Neural Network-Based Skin Detection System

Abstract

This application discloses a skin detection system based on a neural network, belonging to the field of skin detection technology. The neural network-based skin detection system includes a color data acquisition module for acquiring color data of transplanted skin; a temperature data acquisition module for acquiring temperature data of transplanted skin; a blood flow data acquisition module for acquiring blood flow data of transplanted skin; and an analysis module, signal-connected to the color data acquisition module, temperature data acquisition module, and blood flow data acquisition module, for obtaining an activity index of the transplanted skin based on the color data, temperature data, and blood flow data. In the technical solution provided by this application, during skin monitoring, color data, temperature data, and blood flow data of the transplanted skin are collected separately and input into the analysis module. The analysis module integrates the above three sets of data to evaluate the overall condition of the transplanted skin, thereby obtaining the corresponding activity index.

Description

Technical Field

[0001] This application relates to the field of medical technology, and more specifically, to a skin detection system based on neural networks. Background Technology

[0002] After a skin graft is completed, the healing process needs to be monitored to determine if the transplanted skin has a good prognosis. Currently, nurses observe the transplanted skin daily, judging its progress based on changes in color and temperature.

[0003] Because everyone's initial skin color is different, this method of observation is limited by the nurse's experience, which may lead to discrepancies between the judgment and the actual situation. Consequently, problems with the transplanted skin may not be detected in time, affecting the prognosis of the transplanted skin. Summary of the Invention

[0004] The summary section of this application is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description section below. This summary section is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0005] As a first aspect of this application, in order to solve the technical problem of inaccurate assessment of the health status of transplanted skin, this application provides a skin detection system based on a neural network, comprising:

[0006] The color data acquisition module is used to acquire the color data of the transplanted skin;

[0007] Temperature data acquisition module, used to acquire temperature data of the transplanted skin;

[0008] Blood flow data acquisition module, used to acquire blood flow data of transplanted skin;

[0009] The analysis module is connected to the color data acquisition module, temperature data acquisition module, and blood flow data acquisition module respectively, and is used to obtain the activity index of the transplanted skin based on the color data, temperature data, and blood flow data.

[0010] In the technical solution provided in this application, when monitoring the skin, color data, temperature data, and blood flow data of the transplanted skin are collected and input into the analysis module. The analysis module integrates the above three sets of data to evaluate the overall condition of the transplanted skin and obtains the corresponding activity index.

[0011] Different patients have different skin colors, corresponding blood flow conditions, and body temperature changes. If the color, temperature, and blood flow of the transplanted skin are directly used for transplant viability analysis, the calculated viability index may differ from the actual viability index due to individual patient specificities. To address this issue, this application provides the following technical solution:

[0012] Furthermore, the color data is the difference between the current RGB three-channel values ​​of the transplanted skin and the three-channel values ​​of the transplanted skin before transplantation;

[0013] The temperature data is the difference between the current surface temperature of the transplanted skin and the current surface temperature of the patient's skin;

[0014] Blood flow data is the difference between the current blood flow rate of the transplanted skin and the blood flow rate when the transplanted skin was first transplanted.

[0015] In the technical solution provided in this application, the color data, temperature data, and blood flow data are not specific numerical values, but relative values. Therefore, the color data, temperature data, and blood flow data can contain more information about changes in the condition of the transplanted skin, thereby enabling timely detection of changes in the condition of the transplanted skin.

[0016] When analyzing color, temperature, and blood flow data, the analysis module receives data at a minimum level of pixels, or is related to the detection accuracy of the corresponding sensors. This accuracy is often higher than that of human visual observation. However, during actual raw data collection, only observational information from experienced nurses can be input into the analysis module for training. Therefore, when the training data input to the analysis module differs from the actual accuracy, the built-in model of the analysis module may overfit, leading to errors in actual judgment. To address this issue, this application provides the following technical solution:

[0017] The analysis module pre-sets calculation units, and then sends the calculation units to the color data acquisition module, temperature data acquisition module, and blood flow data acquisition module respectively;

[0018] Among them, the color data is the color matrix A of the color parameter a of each computational unit in the transplanted skin;

[0019] The temperature data is the temperature matrix B of the temperature parameter b of each computational unit in the transplanted skin;

[0020] The blood flow data is the blood flow matrix C of the blood flow parameter c of each computational unit in the transplanted skin;

[0021] The size of the calculation unit is 1 to 3 square centimeters.

