An optimized parameter method and system for treating abnormal areas of skin with sod

Abstract

The present application relates to the technical field of computer, especially to a kind of optimization parameter method and system for treating skin abnormal area using SOD, the method can be flexibly selected according to actual application scene with reference to skin area image, SOD basic parameter vector for assisting SOD component parameter adjustment can be determined through image analysis and data processing process, SOD basic parameter vector does not need to be obtained by experimental method, and the application scene of SOD basic parameter vector can be flexibly adjusted by replacing several reference skin area images for processing, to make it possible to optimize SOD component parameter according to SOD basic parameter vector as reference, compared with the way of existing technology close to traversal SOD component parameter experiment, it can reduce the number of experiments in the process of SOD component research, thereby improving the efficiency of SOD component parameter optimization, and SOD basic parameter vector corresponding to different scenes can be provided, further improve the efficiency of SOD component parameter optimization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to an optimized parameter method and system for treating abnormal skin area by SOD. BACKGROUND

[0002] At present, in the prior art, there is a cosmetic containing superoxide dismutase (SOD) component, and the SOD can specifically remove free radicals (such as superoxide anion) and protect cells from oxidative damage.

[0003] However, different proportions of SOD components may cause the cosmetic containing SOD components to have oxidative stress imbalance when acting on the skin area, resulting in abnormal physiological indicators or aggravating the abnormal degree of physiological indicators of the skin area. Therefore, how to select appropriate SOD component parameters has become a difficult and painful point.

[0004] To solve the above problems, the prior art usually determines the SOD component parameters by a large number of cell experiments combined with cell activity curve drawing. However, in order to ensure the reliability of the drawn curve, a relatively complete experiment is usually required, that is, a large number of experimental times are required, thereby resulting in poor efficiency of SOD component parameter optimization.

[0005] In addition, in the prior art, the SOD component parameter optimization is usually targeted at universality. When facing a specific type of skin abnormal area, the determined SOD component parameters may not be suitable for the scene. If parameter optimization needs to be performed according to the scene, the repeated experimental process is required, further resulting in poor efficiency of SOD component parameter optimization.

[0006] Therefore, how to improve the efficiency of SOD component parameter optimization has become a problem to be solved. SUMMARY

[0007] To solve the above technical problems, the technical solution adopted by the present application is an optimized parameter method for treating abnormal skin area by SOD, which comprises the following steps:

[0008] S101, inputting the obtained M reference skin area images into the trained image encoder respectively to obtain reference feature vectors corresponding to the M reference skin area images respectively, wherein M is a positive integer.

[0009] S102, predicting the predicted abnormal class and the predicted physiological indicator corresponding to the M reference feature vectors respectively according to the M reference feature vectors, the trained first full connection layer and the trained second full connection layer.

[0010] S103, for any preset abnormal category, the same as the preset abnormal category corresponding to the same prediction of abnormal category of several reference feature vectors are determined as temporary feature vectors.

[0011] S104, according to the respective prediction physiological index corresponding to each temporary feature vector, the reference physiological index corresponding to the preset abnormal category is determined.

[0012] S105, according to the reference physiological index corresponding to each preset abnormal category, the first index update vector is determined.

[0013] S106, according to the several prediction physiological indexes and reference physiological indexes corresponding to the preset abnormal category satisfying the first preset condition, the second index update vector and the third index update vector are determined.

[0014] S107, according to the first index update vector, the second index update vector and the third index update vector, the target index vector is determined.

[0015] S108, according to the target index vector and the trained SOD parameter optimization model, a plurality of SOD basic parameter vectors are determined, and the SOD basic parameter vector is used to assist in adjusting the component parameter of SOD.

[0016] The application also provides an optimization parameter system for treating abnormal skin area by SOD, which comprises:

[0017] The feature extraction module is used for inputting the obtained M reference skin area images into the trained image encoder respectively, so as to obtain the reference feature vectors corresponding to the M reference skin area images respectively, wherein M is a positive integer.

[0018] The model prediction module is used for predicting the prediction abnormal category and the prediction physiological index corresponding to the M reference feature vectors according to the M reference feature vectors, the trained first full connection layer and the trained second full connection layer.

[0019] The feature selection module is used for determining the temporary feature vectors corresponding to the several reference feature vectors of the same prediction abnormal category as the preset abnormal category for any preset abnormal category.

[0020] The index determination module is used for determining the reference physiological index corresponding to the preset abnormal category according to the respective prediction physiological index corresponding to each temporary feature vector.

[0021] The first vector determination module is used for determining the first index update vector according to the reference physiological index corresponding to each preset abnormal category.

