifically, the system iterates through all support vectors stored in the model, performing kernel function calculations on the test inflammation feature vector with each support vector. If a radial basis function (RBF) kernel is used, the square of the Euclidean distance between the test vector and the support vector is calculated, multiplied by the negative kernel parameter, and then the exponent is calculated to obtain the kernel function value. Each calculated kernel function value is then multiplied together with its corresponding Lagrange multiplier and the class label. Subsequently, the multiplication results for all support vectors are summed, and the bias term from the model parameters is added to the sum. The absolute value of this final algebraic sum, after normalization using the model weight vector norm, represents the vertical geometric distance of the inflammation feature vector to the target classification hyperplane in the mapped high-dimensional space. This classification boundary distance not only includes information about the distance, but the sign in its calculation process also indicates which side of the hyperplane the sample is located on, providing a precise geometric quantification basis for subsequent probability transformation.

[0093] Step 203: Convert the classification boundary distance into an interval probability value through a nonlinear exponential mapping to determine the probability of potential disease induction.

[0094] An interval probability value is a continuous value that, after mathematical mapping, is strictly limited to a closed interval between zero and one; it conforms to the definition of probability in statistics; for example, the calculated value of 0.85.

[0095] Specifically, after receiving the classification boundary distance output by the support vector machine, the system calls a preset nonlinear exponential mapping function for numerical transformation. In the specific calculation process, a logistic regression function is used as the mapping tool. First, the classification boundary distance value is extracted, and its sign is considered before multiplying it by a smooth scaling factor obtained through cross-validation optimization. Next, using the natural constant (approximately 2.718) as the base and the negative value of the distance multiplied by the scaling factor as the exponent, the corresponding exponent value is calculated. Then, a constant one is added to this exponent value, resulting in a denominator greater than one. Finally, the constant one is divided by this denominator, and the quotient is the mapped result. Through this nonlinear exponential mapping calculation, the classification boundary distance, originally ranging from negative infinity to positive infinity, is rigorously and smoothly compressed into a numerical range of zero to one, generating interval probability values. The farther a sample is from the classification hyperplane and located on the affected side, the closer its calculated interval probability value approaches one; conversely, it approaches zero. The system directly assigns the calculated interval probability value to the potential disease trigger probability variable and outputs it as a percentage. This probability value accurately quantifies the risk level of the current skin physiological state triggering a potential disease, providing a standardized decision-making basis for subsequent medical interventions or early warning systems.

[0096] The following describes an atopic dermatitis risk early warning system based on microplastic exposure monitoring from the perspective of hardware processing. Please refer to [link to relevant documentation]. Figure 3 This is a schematic diagram of the structure of an atopic dermatitis risk early warning system based on microplastic exposure monitoring in an embodiment of this application.

[0097] It should be noted that, Figure 3 The structure of an atopic dermatitis risk warning system based on microplastic exposure monitoring shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0098] like Figure 3 As shown, an atopic dermatitis risk early warning system based on microplastic exposure monitoring includes a central processing unit (CPU) 301, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 302 or a program loaded from storage section 308 into random access memory (RAM) 303, such as executing the method in the above embodiment. The RAM 303 also stores various programs and data required for system operation. The CPU 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0099] The following components are connected to I / O interface 305: input section 306 including audio input devices, push-button switches, etc.; output section 307 including a liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 308 including a hard disk, etc.; and communication section 309 including a network interface card such as a LAN (Local Area Network) card, modem, etc. Communication section 309 performs communication processing via a network such as the Internet. Drive 310 is also connected to I / O interface 305 as needed. Removable media 311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 310 as needed so that computer programs read from them can be installed into storage section 308 as needed.

Claims

1. A method for early warning of atopic dermatitis risk based on microplastic exposure monitoring, characterized in that, The method includes: A skin contact signal sequence was acquired, and a convolutional neural network was used to extract features from the skin contact signal sequence to determine the dynamic distribution pattern of microplastic particles on the skin surface. The dynamic distribution pattern is fused with historical exposure data records to calculate the cumulative effect index, and a skin barrier interference risk score is generated based on the cumulative effect index. When the skin barrier interference risk score is higher than a preset risk threshold, real-time skin inflammation marker data is acquired, and the real-time skin inflammation marker data is processed by a support vector machine to determine the probability of potential disease induction. A continuous tracking sequence is generated based on the potential disease induction probability. Risk compensation weights are obtained by identifying mutation signals in the continuous tracking sequence to correct risk measurement bias. The risk compensation weights are used to weight and map the probabilities of potential disease occurrence to generate a target risk distribution map. Accumulated features are extracted from the target risk distribution map, risk warning instructions are generated based on the accumulated features, and warnings are issued according to the risk warning instructions.

2. The method according to claim 1, characterized in that, The step of using a convolutional neural network to extract features from the skin contact signal sequence to determine the dynamic distribution pattern of microplastic particles on the skin surface includes: The skin contact signal sequence is subjected to joint analysis in the time and frequency domains to obtain a multidimensional spatiotemporal feature vector; The multidimensional spatiotemporal feature vector is input into the convolutional layer of the convolutional neural network for feature mapping to extract local concentration features; Concentration gradient evolution map is constructed based on the local concentration features; Calculate the topological difference matrix of the concentration gradient evolution map in adjacent time frames, and determine the dynamic distribution pattern of microplastic particles on the skin surface based on the topological difference matrix.