[0022] In the technical solution provided in this application, a computational unit is pre-defined, and the size of the computational unit is set within 1 to 3 square centimeters. Then, a corresponding matrix is ​​generated using this computational unit as the smallest unit. Therefore, in these matrices, the information of the smallest unit is the relevant information of the corresponding computational unit. This ensures that the training data is basically consistent with the minimum precision, avoiding overfitting of the model.

[0023] When processing color data, directly converting multi-channel information from an image into grayscale values ​​results in the loss of excessive color information and makes it impossible to intuitively assess changes in the patient's skin surface. Furthermore, in practice, it is easily affected by changes in lighting and shadows at different stages. To address this issue, this application provides the following technical solution:

[0024] Furthermore, a = w1R + w2B + w3G, where w1, w2, and w3 are the red weight coefficient, the blue weight coefficient, and the green weight coefficient, respectively.

[0025] R is the average of the differences between the current red channel values ​​of each pixel of the transplanted skin within the calculation unit and the red channel values ​​of each pixel of the transplanted skin before transplantation.

[0026] B is the average of the differences between the current blue channel values ​​of each pixel of the transplanted skin within the calculation unit and the blue channel values ​​of each pixel of the transplanted skin before transplantation;

[0027] G is the average of the differences between the current green channel values ​​of each pixel of the transplanted skin within the calculation unit and the green channel values ​​of each pixel of the transplanted skin before transplantation.

[0028] In the technical solution provided in this application, corresponding weighting coefficients are set for different channels when calculating color parameter 'a', thus highlighting the influence of the corresponding color gamut on the color parameter as needed. In practice, the weighting coefficient for red can be increased to determine the changes in blood color in the transplanted skin.

[0029] Furthermore, b = T0 - T1, where T0 is the current surface temperature of the transplanted skin within the calculation unit, and T1 is the surface temperature of the patient's non-transplanted skin.

[0030] When assessing the health of transplanted skin, relying solely on blood flow rate is susceptible to the influence of the main blood vessels within the transplanted skin. Consequently, changes in blood flow in capillaries cannot be detected in a timely manner. To address this issue, this application provides the following solution:

[0031] c=M1+M 1； Where M1 is the blood flow velocity fraction within the calculation unit, and M2 is the blood flow distribution fraction within the calculation unit;

[0032] k1 is a pre-set first proportional coefficient, V0 is the current blood flow rate of the transplanted skin in the calculation unit, and V1 is the blood flow rate of the transplanted skin in the calculation unit when it was first transplanted.

[0033] Where k2 is a pre-set second proportional coefficient, S0 is the area of ​​the current blood flow region of the transplanted skin in the calculation unit, S is the area of ​​the calculation unit, and S1 is the area of ​​the blood flow region of the transplanted skin in the calculation unit when it was first transplanted.

[0034] In the technical solution provided in this application, when calculating the blood flow parameter c, not only the blood flow velocity within the calculation unit is considered, but also the blood flow region within the calculation unit. Therefore, the blood flow velocity within the transplanted skin and the diffusion of blood flow are comprehensively considered. In other words, the blood flow velocity of the main blood vessels and the blood flow velocity within the capillaries within the transplanted skin are both taken into account, ensuring that the blood flow parameter can well reflect the changes in blood flow of the transplanted skin.

[0035] When generating an activity index using color, temperature, and blood flow data, the analysis module needs to incorporate a regression classification model. Generally, ordinary regression classification models do not perform well. While neural network models offer better classification results, they require manual annotation and training. Therefore, this application provides the following solution:

[0036] Furthermore, the analysis module includes a feature fusion unit and an adversarial generative network unit;

[0037] Color data, temperature data, and blood flow data are input into the feature fusion unit to generate fusion feature F. Fusion feature F is then input into the adversarial generative network unit to generate the activity index of the transplanted skin.

[0038] In the technical solution provided in this application, a feature fusion unit is first used to fuse features, which can reduce the dimensionality of the data. Then, the fused features are input into the Generative Adversarial Network (GAN), which is an unsupervised network model that can complete training without annotations.