[0022] A second vector determination module is configured to determine a second index update vector and a third index update vector according to the predicted physiological indexes and the reference physiological indexes corresponding to the preset abnormality category satisfying the first preset condition.

[0023] A third vector determination module is configured to determine a target index vector according to the first index update vector, the second index update vector and the third index update vector.

[0024] A parameter optimization module is configured to determine a plurality of SOD basic parameter vectors according to the target index vector and a trained SOD parameter optimization model, the SOD basic parameter vectors being used to assist in adjusting the component parameters of the SOD.

[0025] The present application has at least the following beneficial effects: the reference skin area image can be flexibly selected according to the actual application scenario, the SOD basic parameter vector used to assist in adjusting the component parameters of the SOD can be determined through the image analysis and data processing process, the SOD basic parameter vector does not need to be obtained through an experimental method, and the application scenario of the SOD basic parameter vector can be flexibly adjusted by replacing a plurality of reference skin area images processed, so that the SOD component parameter optimization can be performed according to the SOD basic parameter vector as a reference, compared with the prior art of performing experiments close to the traversal of the SOD component parameters, the number of experiments in the SOD component research process can be reduced, thereby improving the efficiency of the SOD component parameter optimization, and the corresponding SOD basic parameter vector can be provided for different scenarios, further improving the efficiency of the SOD component parameter optimization. BRIEF DESCRIPTION OF DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0027] Figure 1 A flowchart of an optimization parameter method for processing skin abnormal area by SOD provided by the first embodiment of the present application is shown in the figure.

[0028] Figure 2 A structural diagram of an optimization parameter system for processing skin abnormal area by SOD provided by the second embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0029] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be clearly and completely described in order to make the technical solutions in the embodiments of the present application apparent to those skilled in the art. Obviously, the described embodiments are only a part rather than all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of the present application.

[0030] It should be noted that the terms "first", "second" and the like in the description and claims of the present application and the above-mentioned accompanying drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It can be understood that the above-mentioned terms for distinguishing similar objects can be interchanged under appropriate circumstances, so that the present application can also be implemented in other embodiments in addition to the above-illustrated embodiments or described embodiments. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or server including a series of steps or units does not necessarily have to include only those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to the process, method, product or device.

[0031] Embodiment one

[0032] The embodiment one provides an optimal parameter method for treating abnormal skin area by SOD, as shown in the figure, a flowchart of an optimal parameter method for treating abnormal skin area by SOD provided by the embodiment one of the present application, the optimal parameter method for treating abnormal skin area by SOD includes the following steps: Figure 1

[0033] S101, input the obtained M reference skin area images into the trained image encoder respectively to obtain the reference feature vectors corresponding to the M reference skin area images respectively, wherein M is a positive integer;

[0034] S102, according to the M reference feature vectors, the trained first full connection layer and the trained second full connection layer, the predicted abnormal categories and the predicted physiological indicators corresponding to the M reference feature vectors are predicted respectively;

[0035] S103, for any preset abnormal category, a plurality of reference feature vectors corresponding to the same predicted abnormal category as the preset abnormal category are determined as temporary feature vectors;

[0036] S104, according to the predicted physiological indicators corresponding to each temporary feature vector, the reference physiological indicators corresponding to the preset abnormal category are determined;

[0037] S105, according to the reference physiological indicators corresponding to each preset abnormal category, a first index update vector is determined;​

[0038] S106, determining a second index update vector and a third index update vector according to the reference physiological indicators and the predicted physiological indicators corresponding to the preset abnormal category satisfying the first preset condition;

[0039] S107, determining a target index vector according to the first index update vector, the second index update vector and the third index update vector;

[0040] S108, determining a plurality of SOD basic parameter vectors according to the target index vector and the trained SOD parameter optimization model, the SOD basic parameter vectors being used to assist in adjusting the component parameters of the SOD.

[0041] The reference skin region image can be a human skin region image collected, the image encoder can be used to extract feature information of an input image, and the structure of the image encoder can at least include a convolution layer, a pooling layer and a normalization layer. The image encoder can be implemented by using an encoder of a ResNet model, a U-Net model or the like.

[0042] The reference feature vector can be used to represent image features of the corresponding reference skin region image. The first full connection layer can be used to implement a classification task, that is, to map the reference feature vector to a corresponding predicted abnormal category. The second full connection layer can be used to implement a regression task, that is, to map the reference feature vector to a corresponding predicted physiological indicator.

