Method and device for analyzing skin moisture by using skin image

Abstract

The invention relates to the technical field of skin moisture analysis methods, in particular to a method and device for analyzing skin moisture by using skin images, and the method comprises the following steps: establishing a skin surface layer parameter library, and acquiring data of skin epidermal layer moisture content, skin epidermal layer thickness and skin pore size by using a multi-mode optical sensor; a neural network model is constructed, the model comprises an input layer, at least one hidden layer and an output layer, and the input layer extracts feature parameters from the skin image; training the neural network model by using the data of the skin surface parameter library to obtain an optimal neural network model; a to-be-detected skin image is input into the optimal neural network model, a corresponding skin deep moisture parameter is output, the skin deep moisture parameter is the skin deep moisture content, and the moisture content prediction accuracy is remarkably improved through the synergistic effect of multi-modal data collection and the deep learning model.

Description

Technical Field

[0001] The present invention relates to the technical field of methods for analyzing skin moisture, and in particular to a method and device for analyzing skin moisture using skin images. Background Art

[0002] As people pay more and more attention to skin health, accurately evaluating skin moisture content has become an important topic in dermatology research and skin care product development. Traditional skin moisture measurement methods mainly include skin conductivity method, skin impedance method and stratum corneum moisture content measurement method. These methods have achieved certain results in practical applications, but there are still many limitations and shortcomings.

[0003] At present, the most widely used method for measuring skin moisture is based on conductivity measurement technology. This method indirectly reflects the water content of the skin by measuring the conductivity of the skin surface. However, this method has several obvious defects: first, it can only measure the moisture content of the top layer of the skin and cannot reflect the moisture status of the deep skin; second, the measurement results are easily affected by environmental factors (such as temperature and humidity), resulting in poor stability and repeatability of the measurement; finally, this method is difficult to distinguish between the content of free water and bound water in the skin, so it cannot fully reflect the hydration status of the skin.

[0004] Another commonly used method is the measurement technology based on near-infrared spectroscopy. This method uses the absorption characteristics of water molecules to light of a specific wavelength to estimate the water content of the skin. Although this method has been improved compared to the conductivity method and can reflect the moisture status of the deep layer of the skin to a certain extent, it still has some problems: the measurement accuracy is affected by factors such as skin pigmentation and blood flow, it is difficult to accurately distinguish the water distribution at different depths, and the equipment is usually expensive, which is not conducive to widespread application.

[0005] In addition, some studies have tried to use technologies such as ultrasound or magnetic resonance imaging to measure skin moisture, but these methods usually require complex equipment and professional operations, making them difficult to be widely used in daily skin care and clinical diagnosis.

[0006] In general, existing skin moisture measurement methods generally have problems such as insufficient detection depth, poor accuracy and stability, and difficulty in providing comprehensive skin moisture distribution information. These limitations seriously affect our accurate assessment of skin moisture status, thereby restricting the formulation of personalized skin care plans and the early diagnosis of certain skin diseases.

[0007] In view of the above problems, the present invention proposes an innovative method and device for analyzing skin moisture using skin images. The method cleverly combines multimodal optical scanning technology and deep learning algorithms to achieve accurate, stable and non-invasive measurement of deep skin moisture content. Summary of the invention

[0008] The core of the present invention is to solve the following key technical problems: first, how to obtain comprehensive information reflecting the moisture status of the deep skin; second, how to accurately identify and quantify the moisture content in the complex skin structure; third, how to improve the stability and repeatability of the measurement; and finally, how to achieve fast and convenient detection while ensuring the measurement accuracy.

[0009] The present invention proposes a method for analyzing skin moisture using skin images, comprising the following steps:

[0010] Establish a skin surface parameter library, which uses a multimodal optical sensor to obtain data on the water content of the skin epidermis, the thickness of the skin epidermis, and the size of skin pores;

[0011] Constructing a neural network model, the model comprising an input layer, at least one hidden layer and an output layer, wherein the input layer extracts feature parameters from the skin image;

[0012] Using the data of the skin surface parameter library to train the neural network model to obtain an optimal neural network model;

[0013] The skin image to be tested is input into the optimal neural network model, and the corresponding deep skin moisture parameter is output.

