Human face skin multi-component concentration analysis method and device based on hyperspectral imaging

Abstract

The invention discloses a human face skin multi-component concentration analysis method and device based on hyperspectral imaging, and relates to the technical field of spectral imaging. The method comprises the following steps: under a standardized condition, sequentially collecting dark background data, standard integrating sphere uniform light field data and standard reflectivity plate reference spectrum data; collecting original face hyperspectral data under the same condition; correcting the original face data and the reference data based on the dark background data and the uniformity correction coefficient, and calculating the reflectivity data of the face skin in combination with the known reflectivity of the standard reflectivity board; establishing correlation between the reflectivity and a tissue absorption coefficient based on a skin optical model, wherein the absorption coefficient is expressed as linear superposition of various target component concentrations and standard spectral characteristics thereof; and solving the relationship by adopting a fitting optimization algorithm to obtain the pixel-level concentration of each component and generate a two-dimensional distribution diagram. By optimizing an acquisition device, a spectral analysis algorithm and a visualization function, high-precision, safe and easy-to-use detection effects are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of spectral imaging technology, in particular to a face skin multi-component concentration analysis method and device based on hyperspectral imaging. BACKGROUND

[0002] Skin detection is the basis of beauty care and skin disease diagnosis and treatment, which can accurately capture the state of skin moisture, pigments, biochemical concentrations, etc., provide scientific basis for personalized skin care program development, early screening of acne, color spots and other problems, and directly affect the skin care effect and diagnosis and treatment accuracy, and is indispensable in medical beauty institutions, dermatology clinics and home scenarios.

[0003] Traditional and existing non-spectral detection methods have significant limitations. Visible light images can only present the surface state of the skin and cannot penetrate the epidermis, making it difficult to support detection needs. Although multispectral detection can cover part of the skin spectral band, it is difficult to perform accurate quantitative analysis due to the small number of wave bands and low resolution. Hyperspectral detection has obvious advantages due to its technical characteristics: it can capture characteristic spectral signals of the epidermis-dermis layer, breaking through the limitations of traditional methods that can only obtain surface information and multispectral methods that are difficult to accurately quantify. In theory, it can realize qualitative and quantitative analysis of multiple biochemical concentrations such as oxyhemoglobin, deoxyhemoglobin, melanin, eumelanin and beta-carotene, providing a more scientific technical path for skin state evaluation. However, there are still significant shortcomings in current spectral detection: first, there is a lack of quantitative analysis methods for the spatial distribution of these concentrations, which cannot accurately present regional differences; second, the calculation of multi-component spectral fitting is large, low in efficiency, and weak in anti-interference ability, and invalid wave bands and noise in hyperspectral data can easily lead to poor result stability; third, halogen lamps are commonly used as light sources, which have high temperature and burn risk, and facial reflection and head movement can easily cause data noise and image distortion, further affecting detection reliability. Therefore, how to break through the bottlenecks of spectral detection in concentration quantitative analysis, efficiency, anti-interference, safety and visualization, while making up for the shortcomings of traditional methods in detection depth and range, to realize more accurate, efficient, safe and easy-to-use skin detection, is a problem to be solved at present. SUMMARY

[0004] Based on the above-mentioned shortcomings of the prior art, the purpose of the present application is to provide a face skin multi-component concentration analysis method and device based on hyperspectral imaging to solve the above technical problems.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solution: a face skin multi-component concentration analysis method based on hyperspectral imaging, comprising: Collecting dark background data, collecting uniform light field data of a standard integrating sphere to calculate uniformity correction coefficients, and collecting reference spectral data of a standard reflectance plate placed at a preset face collection position under a preset standardized optical collection condition; Under the condition of standardized optical acquisition, original human face hyperspectral data is collected; Based on the dark background data, the uniformity correction coefficient, the original human face hyperspectral data and the reference spectral data, and in combination with the standard reflectivity of the standard reflectivity plate, skin reflectivity data of the target human face skin is calculated. Based on the skin optical model, a relationship between the skin reflectivity data and tissue absorption coefficients is established, and the tissue absorption coefficients are expressed as linear superposition of concentrations of a plurality of target skin biochemical components and corresponding standard spectral characteristics. Based on the skin reflectivity data, a fitting optimization algorithm is used to solve the linear superposition relationship, to obtain pixel-level concentrations of each target skin biochemical component, and to generate a two-dimensional concentration distribution map.

[0006] The application is further provided that the uniformity correction coefficient is determined based on a ratio of hyperspectral signals of each pixel position of the standard integrating sphere light emitting surface at a preset wavelength to average hyperspectral signals of all pixels at the wavelength.

[0007] The application is further provided that the standardized optical acquisition conditions include: a multi-dimensional uniform light source with fixed position and light intensity, an orthogonal polarization control component with fixed polarization angle, and a facial adaptive positioning mechanism with fixed positioning points.

[0008] The application is further provided that the calculation of the skin reflectivity data of the target human face skin includes: Based on the dark background data, the original human face hyperspectral data and the reference spectral data are subjected to dark background deduction; The original human face hyperspectral data and the reference spectral data after deduction of the dark background are subjected to uniformity correction by using the uniformity correction coefficient; Based on the original human face hyperspectral data after uniformity correction, the reference spectral data and the standard reflectivity of the standard reflectivity plate, the skin reflectivity data of the human face skin is calculated.

