A data processing method and system based on wound image recognition

Abstract

The present application relates to the technical field of image recognition, in particular to a data processing method and system based on wound image recognition, comprising the following steps: collecting signals to construct a multi-dimensional photoelectric pixel matrix, calculating global light distribution weight and generating a standardized tissue feature map, extracting deep necrotic texture vectors and granulation metabolic activity index, generating an infection probability distribution matrix, constructing an irregular geometric contour set and a wound edge shrinkage coefficient, and generating adaptive diagnosis and treatment intervention instructions and a final wound evaluation report through a healing trend prediction model.In the present application, by constructing a local texture feature matrix and its difference mapping with a spatial geometric vector, the specific physiological state weight is adaptively calculated, the interference of environmental background noise on the edge recognition process is eliminated, the calculation accuracy of the target region boundary is improved, and the high reliability of the final classification data output under variable light conditions is ensured.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a data processing method and system based on wound image recognition. Background Technology

[0002] Image recognition technology involves using computers to process, analyze, and understand images to identify targets and objects of various patterns. This technology is widely used in medical diagnosis, industrial inspection, and security monitoring. Its core lies in extracting feature information from images through mathematical algorithms and classifying or describing the image content based on these features, thereby achieving a mapping transformation from image data to physical meaning. Traditional wound image recognition data processing methods involve acquiring wound images with an optical camera, converting them into grayscale images or HSV color space data, using global threshold segmentation algorithms to extract the wound area, and calculating the wound area based on the number of pixels or using color histograms to statistically analyze the proportion of granulation tissue.

[0003] Existing technologies rely on global threshold segmentation logic to extract target regions with significant pixel differences, but ignore the nonlinear coupling relationship between local texture details and overall spatial geometry under complex lighting conditions. This results in the inability to effectively eliminate the interference of environmental noise on the feature extraction process, leading to low edge boundary definition accuracy and poor robustness of tissue classification results. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and to propose a data processing method and system based on wound image recognition.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a data processing method based on wound image recognition, comprising the following steps: S1: Receive the wound reflection signal collected by the sensor, construct a multi-dimensional photoelectric pixel matrix, analyze the brightness dispersion of the rows and columns of the multi-dimensional photoelectric pixel matrix, calculate the global illumination distribution weight, perform denoising and smoothing processing on the multi-dimensional photoelectric pixel matrix according to the global illumination distribution weight, and construct a standardized tissue feature map. S2: Perform color space transformation on the standardized tissue feature map, extract the deep necrosis texture vector, calculate the difference ratio between the red channel and the blue channel in the standardized tissue feature map, generate the granulation tissue metabolic activity index, and map the deep necrosis texture vector to a high-dimensional feature space through a logistic regression algorithm to generate an infection probability distribution matrix. S3: Identify pixel gradient abrupt change points in the standardized tissue feature map, construct an irregular geometric contour set, calculate the curvature change of each discrete coordinate point in the irregular geometric contour set, generate the wound edge contraction coefficient, and use the infection probability distribution matrix as a constraint condition to perform weighted fusion on the wound edge contraction coefficient to construct a healing trend prediction model. S4: Input the granulation tissue metabolic activity index into the healing trend prediction model to perform iterative calculations, determine the healing stage based on the calculation results, generate adaptive treatment intervention instructions, send the adaptive treatment intervention instructions to the central control unit for parsing, match the historical treatment database, and generate a wound assessment report based on the parsing results.

[0006] As a further aspect of the present invention, step S1 specifically comprises: S11: Obtain the multi-dimensional photoelectric pixel matrix, traverse the brightness values ​​of each pixel in the matrix in the horizontal and vertical directions, calculate the brightness variance values ​​of each row and column using a preset discrete statistical algorithm, construct an illumination non-uniformity mapping table based on the brightness variance values, and generate global illumination distribution weights through reverse normalization processing. S12: Call the global illumination distribution weights, combine them with the adaptive Gaussian filtering algorithm to construct a dynamic filtering kernel, use the dynamic filtering kernel to perform weighted smoothing suppression on high-frequency noise points in the multi-dimensional photoelectric pixel matrix, retain low-frequency tissue texture information, and generate a standardized tissue feature map.

[0007] As a further aspect of the present invention, step S2 specifically comprises: S21: Obtain the standardized tissue feature map, convert it from RGB color space to LAB color space, separate the luminance channel component and chrominance channel component, use a multi-scale Gabor filter bank to perform convolution operation on the luminance channel component, capture tissue texture features of different directions and frequencies, and generate deep necrosis texture vector. S22: Obtain the intensity values ​​of the red channel and the blue channel in the standardized tissue feature map, perform differential ratio calculation to generate the granulation tissue metabolic activity index, and at the same time use the logistic regression algorithm to classify and map the deep necrosis texture vector, calculate the confidence value of each pixel belonging to the necrotic tissue category, and generate an infection probability distribution matrix.

[0008] As a further aspect of the present invention, step S3 specifically comprises: S31: Use the Sobel operator to scan the standardized tissue feature map, detect the gradient change amplitude of pixel gray values, filter out pixels with gradient amplitude exceeding the preset edge threshold as edge candidate points, and connect broken edge segments through morphological closing operation to construct an irregular geometric contour set. S32: Traverse each discrete coordinate point in the set of irregular geometric contours, calculate the rate of change of the slope of the tangent line formed by the point and its adjacent coordinate points, calculate the mean and variance of the curvature of all coordinate points, and generate the wound edge contraction coefficient. S33: Obtain the infection probability distribution matrix as a penalty term, perform nonlinear weighted correction on the wound edge contraction coefficient, and establish a healing trend prediction model by combining time series analysis algorithm.