[0014] The deep skin moisture parameter is the deep skin moisture content.

[0015] Preferably, the multimodal optical sensor uses a light source with a wavelength of 420nm to 1200nm to perform tomography, obtain a section highlight image and a section light intensity image, and establish a three-dimensional skin image through the following steps:

[0016] The biological tissue sample is fixed and aligned perpendicular to the scanning light path;

[0017] Establish a coordinate system on the section perpendicular to the scanning direction;

[0018] Scanning the section to obtain at least two cross-sectional images;

[0019] A three-dimensional skin structure is reconstructed based on the cross-sectional image.

[0020] Preferably, the extraction of the characteristic parameters comprises the following steps:

[0021] Calculate the cortical image histogram;

[0022] Extracting pore areas and their sizes from cortical images;

[0023] Count the number of pores;

[0024] Detect the color information, texture information, pore information and pore shape of the cortical image.

[0025] The color information is converted from RGB to HSV space using the following formula:

[0026]

[0027] Among them, R, G, B represent the pixel values ​​of the red, green, and blue channels respectively, H represents the hue, S represents the saturation, and V represents the brightness.

[0028] Preferably, the extraction of pore information further includes calculating the spatial distribution law of pores by the following formula:

[0029]

[0030] Where D represents the discreteness of pore distribution, d i represents the distance between the i-th pore and its nearest neighbor pore, d avg represents the average value of all pore spacings, and N represents the total number of pores.

[0031] Preferably, the construction of the neural network model comprises the following steps:

[0032] Select histogram equalization, light intensity image, highlight image, and pore data as input layer features;

[0033] Set up multiple hidden layers and use the Sigmoid function as the transfer function:

[0034]

[0035] The root mean square error function is selected as the performance function:

[0036] Among them, y i is the actual value, is the predicted value, N is the number of samples, and the LM algorithm is used as the training function.

[0037] Preferably, the training process of the neural network model comprises the following steps:

[0038] Forward propagation is used to obtain the root mean square error;

[0039] Back propagation is used for parameter adjustment, where the weight update formula is:

[0040]

[0041] Among them, w ijis the weight from the i-th neuron to the j-th neuron, η is the learning rate, and E is the error function; repeat the above steps until convergence to obtain the optimal neural network model.

[0042] Preferably, the calculation formula for the predicted value of the deep skin moisture parameter is:

[0043]

[0044] Among them, P is the predicted value, O is the actual output value, A is the actual deep skin moisture parameter of the training sample, and 0.7 is the standard deep skin moisture parameter.

[0045] Preferably, the method further includes calculating the relative deviation between the predicted value and the actual skin deep moisture parameter, and the deviation calculation formula is:

[0046] δ=|PA| / A*100%

[0047] Among them, δ is the relative deviation, P is the predicted value, and A is the actual deep skin moisture parameter.

[0048] A skin moisture analysis device for implementing the method, comprising:

[0049] A skin surface parameter library storage module is used to store skin images and epidermal three-dimensional data structures;

[0050] Neural network model training module, used to build and train neural network models;

[0051] The skin moisture parameter prediction module is used to input the collected skin images and extracted feature parameters into the neural network model and output the predicted value of the deep skin moisture parameter.

[0052] Preferably, it also includes a computer image analysis component, which is used to:

[0053] Collect three-dimensional images of the epidermis;

[0054] Extract pixel spacing, epidermal highlight image of three-dimensional epidermal layer image, epidermal layer light intensity image, and three-dimensional epidermal layer shape information;

[0055] Calculate the color information, texture information, pore information, and pore shape information of the epidermal highlight image and the epidermal light intensity image;

[0056] Output the actual skin surface parameter information of the sample,

[0057] The three-dimensional epidermal shape information is calculated using the following formula:

[0058]

[0059] Among them, C is the complexity index of the epidermis, V is the volume of the epidermis, and A is the surface area of ​​the epidermis.