[0009] The application is further provided that, before the relationship is established based on the skin optical model, the step of obtaining standard molar extinction coefficients of a plurality of target skin biochemical components as standard spectral characteristics, and matching the wavelength range of the standard spectral characteristics to a preset effective analysis band is further included. The plurality of target skin biochemical components include oxyhemoglobin, deoxyhemoglobin, pheomelanin, eumelanin and beta-carotene.

[0010] The application is further provided that the fitting optimization algorithm includes: The characteristic absorption peak wavelength of the target skin biochemical component is subjected to weighted processing; A nonlinear correction term describing spectral interference between the target skin biochemical components and a physiological proportion constraint are introduced in the linear superposition relationship. A fitting method combining L1 and L2 regularization constraints, multi-starting point parallel search, and dynamic boundary adjustment strategy is adopted to solve the problem; During the solution process, concentration nonnegativity constraints and physiological concentration range constraints are applied.

[0011] The present invention is further configured such that the process of solving the linear superposition relationship to obtain the pixel-level concentration of each target skin biochemical component is achieved by splitting the pixels of the facial region into multiple sub-tasks for parallel computation.

[0012] This invention also provides a device for analyzing the concentration of multiple components of facial skin based on hyperspectral imaging, the device comprising: A dark box is a sealed darkroom designed to isolate external ambient light. A head positioning mechanism, set inside the dark box, is used to fix the head posture and position of the subject being tested, including a forehead fixation bracket and a chin rest; The light source system, located inside the dark box, is used to provide standardized lighting conditions. It includes a first LED array light source, a second LED array light source, and a third LED array light source respectively located on both sides and below the head positioning mechanism. Each of the LED array light sources has an independent first front polarizer, a second front polarizer, and a third front polarizer on its outgoing light path. An image acquisition system, set inside the dark box and located directly above the head positioning mechanism, is used to acquire hyperspectral image data of the face. It includes a hyperspectral data acquisition camera and an internal push-broom hyperspectral camera lens connected to the hyperspectral data acquisition camera. A lens polarizer is provided in the acquisition optical path of the internal push-broom hyperspectral camera lens. A data processing system, communicatively connected to the hyperspectral data acquisition camera, is used to receive and process hyperspectral image data. The data processing system includes a computer connected to the acquisition camera.

[0013] The present invention is further configured such that the front polarizers on the output light paths of each LED array light source in the light source system and the front polarizers on the lens are all linear polarizers with adjustable angles, and can be configured to suppress orthogonal polarization states of facial mirror reflection.

[0014] The present invention is further configured such that the first LED array light source and the second LED array light source are symmetrically arranged on the left and right sides of the subject's face, and the third LED array light source is arranged directly below the subject's face, the three together forming a triangular lighting layout to achieve uniform facial illumination.

[0015] This invention provides a method and apparatus for multi-component concentration analysis of facial skin based on hyperspectral imaging. The method involves acquiring dark background data, acquiring uniform light field data from a standard integrating sphere to calculate a uniformity correction coefficient, and acquiring reference spectral data from a standard reflectance plate placed at a preset face acquisition position under pre-defined standardized optical acquisition conditions. Under standardized optical acquisition conditions, raw facial hyperspectral data is acquired. Based on the dark background data, uniformity correction coefficient, raw facial hyperspectral data, and reference spectral data, and combined with the standard reflectance of the standard reflectance plate, the skin reflectance data of the target facial skin is calculated. Based on a skin optical model, a relationship is established between the skin reflectance data and the tissue absorption coefficient, where the tissue absorption coefficient represents the linear superposition of the concentrations of various target skin biochemical components and their corresponding standard spectral characteristics. Based on the skin reflectance data, a fitting optimization algorithm is used to solve the linear superposition relationship, obtaining the pixel-level concentration of each target skin biochemical component and generating a two-dimensional concentration distribution map. The beneficial effects include: 1. By designing a customized acquisition device, we solve the problems of poor data quality, insufficient security, and image distortion associated with traditional detection methods, while ensuring data validity. On one hand, we use multiple LED array light sources instead of halogen lamps to accurately provide an effective detection wavelength range of 400-1000nm. This meets the wavelength range requirements of multi-component spectral analysis while avoiding the risk of skin burns from the high temperatures of halogen lamps, thus balancing detection safety and data specificity. On the other hand, we add polarizers in front of the light source and lens. By adjusting the angle of the polarizers, we suppress localized facial reflections and reduce invalid data interference caused by reflections. We also equip the device with a forehead fixation bracket and chin rest adapted to the shape of the face to fix the head position and prevent image distortion caused by facial movement during scanning. This ensures the authenticity and stability of hyperspectral reflectance data from the source of acquisition.