[0039] Merging color, temperature, and blood flow data into a single fusion feature effectively reduces feature dimensionality. However, this hides much information, thus reducing the accuracy of the generated activity index. To address this issue, this application provides the following technical solution:

[0040] Furthermore, the feature fusion unit includes a convolutional layer, a normalization layer, an activation function layer, and a fusion layer;

[0041] The convolutional layer includes three parallel convolution inputs. Color data, temperature data, and blood flow data are input into the convolution inputs respectively. The dilation rates of the three parallel convolution inputs are 5, 5, and 1, respectively.

[0042] Normalization layers are connected to convolutional layers, activation layers are connected to normalization layers, and fusion layers are connected to activation layers.

[0043] The technical solution provided in this application treats color data, temperature data, and blood flow data as features similar to grayscale values ​​in an image, and then employs different convolutional inputs with different dilation rates for feature extraction. For color and temperature data, a more global perspective can be considered to extract global features, while for blood flow data, a more local perspective can be considered to extract more local texture features. Therefore, in practice, the most critical information can be found from the corresponding data, avoiding the blurring of key information in color, temperature, and blood flow data during the information dimensionality reduction process. Attached Figure Description

[0044] The accompanying drawings, which form part of this application, are used to provide a further understanding of the application and to make other features, objects, and advantages of the application more apparent. The illustrative embodiments and descriptions of this application are used to explain the application and do not constitute an undue limitation of the application.

[0045] Furthermore, throughout the accompanying drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the elements are not necessarily drawn to scale.

[0046] In the attached diagram:

[0047] Figure 1 This is a schematic diagram of a neural network-based skin detection system.

[0048] Figure 2 This is a schematic diagram of the structure of the fused feature unit.

[0049] Figure 3 This is a temperature distribution map obtained by the temperature data acquisition module.

[0050] Figure 4 This is a blood flow distribution map obtained from the blood flow data module. Detailed Implementation

[0051] Embodiments of this application will now be described in more detail with reference to the accompanying drawings. While some embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this application. It should be understood that the drawings and embodiments of this application are for illustrative purposes only and are not intended to limit the scope of protection of this application.

[0052] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0053] The present application will now be described in detail with reference to the accompanying drawings and embodiments.

[0054] Reference Figure 1 A neural network-based skin detection system includes a color data acquisition module, a temperature data acquisition module, a blood flow data acquisition module, and an analysis module. The analysis module is signal-connected to the color data acquisition module, temperature data acquisition module, and blood flow data acquisition module, respectively. The color data acquisition module, temperature data acquisition module, blood flow data acquisition module, and analysis module can each be four independent devices connected together via a local area network.

[0055] The color data acquisition module is used to obtain the color data of the transplanted skin. Specifically, this module is essentially a camera. After the color data acquisition module photographs the patient's transplanted skin, medical staff manually outline the edges of the transplanted skin in the image. This produces an image of the transplanted skin, which, after processing, yields the required color data.

[0056] The temperature data acquisition module is used to obtain the temperature data of the transplanted skin. This module uses a commercially available infrared thermometer. In practice, the infrared thermometer can be combined with a front-facing camera to obtain images of the temperature distribution in different areas. Therefore, by simply having the temperature data acquisition module and the color data acquisition module photograph the same area, and then outlining it, the color data and corresponding temperature data of the transplanted skin can be obtained.

[0057] The blood flow data acquisition module is used to acquire blood flow data from the transplanted skin. This blood flow data actually refers to information about the flow rate of blood within the transplanted skin.

[0058] In practice, this blood flow data acquisition module can acquire blood flow data in two ways. The first method is to utilize the commonly used laser speckle imaging technology for detecting blood flow in small blood vessels. For details on the technical principles, please refer to the following technical documents:

[0059] [1] Tang Xin, Tan Jie. Research progress on the application of laser speckle contrast imaging technology in animal models of chronic cerebral ischemia [J]. Chinese Journal of Comparative Medicine, 2024, 34(02):154-160.

[0060] [2] Chen Ziqing. Clinical study on the effect of the degree of vascularization of artificial dermis on the treatment of wounds using laser speckle contrast imaging technology [D]. Nanchang University, 2023. DOI:10.27232 / d.cnki.gnchu.2023.001343.