3. The method according to claim 2, characterized in that, The step of inputting the multidimensional spatiotemporal feature vector into the convolutional layer of a convolutional neural network for feature mapping and extracting local concentration features includes: A multi-scale convolutional kernel is used to perform a sliding window scan on the multi-dimensional spatiotemporal feature vector to generate a multi-scale receptive field feature map. The multi-scale receptive field feature map is subjected to nonlinear activation processing to obtain the activation response value in the feature space; Spatial pooling dimensionality reduction is performed based on the activation response value to obtain the local concentration feature.

4. The method according to claim 3, characterized in that, The step of performing spatial pooling dimensionality reduction based on the activation response value to obtain the local concentration features includes: Calculate the information entropy of the activation response value in different channel dimensions, and determine the channel attention weights based on the information entropy; The activation response values ​​are weighted using the channel attention weights to obtain a weighted response matrix; The weighted response matrix is ​​subjected to global average pooling to output the local concentration features.

5. The method according to claim 1, characterized in that, The process of fusing the dynamic distribution pattern with historical exposure data records, calculating a cumulative effect index, and generating a skin barrier disturbance risk score based on the cumulative effect index includes: Align the dynamic distribution pattern with the historical exposure data records along the timeline to extract the historical cumulative exposure amount; The cumulative effect index characterizing the degree of skin barrier damage is calculated by weighting and integrating the historical cumulative exposure amount according to a preset time decay factor. Calculate the degree of deviation between the cumulative effect index and the preset inflammation trigger threshold; The degree of deviation is mapped to a risk score mapping table to obtain the corresponding skin barrier interference risk score.

6. The method according to claim 1, characterized in that, The process of acquiring real-time skin inflammation marker data and processing the data using a support vector machine to determine the probability of potential disease induction includes: Real-time skin inflammation biomarker data, including skin temperature and transdermal water loss rate, are collected using physiological monitoring sensors. The real-time skin inflammation biomarker data are mapped to a multi-dimensional feature space to construct an inflammation feature vector; The inflammation feature vector is input into a support vector machine to calculate the distance from the inflammation feature vector to the classification boundary of the target classification hyperplane; The classification boundary distance is converted into an interval probability value by nonlinear exponential mapping to determine the potential disease induction probability.

7. The method according to claim 6, characterized in that, The step of mapping the real-time skin inflammation biomarker data to a multi-dimensional feature space to construct an inflammation feature vector includes: Extract the skin temperature time series and transdermal water loss rate time series from the real-time skin inflammation marker data, respectively. Calculate the covariance matrix of the skin temperature time series and the transdermal water loss rate time series, and extract the eigenvalues ​​of the covariance matrix to obtain the cross-eigenvalues ​​characterizing the imbalance state of skin hydrothermal coupling. Fourier transform was performed on the skin temperature time series to extract the frequency band energy spectrum characterizing the subcutaneous microcirculation metabolic state; The cross-feature values ​​and the frequency band energy spectrum are vector-concatenated to construct the inflammation feature vector.

8. The method according to claim 1, characterized in that, The process of generating a continuous tracking sequence based on the potential disease induction probability, correcting risk assessment bias by identifying mutation signals in the continuous tracking sequence, and obtaining risk compensation weights includes: Record the probability of potential disease induction according to a preset sampling frequency to generate a continuous tracking sequence that reflects the trajectory of risk evolution; Extreme value detection is performed on the continuously tracked sequence to identify abrupt signals in the sequence that deviate from the baseline fluctuation range; Obtain the sampling density corresponding to the mutation signal. When the sampling density exceeds the sampling density reference range, determine that there is a risk of measurement deviation and determine the measurement correction coefficient based on the deviation rate of the sampling density exceeding the sampling density reference range. The initial risk assessment parameters are corrected using the calculated correction coefficient to obtain the risk compensation weight.

9. The method according to claim 1, characterized in that, The step of extracting accumulated features based on the target risk distribution map, generating risk warning instructions based on the accumulated features, and issuing warnings according to the risk warning instructions includes: The target monitoring area is determined based on the target risk distribution map, and the spatial heterogeneity gradient vector is extracted from the distribution data corresponding to the target monitoring area. The spatial heterogeneity gradient vector is normalized to construct the accumulated feature; Calculate the target similarity between the accumulated features and each reference pattern in the preset skin reaction triggering pattern library; When the target similarity reaches a preset matching threshold, it is confirmed that the accumulated features are associated with and matched with the skin reaction, and a risk warning instruction containing the corresponding reference pattern is generated, and an early warning is issued according to the risk warning instruction.

10. An atopic dermatitis risk early warning system based on microplastic exposure monitoring, characterized in that, The atopic dermatitis risk warning system based on microplastic exposure monitoring includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the atopic dermatitis risk warning system based on microplastic exposure monitoring to perform the method as described in any one of claims 1-9.

Images