[0016] 2. By employing precise spectral analysis technology, we overcome the challenges of traditional detection methods, such as the inability to quantitatively analyze multiple concentrations and low fitting efficiency, thereby improving detection accuracy and real-time performance. On one hand, we construct a multi-component fitting model based on a reference spectral library, introducing reference extinction coefficient spectral curves for various skin biochemical concentrations, such as oxyhemoglobin and deoxyhemoglobin. After interpolation and matching to the effective wavelength band, we combine least squares and regularization techniques to achieve quantitative calculation of the spatial distribution of each concentration, filling the gap in traditional methods' inability to deeply analyze multiple concentrations. On the other hand, we adopt a multi-threaded parallel computing framework to process facial region pixels row by row, simultaneously advancing the extremely time-consuming fitting task of a single thread, significantly shortening the computation time, meeting the real-time requirements of clinical and cosmetic testing, and solving the industry pain point of difficulty in balancing efficiency and accuracy.

[0017] 3. Through user-friendly functions and adaptable design, it addresses the shortcomings of traditional detection methods in terms of visualization and practicality, while expanding application scenarios. On one hand, it generates a two-dimensional heatmap of the entire face at various concentrations, visually presenting the differences in concentration distribution using color gradients. Coupled with the pixel-click interaction function, users can view the concentration value of any pixel in real time, lowering the analysis threshold for professionals. On the other hand, it supports integrated data output, simultaneously saving numerical results and visualized images, facilitating subsequent medical diagnostic comparisons, cosmetic effect tracking, and research data retrospection. Furthermore, the acquisition device and algorithm can be adaptively adjusted through parameters to adapt to detection subjects with different skin tones and face shapes, ensuring consistency of detection results for different individuals.

[0018] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 A flowchart illustrating a method for multi-component concentration analysis of facial skin based on hyperspectral imaging, as shown in an exemplary embodiment of the present invention; Figure 2 A schematic diagram of a multi-component concentration analysis device for human facial skin based on hyperspectral imaging, as shown in an exemplary embodiment of the present invention; Among them, 1 is the dark box, 2 is the hyperspectral data acquisition camera, 3 is the forehead fixing bracket, 4 is the internal push-scan hyperspectral camera lens, 5 is the lens front polarizer, 6 is the first front polarizer, 7 is the third front polarizer, 8 is the first LED array light source, 9 is the third LED array light source, 10 is the acquisition connection computer, 11 is the second front polarizer, 12 is the second LED array light source, and 13 is the chin rest. Detailed Implementation

[0020] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0021] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0022] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention. Example

[0023] Methods for multi-component concentration analysis of facial skin based on hyperspectral imaging, such as Figure 1 As shown, it includes: Collect dark background data, collect uniform light field data of a standard integrating sphere to calculate uniformity correction coefficient, and collect reference spectral data of a standard reflectivity plate placed at a preset face acquisition position under preset standardized optical acquisition conditions. Under standardized optical acquisition conditions, raw hyperspectral data of human faces were acquired. Based on dark background data, uniformity correction coefficient, original face hyperspectral data and reference spectral data, and combined with the standard reflectance of the standard reflectance plate, the skin reflectance data of the target face skin is calculated. Based on the skin optical model, the relationship between skin reflectance data and tissue absorption coefficient is established. The tissue absorption coefficient is expressed as the linear superposition of the concentrations of various target skin biochemical components and their corresponding standard spectral characteristics. Based on skin reflectivity data, a fitting optimization algorithm is used to solve the linear superposition relationship, obtain the pixel-level concentration of each target skin biochemical component, and generate a two-dimensional concentration distribution map.

[0024] The invention is further configured such that the uniformity correction coefficient is determined based on the ratio of the hyperspectral signal at each pixel position on the light-emitting surface of the standard integrating sphere at a preset wavelength to the average hyperspectral signal of all pixels at that wavelength. Specifically, firstly, a standard integrating sphere with a uniform internal diffuse reflection coating and a precisely designed light-emitting aperture is securely mounted in front of the lens of a hyperspectral camera, ensuring that the light-emitting aperture of the integrating sphere completely covers the entire imaging field of view of the lens. This eliminates interference from external stray light, ensuring that the light signal entering the imaging optical path originates entirely from a standard uniform light field that is spatially highly uniform and spectrally continuous and stable, formed after multiple diffuse reflections within the integrating sphere. Subsequently, the hyperspectral data acquisition camera and the internal push-broom hyperspectral phase... The precision displacement stage built into the internal push-broom hyperspectral imaging unit, composed of the camera lens, drives the imaging module to perform a complete line scan of the integrating sphere's light-emitting surface along a preset track, thereby acquiring a complete frame of hyperspectral image data. This data completely records the original grayscale response of each pixel on the camera sensor to the aforementioned standard uniform light field within the entire effective spectral band. For each discrete wavelength point within the acquired spectral range, the arithmetic mean of the grayscale values ​​of all effective pixels in the entire image at that wavelength is calculated to characterize the standard response level of the imaging system to an ideal uniform light field. The original grayscale value of each pixel at that wavelength is obtained, and the arithmetic mean is divided by the original grayscale value; the quotient is the grayscale value of that pixel at that wavelength. The uniformity correction coefficient is defined as follows: when the original response intensity of a pixel is lower than the average response intensity of the entire imaging field of view, its corresponding uniformity correction coefficient is greater than 1, and the signal of that pixel will be positively compensated in subsequent data processing; when the original response intensity of a pixel is higher than the average response intensity, its corresponding uniformity correction coefficient is less than 1, and the signal of that pixel will be negatively corrected in subsequent data processing. The above calculation process is performed sequentially for each effective wavelength point, generating a wavelength-related uniformity correction coefficient curve for each pixel on the imaging sensor; the uniformity correction coefficient of each pixel is stored in the data processing terminal in the form of a structured data file. In the non-volatile memory, a permanent calibration file corresponding to the hyperspectral data acquisition camera, hyperspectral camera lens and its optical configuration parameters are formed. When actually acquiring raw hyperspectral data of facial skin or standard reflectivity plate, the pre-stored uniformity correction coefficient file is called to perform real-time or post-processing correction operations on the raw hyperspectral data after dark background subtraction. Specifically, the signal value of each pixel at each wavelength is multiplied by the uniformity correction coefficient value of the corresponding pixel and the corresponding wavelength in the uniformity correction coefficient file to correct the spatial response non-uniformity caused by differences in camera sensor pixel sensitivity and vignetting at the edge of the optical lens, thereby ensuring the consistency of spectral signals acquired from different positions in the imaging field of view.