[0060] The beneficial effects of the present invention are mainly reflected in the following aspects:

[0061] By innovatively integrating multimodal optical scanning and deep learning technology, the present invention has achieved breakthroughs in many aspects and brought significant beneficial effects:

[0062] First, in terms of detection depth, the method of the present invention can effectively detect deep into the dermis (about 3.5mm depth), far exceeding the detection range of traditional methods. This breakthrough enables us to comprehensively evaluate the moisture status of the epidermis and dermis, providing a more reliable basis for the overall assessment of skin health.

[0063] Secondly, in terms of prediction accuracy, the present invention significantly improves the accuracy of moisture content prediction through the synergy of multimodal data acquisition and deep learning models. Multimodal optical scanning provides rich skin feature information, while deep learning models can extract features that are highly correlated with moisture content from these complex data. The combination of the two produces a synergistic effect, making the prediction accuracy reach a high level of 95.6%.

[0064] Furthermore, the present invention has also made significant progress in measurement stability. By comprehensively analyzing multiple optical features and introducing advanced data processing algorithms, the present method greatly reduces the impact of environmental factors on the measurement results. This not only improves the repeatability of the measurement, but also provides a reliable tool for long-term monitoring of skin moisture changes.

[0065] In addition, the present invention also achieves a good balance between detection speed and convenience. Although more data is collected, through optimization of algorithms and hardware design, the present method can still complete a comprehensive detection within 45 seconds, which basically meets the needs of practical applications.

[0066] Finally, it is worth mentioning that the method of the present invention can not only measure the overall water content, but also has the potential to distinguish the proportion of free water and bound water, which provides a deeper understanding of the hydration state of the skin. This feature has important application value in dermatological research and the formulation of personalized skin care programs.

[0067] In summary, the present invention has innovatively solved several key problems existing in the prior art and achieved significant improvements in detection depth, accuracy, stability and comprehensiveness. These advances not only provide new and powerful tools for dermatological research, but also open up new possibilities for personalized skin care and early diagnosis of skin diseases, and have broad application prospects and important social value. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a logic block diagram of the overall method of the present invention.

[0069] Figure 2 This is a logic block diagram of the image preprocessing algorithm of the present invention.

[0070] Figure 3 This is a logic block diagram of the feature extraction algorithm of the present invention.

[0071] Figure 4 This is a logic block diagram of the neural network model training algorithm of the present invention.

[0072] Figure 5 This is a logic block diagram of the deep moisture content prediction algorithm of the present invention.

[0073] Figure 6 The figure is a logic block diagram of the result correction algorithm of the present invention. DETAILED DESCRIPTION

[0074] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects thereof are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0075] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0076] Example 1

[0077] See also Figure 1-6 The present invention relates to a method and device for analyzing skin moisture using skin images, which is specifically applied in the field of skin health assessment and care.

[0078] First, the present invention proposes a method for analyzing skin moisture using skin images. The method includes the following key steps: establishing a skin surface parameter library, constructing a neural network model, training the neural network model, and using the trained model to predict skin deep moisture parameters.

[0079] Specifically, in the step of establishing the skin surface parameter library, the present invention uses a multimodal optical sensor to obtain data on the water content of the skin epidermis, the thickness of the skin epidermis, and the size of the skin pores. This multimodal data acquisition method can comprehensively capture the various characteristics of the skin surface and lay a solid foundation for subsequent analysis. Preferably, the multimodal optical sensor used in the present invention has an operating wavelength range of 420nm to 1200nm. The selection of this wavelength range is deliberate because it can effectively penetrate the skin surface without causing damage to the deep tissue.