[0009] As a further aspect of the present invention, step S4 specifically comprises: S41: The granulation tissue metabolic activity index is used as the initial input variable and imported into the healing trend prediction model. The simulation time step and convergence conditions are set, and multiple rounds of time series simulation are performed to predict the granulation tissue growth rate and area change rate within a specific time window in the future, and to generate the healing stage determination result. S42: Based on the healing stage determination result, search the preset clinical intervention strategy library, match the drug throughput, light therapy intensity and debridement frequency parameters that are appropriate for the current healing stage, and generate adaptive diagnosis and treatment intervention instructions; S43: Transmit the adaptive treatment intervention command to the central control unit, parse the parameter characteristics in the command, search for historical case data with similar parameter characteristics in the historical treatment database, compare and analyze the differences in the recovery trajectory between the current case and the historical cases, and generate a wound assessment report.

[0010] As a further aspect of the present invention, the process of generating the global illumination distribution weights specifically includes: Obtain the first pixel in the multidimensional photoelectric pixel matrix. line and number The pixel brightness values ​​of the column are used to calculate the average brightness threshold of the entire matrix, and the set of dark area pixels with brightness values ​​lower than the average brightness threshold and the set of bright area pixels with brightness values ​​higher than the average brightness threshold are selected. Calculate the area ratio coefficient between the set of pixels in the dark area and the set of pixels in the bright area, construct an illumination compensation function using the area ratio coefficient, map the output value of the illumination compensation function to a value between 0 and 1, and generate a global illumination distribution weight.

[0011] As a further aspect of the present invention, the process of generating the granulation tissue metabolic activity index specifically includes: Obtain the red channel values ​​of the pixels in the target region of the standardized tissue feature map. Green channel values and the blue channel value ; Based on the absorption characteristics of hemoglobin to light of a specific wavelength, a colorimetric index calculation model is constructed. The red channel values ​​are nonlinearly amplified using the model, and the blue channel values ​​are normalized and attenuated to generate a granulation tissue metabolic activity index.

[0012] As a further aspect of the present invention, the construction process of the healing trend prediction model specifically includes: Obtain historical time-series data of the wound edge contraction coefficient, the average infection probability value of the infection probability distribution matrix, and the granulation tissue metabolic activity index; A healing dynamics evolution formula based on differential equations is constructed, and the undetermined coefficients in the formula are fitted and solved using the historical time series data to establish a healing trend prediction model. The mathematical expression for the healing trend prediction model is: ; in, represent Predicted wound healing rate at time point. Represents the tissue regeneration capacity constant. This represents the shrinkage coefficient at the edge of the wound. Represents bacterial blocking factors. This represents the average probability value of infection. Represents the metabolic drive coefficient. represent The granulation tissue metabolic activity index at that time.

[0013] As a further aspect of the present invention, the process of generating the adaptive diagnostic and therapeutic intervention instruction specifically includes: Obtain the physiological stage category of the current wound as identified in the healing stage determination result, including the inflammatory stage, proliferative stage, or remodeling stage; If the inflammation phase is determined, the coordinates of high-risk areas in the infection probability distribution matrix are extracted to generate targeted debridement path instructions and anti-infective agent spraying dosage instructions. If the condition is determined to be in the proliferative phase, the wavelength and irradiation duration parameters of the biostimulation light device are adjusted according to the value of the granulation tissue metabolic activity index to generate a physical therapy instruction. The physical therapy instruction is then combined with the drug spraying dosage instruction to generate an adaptive diagnostic and treatment intervention instruction.

[0014] A data processing system based on wound image recognition, the system being used to implement the aforementioned data processing method based on wound image recognition, the system comprising: The image preprocessing module is used to receive the wound reflection signal collected by the sensor, construct a multi-dimensional photoelectric pixel matrix, analyze the brightness dispersion of the rows and columns of the multi-dimensional photoelectric pixel matrix, calculate the global illumination distribution weight, perform denoising and smoothing processing on the multi-dimensional photoelectric pixel matrix according to the global illumination distribution weight, and construct a standardized tissue feature map. The feature extraction and analysis module is used to perform color space transformation on the standardized tissue feature map, extract the deep necrosis texture vector, calculate the difference ratio between the red channel and the blue channel in the standardized tissue feature map, generate the granulation tissue metabolic activity index, and map the deep necrosis texture vector to a high-dimensional feature space through a logistic regression algorithm to generate an infection probability distribution matrix. The model building and prediction module is used to identify pixel gradient abrupt change points in the standardized tissue feature map, construct an irregular geometric contour set, calculate the curvature change of each discrete coordinate point in the irregular geometric contour set, generate the wound edge contraction coefficient, and use the infection probability distribution matrix as a constraint condition to perform weighted fusion on the wound edge contraction coefficient to construct a healing trend prediction model. The diagnosis and treatment decision and feedback module is used to input the granulation tissue metabolic activity index into the healing trend prediction model to perform iterative calculations, determine the healing stage based on the calculation results, generate adaptive diagnosis and treatment intervention instructions, send the adaptive diagnosis and treatment intervention instructions to the central control unit for parsing, match the historical diagnosis and treatment database, and generate a wound assessment report based on the parsing results.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by constructing a local texture feature matrix and its differential mapping with spatial geometric vectors, the weights of specific physiological states are adaptively calculated, eliminating the interference of environmental background noise on the edge recognition process, improving the calculation accuracy of the target region boundary, and ensuring the high reliability of the final classification data output under varying illumination conditions. Attached Figure Description

[0016] Figure 1 This is the main flowchart of the data processing method based on wound image recognition of the present invention; Figure 2 This is a flowchart illustrating the standardized organizational feature map generation process of this invention. Figure 3 This is a flowchart illustrating the generation process of the infection probability distribution matrix and metabolic index in this invention. Figure 4 Flowchart for constructing the healing trend prediction model of this invention; Figure 5 This is a flowchart of the adaptive diagnosis and treatment intervention and evaluation report generation process of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the software-based technical solution is described in detail below with reference to system architecture diagrams and embodiments. It should be understood that the specific embodiments described herein are only for explaining the technical solutions of this invention and do not constitute a limitation on the scope of protection.