[0025] The present invention is further configured such that the standardized optical acquisition conditions include: a multi-dimensional uniform light source with fixed position and light intensity, an orthogonal polarization control component with fixed polarization angle, and a facial adaptation positioning mechanism with fixed positioning points; specifically, the multi-dimensional uniform light source is constructed and fixed in the following manner: three independent LED array light sources are respectively set as the left light source, the right light source, and the lower light source, and are installed in a preset three-dimensional spatial position around the facial adaptation positioning mechanism by an adjustable mechanical bracket; the illumination angle of each light source is adjusted so that the central axis of its emitted light beam intersects at a preset calibration point in the facial area to form an overlapping and uniform illumination field; and the light sources are connected... A programmable power supply with constant current drive and feedback control functions independently adjusts the drive current of each light source during the system debugging phase, and measures it using a spectroradiometer located at the facial calibration point until the synthetic illuminance of the target facial area reaches the preset value and the spatial non-uniformity meets the preset tolerance. After debugging, the drive current parameters of each light source are saved as a configuration file, and each mechanical support is locked. During subsequent standard acquisition, the corresponding drive current is loaded by calling the configuration file to reproduce the spectral and light intensity states during calibration. The orthogonal polarization control component is constructed and fixed in the following way: a rotatable linear polarization device is installed in front of the light outlet of each LED array light source. A polarizer is used as a pre-polarizer for the light source, and a rotatable linear polarizer is installed at the front of the hyperspectral imaging lens as a pre-polarizer for the lens. A standard reference plate with high specular reflection characteristics is placed on the facial positioning mechanism, and the image is previewed through a hyperspectral camera under light source illumination. First, the angle of the pre-polarizer for the lens is fixed, and then the rotation angle of each pre-polarizer for the light source is adjusted sequentially to suppress the brightness of specular reflections on the reference plate caused by each light source in the preview image. When the brightness of all reflections is suppressed to the minimum, an orthogonal polarization state is achieved. Finally, the angles of all polarizers are locked to maintain the polarization relationship between illumination and imaging in subsequent data acquisition, thereby stably suppressing specular reflections on the skin surface. The facial adaptation and positioning mechanism consists of a three-dimensionally adjustable forehead fixation bracket and a biomimetic ergonomic chin rest. During mechanism debugging, a standard reflectivity plate is placed on a preset acquisition plane. The padding of the forehead fixation bracket is adjusted to contact the upper edge of the reflectivity plate, and the chin rest is adjusted so that its supporting surface fits against the lower edge of the reflectivity plate, thereby establishing a standardized facial spatial posture and imaging distance. After adjustment, all movable parts of the mechanism are locked to make it a fixed spatial reference. During subsequent face acquisition, the subject's forehead and chin are respectively fitted onto the locked bracket and rest to reproduce the same spatial position and orientation as the reflectivity plate, thereby eliminating optical errors caused by differences in sample position.By constructing, adjusting, and locking a multi-dimensional uniform light source, orthogonal polarization control components, and a facial adaptation and positioning mechanism, a standardized set of optical acquisition conditions was established and maintained. These conditions ensure that spectral measurements from a standard reflectivity plate to facial skin are performed under a consistent optical environment, thus providing a foundation for obtaining accurate, reliable, and comparable absolute skin reflectivity data.