[0080] Next, the present invention constructs a neural network model including an input layer, at least one hidden layer and an output layer. In the input layer, we extract feature parameters from the skin image. The selection of these feature parameters is an innovation of the present invention, and they include information such as the cortical image histogram, the pore area and its size, and the number of pores. Through this comprehensive feature extraction, we can make full use of various information in the skin image and improve the prediction accuracy of the model.

[0081] Then, the present invention uses the established skin surface parameter library to train the neural network model to obtain the optimal neural network model. This training process is an iterative optimization process, and the error between the model output and the actual skin deep moisture parameter is minimized by continuously adjusting the network parameters.

[0082] Finally, the skin image to be tested is input into the trained optimal neural network model to obtain the corresponding predicted value of the deep skin moisture parameter. It should be emphasized that the deep skin moisture parameter predicted by the present invention specifically refers to the deep skin moisture content, which is of great significance for evaluating skin health status.

[0083] Secondly, the present invention describes in detail the working principle of the multimodal optical sensor. Specifically, the sensor uses a light source with a wavelength of 420nm to 1200nm to perform tomographic scanning to obtain a section highlight image and a section light intensity image. In actual operation, the biological tissue sample is first fixed and aligned perpendicular to the scanning light path. This step ensures the accuracy of the scan. Then, a coordinate system is established on the section perpendicular to the scanning direction, which provides a spatial reference for subsequent image analysis. Next, the section is scanned to obtain at least two cross-sectional images. Finally, the three-dimensional skin structure is reconstructed based on these cross-sectional images. This three-dimensional reconstruction technology enables us to have a more comprehensive understanding of the three-dimensional structure of the skin, thereby improving the accuracy of moisture content prediction.

[0084] Extracting characteristic parameters of the cortical image includes detecting color information, texture information, pore information, and pore shape of the cortical image, and specifically refers to extracting corresponding characteristic parameters using a color space model, an image texture algorithm, computer image analysis, and pattern recognition; extracting hair follicle information specifically refers to identifying the various positions of hair follicles in the cortical image, and extracting the shape (aspect ratio), size (area), and spatial distribution law of the hair follicles corresponding to each position, and the cortical image includes an RGB image, a light intensity image, and a highlight image.

[0085] Again, the present invention describes in detail the process of extracting feature parameters. In addition to calculating the cortical image histogram, extracting the pore area and its size, and counting the number of pores, it also includes detecting the color information, texture information, pore information, and pore shape of the cortical image. Among them, the processing of color information adopts the conversion from RGB to HSV space. The advantage of this conversion is that the HSV space is closer to the human eye's perception of color, which helps to extract more meaningful color features. The conversion formula is as follows:

[0086]

[0087]

[0088] Among them, R, G, B represent the pixel values ​​of the red, green, and blue channels respectively, H represents the hue, S represents the saturation, and V represents the brightness.

[0089] In addition, the present invention also innovatively introduces the calculation of the spatial distribution of pores. The introduction of this feature is based on the observation that the moisture content of the skin may affect the distribution of pores. The specific calculation formula is as follows:

[0090]

[0091] Where D represents the discreteness of pore distribution, d i represents the distance between the i-th pore and its nearest neighbor pore, d avg represents the average value of all pore spacing, and N represents the total number of pores. Through this formula, we can quantify the uniformity of pore distribution and provide additional information for the prediction of skin moisture content.

[0092] Next, the present invention elaborates on the construction process of the neural network model. In the input layer, we selected histogram equalization, light intensity image, highlight image, and pore data as features. The selection of these features is obtained through repeated experiments and optimization, and they can fully reflect the various characteristics of the skin. In the hidden layer, we set multiple levels and selected the Sigmoid function as the transfer function:

[0093]

[0094] The choice of Sigmoid function is based on its good nonlinear mapping ability, which can effectively capture the complex relationship between features. At the same time, we choose the root mean square error function as the performance function:

[0095]

[0096] Among them, y i is the actual value, is the predicted value, N is the number of samples, and the LM algorithm is used as the training function. The root mean square error function can effectively measure the gap between the predicted value and the actual value, and is a commonly used performance evaluation indicator in regression problems.