[0043] A single predicted abnormal category is one of a plurality of preset abnormal categories. The predicted physiological indicator can be represented by a 1xK dimension vector, where K is an index dimension. Different index dimensions correspond to different index types. The index types can include oxidative stress level, cell activity, degree of vascular dilation, trans-epidermal water loss and the like.

[0044] A single preset abnormal category corresponds to a plurality of temporary feature vectors classified as the preset abnormal category. The reference physiological indicator is also represented by a 1xK dimension vector.

[0045] The first index update vector can be used to represent an optimization direction of the physiological indicator. The second index update vector and the third index update vector can be used to represent optimization values of the physiological indicator in different optimization directions.

[0046] The target index vector can be used to obtain an expected physiological indicator result. The SOD basic parameter vector can be used to provide an initial SOD component parameter for a cosmetic researcher, so that the optimization of the SOD component parameter can be based on the SOD basic parameter vector, thereby reducing a large number of experiment times and improving the optimization efficiency of the SOD component parameter.

[0047] In a specific embodiment, the predicting, according to the M reference feature vectors, the trained first fully connected layer and the trained second fully connected layer, the predicted abnormal class and the predicted physiological index corresponding to each of the M reference feature vectors comprises:

[0048] inputting each of the M reference feature vectors into the trained first fully connected layer for feature mapping to obtain the predicted abnormal class corresponding to each of the M reference feature vectors;

[0049] inputting each of the M reference feature vectors into the trained second fully connected layer for feature mapping to obtain the predicted physiological index corresponding to each of the M reference feature vectors.

[0050] For any reference feature vector, inputting the reference feature vector into the trained first fully connected layer for feature mapping to obtain the predicted value corresponding to each preset abnormal class, normalizing and mapping the predicted value corresponding to each preset abnormal class by using a Softmax function to obtain the predicted probability corresponding to each preset abnormal class, and taking the preset abnormal class corresponding to the maximum predicted probability as the predicted abnormal class of the reference feature vector.

[0051] In a specific embodiment, the training process of the image encoder, the first fully connected layer and the second fully connected layer comprises the following steps:

[0052] obtaining a plurality of sample skin region images, wherein each sample skin region image corresponds to a sample abnormal class and a sample physiological index;

[0053] optionally selecting two sample skin region images, inputting the selected two sample skin region images into the image encoder and a preset auxiliary encoder respectively to obtain the first sample feature output by the image encoder and the second sample feature output by the auxiliary encoder, wherein the auxiliary encoder is the same as the image encoder;

[0054] calculating a first training loss according to the first sample feature, the second sample feature, the sample abnormal class corresponding to each of the selected two sample skin region images and a preset first loss function;

[0055] inputting the first sample feature into the first fully connected layer and the second fully connected layer respectively to obtain the first predicted sample class output by the first fully connected layer and the first predicted sample index output by the second fully connected layer;

[0056] inputting the second sample feature into the first fully connected layer and the second fully connected layer respectively to obtain the second predicted sample class output by the first fully connected layer and the second predicted sample index output by the second fully connected layer.

[0057] a second training loss is calculated according to the first predicted sample category, the second predicted sample category, and a preset second loss function;

[0058] a third training loss is calculated according to the first predicted sample index, the second predicted sample index, and a preset third loss function;

[0059] a target training loss is determined according to the first training loss, the second training loss, and the third training loss, and model parameters corresponding to the image encoder, the first fully connected layer, and the second fully connected layer are updated according to the target training loss until the target training loss converges, thereby obtaining the trained image encoder, the trained first fully connected layer, and the trained second fully connected layer.

[0060] wherein a single sample abnormality category is one of a plurality of preset abnormality categories, and the sample abnormality category and the sample physiological index corresponding to the sample skin region image can be directly obtained through existing public information.

[0061] The auxiliary encoder has the same architecture as the image encoder, and the parameters of the auxiliary encoder and the image encoder are always the same, that is, the auxiliary encoder and the image encoder form a twin network structure.

[0062] Specifically, the preset first loss function can adopt a contrast loss function, and the preset first loss function can be represented as L1=yd 2 +(1-y)max(margin-d,0) 2 wherein L1 is the first training loss, y is the category matching information, when the sample abnormality categories corresponding to the two selected sample skin region images are the same, y=1, otherwise, y=0, d is the Euclidean distance between the first sample feature and the second sample feature, and margin is a distance threshold value. It can be known that when y=1, L1=d 2 that is, when the sample abnormality categories corresponding to the two input sample skin region images are the same, the first training loss supervises the Euclidean distance between the first sample feature and the second sample feature to be as small as possible, so that the first sample feature and the second sample feature are as close as possible, and when y=0, L1=max(margin-d,0) 2That is, when the two input sample skin region images correspond to different sample abnormal categories, the first training loss supervises the Euclidean distance between the first sample feature and the second sample feature to be as close to the distance threshold as possible, so that the first sample feature and the second sample feature have as large a difference as possible. The first training loss can enable the image encoder to better learn the feature information capable of distinguishing the sample abnormal categories, so that when the intermediate feature vectors are screened according to the temporary feature vectors, the edges can be better removed, and thus the reliability of the reference physiological indicators can be improved.