[0026] The present invention is further configured such that the calculation of the skin reflectance data of the target human face skin includes: Based on dark background data, dark background subtraction is performed on the original hyperspectral data of the face and the reference spectral data; Uniformity correction coefficients are used to perform uniformity correction on the original hyperspectral data of the face after removing the dark background and the reference spectral data; Based on the homogeneous-corrected original hyperspectral data of the face, reference spectral data, and the standard reflectance of the standard reflectance plate, the skin reflectance data of the facial skin is calculated. Specifically, during the system initialization phase, dark background data collected and stored in advance under no-light conditions is called. This dark background data includes the inherent dark current, thermal noise, and readout noise of the camera sensor. From the original hyperspectral data of the target face and the reference spectral data of the standard reflectance plate, the gray values ​​of the corresponding coordinates and wavelengths in the dark background data are subtracted pixel by pixel and wavelength by wavelength according to the spatial and spectral coordinate correspondence, so as to separate and remove the systematic background noise generated by the sensor itself, thereby obtaining the original hyperspectral data and reference spectral data of the face after background subtraction that only characterize the response to external light signals. The pre-stored homogeneity correction coefficient file is called, and the original hyperspectral data and reference spectral data of the face after dark background subtraction are multiplied pixel by pixel and wavelength by the corresponding coordinates in the homogeneity correction coefficient file according to their respective spatial pixel index and spectral band index. The uniformity correction coefficients for the standard and wavelength are used to compensate for the spatial response deviations introduced by the non-uniformity of the photoelectric response of sensor pixels and the attenuation of the image field illumination of the optical lens. This ensures that the processed signal theoretically has a consistent response level to the same uniform light field, resulting in the original hyperspectral data and reference spectral data of the face after uniformity correction. The calibration file of the standard reflectivity plate is read to obtain its known standard reflectivity at each wavelength. For each effective pixel in the face region and each wavelength, the ratio between the original hyperspectral signal value of the face after uniformity correction at that wavelength and the signal value in the reference spectral data that is at a completely consistent spatial coordinate position is calculated. This ratio is then multiplied by the known standard reflectivity value of the standard reflectivity plate at the corresponding wavelength, thus completing the conversion of signal intensity relative to the standard reflectivity plate to absolute reflectivity. By traversing all effective pixels and wavelengths of the face, skin reflectivity data of the facial skin is generated. This skin reflectivity data characterizes the absolute reflectivity of specific microscopic locations on the facial skin surface to incident light of a specific wavelength.

[0027] The present invention is further configured to include the following steps before establishing the relationship based on the skin optical model: obtaining the standard molar extinction coefficients of various target skin biochemical components as their standard spectral characteristics, and matching the wavelength range of the standard spectral characteristics to a preset effective analysis band. The target skin biochemical components include oxyhemoglobin, deoxyhemoglobin, pheomelanin, eumelanin, and β-carotene. Specifically, raw spectral data of the standard molar extinction coefficients for each of the five target skin biochemical components—oxyhemoglobin, deoxyhemoglobin, pheomelanin, eumelanin, and β-carotene—are extracted from a validated database of skin optical and biochemical standard parameters. This set of raw spectral data records the characteristic absorption spectra of each component in the ultraviolet, visible, and near-infrared bands in the form of discrete wavelength points and their corresponding extinction coefficient values. To ensure that this standard spectral data is completely consistent with the preprocessed facial skin reflectance spectrum used in subsequent modeling and analysis in terms of wavelength coordinates, a wavelength range matching operation is performed: based on the effective analysis bands and their wavelength sampling intervals previously defined from the facial reflectance data, the obtained... The original extinction coefficient data of each component were recalculated using a linear interpolation algorithm. The specific implementation process of this algorithm is as follows: for each wavelength sampling point within the effective analysis band, its interval in the wavelength sequence of the corresponding component's original spectral data is first determined based on its wavelength value. Then, based on the wavelength values ​​and extinction coefficient values ​​of the two adjacent original data points before and after this interval, a new standard extinction coefficient value corresponding to that sampling point is calculated through a linear relationship. This process traverses all wavelength sampling points within the effective analysis band, ultimately generating a new standard spectral characteristic curve for each target skin biochemical component. The wavelength coordinate sequence of this curve is perfectly aligned with the wavelength coordinate sequence of the measured human face reflectance spectrum. These precisely matched standard spectral characteristic data constitute the basic spectral library used in subsequent skin optical models to quantitatively describe tissue absorption characteristics.