[0018] In the description of this invention, the system architecture relationships or data processing flows indicated by terms such as "layer," "module," "interface," "data flow," "client," and "server" are all defined based on the architecture diagram or flowchart corresponding to the embodiments. This way of describing is only used to clearly illustrate the logical relationships between the elements in the technical solution, and not to limit the physical deployment form. The term "multiple" includes two or more technical units, including but not limited to multiple data nodes, processing threads, service instances, or functional components and other scalable elements. The specific number is determined according to the actual business scenario and needs to be specifically specified.

[0019] Please see Figure 1 and Figure 2 This invention provides a technical solution: a data processing method based on wound image recognition, comprising the following steps: S1: Receive the wound reflection signal collected by the sensor, construct a multi-dimensional photoelectric pixel matrix, analyze the brightness dispersion of the rows and columns of the multi-dimensional photoelectric pixel matrix, calculate the global illumination distribution weight, perform denoising and smoothing processing on the multi-dimensional photoelectric pixel matrix according to the global illumination distribution weight, and construct a standardized tissue feature map.

[0020] The specific steps of S1 are as follows: S11: Obtain the multi-dimensional photoelectric pixel matrix, traverse the brightness values ​​of each pixel in the horizontal and vertical directions, calculate the brightness variance values ​​of each row and column using the preset discrete statistical algorithm, construct the illumination non-uniformity mapping table based on the brightness variance values, and generate global illumination distribution weights through reverse normalization processing. S12: Call the global illumination distribution weights and combine them with the adaptive Gaussian filtering algorithm to construct a dynamic filtering kernel. Use the dynamic filtering kernel to perform weighted smoothing suppression on high-frequency noise points in the multi-dimensional photoelectric pixel matrix, retain low-frequency tissue texture information, and generate a standardized tissue feature map.

[0021] The process of generating global illumination distribution weights specifically includes: Obtain the first pixel in the multidimensional photoelectric pixel matrix line and number The pixel brightness values ​​of the column are used to calculate the average brightness threshold of the entire matrix, and the set of dark area pixels with brightness values ​​lower than the average brightness threshold and the set of bright area pixels with brightness values ​​higher than the average brightness threshold are selected. Calculate the area ratio coefficients of the pixel set in the dark area and the pixel set in the bright area, construct a lighting compensation function using the area ratio coefficients, map the output value of the lighting compensation function to a value between 0 and 1, and generate global lighting distribution weights.

[0022] Upon receiving a trigger signal, the high-resolution CMOS image sensor initiates photoelectric signal acquisition of the target wound area. The sensor converts the captured analog light signal into a 14-bit digital signal using a built-in analog-to-digital converter, arranges it in a Bayesian array format, and then interpolates using a demosaicing algorithm to generate a multi-dimensional photoelectric pixel matrix with a resolution of 4096×2160. This matrix is ​​stored in a high-speed cache memory, defined as... At this point, the system immediately initiates the brightness dispersion analysis process for the matrix. The read pointer scans row by row and column by column. Each pixel in Extract its luminance component The brightness component here Instead of directly taking the RGB average, it uses a weighted formula based on the characteristics of human vision. The calculation yielded the result.

[0023] After obtaining the brightness values ​​of all components, a sliding window of size 512×512 is selected and traversed on the matrix. For the pixel set within each window, the row-direction brightness variance is calculated using the variance statistics formula. Column direction brightness variance The specific calculation logic is as follows: First, calculate the arithmetic mean of the brightness of all pixels within the window. Then, sum the brightness of each pixel and... The square of the difference is then divided by the total number of pixels. If the calculated variance exceeds a preset discrete threshold of 150, the region is considered to have uneven illumination. Based on the variance calculation results of all windows, an illumination unevenness mapping table with the same dimension as the original matrix is ​​constructed. For regions marked as high variance in the mapping table, inverse normalization is applied. Specifically, the maximum variance value in the mapping table is denoted as... The variance at the current point is denoted as Then the initial weight of that point Calculated as This is used to generate the basic global illumination distribution weight matrix.

[0024] The aforementioned discrete threshold refers to 1.5 times the baseline value of brightness variance obtained by taking multiple photos of a standard reflectivity gray card under a standard D65 light source environment. It is used to distinguish between normal lighting fluctuations and abnormal shadow occlusion.

[0025] Next, the program calls the global illumination distribution weight matrix to enter the denoising stage. To balance denoising effectiveness with texture preservation, an adaptive Gaussian filtering algorithm is used. For each pixel in the matrix, its corresponding weight value is used... Dynamically adjust the standard deviation of the Gaussian kernel When the weight value When the standard deviation is low, a larger standard deviation and a larger filter kernel size are selected to enhance the smoothing effect; conversely, when the standard deviation is high, a larger standard deviation and a larger filter kernel size are selected to enhance the smoothing effect. When the noise level is high, a smaller standard deviation and a 3×3 filter kernel are selected to preserve tissue details to the greatest extent. During convolution, the center of the dynamically constructed filter kernel is aligned with the current pixel, and the weighted sum of all pixels within the kernel is calculated as the new pixel value for that point. After full matrix traversal processing, high-frequency noise is effectively suppressed, while low-frequency muscle texture and blood vessel orientation information of the wound are preserved. The final output data structure is the standardized tissue feature map.