[0097] In the training process, the present invention adopts a method combining forward propagation and back propagation. First, the error between the current network output and the actual value is calculated by forward propagation, and then the network parameters are adjusted by back propagation. Among them, the weight update formula is:

[0098]

[0099] Among them, w ij is the weight from the i-th neuron to the j-th neuron, η is the learning rate, and E is the error function. This process will be repeated until the network converges and the optimal neural network model is obtained.

[0100] After obtaining the trained model, the present invention proposes a novel method for predicting deep skin moisture parameters. The calculation formula for the predicted value is:

[0101]

[0102] Among them, P is the predicted value, O is the actual output value, A is the actual deep skin moisture parameter of the training sample, and 0.7 is the standard deep skin moisture parameter. The design of this formula takes individual differences into consideration, and by introducing the standard deep skin moisture parameter, the prediction results have better universality. The standard deep skin moisture parameter (70%) is introduced as a reference and corrected according to the actual situation of the training sample. This method takes individual differences into consideration and improves the applicability of the prediction results.

[0103] In order to evaluate the accuracy of the prediction, the present invention also introduces the calculation of relative deviation:

[0104] δ=|PA| / A*100%

[0105] Among them, δ is the relative deviation, P is the predicted value, and A is the actual skin deep moisture parameter. By calculating the relative deviation, we can intuitively evaluate the accuracy of the prediction results and provide a basis for further optimization of the model.

[0106] Finally, the present invention also proposes a skin moisture analysis device for implementing the above method. The device includes a skin surface parameter library storage module, a neural network model training module and a skin moisture parameter prediction module. Among them, the skin surface parameter library storage module is used to store skin images and epidermal three-dimensional data structures; the neural network model training module is used to build and train the neural network model; the skin moisture parameter prediction module is responsible for inputting the collected skin images and extracted feature parameters into the neural network model, and outputting the predicted value of the deep skin moisture parameter.

[0107] In addition, the device also includes an innovative computer image analysis component. The main functions of this component include collecting three-dimensional images of the epidermis, extracting pixel spacing, epidermis highlight images of the three-dimensional epidermis image, epidermis light intensity images, and three-dimensional epidermis shape information, as well as calculating the color information, texture information, pore information, and pore shape information of the epidermis highlight image and epidermis light intensity image. Among them, the three-dimensional epidermis shape information is calculated by the following formula:

[0108]

[0109] Among them, C is the epidermal complexity index, V is the epidermal volume, and A is the epidermal surface area. The introduction of this index enables us to quantify the morphological characteristics of the epidermis and provide additional reference information for the prediction of skin moisture content.

[0110] In general, the method and device for analyzing skin moisture using skin images proposed in the present invention achieve accurate prediction of deep skin moisture content through multimodal data acquisition, comprehensive feature extraction, application of deep learning models and innovative prediction formulas. This technology not only has important application value in the field of skin health assessment, but also provides a scientific basis for the formulation of personalized skin care plans. In the future, with the continuous expansion of the database and further optimization of the algorithm, we believe that this technology will play an important role in a wider range of fields.

[0111] Example 1: In a preferred embodiment of the present invention, we selected 100 volunteers aged between 20 and 50, including subjects with different skin types (dry, oily, neutral and mixed). We used the multimodal optical sensor proposed in the present invention to collect skin images of the T zone and U zone of each volunteer's face, with a wavelength range of 420nm-1200nm. Then, we applied the neural network model of the present invention to predict the deep skin moisture content.

[0112] At the same time, we use the currently recognized skin conductivity method as a comparison. The skin conductivity method is a traditional skin moisture measurement method that indirectly reflects the moisture content by measuring the conductivity of the skin.

[0113] In order to comprehensively evaluate the performance of the method of the present invention, we selected the following key indicators for testing:

[0114] 1. Forecast accuracy: refers to the degree of consistency between the predicted value and the actual value.