[0028] The present invention is further configured such that the fitting optimization algorithm includes: Weighted processing of the characteristic absorption peak wavelengths of the target skin biochemical components; A nonlinear correction term and physiological proportion constraint describing the spectral interference between biochemical components of the target skin are introduced into the linear superposition relationship; A fitting method combining L1 and L2 regularization constraints, multi-starting point parallel search, and dynamic boundary adjustment strategy is adopted to solve the problem; During the solution process, concentration nonnegativity constraints and physiological concentration range constraints are applied. Specifically, based on prior knowledge of the characteristic absorption spectra of the target skin biochemical components, specific wavelengths corresponding to key characteristic absorption peaks are identified and determined within a pre-defined effective analytical band. For example, the absorption peaks of oxyhemoglobin and deoxyhemoglobin near 542 nm and 577 nm, respectively, are assigned higher weight coefficients than other bands. This weighting strategy assigns greater weight to the characteristic absorption peak wavelengths when constructing the objective function that minimizes the spectral fitting residuals, thereby prioritizing the consistency between the calculated reflectance and the measured reflectance at these key spectral positions. This enhances the analytical sensitivity and inversion accuracy for changes in the concentration of core components, while suppressing noise in non-characteristic bands. To mitigate interference, a nonlinear correction term is introduced on the basic model established based on skin optics theory, which is based on the principle that the total absorption coefficient of a tissue is equal to the linear superposition of the concentrations of each component and their standard spectral characteristics. This term is constructed through the product or higher-order exponential mathematical form of the target skin biochemical component concentrations. It is used to quantify the possible spectral interference effects and non-independent scattering contributions between component molecules in the high concentration range, so as to more accurately simulate the actual optical response characteristics of skin tissue. At the same time, a constraint mechanism based on physiological knowledge is integrated. For example, a reasonable dynamic fluctuation range for the concentration ratio of oxyhemoglobin to deoxyhemoglobin is preset according to the physiological state of skin microcirculation, and the concentration ratio of eumelanin to pheomelanin is preset according to the biological laws of skin pigment synthesis and metabolism. For example, constraints are imposed to ensure that the concentration relationship obtained by inversion conforms to the basic physiological characteristics of human skin. To solve the inversion model integrating linear basis, nonlinear correction, and physiological constraints, an objective function is constructed with minimizing the spectral fitting residual as its core. To suppress overfitting and improve the robustness of the solution, L1 and L2 regularization terms are introduced into the objective function. The L1 regularization term, by penalizing the absolute value of the concentration parameter, helps to obtain sparse solutions, thereby identifying the main contributing components. The L2 regularization term, by penalizing the square value of the concentration parameter, achieves smooth constraint on the solution vector and enhances numerical stability. To avoid the risk of nonlinear optimization problems getting trapped in local optima, a multi-starting point parallel search strategy is implemented, that is, starting from multiple preset different initial concentrations... The degree vector simultaneously initiates an iterative optimization process. During the iteration, the search boundaries of each component concentration are dynamically adjusted based on the convergence of the current fitting error to accelerate the optimization process. The optimization processes of all starting points are executed in parallel. Finally, the solution that minimizes the objective function value is selected from all converged results as the optimal concentration estimate for that pixel. Throughout the iterative optimization process, dual hard constraints of physical feasibility and physiological rationality are enforced. The physical constraint is that the concentration estimates of all target skin biochemical components must meet the non-negativity requirement. The physiological constraint is based on existing skin physiology research data to pre-set effective physiological ranges for the concentration of each component. For example, the total hemoglobin concentration and the total melanin concentration must fall within the known typical concentration range of human skin.After each iteration, the concentration components are checked to ensure they meet the constraints outlined above. If any concentration value exceeds the preset non-negative boundary or physiological range, it is immediately corrected to the preset feasible region using numerical projection or boundary truncation methods. This dual constraint mechanism ensures that the concentration values ​​of each component at each pixel obtained in the final inversion are not only mathematically optimal solutions to the optimization problem, but also satisfy the basic definitions of physical quantities and conform to the known physiological characteristics of human skin, thus guaranteeing the physical interpretability and biomedical reliability of the inversion results.

[0029] The present invention is further configured such that the process of solving the linear superposition relationship to obtain the pixel-level concentration of each target skin biochemical component is achieved by splitting the pixels of the facial region into multiple sub-tasks for parallel computation; specifically, after completing the extraction of the effective skin region of the face, a dataset containing a large-scale spatial pixel set is obtained, where each pixel corresponds to a reflectance spectrum curve that needs to be independently inverted for multi-component concentration; to achieve efficient solution, a parallel computing architecture is adopted to accelerate the overall processing. In specific implementation, the computational tasks are first divided according to the resource scale of the parallel computing hardware: the effective pixel region of the face is regarded as a The image is divided into multiple similarly sized, spatially contiguous sub-blocks, each containing several adjacent rows of pixels or a rectangular set of pixels, forming an independent computational subtask. The goal of task partitioning is to match the number of generated subtasks with the number of available parallel computing units, thereby achieving a balanced distribution of computational load. Subsequently, the parallel computing process is initiated: the main control program creates a task queue and loads all subtasks into it, while simultaneously creating multiple worker threads corresponding to the number of computing resources. Each activated worker thread dynamically obtains a subtask to be processed from the task queue and then enters its independent, enclosed process. The computation loop is as follows: Within this loop, the worker thread independently executes a complete iterative optimization computation process for each pixel contained in the acquired subtask. This process includes reading the reflectance data of that pixel, calling pre-stored standard spectral features and algorithm parameters, running a fitting inversion algorithm that integrates spectral weighting, nonlinear correction, regularization constraints, multi-start point search, and dynamic boundary adjustment, until convergence and obtaining the optimal concentration estimate of all target skin biochemical components at that pixel. This result is then written to the corresponding storage location in a pre-allocated, thread-safe shared result array. After completing all computations for a subtask, the worker thread... The worker thread immediately returns to the task queue to obtain the next task to be processed. This dynamic scheduling mechanism ensures that all computing units are continuously and efficiently utilized. When all worker threads have finished executing and the task queue is empty, the main thread determines that the parallel computation is complete. Then, it collects and reassembles the concentration inversion results of all pixels from the shared result array according to the original spatial order, thereby generating a two-dimensional concentration distribution map of multi-target biochemical components covering the entire face area. This parallel processing strategy significantly reduces the time consumption of the originally time-consuming serial computation process, enabling pixel-level quantitative analysis of high spatial resolution hyperspectral facial images to meet the strict requirements of processing efficiency in practical application scenarios. Example