[0061] [3] Guo Le, Hairegu Tusun, Zhou Jiaqi, et al. Application of laser speckle contrast imaging technology in blood flow monitoring [J]. Journal of Laser Biology, 2023, 32(01):8-14.

[0062] [4] Zhai Linjun, Fu Yuqing, Du Yongzhao. Research progress on key technologies and applications of laser speckle contrast blood flow imaging [J]. Chinese Journal of Lasers, 2023, 50(09): 52-79.

[0063] Generally, laser speckle technology produces a blood flow map, where areas with fast blood flow are represented by brighter spots, areas with slow blood flow by brighter spots, and areas with no blood flow by pixels with zero brightness. Therefore, laser speckle technology can obtain a network map of blood vessel distribution.

[0064] To facilitate understanding of the application of laser speckle imaging (LSCI) in the measurement of microcirculatory blood flow parameters, the following are some existing experimental examples and their detailed experimental specifications:

[0065] 1. Cerebral blood flow monitoring experiment:

[0066] Experimental parameters: Cerebral blood flow (CBF) was monitored using laser speckle contrast imaging, and the measurement results of laser speckle technology and laser Doppler technology were compared.

[0067] Experimental procedure: Boas's research group used laser speckle imaging to monitor changes in blood flow in the cortex and pia mater during cortical diffusion inhibition (CSD).

[0068] Experimental results: The effectiveness of laser speckle blood flow monitoring technology was verified, and changes in blood flow were monitored in CSD.

[0069] 2. Skin microcirculation measurement experiment:

[0070] Experimental parameters: Observe the changes in blood flow in the epidermis and subepidermis of the dorsal skin of rodents.

[0071] Experimental procedure: Choi et al. used laser speckle technology to observe changes in blood flow in the skin microcirculation of animals.

[0072] Experimental results: Bray compared the measurement of skin microcirculation blood flow using laser Doppler and laser speckle, providing a new measurement method for skin microcirculation research.

[0073] 3. Real-time imaging study of blood flow using laser speckle technology

[0074] Experimental parameters: A 785 nm laser with a power of 60 mW was used, and the imaging system included 8-bit and 16-bit cameras.

[0075] Experimental procedure: The rabbit mesenteric vascular experiment was conducted at Shanghai Sixth People's Hospital, during which the blood flow in the rabbit mesentery was monitored.

[0076] Experimental results: LSCI can accurately determine whether there is blood flow in blood vessels, and high grayscale and high resolution imaging helps to improve the monitoring effect.

[0077] The second method involves using needle insertion to obtain blood flow data from the transplanted skin. Specifically, needles are inserted into the target area, and the blood flow data within the skin is obtained based on the bleeding rate at the needle insertion site. For example, if the transplanted skin has severe ischemia, no blood will seep out from the needle puncture site for a long time after insertion; if the ischemia is not severe, blood will seep out quickly. Therefore, the bleeding rate and amount of bleeding can be used to describe the blood flow at the needle insertion site. In practice, needles can be inserted into the corresponding area of ​​the patient to obtain the bleeding rate and amount of bleeding in that area.

[0078] Bleeding rate and bleeding volume can be obtained using machine vision algorithms or by manual input; specific methods will not be elaborated here.

[0079] In practice, to avoid repeated needle pricks on patients, when acquiring blood flow data, medical staff only prick areas with a larger area compared to the previous examination and update the blood flow data for that area. Blood flow data for previous areas is then used. In this solution, however, the corresponding needle prick area can be obtained by using the data difference between the temperature data acquisition modules.

[0080] Therefore, in the technical solution provided in this application, the initial data obtained by the color data acquisition module, temperature data acquisition module, and blood flow data acquisition module are all in image format, only the information represented by the pixels in the images differs. In the image obtained by the color data acquisition module, each pixel has values ​​for RGB three channels, used to represent color information. For example... Figure 3 As shown, in the image obtained by the temperature data acquisition module, each pixel only contains temperature-related brightness or color information. The higher the temperature of the corresponding area, the brighter it appears in the image, or the more it tends towards a reddish hue. In common designs, low temperatures are represented by blue, medium temperatures by yellow, and high temperatures by red. For example... Figure 4 As shown, in the image obtained by the blood flow data acquisition module, the brightness information of each pixel is only related to the blood flow rate; the faster the blood flow rate, the higher the corresponding brightness.