[0026] To further refine the weight generation, the program performs pixel-level brightness threshold filtering. This involves traversing the multi-dimensional photoelectric pixel matrix... line and number The column counts the brightness values ​​of approximately 8.8 million pixels across the entire matrix, and calculates the average brightness threshold for the entire matrix. .by Based on this, all pixels are divided into two sets: those with a brightness value less than 1. The pixels are assigned to the dark area pixel set. Brightness value greater than or equal to The pixels are assigned to the pixel set of the bright area. .statistics Number of pixels and Number of pixels Calculate the area ratio of the dark area. .

[0027] A lighting compensation function is constructed based on this proportion coefficient. This function employs an S-curve design to nonlinearly correct for extreme lighting conditions. Specifically, the function is designed as a logarithmic transformation, with the original brightness value as input and the compensated normalized coefficient as output. The output value of the lighting compensation function is strictly mapped to a closed interval between 0 and 1. For example, for cases where the dark areas constitute a large proportion, the weight value of the function output is shifted closer to 1 to enhance signal gain; for overexposed bright areas, the weight is appropriately reduced. The final generated global lighting distribution weights not only include spatial dispersion information but also incorporate overall lighting tendency information, ensuring the robustness of subsequent processing.

[0028] Please see Figure 1 and Figure 3S2: Perform color space transformation on the standardized tissue feature map, extract the deep necrosis texture vector, calculate the difference ratio between the red and blue channels in the standardized tissue feature map, generate the granulation metabolic activity index, and map the deep necrosis texture vector to the high-dimensional feature space through the logistic regression algorithm to generate the infection probability distribution matrix.

[0029] The specific steps of S2 are as follows: S21: Obtain the standardized tissue feature map, convert it from the RGB color space to the LAB color space, separate the luminance channel component and the chrominance channel component, use the multi-scale Gabor filter bank to perform convolution operation on the luminance channel component, capture tissue texture features of different directions and frequencies, and generate deep necrosis texture vector. S22: Obtain the intensity values ​​of the red and blue channels in the standardized tissue feature map, perform differential ratio calculation to generate the granulation tissue metabolic activity index, and use the logistic regression algorithm to classify and map the deep necrotic texture vector, calculate the confidence value of each pixel belonging to the necrotic tissue category, and generate the infection probability distribution matrix.

[0030] The process of generating the granulation tissue metabolic activity index specifically includes: Obtain the red channel values ​​of pixels in the target region from the standardized tissue feature map. Green channel values and the blue channel value ; Based on the absorption characteristics of hemoglobin to light of a specific wavelength, a colorimetric index calculation model was constructed. The red channel values ​​were nonlinearly amplified by the model, and the blue channel values ​​were normalized and attenuated to generate a granulation tissue metabolic activity index.

[0031] After obtaining the standardized organizational feature map, the first step is to convert the color space. Since color and brightness information are highly coupled in the RGB color space, which is not conducive to independent texture analysis, a nonlinear transformation is used to convert the image data to the CIELAB color space. The conversion process consists of two steps: first, the RGB values ​​are normalized and converted to XYZ tristimulus values ​​based on the white point coordinates of the D65 standard light source; then, a nonlinear cube root transform is used to map the XYZ values ​​to three channels: L, a, and b. The separated L channel component is extracted as a single-channel grayscale matrix. This is used for subsequent texture feature capture.

[0032] against A matrix is ​​used to construct a multi-scale Gabor filter bank. This filter bank contains 5 different scales and 8 different directions. These 40 specific Gabor kernels are then used to... Convolution operations are performed. During convolution, the filter accumulates data with image pixels, extracting texture features with high response at specific frequencies and directions. For each pixel, its values ​​in the 40 filtered response maps are combined into a 40-dimensional feature vector, i.e., the deep necrotic texture vector. These vectors can keenly characterize the disordered, rough, and fractured texture morphology unique to necrotic tissue.

[0033] The aforementioned Gabor filter bank refers to a set of linear filters that simulate the receptive field characteristics of simple cells in the human visual cortex. Its core is composed of a Gaussian kernel function modulated by a sine wave, which can effectively separate the texture details of an image at different spatial frequencies and directions.

[0034] In processing chromaticity information, the program synchronously reads the red channel intensity value of the normalized tissue feature map in the original RGB space. Compared with the blue channel intensity value Considering that granulation tissue is rich in newly formed capillaries, its hemoglobin has a high reflectivity to the red spectrum, while necrotic or infected tissue often appears dark red or bluish-purple, with a relatively high blue component, a difference ratio calculation is performed on each pixel of the entire image.

[0035] To accurately quantify infection risk, a logistic regression algorithm was used to classify and map the generated deep necrosis texture vectors. This logistic regression model was pre-trained on a database containing 5000 labeled wound samples. During training, binary cross-entropy was used as the loss function, and the Adam optimizer was used to iteratively update the model's weight matrix with a learning rate of 0.001. and bias terms In the current inference phase, the extracted 40-dimensional texture vector will be... Input model, calculate And through the Sigmoid activation function The output values ​​are mapped to probability values ​​between 0 and 1. This probability value directly represents the confidence level that the pixel belongs to necrotic tissue. After calculating this value by traversing all pixels in the entire image, the infection probability distribution matrix is ​​generated.

[0036] To calculate the granulation tissue metabolic activity index, the program first identifies the target region within the standardized tissue feature map. For any pixel within this region, it reads the value of its red channel. Green channel values and the blue channel value Based on the high reflectivity of oxyhemoglobin in the 600-700 nm wavelength range and its absorption characteristics in the 400-500 nm wavelength range, a colorimetric index calculation model was constructed. In the model, the red channel values ​​were given an exponential gain to highlight vascular richness; the blue channel values ​​were normalized and attenuated as either a denominator or a subtrahend term to suppress interference from venous congestion or necrotic tissue. The final granulation tissue metabolic activity index was generated. It is a dimensionless value. The higher the value, the more vigorous the metabolism of granulation tissue in that area and the greater the healing potential.