[0063] The preset second loss function can be a cross-entropy loss function, and the preset third loss function can be a mean square error loss function.

[0064] In a specific embodiment, the reference physiological indicators corresponding to the preset abnormal category are determined according to the predicted physiological indicators respectively corresponding to each temporary feature vector, including:

[0065] Each temporary feature vector is clustered by using a preset clustering algorithm to obtain a plurality of clustering sets;

[0066] According to the preset number threshold and the number of temporary feature vectors respectively contained in each clustering set, a plurality of intermediate sets are determined from the plurality of clustering sets;

[0067] Each temporary feature vector respectively contained in each intermediate set is taken as an intermediate feature vector;

[0068] The reference physiological indicators corresponding to the preset abnormal category are determined according to the predicted physiological indicators respectively corresponding to each intermediate feature vector.

[0069] The preset clustering algorithm uses a DBSCAN clustering algorithm to perform clustering processing according to the Euclidean distances between the temporary feature vectors.

[0070] The preset number threshold can be determined according to the number of temporary feature vectors respectively contained in each clustering set, for example, using the median, mean, etc. of the number of temporary feature vectors respectively contained in each clustering set, or combining Otsu thresholding to process the number of temporary feature vectors respectively contained in each clustering set to obtain the preset number threshold.

[0071] Specifically, each clustering set containing a number of temporary feature vectors greater than the preset number threshold is taken as an intermediate set.

[0072] In a specific embodiment, the reference physiological indicators corresponding to the preset abnormal category are determined according to the predicted physiological indicators respectively corresponding to each temporary feature vector, including:

[0073] The mean value calculation result is taken as the reference physiological index corresponding to the preset abnormal category.

[0074] The implementer can adjust the calculation method of the reference physiological index according to the actual situation, for example, using the median, quartile, etc. It should be noted that the calculation of the reference physiological index needs to be calculated respectively according to the index dimension.

[0075] In a specific embodiment, the preset abnormal category is a normal skin category or an abnormal skin category.

[0076] The first index update vector is determined according to the reference physiological index corresponding to each preset abnormal category, including:

[0077] The reference physiological index corresponding to the abnormal skin category is subtracted from the reference physiological index corresponding to the normal skin category to obtain a subtraction result.

[0078] The subtraction result is processed by using a sign function to obtain the first index update vector.

[0079] In this embodiment, the preset abnormal category is set to two categories of normal skin category and abnormal skin category.

[0080] Specifically, the reference physiological index corresponding to the normal skin category can be considered as a relatively ideal reference physiological index, and it is expected that the reference physiological index corresponding to the abnormal skin category can be adjusted to be close to the reference physiological index corresponding to the normal skin category. Therefore, the reference physiological index corresponding to the abnormal skin category is subtracted from the reference physiological index corresponding to the normal skin category to obtain a subtraction result, and then the subtraction result is processed by using a sign function to obtain a first index update vector. It can be known that the first index update vector only contains elements with element values of -1 or 1, and the element value can represent the optimization direction of the corresponding index dimension.

[0081] In a specific embodiment, the first preset condition is to belong to the normal skin category.

[0082] The second index update vector and the third index update vector are determined according to the plurality of predicted physiological indexes and the reference physiological index corresponding to the preset abnormal category satisfying the first preset condition, including:

[0083] The lower limit and the upper limit of the physiological index are determined according to the plurality of predicted physiological indexes corresponding to the normal skin category.

[0084] The second index update vector is determined according to the lower limit of the physiological index and the reference physiological index.

[0085] determine the third index update vector according to the physiological index upper limit and the reference physiological index.

[0086] The physiological index upper limit and the physiological index lower limit are also determined according to the index dimensions, that is, for any index dimension, the maximum element value of each predicted physiological index corresponding to the normal skin category in the index dimension is taken as the upper limit value, and the minimum element value of each predicted physiological index corresponding to the normal skin category in the index dimension is taken as the lower limit value, and the upper limit value corresponding to each index dimension forms the physiological index upper limit, and the lower limit value corresponding to each index dimension forms the physiological index lower limit.