[0081] The blood flow data obtained by using needles is rendered from the bleeding rate and amount in each area, so it can actually be used to obtain a rendering of the amount of bleeding in the area where the transplanted skin is located.

[0082] The above is an introduction to the functions of the color data acquisition module, temperature data acquisition module, and blood flow data acquisition module. The technical principles behind these three modules are all existing, mature technologies. In practice, to ensure that the color data acquisition module, temperature data acquisition module, and blood flow data acquisition module obtain images with the same scale, the three modules are integrated as much as possible, synchronously acquiring color data, temperature data, and blood flow data for the corresponding areas. This ensures that the color data, temperature data, and blood flow data can be directly registered, avoiding the need for additional registration work from the analysis module.

[0083] The above explains the principles of the color data acquisition module, temperature data acquisition module, and blood flow data acquisition module. However, the data directly measured by these modules are not the corresponding color data, temperature data, and blood flow data. The specific methods for acquiring color data, temperature data, and blood flow data are provided below.

[0084] Color data refers to the difference between the current RGB three-channel values ​​of the transplanted skin and the values ​​of the three channels before transplantation. In other words, color data is not the most direct information about the color change of the transplanted skin. Instead, it is the difference between the current color information of the transplanted skin and its color information before transplantation. This avoids misdiagnosis caused by different initial color data values ​​due to differences in skin tone and complexion among different patients.

[0085] Therefore, in practice, it is necessary to obtain the requested color data before a patient undergoes skin grafting surgery. For example, if skin is grafted from a patient's thigh to their face, the color information of the patient's thigh area needs to be obtained before the grafting surgery. Then, when monitoring the grafted skin, the current color information of the grafted skin is compared with the color information before the grafting surgery to obtain the required color data.

[0086] Temperature data represents the difference between the current surface temperature of the transplanted skin and the current surface temperature of the patient's skin. Similarly, the color data represents the difference between the current color of the transplanted skin and its color before transplantation. Temperature data requires the same approach. However, in practice, a patient's body temperature is not constant daily under normal circumstances. Therefore, it is necessary to compare the temperature of the transplanted skin with the temperature of the surrounding skin. For this reason, when collecting temperature data, it is necessary to collect both the current temperature of the transplanted skin and the temperature of the surrounding skin, using the difference between the two temperatures as the final temperature reading.

[0087] Correspondingly, blood flow data, like temperature and color data, is not a state indicator, but rather a comparative indicator. Blood flow data represents the difference between the current blood flow rate of the transplanted skin and its blood flow rate immediately after transplantation.

[0088] Because the blood flow of transplanted skin is good before transplantation, it is generally not ideal after transplantation. Comparing blood flow data before and after transplantation will result in a large baseline difference, which will obscure subtle changes in blood flow. For example, the blood flow rate of transplanted skin may vary between 10 and 20. The maximum variation (20-10) is 100% different from the minimum blood flow rate (10). However, the blood flow rate of skin before transplantation is generally around 100. Therefore, the final difference will vary between 90 and 80. The maximum variation (90-80) is 12.5% ​​different from the minimum blood flow rate (80), which may cause subtle information to go unnoticed. Therefore, in the solution provided in this application, the blood flow data is the difference between the current blood flow rate of the transplanted skin immediately after transplantation and the blood flow rate of the transplanted skin immediately after transplantation.

[0089] When using laser speckle technology to acquire blood flow data, only the corresponding equipment is needed for monitoring. However, when using needle puncture to acquire blood flow data, to avoid excessive needle punctures on the transplanted skin, the following method is required: For a period immediately after skin transplantation, needles are periodically punctured in various areas of the transplanted skin to obtain the blood flow rate. For example, within 4 hours after transplantation, needle punctures are periodically performed in various areas of the patient's transplanted skin to obtain the blood flow rate immediately after transplantation. Correspondingly, when collecting blood flow data, only needle punctures are needed to obtain the blood flow rate of the corresponding area, thus obtaining the patient's blood flow data.