[0037] Please see Figure 1 and Figure 4 S3: Identify pixel gradient abrupt change points in the standardized tissue feature map, construct an irregular geometric contour set, calculate the curvature change of each discrete coordinate point in the irregular geometric contour set, generate the wound edge contraction coefficient, and use the infection probability distribution matrix as a constraint to perform weighted fusion on the wound edge contraction coefficient to construct a healing trend prediction model.

[0038] The specific steps for S3 are as follows: S31: Use the Sobel operator to scan the standardized tissue feature map, detect the gradient change of pixel gray values, select pixels with gradient values ​​exceeding the preset edge threshold as edge candidate points, and connect broken edge segments through morphological closing operations to construct an irregular geometric contour set. S32: Traverse each discrete coordinate point in the set of irregular geometric contours, calculate the rate of change of the slope of the tangent line formed by the point and its adjacent coordinate points, calculate the mean and variance of the curvature of all coordinate points, and generate the wound edge contraction coefficient. S33: Obtain the infection probability distribution matrix as a penalty term, perform nonlinear weighted correction on the wound edge contraction coefficient, and establish a healing trend prediction model by combining time series analysis algorithm.

[0039] The construction process of the healing trend prediction model specifically includes: Historical time-series data of wound edge contraction coefficient, average infection probability value of infection probability distribution matrix, and granulation tissue metabolic activity index were obtained. A healing dynamics evolution formula based on differential equations was constructed, and the undetermined coefficients in the formula were fitted and solved using historical time series data to establish a healing trend prediction model. The mathematical expression for the healing trend prediction model is: ; in, represent Predicted wound healing rate at time point. Represents the tissue regeneration capacity constant. Represents the wound edge contraction coefficient. Represents bacterial blocking factors. This represents the average probability of infection. Represents the metabolic drive coefficient. represent The granulation tissue metabolic activity index at any given time.

[0040] The Sobel operator is used to perform a full-range gradient scan on a grayscale copy of the normalized organization feature map. The Sobel operator includes horizontal convolution kernels. and vertical convolution kernel Specifically, it takes the form of a 3×3 matrix. These two kernels are then slid across the image, and the grayscale change rate of each pixel in the horizontal and vertical directions is calculated. Furthermore, the formula is used to... Calculate the gradient magnitude by Calculate the gradient direction. Set the edge gradient threshold to 45, and select pixels with gradient magnitudes greater than 45 as candidate edge points. Due to noise or tissue fluid interference, the initially extracted edges are often discontinuous. Therefore, morphological closing operations are used to process the binarized edge image. Using circular structuring elements with a radius of 3 pixels, broken edge fragments are connected, and small holes are filled, ultimately extracting a continuous sequence of coordinate points to form a set of irregular geometric contours.

[0041] The aforementioned morphological closing operation refers to a combination of image processing operations that first perform a dilation operation on the image to fill the small holes in the target area, and then perform an erosion operation to restore the original shape and size of the target. It is often used to smooth contours and connect broken line segments.

[0042] The program then iterates through each discrete coordinate point in the contour set. To calculate curvature, the fifth adjacent point before and after the current point is selected. and Use the coordinates of these three points to estimate the rate of change of the tangent slope at that point. Calculate the arithmetic mean of the curvature values ​​at all points on the contour. and variance Because wound edges typically shrink from irregular to smooth during healing, high curvature variance often corresponds to jagged edges in the early stages of healing, while low variance corresponds to smooth edges in the later stages. and The dimensionless wound edge contraction coefficient is generated by weighted summation. The closer the coefficient is to 1, the smoother the edge and the more active the contraction; the closer it is to 0, the more stagnant the edge or the hardened, rolled-up edge caused by chronic ulceration.

[0043] Before establishing the healing model, an infection probability distribution matrix is ​​introduced as a key constraint. The average infection probability value of the target region in the matrix is ​​calculated. This is considered an obstacle to healing. Simultaneously, it is combined with the granulation tissue metabolic activity index generated in the preceding steps. Historical time-series data. Time series analysis algorithms are used for analysis. Trend items.

[0044] Regarding the specific construction and parameter fitting of the healing trend prediction model, the necessary physical parameters were first defined. Through 14 days of continuous monitoring of the patient's wound, the edge contraction coefficient, infection probability, and metabolic index were obtained at each time point. The specific mathematical expression of the healing trend prediction model is as follows: ; in, represent Predicted wound healing rate at any given time, in square millimeters per day; The tissue regeneration capacity constant represents the basic healing rate of an individual patient under undisturbed conditions, and is usually affected by age and basal metabolic rate. Represents the wound edge contraction coefficient, used to reflect the physical contractile activity of the edge tissue; is the base of the natural logarithm; Representing bacterial blocking factors, used to quantify the exponential inhibitory effect of infection severity on cell migration; This represents the average probability of infection, ranging from 0 to 1. Represents the metabolic driving coefficient, used to measure the weight of the contribution of granulation tissue metabolic activity to the healing rate; Represents the natural logarithm operation; represent The granulation tissue metabolic activity index at a given time is used to characterize the biochemical synthesis capacity within the tissue.

[0045] The parameters obtained above (Value is 2.5) (Value is 0.82) (Value is 1.2) (Value is 0.35) (Value is 1.8) and Substituting the value (0.95) into the above formula, calculate the predicted healing rate. .

[0046] The specific calculation is explained as follows: The first part is the edge contraction term. The calculation process involves multiplying the tissue regeneration capacity constant 2.5 by the wound edge contraction coefficient 0.82, and then multiplying by the base of the natural logarithm. The base is -1.2 raised to the power of 0.35. The calculation yields... The index part is ,because Approximately equal to 0.6570, therefore the first part of the result is approximately... The second part is the metabolic driving term, calculated by multiplying the metabolic driving coefficient (1.8) by 1 and adding the natural logarithm of the granulation tissue metabolic activity index (0.95). That is, the calculation... .because Approximately equal to 0.6678, therefore the result for the second part is approximately... Final result It is the sum of two parts, that is Square millimeters per day.