[0087] Specifically, the second index update vector can be obtained by subtracting the reference physiological index from the physiological index lower limit, and the third index update vector can be obtained by subtracting the reference physiological index from the physiological index upper limit.

[0088] The second index update vector and the third index update vector can be used to ensure that the physiological index optimization result will not cause the normal skin area to become an abnormal skin area, and improve the scene adaptability of the SOD component parameter optimization.

[0089] In a specific embodiment, the first index update vector, the second index update vector and the third index update vector each correspond to K index dimensions, and K is a positive integer.

[0090] The determination of the target index vector according to the first index update vector, the second index update vector and the third index update vector includes:

[0091] For any index dimension, if the element corresponding to the index dimension in the first index update vector is a first preset value, the element corresponding to the index dimension in the second index update vector is taken as the target element corresponding to the index dimension.

[0092] If the element corresponding to the index dimension in the first index update vector is a second preset value, the element corresponding to the index dimension in the third index update vector is taken as the target element corresponding to the index dimension.

[0093] The target index vector is formed by the target elements corresponding to each index dimension.

[0094] The first preset value is -1, and the second preset value is 1.

[0095] In a specific embodiment, the determination of the plurality of SOD basic parameter vectors according to the target index vector and the trained SOD parameter optimization model includes:

[0096] A plurality of SOD initial parameter vectors are randomly generated.

[0097] inputting the SOD initial parameter vector into the trained SOD parameter optimization model to obtain an initial index vector;

[0098] According to the initial index vector, the target index vector, and a preset fourth loss function, an optimization loss is calculated, and the SOD initial parameter vector is updated according to the optimization loss until the optimization loss converges, so as to obtain an SOD temporary parameter vector corresponding to the SOD initial parameter vector;

[0099] Each SOD initial parameter vector is traversed to obtain an SOD temporary parameter vector corresponding to each SOD initial parameter vector respectively.

[0100] Each SOD temporary parameter vector is processed to obtain a plurality of SOD basic parameter vectors.

[0101] The preset fourth loss function can be a mean square error loss function. When the input SOD initial parameter vectors are different, the parameter optimization based on the optimization loss can have different optimization results, so as to obtain a plurality of SOD basic parameter vectors.

[0102] Specifically, when the cosmetic researcher determines the SOD component parameters, the SOD basic parameter vector can be selected as the initial value, and the parameter optimization is performed through experiments. Since the initial value that satisfies the expected physiological index change in a specific scene is given, the number of experiments can be greatly reduced, and the optimization efficiency of the SOD component parameters is improved. Moreover, the implementer can select different reference skin region images according to actual needs. For example, if it is expected to adjust the SOD component parameters for a specific skin abnormality category, only a plurality of reference skin region images belonging to the specific skin abnormality category and a plurality of reference skin region images belonging to the normal skin category are needed to execute the optimization parameter method for processing the skin abnormality region by the SOD provided in this embodiment, so as to obtain the SOD basic parameter vector for the specific skin abnormality category, and then perform experimental optimization, so as to flexibly adapt to different needs and efficiently optimize the SOD component parameters when the scene changes.

[0103] In the first embodiment, the reference skin area image can be flexibly selected according to the actual application scene, and the SOD basic parameter vector used for assisting the SOD component parameter adjustment can be determined through the image analysis and data processing process. The SOD basic parameter vector does not need to be obtained through an experimental method, and the application scene of the SOD basic parameter vector can be flexibly adjusted by replacing a plurality of reference skin area images processed. Therefore, the SOD component parameter optimization can be performed according to the SOD basic parameter vector as a reference. Compared with the prior art, the experimental method of approaching the traversal of the SOD component parameter can reduce the number of experiments in the SOD component research process, thereby improving the efficiency of the SOD component parameter optimization. In addition, the SOD basic parameter vector corresponding to different scenes can be provided, and the efficiency of the SOD component parameter optimization is further improved.

[0104] Embodiment two

[0105] The second embodiment provides an optimization parameter system for processing a skin abnormal area by using SOD, as shown in the following figure, which is a structural schematic diagram of an optimization parameter system for processing a skin abnormal area by using SOD provided by the second embodiment of the present application. The optimization parameter system for processing a skin abnormal area by using SOD comprises: Figure 2

[0106] The feature extraction module 201 is configured to input the obtained M reference skin area images into the trained image encoder respectively, to obtain reference feature vectors corresponding to the M reference skin area images respectively, wherein M is a positive integer.