[0090] The above outlines a general approach for acquiring color data, temperature data, and blood flow data. To balance the granularity and annotability of color data, temperature data, and blood flow data, this application provides the following approach:

[0091] The analysis module pre-sets calculation units, and then sends these calculation units to the color data acquisition module, temperature data acquisition module, and blood flow data acquisition module respectively; the size of the calculation unit is 1-3 square centimeters. Specifically, the color data is the color matrix A of color parameter 'a' for each calculation unit in the transplanted skin; the temperature data is the temperature matrix B of temperature parameter 'b' for each calculation unit in the transplanted skin; and the blood flow data is the blood flow matrix C of blood flow parameter 'c' for each calculation unit in the transplanted skin.

[0092] In the previous description, the color data acquisition module, temperature data acquisition module, and blood flow data acquisition module all obtained image information, but the meanings represented by the corresponding pixels in the images were not the same. Therefore, the minimum resolution of the image information obtained by the color data acquisition module, temperature data acquisition module, and blood flow data acquisition module is related to the accuracy of the selected device, but regardless of the device, its accuracy is far superior to that of the human eye. In order to simulate the doctor's experience in judging the patient's transplanted skin, a calculation unit was pre-set, which is actually an area of ​​1-3 cm in size. In this solution, 1 square centimeter is preferred. How to convert the corresponding area in the image to 1 square meter is a common technique in this field. In practice, each time information is acquired, the color data acquisition module, temperature data acquisition module, and blood flow data acquisition module can be placed at the same distance from the transplanted skin, thereby calculating the actual area size represented by each pixel.

[0093] Therefore, after the calculation units are pre-set, the image information obtained by the color data acquisition module, temperature data acquisition module, and blood flow data acquisition module can be divided into individual calculation units.

[0094] The following describes the specific calculation methods for color data, temperature data, and blood flow data after the calculation unit has been set up:

[0095] The color data is the color matrix A of the color parameter 'a' for each computational unit in the transplanted skin. a = w1R + w2B + w3G, where w1, w2, and w3 are the weighting coefficients for red, blue, and green, respectively.

[0096] R is the average of the differences between the current red channel values ​​of each pixel of the transplanted skin within the calculation unit and the red channel values ​​of each pixel of the transplanted skin before transplantation.

[0097] B is the average of the differences between the current blue channel values ​​of each pixel of the transplanted skin within the calculation unit and the blue channel values ​​of each pixel of the transplanted skin before transplantation;

[0098] G is the average of the differences between the current green channel values ​​of each pixel of the transplanted skin within the calculation unit and the green channel values ​​of each pixel of the transplanted skin before transplantation.

[0099] In practice, the weighting coefficient for red is much larger than that for blue and green. In fact, the color parameter 'a' is the average value of the differences between pixels within a calculation unit. It represents the average difference in color features within the calculation unit.

[0100] The temperature data is a temperature matrix B representing the temperature parameter b of each computational unit within the transplanted skin. b = T0 - T1, where T0 is the current surface temperature of the transplanted skin within the computational unit, and T1 is the surface temperature of the patient's non-transplanted skin. The temperature parameter is essentially the temperature difference between each pixel within the computational unit.

[0101] Blood flow data is the blood flow matrix C of the blood flow parameter c of each computational unit in the transplanted skin; c = M1 + M 1； Where M1 is the blood flow velocity fraction within the calculation unit, and M2 is the blood flow distribution fraction within the calculation unit;

[0102] When using laser speckle technology:

[0103] k1 is a pre-set first proportional coefficient, V0 is the sum of the current blood flow rates of the transplanted skin in the calculation unit, and V1 is the sum of the blood flow rates of the transplanted skin in the calculation unit when it was first transplanted.

[0104] Where k2 is a pre-set second proportional coefficient, S0 is the area of ​​the current blood flow region of the transplanted skin in the calculation unit, S is the area of ​​the calculation unit, and S1 is the area of ​​the blood flow region of the transplanted skin in the calculation unit when it was first transplanted.

[0105] Compared to color and temperature parameters, blood flow parameters introduce a blood flow distribution fraction in addition to the average blood flow velocity. Therefore, blood flow parameters are affected by both the blood flow velocity within the computational unit and the area where blood flow exists.