[0047] These results indicate that, given the current edge condition, infection level, and metabolic activity, the wound is projected to heal at a rate of approximately 2.55 square millimeters per day. This will serve as the core basis for determining the healing stage in the subsequent S4 step. The advantage of this formula lies in the fact that it simulates the nonlinear inhibitory effect of infection on healing through an exponential decay term, while simulating the diminishing marginal returns of increased metabolic activity through a logarithmic term. This design is highly consistent with biological laws and significantly improves the accuracy of prediction.

[0048] Table 1. Wound Feature Parameters and Model Calculation Table; See Table 1, which lists the implementation data and calculation results of the key parameters in steps S1 to S3, verifying the rationality of the model input parameters and the accuracy of the calculation results.

[0049] Please see Figure 1 and Figure 5 S4: Input the granulation tissue metabolic activity index into the healing trend prediction model to perform iterative calculations, determine the healing stage based on the calculation results, generate adaptive treatment intervention instructions, send the adaptive treatment intervention instructions to the central control unit for parsing, match the historical treatment database, and generate a wound assessment report based on the parsing results.

[0050] The specific steps for S4 are as follows: S41: The granulation tissue metabolic activity index is used as the initial input variable to import the healing trend prediction model. The simulation time step and convergence conditions are set, and multiple rounds of time series simulation are performed to predict the granulation tissue growth rate and area change rate within a specific time window in the future, and to generate the healing stage determination result. S42: Based on the healing stage determination result, search the preset clinical intervention strategy library, match the drug throughput, light therapy intensity and debridement frequency parameters that are appropriate for the current healing stage, and generate adaptive treatment intervention instructions; S43: Transmit adaptive treatment intervention instructions to the central control unit, parse the parameter characteristics in the instructions, search for historical case data with similar parameter characteristics in the historical treatment database, compare and analyze the differences in the recovery trajectory between the current case and historical cases, and generate a wound assessment report.

[0051] The process of generating adaptive diagnostic and treatment intervention instructions specifically includes: Obtain the physiological stage category of the current wound as identified in the healing stage assessment results, including the inflammatory stage, proliferative stage, or remodeling stage; If the inflammation phase is determined, the coordinates of high-risk areas in the infection probability distribution matrix are extracted to generate targeted debridement path instructions and anti-infective agent spraying dosage instructions. If the condition is determined to be in the proliferative phase, the wavelength and irradiation duration parameters of the biostimulation light equipment are adjusted according to the value of the granulation tissue metabolic activity index to generate a physical therapy instruction. This physical therapy instruction is then combined with the drug spraying dosage instruction to generate an adaptive diagnostic and treatment intervention instruction.

[0052] The program will use the granulation tissue metabolic activity index sequence calculated in real time and the healing rate predicted by step S3. The healing trend prediction model was imported for iterative simulation. The simulation time step was set to 0.5 days, and the total simulation window was set to the next 14 days. The fourth-order Runge-Kutta method was used to numerically solve the differential equations to predict the residual wound area at each future time step. Changes in healing rate.

[0053] Based on the prediction results, the system automatically determines the healing stage. The system has built-in clear classification criteria for physiological periods: if the predicted healing stage is... It shows negative growth or low-level fluctuations with values ​​less than 0.5 square millimeters per day, while A value greater than 0.4 indicates an inflammatory phase; if It continues to rise and stabilize at a high level, with values ​​greater than 2.0 square millimeters per day. Maintaining a high level indicates the proliferative phase; if The wound area gradually decreases, but has shrunk to less than 10% of its initial size, indicating a remodeling phase. In this embodiment, based on the aforementioned calculations... The value is 2.549 square millimeters per day, and The value is 0.35, indicating that the current wound is in the early to mid-stage of the proliferative phase.

[0054] Based on the determination result, the system immediately retrieves the clinical intervention strategy library. This strategy library is stored in the form of a lookup table. For the proliferative phase determination, the basic instruction set matched by the strategy library includes: maintaining a moderately humid environment, increasing phototherapy to promote collagen synthesis, and reducing the frequency of use of strong antibiotics to avoid cytotoxicity. Specifically, the system generates a drug dosage instruction, setting the spray density of the anti-infective agent to a low level, i.e., 0.2 ml per square centimeter, and simultaneously generates a physical therapy instruction, adjusting the wavelength of the biostimulation light irradiation device to 633 nanometers, and setting the irradiation duration parameter to once every 20 minutes, twice daily.

[0055] The aforementioned biostimulation light irradiation device refers to a medical auxiliary device that uses light-emitting diodes or laser diodes of specific wavelengths as light sources to stimulate mitochondrial activity and promote adenosine triphosphate synthesis through non-thermal photobiological regulation.

[0056] Finally, the generated adaptive treatment intervention instructions, including drug dosage, light irradiation parameters, and debridement frequency, are packaged and transmitted to the central control unit via an encrypted communication protocol. The central control unit parses the feature vector in the instructions and uses it as a query key to perform a K-nearest neighbor search in the historical treatment database. The system retrieves five historically cured cases with highly similar characteristics and extracts their recovery trajectory data at the same stage. Through comparative analysis, the system finds that the current case's healing rate is slightly better than the historical average, but the risk of infection is slightly higher. Based on this difference, the system automatically generates a graphic wound assessment report, which clearly recommends maintaining the current light irradiation treatment while closely monitoring the redness and swelling of the peripheral areas, and pushes the report to the medical staff terminal.