[0107] The model prediction module 202 is configured to predict the predicted abnormal categories and the predicted physiological indicators corresponding to the M reference feature vectors respectively according to the M reference feature vectors, the trained first full connection layer, and the trained second full connection layer.

[0108] The feature selection module 203 is configured to determine a plurality of reference feature vectors corresponding to the same predicted abnormal category as any preset abnormal category as temporary feature vectors for the preset abnormal category.

[0109] The index determination module 204 is configured to determine a reference physiological indicator corresponding to the preset abnormal category according to the predicted physiological indicators corresponding to the temporary feature vectors respectively.

[0110] The first vector determination module 205 is configured to determine a first index update vector according to the reference physiological indicators corresponding to the preset abnormal categories respectively.

[0111] The second vector determination module 206 is configured to determine a second index update vector and a third index update vector according to a plurality of predicted physiological indicators and reference physiological indicators corresponding to the preset abnormal categories satisfying a first preset condition. ​

[0112] The third vector determination module 207 is configured to determine a target index vector according to the first index update vector, the second index update vector, and the third index update vector.

[0113] The parameter optimization module 208 is configured to determine a plurality of SOD basic parameter vectors according to the target index vector and a trained SOD parameter optimization model, and the SOD basic parameter vectors are used to assist in adjusting component parameters of the SOD.

[0114] It should be noted that the specific definition of the optimization parameter system for processing the skin abnormal area by using the SOD can refer to the definition of the optimization parameter method for processing the skin abnormal area by using the SOD in the foregoing, and details are not described herein again. The information interaction and execution process between the above modules can be based on the same concept as the method embodiments, and the specific functions and the technical effects brought by the method embodiments can be referred to the method embodiments part, and details are not described herein again.

[0115] The above is only a preferred embodiment of the present application, and does not limit the present application in any form. Although the present application has been disclosed as above with the preferred embodiment, it is not intended to limit the present application. Any person skilled in the art can make some changes or modifications to the above disclosed technical content without departing from the technical solution of the present application, and the equivalent embodiments with equivalent changes are equivalent to the above embodiments. Any simple modification, equivalent change and modification of the above embodiments according to the technical essence of the present application are still within the scope of the technical solution of the present application.

Claims

1. A method for optimizing parameters in treating abnormal skin areas using SOD, characterized in that, The method for optimizing parameters in treating abnormal skin areas using SOD includes the following steps: S101, input the acquired M reference skin region images into the trained image encoder respectively to obtain the reference feature vectors corresponding to the M reference skin region images respectively, where M is a positive integer; S102, based on the M reference feature vectors, the trained first fully connected layer, and the trained second fully connected layer, predict the predicted abnormality category and predicted physiological index corresponding to the M reference feature vectors respectively; S103, for any preset anomaly category, determine several reference feature vectors corresponding to the predicted anomaly category that is the same as the preset anomaly category as temporary feature vectors; S104, based on the predicted physiological indicators corresponding to each temporary feature vector, determine the reference physiological indicator corresponding to the preset abnormality category, wherein determining the reference physiological indicator corresponding to the preset abnormality category based on the predicted physiological indicators corresponding to each temporary feature vector includes: A pre-defined clustering algorithm is used to cluster each temporary feature vector to obtain several cluster sets; Based on the preset quantity threshold and the number of temporary feature vectors contained in each cluster set, several intermediate sets are determined from each cluster set; Each intermediate set contains a temporary feature vector, which is then used as an intermediate feature vector. Based on the predicted physiological indicators corresponding to each intermediate feature vector, the reference physiological indicators corresponding to the preset abnormality category are determined. S105, determine the first indicator update vector according to the reference physiological indicators corresponding to each preset abnormality category, wherein the preset abnormality category is a normal skin category or an abnormal skin category; The step of determining the first indicator update vector based on the reference physiological indicators corresponding to each preset abnormality category includes: Subtract the reference physiological indicators corresponding to the abnormal skin category from the reference physiological indicators corresponding to the normal skin category to obtain the subtraction result; The subtraction result is processed using a sign function to obtain the first index update vector; S106, based on several predicted physiological indicators and reference physiological indicators corresponding to the preset abnormality category that meets the first preset condition, determine the second indicator update vector and the third indicator update vector, wherein the first preset condition is that it belongs to the normal skin category. The step of determining the second indicator update vector and the third indicator update vector based on a number of predicted physiological indicators and reference physiological indicators corresponding to preset abnormality categories that meet the first preset condition includes: Based on several predictive physiological indicators corresponding to normal skin types, determine the lower limit and upper limit of physiological indicators; The second indicator update vector is determined based on the lower limit of the physiological indicator and the reference physiological indicator; The third indicator update vector is determined based on the upper limit of the physiological indicator and the reference physiological indicator; S107, determine the target indicator vector based on the first indicator update vector, the second indicator update vector and the third indicator update vector, wherein the first indicator update vector, the second indicator update vector and the third indicator update vector each correspond to K indicator dimensions, and K is a positive integer; The step of determining the target indicator vector based on the first indicator update vector, the second indicator update vector, and the third indicator update vector includes: For any indicator dimension, if the element corresponding to that indicator dimension in the first indicator update vector is a first preset value, then the element corresponding to that indicator dimension in the second indicator update vector is taken as the target element corresponding to that indicator dimension. If the element corresponding to the indicator dimension in the first indicator update vector is the second preset value, then the element corresponding to the indicator dimension in the third indicator update vector is taken as the target element corresponding to the indicator dimension. The target indicator vector is formed by the target elements corresponding to each indicator dimension. S108, Based on the target index vector and the trained SOD parameter optimization model, determine several SOD basic parameter vectors. These SOD basic parameter vectors are used to assist in adjusting the component parameters of SOD. Specifically, determining these several SOD basic parameter vectors based on the target index vector and the trained SOD parameter optimization model includes: Randomly generate several initial parameter vectors for SOD; For any SOD initial parameter vector, input the SOD initial parameter vector into the trained SOD parameter optimization model to obtain the initial index vector. Based on the initial index vector, the target index vector, and the preset fourth loss function, the optimization loss is calculated. Based on the optimization loss, the initial SOD parameter vector is updated until the optimization loss converges, and the temporary SOD parameter vector corresponding to the initial SOD parameter vector is obtained. Iterate through each SOD initial parameter vector to obtain the SOD temporary parameter vector corresponding to each SOD initial parameter vector; The temporary SOD parameter vectors are deduplicated to obtain several basic SOD parameter vectors.