[0106] When using needle insertion techniques to obtain blood flow:

[0107] k1 is a pre-set first proportional coefficient, V0 is the current blood flow rate of the transplanted skin in the calculation unit, and V1 is the blood flow rate of the transplanted skin in the calculation unit when it was first transplanted.

[0108] Where k2 is a pre-set second proportional coefficient, S0 is the current bleeding area of ​​the transplanted skin in the calculation unit, S is the area of ​​the calculation unit, and S1 is the bleeding area of ​​the transplanted skin in the calculation unit when it was first transplanted.

[0109] Compared to color and temperature parameters, blood flow parameters introduce the bleeding area of ​​blood flow in addition to the average blood flow rate. Therefore, blood flow parameters are affected by both the blood flow rate and the bleeding area within the calculation unit.

[0110] Specifically, blood flow velocity actually indicates the bleeding rate, while the bleeding area is actually the bleeding volume of the blood flow. However, calculating the blood flow volume is very difficult in machine vision models, so the blood flow area is used instead of the bleeding volume.

[0111] Color data, temperature data, and blood flow data are essentially an image matrix. This can be understood as a special grayscale image, where each pixel is a computational unit, and the value of the relevant parameters within that unit is the grayscale value. Therefore, the analysis module processes the color, temperature, and blood flow parameters using the following scheme: The analysis module includes a feature fusion unit and a generative adversarial network (GAN) unit; the color, temperature, and blood flow data are input to the feature fusion unit to generate a fused feature F, and the fused feature F is input to the GAN unit to generate the activity index of the transplanted skin.

[0112] The feature fusion unit includes a convolutional layer, a normalization layer, an activation function layer, and a fusion layer. The convolutional layer includes three parallel convolutional inputs. Color data, temperature data, and blood flow data are input into the convolutional inputs respectively. The dilation rates of the three parallel convolutional inputs are 5, 5, and 1, respectively. The normalization layer is connected to the convolutional layer, the activation layer is connected to the normalization layer, and the fusion layer is connected to the activation layer.

[0113] refer to Figure 2The feature fusion unit essentially processes and fuses color data, temperature data, and blood flow data as image information. The expansion rate for color and temperature data is 5, while the expansion rate for blood flow data is 1. Therefore, during feature fusion, color and temperature data incorporate more global features, while blood flow data incorporates more local features.

[0114] To facilitate understanding, the following explanations are provided for convolutional layers, normalization layers, and activation function layers:

[0115] Convolutional Layer: This convolutional layer is a smoothly dilated convolutional layer, employing the technique of dilated convolution (also known as attenuated convolution). Dilated convolution increases the receptive field of the convolutional kernel without increasing the number of parameters by inserting spaces (zero values) between the elements of the kernel. The receptive field is the spatial range of input data that a neuron in a convolutional neural network can "see." The dilation rate is a key parameter in the convolutional layer. The dilation rate defines the spacing between elements in the kernel. For example, a dilation rate of 1 results in ordinary convolution; a dilation rate of 3 means inserting two spaces (zero values) between every two adjacent elements of the kernel; similarly, a dilation rate of 5 inserts four spaces.

[0116] The normalization layer is responsible for standardizing the data, that is, adjusting the scale of the data to have zero mean and unit variance. The activation layer processes the linear output of the previous layer through a non-linear activation function. In this scheme, the main change is to adjust the dilation rate of the convolutional layer; the remaining normalization and activation layers are existing technologies.

[0117] The fusion layer primarily fuses the features output by the activation function layer. Specifically, the feature map F1 corresponding to color data, the feature map F2 corresponding to temperature data, and the feature map F3 corresponding to blood flow data each undergo a 1×1 convolution operation. After the convolution operation, the three feature maps are added together, that is, the corresponding elements in the matrix are summed to obtain the fused feature map. This fused feature map then undergoes a final 3×3 convolution operation to obtain the final fused feature F. This process increases the smoothness of the feature fusion process.

[0118] Specifically, Among them, Conv 3×3 and Conv 1×1 F represents a 3×3 convolution operation and a 1×1 convolution operation, respectively. i Let i represent the i-th feature map, where i represents the index of the feature map, i∈{1,2,3}; This indicates that the summation is performed on all feature maps.