[0115] 2. Detection depth: refers to the skin depth that can be effectively detected.

[0116] 3. Detection stability: refers to the measurement consistency under different environmental conditions (such as temperature and humidity changes).

[0117] 4. Detection speed: refers to the time required to complete a complete test.

[0118] The selection of these indicators fully considers the actual needs of skin moisture detection, focusing on both the accuracy of the results and the practicality and reliability of the methods.

[0119] Testing methods and standards:

[0120] 1. Prediction accuracy: We use subcutaneous microdialysis technology as the gold standard to measure the actual water content in the deep layer of the skin. The predicted value is compared with the actual value and the relative error is calculated. The relative error is considered accurate within ±5%.

[0121] 2. Detection depth: The detection depth is determined by an ultrasonic skin thickness gauge. Methods that can effectively detect the depth of the dermis layer (approximately 1-4 mm) are considered to have a good detection depth.

[0122] 3. Test stability: The initial measurement is carried out under the standard environment of 20℃ and 60% relative humidity, and then repeated under the conditions of 15℃ / 40% humidity and 30℃ / 80% humidity. The coefficient of variation of the three measurement results is less than 5%, which is considered to have good stability.

[0123] 4. Detection speed: The time from starting measurement to obtaining the final result. Less than 1 minute is considered to have a good detection speed.

[0124] The test results are shown in the following table:

[0125]

[0126]

[0127] From the above results, it can be seen that the method of the present invention is significantly superior to the traditional skin conductivity method in many key indicators. Let us analyze these results in depth:

[0128] First, in terms of prediction accuracy, the method of the present invention achieved a high accuracy of 95.6%, far exceeding the 82.3% of the comparison method. This result fully demonstrates the superiority of our multimodal optical scanning technology combined with the deep learning model. By collecting richer skin feature information and using the neural network model for intelligent analysis, we can more accurately predict the deep moisture content of the skin.

[0129] Secondly, in terms of detection depth, the method of the present invention can reach a detection depth of 3.5mm, almost covering the entire dermis. In contrast, the traditional skin conductivity method can only detect 0.5mm depth, which is mainly limited to the epidermis. This advantage enables our method to provide a more comprehensive assessment of skin moisture status, not only reflecting the surface moisture, but also detecting the deep moisture content, which is of great significance for the diagnosis of dry skin and some skin diseases.

[0130] In terms of detection stability, the coefficient of variation of the method of the present invention is only 2.8%, which is much lower than 8.7% of the comparison method. This shows that our method is less affected by environmental factors and can maintain stable measurement results under different conditions. This feature is particularly important for practical applications because it ensures that reliable measurement results can be obtained in various practical situations (such as different seasons and different indoor environments).

[0131] Finally, although the detection speed of the proposed method is slightly slower than that of the traditional method (45 seconds vs 30 seconds), this time difference is acceptable considering that we provide more comprehensive and in-depth skin moisture information. Moreover, with the optimization of hardware and further improvement of algorithms, we are confident that this time gap can be shortened in the future.

[0132] In summary, the skin moisture detection method proposed in the present invention is significantly superior to traditional methods in key indicators such as accuracy, detection depth and stability. These advantages are due to our innovative combination of multimodal optical scanning technology and deep learning algorithms, which enables us to evaluate the moisture status of the skin more comprehensively and accurately. This method not only provides a new tool for dermatological research, but also opens up new possibilities for the formulation of personalized skin care plans and early diagnosis of skin diseases.

[0133] In future research, we will further expand the sample library to include subjects of more age groups and different skin types to improve the generalization ability of the model. At the same time, we will also explore combining this technology with the detection of other skin parameters (such as oil content, elasticity, etc.) to achieve a more comprehensive skin health assessment.