2. The method for optimizing parameters for treating abnormal skin areas using SOD according to claim 1, characterized in that, The step of predicting the predicted abnormality category and predicted physiological index corresponding to the M reference feature vectors based on the M reference feature vectors, the trained first fully connected layer, and the trained second fully connected layer includes: The M reference feature vectors are respectively input into the trained first fully connected layer for feature mapping to obtain the predicted anomaly category corresponding to the M reference feature vectors respectively; The M reference feature vectors are respectively input into the trained second fully connected layer for feature mapping to obtain the predicted physiological indicators corresponding to the M reference feature vectors.

3. The method for optimizing parameters for treating abnormal skin areas using SOD according to claim 1, characterized in that, The training process of the image encoder, the first fully connected layer, and the second fully connected layer includes the following steps: Acquire several sample skin region images, where each sample skin region image corresponds to a sample abnormality category and a sample physiological index; Two sample skin region images are randomly selected and input into the image encoder and the preset auxiliary encoder respectively to obtain the first sample feature output by the image encoder and the second sample feature output by the auxiliary encoder, wherein the auxiliary encoder is the same as the image encoder; The first training loss is calculated based on the first sample features, the second sample features, the sample anomaly categories corresponding to the two selected sample skin region images respectively, and the preset first loss function. The first sample features are input into the first fully connected layer and the second fully connected layer respectively to obtain the first predicted sample category output by the first fully connected layer and the first predicted sample index output by the second fully connected layer. The second sample features are input into the first fully connected layer and the second fully connected layer respectively to obtain the second predicted sample category output by the first fully connected layer and the second predicted sample index output by the second fully connected layer. The second training loss is calculated based on the first predicted sample category, the second predicted sample category, the sample anomaly categories corresponding to the two selected sample skin region images, and a preset second loss function. The third training loss is calculated based on the first predicted sample index, the second predicted sample index, the sample physiological index corresponding to the two selected sample skin region images, and the preset third loss function. Based on the first training loss, the second training loss, and the third training loss, a target training loss is determined. Based on the target training loss, the model parameters corresponding to the image encoder, the first fully connected layer, and the second fully connected layer are updated until the target training loss converges, thereby obtaining the trained image encoder, the trained first fully connected layer, and the trained second fully connected layer.

4. The method for optimizing parameters for treating abnormal skin areas using SOD according to claim 1, characterized in that, The step of determining the reference physiological index corresponding to the preset abnormality category based on the predicted physiological index corresponding to each intermediate feature vector includes: The mean of the predicted physiological indicators corresponding to each intermediate feature vector is calculated, and the result of the mean calculation is used as the reference physiological indicator corresponding to the preset abnormality category.