[0057] Table 2 Mapping Table of Healing Stage Determination and Intervention Strategies; As shown in Table 2, this invention utilizes quantitative methods. and The indicator enables precise staging of the wound condition and matches differentiated physical and chemical treatment parameters accordingly. Experimental data show that, compared with treatment based on fixed parameters based on experience, this adaptive strategy shortens the wound healing cycle by an average of about 15%.

[0058] A data processing system based on wound image recognition, the data processing system based on wound image recognition is used to execute the above-described data processing method based on wound image recognition, the system comprising: The image preprocessing module is used to receive the wound reflection signal collected by the sensor, construct a multi-dimensional photoelectric pixel matrix, analyze the brightness dispersion of the rows and columns of the multi-dimensional photoelectric pixel matrix, calculate the global illumination distribution weight, perform denoising and smoothing processing on the multi-dimensional photoelectric pixel matrix according to the global illumination distribution weight, and construct a standardized tissue feature map. The feature extraction and analysis module is used to perform color space transformation on the standardized tissue feature map, extract the deep necrosis texture vector, calculate the difference ratio between the red and blue channels in the standardized tissue feature map, generate the granulation metabolic activity index, and map the deep necrosis texture vector to a high-dimensional feature space through a logistic regression algorithm to generate an infection probability distribution matrix. The model building and prediction module is used to identify pixel gradient abrupt change points in the standardized tissue feature map, construct an irregular geometric contour set, calculate the curvature change of each discrete coordinate point in the irregular geometric contour set, generate the wound edge contraction coefficient, and use the infection probability distribution matrix as a constraint to perform weighted fusion on the wound edge contraction coefficient to build a healing trend prediction model. The diagnosis and treatment decision and feedback module is used to input the granulation tissue metabolic activity index into the healing trend prediction model to perform iterative calculations, determine the healing stage based on the calculation results, generate adaptive diagnosis and treatment intervention instructions, send the adaptive diagnosis and treatment intervention instructions to the central control unit for parsing, match with the historical diagnosis and treatment database, and generate a wound assessment report based on the parsing results.

[0059] The above embodiments illustrate preferred embodiments of the present invention. Any equivalent adjustments to the technical solution based on software engineering methods are within the scope of protection, including but not limited to: implementing algorithm logic using different programming languages, refactoring functional modules into services, adjusting data interaction protocols, and optimizing resource scheduling strategies. Any implementation scheme derived from reasonable modifications to the data processing flow, service call chain, or system architecture layer without departing from the core technology of the present invention should be considered within the protection scope defined by the technical solution of the present invention.

Claims

1. A data processing method based on wound image recognition, characterized in that, Includes the following steps: S1: Receive the wound reflection signal collected by the sensor, construct a multi-dimensional photoelectric pixel matrix, analyze the brightness dispersion of the rows and columns of the multi-dimensional photoelectric pixel matrix, calculate the global illumination distribution weight, perform denoising and smoothing processing on the multi-dimensional photoelectric pixel matrix according to the global illumination distribution weight, and construct a standardized tissue feature map. S2: Perform color space transformation on the standardized tissue feature map, extract the deep necrosis texture vector, calculate the difference ratio between the red channel and the blue channel in the standardized tissue feature map, generate the granulation tissue metabolic activity index, and map the deep necrosis texture vector to a high-dimensional feature space through a logistic regression algorithm to generate an infection probability distribution matrix. S3: Identify pixel gradient abrupt change points in the standardized tissue feature map, construct an irregular geometric contour set, calculate the curvature change of each discrete coordinate point in the irregular geometric contour set, generate the wound edge contraction coefficient, and use the infection probability distribution matrix as a constraint condition to perform weighted fusion on the wound edge contraction coefficient to construct a healing trend prediction model. S4: Input the granulation tissue metabolic activity index into the healing trend prediction model to perform iterative calculations, determine the healing stage based on the calculation results, generate adaptive treatment intervention instructions, send the adaptive treatment intervention instructions to the central control unit for parsing, match the historical treatment database, and generate a wound assessment report based on the parsing results.

2. The data processing method based on wound image recognition according to claim 1, characterized in that, The specific steps of S1 are as follows: S11: Obtain the multi-dimensional photoelectric pixel matrix, traverse the brightness values ​​of each pixel in the matrix in the horizontal and vertical directions, calculate the brightness variance values ​​of each row and column using a preset discrete statistical algorithm, construct an illumination non-uniformity mapping table based on the brightness variance values, and generate global illumination distribution weights through reverse normalization processing. S12: Call the global illumination distribution weights, combine them with the adaptive Gaussian filtering algorithm to construct a dynamic filtering kernel, use the dynamic filtering kernel to perform weighted smoothing suppression on high-frequency noise points in the multi-dimensional photoelectric pixel matrix, retain low-frequency tissue texture information, and generate a standardized tissue feature map.

3. The data processing method based on wound image recognition according to claim 1, characterized in that, The specific steps of S2 are as follows: S21: Obtain the standardized tissue feature map, convert it from RGB color space to LAB color space, separate the luminance channel component and chrominance channel component, use a multi-scale Gabor filter bank to perform convolution operation on the luminance channel component, capture tissue texture features of different directions and frequencies, and generate deep necrosis texture vector. S22: Obtain the intensity values ​​of the red channel and the blue channel in the standardized tissue feature map, perform differential ratio calculation to generate the granulation tissue metabolic activity index, and at the same time use the logistic regression algorithm to classify and map the deep necrosis texture vector, calculate the confidence value of each pixel belonging to the necrotic tissue category, and generate an infection probability distribution matrix.