[0134] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for analyzing skin moisture using skin images, characterized in that: The following steps are involved: Establish a skin surface parameter library, which uses a multimodal optical sensor to obtain data on the water content of the skin epidermis, the thickness of the skin epidermis, and the size of skin pores; Constructing a neural network model, the model comprising an input layer, at least one hidden layer and an output layer, wherein the input layer extracts feature parameters from the skin image; Using the data of the skin surface parameter library to train the neural network model to obtain an optimal neural network model; The skin image to be tested is input into the optimal neural network model, and the corresponding deep skin moisture parameter is output. The deep skin moisture parameter is the deep skin moisture content.

2. The method according to claim 1, characterized in that The multimodal optical sensor uses a light source with a wavelength of 420nm to 1200nm to perform tomography, obtain a section highlight image and a section light intensity image, and establish a three-dimensional skin image through the following steps: The biological tissue sample is fixed and aligned perpendicular to the scanning light path; Establish a coordinate system on the section perpendicular to the scanning direction; Scanning the section to obtain at least two cross-sectional images; A three-dimensional skin structure is reconstructed based on the cross-sectional image.

3. The method according to claim 1, characterized in that The extraction of the characteristic parameters comprises the following steps: Calculate the cortical image histogram; Extracting pore areas and their sizes from cortical images; Count the number of pores; Detect the color information, texture information, pore information and pore shape of the cortical image. The color information is converted from RGB to HSV space using the following formula: Among them, R, G, B represent the pixel values ​​of the red, green, and blue channels respectively, H represents the hue, S represents the saturation, and V represents the brightness.

4. The method according to claim 3, characterized in that The extraction of pore information also includes calculating the spatial distribution law of pores, which is calculated by the following formula: Where D represents the discreteness of pore distribution, d i represents the distance between the i-th pore and its nearest neighbor pore, d avg represents the average value of all pore spacings, and N represents the total number of pores.

5. The method according to claim 1, characterized in that The construction of the neural network model includes the following steps: Select histogram equalization, intensity image, highlight image, and pore data as input layer features; Set up multiple hidden layers and use the Sigmoid function as the transfer function: The root mean square error function is selected as the performance function: Among them, y i is the actual value, is the predicted value, N is the number of samples, and the LM algorithm is used as the training function.

6. The method according to claim 5, characterized in that The training process of the neural network model includes the following steps: Forward propagation is used to obtain the root mean square error; Back propagation is used for parameter adjustment, where the weight update formula is: Among them, w ij is the weight from the i-th neuron to the j-th neuron, η is the learning rate, and E is the error function; repeat the above steps until convergence to obtain the optimal neural network model.

7. The method according to claim 1, characterized in that The calculation formula for the predicted value of the deep skin moisture parameter is: Among them, P is the predicted value, O is the actual output value, A is the actual deep skin moisture parameter of the training sample, and 0.7 is the standard deep skin moisture parameter.

8. The method according to claim 7, characterized in that It also includes calculating the relative deviation between the predicted value and the actual skin deep moisture parameter. The deviation calculation formula is: δ=|PA| / A*100% Among them, δ is the relative deviation, P is the predicted value, and A is the actual deep skin moisture parameter.

9. A skin moisture analysis device for implementing the method according to any one of claims 1 to 8, characterized in that: include: A skin surface parameter library storage module is used to store skin images and epidermal three-dimensional data structures; Neural network model training module, used to build and train neural network models; The skin moisture parameter prediction module is used to input the collected skin images and extracted feature parameters into the neural network model and output the predicted value of the deep skin moisture parameter.

10. The device according to claim 9, characterized in that Also included is a computer image analysis component that is used to: Collect three-dimensional images of the epidermis; Extract pixel spacing, epidermal highlight image of three-dimensional epidermal layer image, epidermal layer light intensity image, and three-dimensional epidermal layer shape information; Calculate the color information, texture information, pore information, and pore shape information of the epidermal highlight image and the epidermal light intensity image; Output the actual skin surface parameter information of the sample, The three-dimensional epidermal shape information is calculated using the following formula: Among them, C is the complexity index of the epidermis, V is the volume of the epidermis, and A is the surface area of ​​the epidermis.

Images