[0030] Please see Figure 2 The exemplary hyperspectral imaging-based multi-component concentration analysis device for facial skin includes: Dark box 1, whose interior is a sealed darkroom for isolating external ambient light; A head positioning mechanism, set inside the dark box 1, is used to fix the head posture and position of the subject being tested, including a forehead fixing bracket 3 and a chin rest 13; A light source system, housed within the dark box 1, provides standardized illumination conditions. It includes a first LED array light source 8, a second LED array light source 12, and a third LED array light source 9, respectively positioned on both sides and below the head positioning mechanism. Each LED array light source has an independent first front polarizer 6, a second front polarizer 11, and a third front polarizer 7 on its outgoing light path. Further, the front polarizers on the outgoing light paths of each LED array light source and the lens front polarizer 5 in the light source system are all angle-adjustable linear polarizers, and can be configured to suppress orthogonal polarization of facial specular reflections. An image acquisition system is set inside the dark box 1 and located directly above the head positioning mechanism. It is used to acquire hyperspectral image data of the face. The system includes a hyperspectral data acquisition camera 2 and an internal push-broom hyperspectral camera lens 4 connected to the hyperspectral data acquisition camera 2. A lens polarizer 5 is provided in the acquisition optical path of the internal push-broom hyperspectral camera lens 4. The data processing system, communicatively connected to the hyperspectral data acquisition camera 2, is used to receive and process hyperspectral image data. The data processing system includes a computer 10 connected to the acquisition system. Specifically, after the user enters the detection position, their forehead and chin are respectively placed against the mechanically locked forehead fixation bracket 3 and ergonomic chin rest. The head positioning mechanism standardizes and fixes the head posture and three-dimensional spatial position. The darkroom 1 is closed to form a completely sealed acquisition darkroom that isolates external ambient light. When detection is initiated, multiple LED array light sources in the light source system, respectively located on both sides and below the head positioning mechanism, activate according to pre-stored... The driving parameters are synchronously illuminated, providing illumination from multiple directions to form a uniform triangular lighting layout. This spatial layout, through multi-directional light supplementation, is used to minimize shadows caused by facial contours, such as the nostrils and eye sockets. Adjustable linear polarizers installed on the output light paths of each LED array light source and the acquisition light path of the internal push-scan hyperspectral camera lens 4 are pre-configured to have mutually orthogonal polarization directions. This configuration utilizes the principle of polarization extinction to effectively suppress specular reflection light from the smooth skin surface, ensuring that the light signal entering the imaging system is primarily diffuse reflection carrying subcutaneous biochemical information after scattering by skin tissue. The light source enhances the signal-to-noise ratio; under the established standardized illumination and polarization conditions, the image acquisition system begins operation: its internal push-broom hyperspectral camera lens 4 performs a one-dimensional scan driven by a precision displacement stage, while the hyperspectral data acquisition camera 2 simultaneously performs spectral imaging, thereby acquiring a hyperspectral data cube covering the entire face area and containing spatial and spectral three-dimensional information; throughout the acquisition process, head positioning, light source intensity and spectrum, and polarization state are strictly consistent with the system calibration phase; the acquired raw hyperspectral data is transmitted to the data processing system in real time, which automatically processes the data according to a predetermined workflow. The system retrieves and stores dark background data, uniformity correction coefficients, and standard reflectance plate reference data acquired during the system calibration phase. It then sequentially performs dark background subtraction, uniformity correction, and reflectance calculation on the raw data. Finally, it performs multi-component spectral unmixing and inversion based on the skin optical model to generate a spatial concentration distribution map of each target biochemical component on the human face skin. The entire process, from environmental shading, standardized optical condition control, data acquisition to processing and analysis, is automatically completed in this integrated device. Through the strict reproduction of hardware states and the standardized execution of processing algorithms, the high controllability of the measurement process and the repeatability and comparability of the results are ensured.

[0031] It should be noted that the facial skin multi-component concentration analysis device based on hyperspectral imaging provided in the above embodiments and the facial skin multi-component concentration analysis method based on hyperspectral imaging provided in the above embodiments belong to the same concept. The specific operation methods of each module and unit have been described in detail in the method embodiments and will not be repeated here. In practical applications, the facial skin multi-component concentration analysis device based on hyperspectral imaging provided in the above embodiments can be assigned to different functional modules as needed, that is, the internal structure of the system can be divided into different functional modules to complete all or part of the functions described above. This is not a limitation here.