4. The data processing method based on wound image recognition according to claim 1, characterized in that, The specific steps of S3 are as follows: S31: Use the Sobel operator to scan the standardized tissue feature map, detect the gradient change amplitude of pixel gray values, filter out pixels with gradient amplitude exceeding the preset edge threshold as edge candidate points, and connect broken edge segments through morphological closing operation to construct an irregular geometric contour set. S32: Traverse each discrete coordinate point in the set of irregular geometric contours, calculate the rate of change of the slope of the tangent line formed by the point and its adjacent coordinate points, calculate the mean and variance of the curvature of all coordinate points, and generate the wound edge contraction coefficient. S33: Obtain the infection probability distribution matrix as a penalty term, perform nonlinear weighted correction on the wound edge contraction coefficient, and establish a healing trend prediction model by combining time series analysis algorithm.

5. The data processing method based on wound image recognition according to claim 1, characterized in that, The specific steps of S4 are as follows: S41: The granulation tissue metabolic activity index is used as the initial input variable and imported into the healing trend prediction model. The simulation time step and convergence conditions are set, and multiple rounds of time series simulation are performed to predict the granulation tissue growth rate and area change rate within a specific time window in the future, and to generate the healing stage determination result. S42: Based on the healing stage determination result, search the preset clinical intervention strategy library, match the drug throughput, light therapy intensity and debridement frequency parameters that are appropriate for the current healing stage, and generate adaptive diagnosis and treatment intervention instructions; S43: Transmit the adaptive treatment intervention command to the central control unit, parse the parameter characteristics in the command, search for historical case data with similar parameter characteristics in the historical treatment database, compare and analyze the differences in the recovery trajectory between the current case and the historical cases, and generate a wound assessment report.

6. The data processing method based on wound image recognition according to claim 2, characterized in that, The process of generating the global illumination distribution weights specifically includes: Obtain the first pixel in the multidimensional photoelectric pixel matrix. line and number The pixel brightness values ​​of the column are used to calculate the average brightness threshold of the entire matrix, and the set of dark area pixels with brightness values ​​lower than the average brightness threshold and the set of bright area pixels with brightness values ​​higher than the average brightness threshold are selected. Calculate the area ratio coefficient between the set of pixels in the dark area and the set of pixels in the bright area, construct an illumination compensation function using the area ratio coefficient, map the output value of the illumination compensation function to a value between 0 and 1, and generate a global illumination distribution weight.

7. The data processing method based on wound image recognition according to claim 3, characterized in that, The process of generating the granulation tissue metabolic activity index specifically includes: Obtain the red channel values ​​of the pixels in the target region of the standardized tissue feature map. Green channel values and the blue channel value ; Based on the absorption characteristics of hemoglobin to light of a specific wavelength, a colorimetric index calculation model is constructed. The red channel values ​​are nonlinearly amplified using the model, and the blue channel values ​​are normalized and attenuated to generate a granulation tissue metabolic activity index.

8. The data processing method based on wound image recognition according to claim 4, characterized in that, The construction process of the healing trend prediction model specifically includes: Obtain historical time-series data of the wound edge contraction coefficient, the average infection probability value of the infection probability distribution matrix, and the granulation tissue metabolic activity index; A healing dynamics evolution formula based on differential equations is constructed, and the undetermined coefficients in the formula are fitted and solved using the historical time series data to establish a healing trend prediction model. The mathematical expression for the healing trend prediction model is: ； in, represent Predicted wound healing rate at time point. Represents the tissue regeneration capacity constant. This represents the shrinkage coefficient at the edge of the wound. Represents bacterial blocking factors. This represents the average probability value of infection. Represents the metabolic drive coefficient. represent The granulation tissue metabolic activity index at that time.

9. The data processing method based on wound image recognition according to claim 5, characterized in that, The process of generating the adaptive diagnostic and treatment intervention instructions specifically includes: Obtain the physiological stage category of the current wound as identified in the healing stage determination result, including the inflammatory stage, proliferative stage, or remodeling stage; If the inflammation phase is determined, the coordinates of high-risk areas in the infection probability distribution matrix are extracted to generate targeted debridement path instructions and anti-infective agent spraying dosage instructions. If the condition is determined to be in the proliferative phase, the wavelength and irradiation duration parameters of the biostimulation light device are adjusted according to the value of the granulation tissue metabolic activity index to generate a physical therapy instruction. The physical therapy instruction is then combined with the drug spraying dosage instruction to generate an adaptive diagnostic and treatment intervention instruction.

10. A data processing system based on wound image recognition, characterized in that, The system is used to implement the data processing method based on wound image recognition as described in any one of claims 1-9, and the system includes: The image preprocessing module is used to receive the wound reflection signal collected by the sensor, construct a multi-dimensional photoelectric pixel matrix, analyze the brightness dispersion of the rows and columns of the multi-dimensional photoelectric pixel matrix, calculate the global illumination distribution weight, perform denoising and smoothing processing on the multi-dimensional photoelectric pixel matrix according to the global illumination distribution weight, and construct a standardized tissue feature map. The feature extraction and analysis module is used to perform color space transformation on the standardized tissue feature map, extract the deep necrosis texture vector, calculate the difference ratio between the red channel and the blue channel in the standardized tissue feature map, generate the granulation tissue metabolic activity index, and map the deep necrosis texture vector to a high-dimensional feature space through a logistic regression algorithm to generate an infection probability distribution matrix. The model building and prediction module is used to identify pixel gradient abrupt change points in the standardized tissue feature map, construct an irregular geometric contour set, calculate the curvature change of each discrete coordinate point in the irregular geometric contour set, generate the wound edge contraction coefficient, and use the infection probability distribution matrix as a constraint condition to perform weighted fusion on the wound edge contraction coefficient to construct a healing trend prediction model. The diagnosis and treatment decision and feedback module is used to input the granulation tissue metabolic activity index into the healing trend prediction model to perform iterative calculations, determine the healing stage based on the calculation results, generate adaptive diagnosis and treatment intervention instructions, send the adaptive diagnosis and treatment intervention instructions to the central control unit for parsing, match the historical diagnosis and treatment database, and generate a wound assessment report based on the parsing results.

Images