[0119] The feature fusion unit can fuse color data, temperature data, and blood flow data into a single fusion feature F. The generative adversarial network unit can then output the input fusion feature F as a corresponding health indicator.

[0120] Specifically, the Generative Adversarial Network (GAN) unit includes a generator and a discriminator. The generator receives the fused feature F and generates a corresponding health indicator U. The health indicator U is then input into the discriminator to generate the probability that the health indicator U is accurate. When the probability of the health indicator U exceeds a threshold, the health indicator U is considered accurate.

[0121] The above description is merely a selection of preferred embodiments of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this application.

Claims

1. A neural network-based skin detection system, characterized by, include: The color data acquisition module is used to acquire the color data of the transplanted skin; Color data is the difference between the current RGB three-channel values ​​of the transplanted skin and the three-channel values ​​of the transplanted skin before transplantation; The temperature data is the difference between the current surface temperature of the transplanted skin and the current surface temperature of the patient's skin; Blood flow data is the difference between the current blood flow rate of the transplanted skin and the blood flow rate when the transplanted skin was first transplanted; Temperature data acquisition module, used to acquire temperature data of the transplanted skin; Blood flow data acquisition module, used to acquire blood flow data of transplanted skin; The analysis module is connected to the color data acquisition module, temperature data acquisition module, and blood flow data acquisition module respectively, and is used to obtain the activity index of the transplanted skin based on the color data, temperature data, and blood flow data. The analysis module pre-sets calculation units, and then sends the calculation units to the color data acquisition module, temperature data acquisition module, and blood flow data acquisition module respectively; Among them, the color data is the color matrix A of the color parameter a of each computational unit in the transplanted skin; The temperature data is the temperature matrix B of the temperature parameter b of each computational unit in the transplanted skin; The blood flow data is the blood flow matrix C of the blood flow parameter c of each computational unit in the transplanted skin; The size of the calculation unit is 1 to 3 square centimeters; c = M1 + M 1； where M1 is the fraction of blood flow rate within the calculation unit, and M2 is the fraction of blood flow distribution within the calculation unit. ; k1 is a preset first proportional coefficient, V0 is the current blood flow rate of the graft skin in the calculation unit, and V1 is the blood flow rate of the graft skin immediately after the grafting. ; wherein k2 is a second preset proportionality coefficient, S0 is the area of the current blood flow region of the transplanted skin in the calculation unit, S is the area of the calculation unit, and S1 is the area of the blood flow region of the transplanted skin just after the transplantation in the calculation unit.

2. The neural network-based skin detection system of claim 1, wherein: a = w1R + w2B + w3G, where w1, w2, and w3 are the red weight coefficient, the blue weight coefficient, and the green weight coefficient, respectively. R is the average of the differences between the current red channel values ​​of each pixel of the transplanted skin within the calculation unit and the red channel values ​​of each pixel of the transplanted skin before transplantation. B is the average of the differences between the current blue channel values ​​of each pixel of the transplanted skin within the calculation unit and the blue channel values ​​of each pixel of the transplanted skin before transplantation; G is the average of the differences between the current green channel values ​​of each pixel of the transplanted skin within the calculation unit and the green channel values ​​of each pixel of the transplanted skin before transplantation.

3. The neural network-based skin detection system of claim 2, wherein: b = T0 - T1, where T0 is the current surface temperature of the transplanted skin within the calculation unit, and T1 is the surface temperature of the patient's non-transplanted skin.

4. The neural network-based skin detection system of claim 1, wherein: The analysis module includes a feature fusion unit and an adversarial generative network unit; Color data, temperature data, and blood flow data are input into the feature fusion unit to generate fused features. The fusion feature input is fed into the generative adversarial network unit to generate the activity index of the transplanted skin.

5. The neural network-based skin detection system of claim 4, wherein: The feature fusion unit includes a convolutional layer, a normalization layer, an activation function layer, and a fusion layer; The convolutional layer includes three parallel convolution inputs. Color data, temperature data, and blood flow data are input into the convolution inputs respectively. The dilation rates of the three parallel convolution inputs are 5, 5, and 1, respectively. Normalization layers are connected to convolutional layers, activation layers are connected to normalization layers, and fusion layers are connected to activation layers.

Images