[0032] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for multi-component concentration analysis of facial skin based on hyperspectral imaging, characterized in that, include: Collect dark background data, collect uniform light field data of standard integrating sphere to calculate uniformity correction coefficient, and collect reference spectral data of standard reflectivity plate placed at preset face acquisition position under preset standardized optical acquisition conditions. Under standardized optical acquisition conditions, raw hyperspectral data of human faces were acquired. Based on dark background data, uniformity correction coefficient, original face hyperspectral data and reference spectral data, and combined with the standard reflectance of the standard reflectance plate, the skin reflectance data of the target face skin is calculated. Based on the skin optical model, the relationship between skin reflectance data and tissue absorption coefficient is established. The tissue absorption coefficient is expressed as the linear superposition of the concentrations of various target skin biochemical components and their corresponding standard spectral characteristics. Based on skin reflectivity data, a fitting optimization algorithm is used to solve the linear superposition relationship, obtain the pixel-level concentration of each target skin biochemical component, and generate a two-dimensional concentration distribution map.

2. The method for multi-component concentration analysis of facial skin based on hyperspectral imaging according to claim 1, characterized in that, The uniformity correction coefficient is determined based on the ratio of the hyperspectral signal at each pixel position on the light-emitting surface of the standard integrating sphere at a preset wavelength to the average hyperspectral signal of all pixels at that wavelength.

3. The method for multi-component concentration analysis of facial skin based on hyperspectral imaging according to claim 1, characterized in that, The standardized optical acquisition conditions include: using a multi-dimensional uniform light source with fixed position and light intensity, using an orthogonal polarization control component with fixed polarization angle, and using a facial adaptation positioning mechanism with fixed positioning points.

4. The method for multi-component concentration analysis of facial skin based on hyperspectral imaging according to claim 1, characterized in that, The skin reflectance data for calculating the target facial skin includes: Based on dark background data, dark background subtraction is performed on the original hyperspectral data of the face and the reference spectral data; Uniformity correction coefficients are used to perform uniformity correction on the original hyperspectral data of the face after removing the dark background and the reference spectral data; Based on the original hyperspectral data of the face after uniformity correction, the reference spectral data, and the standard reflectance of the standard reflectance plate, the skin reflectance data of the face skin are calculated.

5. The method for multi-component concentration analysis of facial skin based on hyperspectral imaging according to claim 1, characterized in that, Before establishing the relationship based on the skin optical model, the steps include: obtaining the standard molar extinction coefficients of various target skin biochemical components as their standard spectral characteristics, and matching the wavelength range of the standard spectral characteristics to the preset effective analysis band; The various target skin biochemical components include oxyhemoglobin, deoxyhemoglobin, pheomelanin, eumelanin, and β-carotene.

6. The method for multi-component concentration analysis of facial skin based on hyperspectral imaging according to claim 1, characterized in that, The fitting optimization algorithm includes: Weighted processing of the characteristic absorption peak wavelengths of the target skin biochemical components; A nonlinear correction term and physiological proportion constraint describing the spectral interference between biochemical components of the target skin are introduced into the linear superposition relationship; A fitting method combining L1 and L2 regularization constraints, multi-starting point parallel search, and dynamic boundary adjustment strategy is adopted to solve the problem; During the solution process, concentration nonnegativity constraints and physiological concentration range constraints are applied.

7. The method for multi-component concentration analysis of facial skin based on hyperspectral imaging according to claim 1, characterized in that, The process of solving the linear superposition relationship to obtain the pixel-level concentration of each target skin biochemical component is achieved by splitting the pixels of the facial region into multiple sub-tasks for parallel computation.

8. A multi-component concentration analysis device for facial skin based on hyperspectral imaging, used to implement the multi-component concentration analysis method for facial skin based on hyperspectral imaging as described in any one of claims 1-7, characterized in that, include: A dark box is a sealed darkroom designed to isolate external ambient light. A head positioning mechanism, set inside the dark box, is used to fix the head posture and position of the subject being tested, including a forehead fixation bracket and a chin rest; The light source system, located inside the dark box, is used to provide standardized lighting conditions. It includes a first LED array light source, a second LED array light source, and a third LED array light source respectively located on both sides and below the head positioning mechanism. Each of the LED array light sources has an independent first front polarizer, a second front polarizer, and a third front polarizer on its outgoing light path. An image acquisition system, set inside the dark box and located directly above the head positioning mechanism, is used to acquire hyperspectral image data of the face. It includes a hyperspectral data acquisition camera and an internal push-broom hyperspectral camera lens connected to the hyperspectral data acquisition camera. A lens polarizer is provided in the acquisition optical path of the internal push-broom hyperspectral camera lens. A data processing system, communicatively connected to the hyperspectral data acquisition camera, is used to receive and process hyperspectral image data. The data processing system includes a computer connected to the acquisition camera.

9. The facial skin multi-component concentration analysis device based on hyperspectral imaging according to claim 8, characterized in that, The front polarizers on the output optical paths of each LED array light source in the light source system and the front polarizers on the lens are all linear polarizers with adjustable angles, and can be configured to suppress orthogonal polarization states for facial mirror reflection.

10. The facial skin multi-component concentration analysis device based on hyperspectral imaging according to claim 8, characterized in that, The first LED array light source and the second LED array light source are symmetrically arranged on the left and right sides of the subject's face, and the third LED array light source is arranged directly below the subject's face. Together, they form a triangular lighting layout to achieve uniform facial illumination.

Images