5. An optimized parameter system for treating abnormal skin areas using SOD, characterized in that, The optimized parameter system for treating abnormal skin areas using SOD includes: The feature extraction module is used to input the acquired M reference skin region images into the trained image encoder to obtain the reference feature vectors corresponding to the M reference skin region images respectively, where M is a positive integer; The model prediction module is used to predict the predicted abnormality category and predicted physiological index corresponding to the M reference feature vectors based on the M reference feature vectors, the trained first fully connected layer and the trained second fully connected layer. The feature selection module is used to determine several reference feature vectors corresponding to the predicted anomaly category that is the same as the preset anomaly category as temporary feature vectors for any preset anomaly category. The indicator determination module is used to determine the reference physiological indicator corresponding to the preset abnormality category based on the predicted physiological indicators corresponding to each temporary feature vector. The step of determining the reference physiological indicator corresponding to the preset abnormality category based on the predicted physiological indicators corresponding to each temporary feature vector includes: A pre-defined clustering algorithm is used to cluster each temporary feature vector to obtain several cluster sets; Based on the preset quantity threshold and the number of temporary feature vectors contained in each cluster set, several intermediate sets are determined from each cluster set; Each intermediate set contains a temporary feature vector, which is then used as an intermediate feature vector. Based on the predicted physiological indicators corresponding to each intermediate feature vector, the reference physiological indicators corresponding to the preset abnormality category are determined. The first vector determination module is used to determine the first indicator update vector based on the reference physiological indicators corresponding to each preset abnormality category, wherein the preset abnormality category is a normal skin category or an abnormal skin category. The step of determining the first indicator update vector based on the reference physiological indicators corresponding to each preset abnormality category includes: Subtract the reference physiological indicators corresponding to the abnormal skin category from the reference physiological indicators corresponding to the normal skin category to obtain the subtraction result; The subtraction result is processed using a sign function to obtain the first index update vector; The second vector determination module is used to determine the second indicator update vector and the third indicator update vector based on a number of predicted physiological indicators and reference physiological indicators corresponding to a preset abnormality category that meets the first preset condition, wherein the first preset condition is that it belongs to the normal skin category. The step of determining the second indicator update vector and the third indicator update vector based on a number of predicted physiological indicators and reference physiological indicators corresponding to preset abnormality categories that meet the first preset condition includes: Based on several predictive physiological indicators corresponding to normal skin types, determine the lower limit and upper limit of physiological indicators; The second indicator update vector is determined based on the lower limit of the physiological indicator and the reference physiological indicator; The third indicator update vector is determined based on the upper limit of the physiological indicator and the reference physiological indicator; The third vector determination module is used to determine the target indicator vector based on the first indicator update vector, the second indicator update vector and the third indicator update vector, wherein the first indicator update vector, the second indicator update vector and the third indicator update vector each correspond to K indicator dimensions, and K is a positive integer; The step of determining the target indicator vector based on the first indicator update vector, the second indicator update vector, and the third indicator update vector includes: For any indicator dimension, if the element corresponding to that indicator dimension in the first indicator update vector is a first preset value, then the element corresponding to that indicator dimension in the second indicator update vector is taken as the target element corresponding to that indicator dimension. If the element corresponding to the indicator dimension in the first indicator update vector is the second preset value, then the element corresponding to the indicator dimension in the third indicator update vector is taken as the target element corresponding to the indicator dimension. The target indicator vector is formed by the target elements corresponding to each indicator dimension. The parameter optimization module is used to determine several SOD basic parameter vectors based on the target index vector and the trained SOD parameter optimization model. These SOD basic parameter vectors are used to assist in adjusting the component parameters of SOD. The process of determining several SOD basic parameter vectors based on the target index vector and the trained SOD parameter optimization model includes: Randomly generate several initial parameter vectors for SOD; For any SOD initial parameter vector, input the SOD initial parameter vector into the trained SOD parameter optimization model to obtain the initial index vector. Based on the initial index vector, the target index vector, and the preset fourth loss function, the optimization loss is calculated. Based on the optimization loss, the initial SOD parameter vector is updated until the optimization loss converges, and the temporary SOD parameter vector corresponding to the initial SOD parameter vector is obtained. Iterate through each SOD initial parameter vector to obtain the SOD temporary parameter vector corresponding to each SOD initial parameter vector; The temporary SOD parameter vectors are deduplicated to obtain several basic SOD parameter vectors.

Images