WO2024041524A1

WO2024041524A1 - Scalp hair detection method, system and device

Info

Publication number: WO2024041524A1
Application number: PCT/CN2023/114216
Authority: WO
Inventors: 蔡权; 杨建辉; 卢伟; 严靖宇
Original assignee: 漳州松霖智能家居有限公司
Priority date: 2022-08-24
Filing date: 2023-08-22
Publication date: 2024-02-29
Also published as: CN117710686A

Abstract

Disclosed in the present invention are a scalp hair detection method, system and device. The method comprises: acquiring different scalp hair images; according to scalp hair properties, labeling and classifying the scalp hair images to form scalp hair property-based classification data sets; inputting the labeled images of the classification data sets into an improved MobileNet deep network model for training to obtain a trained scalp hair property-based deep network model; and inputting a scalp hair image under detection into the trained deep network model to obtain a detection result corresponding to the scalp hair properties, wherein the detection result comprises a category and a confidence coefficient corresponding to the category. The present invention realizes scalp hair property detection on scalp hair images by means of an improved MobileNet deep network model, and categories and confidence coefficients corresponding to scalp hair properties are finally output; the operational speed is increased, and terminal-side deployment is more convenient.

Description

Scalp hair detection method, system and equipment

Technical field

The present invention relates to the technical field of scalp and hair detection, in particular to a method, system and equipment for scalp and hair detection.

Background technique

The scalp is one of the sensitive skins of the human body. Due to living habits and work pressure, more and more people are currently suffering from scalp and hair problems. Many people suffer from damaged hair, greasy hair, thick scalp cuticles, and red scalp. Problems such as excess hair follicles and subcutaneous oil. There are many chain hairdressing institutions and hair management centers on the market today. Many of the hair tests are based on single-point photography of the scalp, and manual interpretation is used to obtain the condition of the subject's scalp and hair. , this method is often affected by the subjective consciousness of the interpreter and cannot obtain objective and accurate results, resulting in the subject being unable to correctly understand the condition of his scalp and hair. How to objectively and accurately detect the condition of scalp and hair is an urgent problem that needs to be solved.

The patent application number 202010228550.4 discloses a scalp detection method based on deep learning, which includes the following steps: Step S1: Collect scalp image data; Step S2: Label and classify scalp images according to scalp attributes to form classification data for each scalp attribute. Set; Step S3: Use the ImageNet image database to pre-train the SqueezeNet model to obtain the pre-trained SqueezeNet model; Step S4: Modify the pre-trained SqueezeNet model to adapt it to the regression task and obtain the improved SqueezeNet model; Step S5: Formulate the scalp detection accuracy Determine the rules, use the classification data set in step S2 to retrain the improved SqueezeNet model, and obtain scalp detection models for various scalp attributes; step S6, classify the scalp image to be tested according to the scalp attributes, and input the corresponding scalp detection model Get prediction results. Compared with the SqueezeNet model, Mobilenet reduces the number of parameters and improves the computing speed, making it more convenient for device-side deployment.

Contents of the invention

The main purpose of the present invention is to propose a method, system and equipment for scalp and hair detection, which overcomes the shortcomings of the existing technology and detects scalp and hair through an improved MobileNet deep network model. The scalp and hair attributes in the image are detected, and the category and confidence level corresponding to the scalp and hair attributes are finally output, which improves the computing speed and makes terminal-side deployment more convenient.

The present invention adopts the following technical solutions:

In one aspect, a scalp hair detection method includes:

Acquire different scalp and hair images;

According to scalp and hair attributes, scalp and hair images are annotated and classified to form a classification data set based on scalp and hair attributes;

Input the labeled classification data set images into the improved MobileNet deep network model for training, and obtain the trained deep network model based on scalp and hair attributes;

Input the scalp and hair image to be tested into the trained deep network model to obtain detection results corresponding to scalp and hair attributes; the detection results include categories and confidence levels corresponding to the categories.

Preferably, the improved MobileNet deep network model includes in order: a first convolution layer, several block layers, a pooling layer, a second convolution layer and a third convolution layer; each block layer includes in order: a fourth convolution layer. The accumulation layer, the depth convolution layer and the fifth convolution layer; after each block layer, there is a skip connection layer connected to the last block layer.

Preferably, the improved MobileNet deep network model also includes several first activation function layers, and each convolutional layer is connected to a first activation function layer to perform feature information on the scalp and hair images extracted by the convolutional layer. Nonlinear operations.

Preferably, the first activation function layer includes a ReLU layer.

Preferably, the end of the improved MobileNet deep network model includes a fully connected layer, the fully connected layer outputs three 1*1 channel images, and the second activation function layer connected to the fully connected layer activates and outputs each Classification confidence.

Preferably, the second activation function layer includes a Softmax layer.

Preferably, the loss calculation function of the improved MobileNet deep network model is as follows:

Among them, H(y,p) represents the model loss; y represents the true value of the image label in the test set; p represents the predicted value of the output label after being sent to the model; N represents the number of images in the test set; M represents the number of categories; c represents Current output category; y _ic represents the c-th category true value of the i sample; p _ic represents the c-th category of the i sample The predicted value output after being fed into the model.

Preferably, the accuracy calculation function is also included as follows:

Among them, Precision represents the accuracy of this round of weights in this test set; TP represents the number of correct judgments; FP represents the number of incorrect judgments.

Preferably, after obtaining the detection result corresponding to the scalp hair attribute, the method further includes: inputting the category and confidence into the constructed score mapping function to obtain a score corresponding to the category and confidence.

Preferably, the score mapping function is as follows:

Among them, x is the confidence of the detection result output; cls is the category of the detection result output; sigmoid(x) represents the mapping intermediate function; f(x, cls) represents the mapping function; f represents the score corresponding to the confidence.

Preferably, the scalp and hair attributes include at least one of hair thickness, hair damage degree, hair oil, scalp cuticle, scalp red blood filaments, and hair follicle subcutaneous oil; each scalp and hair attribute corresponds to an improved MobileNet deep network model.

In another aspect, a scalp hair detection system includes:

Image acquisition module, used to acquire different scalp and hair images;

The classification data set labeling module is used to label and classify scalp and hair images according to scalp and hair attributes, forming a classification data set based on scalp and hair attributes;

The deep network model training module is used to input the labeled classification data set images into the improved MobileNet deep network model for training, and obtain the trained deep network model based on scalp and hair attributes;

The detection result output module is used to input the scalp and hair image to be tested into the trained deep network model to obtain detection results corresponding to scalp and hair attributes; the detection results include categories and confidence levels corresponding to the categories.

In another aspect, a scalp hair detection device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the scalp hair detection device is implemented. Skin and hair detection methods.

Compared with the prior art, the beneficial effects of the present invention are as follows:

(1) The present invention uses an improved MobileNet deep network model to detect scalp and hair attributes in scalp and hair images, and finally outputs the category and confidence level corresponding to the scalp and hair attributes, which improves the computing speed and makes terminal-side deployment more convenient;

(2) The improved MobileNet deep network model of the present invention reduces the number of block layers of the original Mobilenet. Since the scalp and hair features are relatively obvious, there is no need for too many block layers to extract information, so several block layers in the middle are removed. Thereby reducing the amount of calculation and speeding up the operation;

(3) The improved MobileNet deep network model of the present invention adds a skip connection layer to enhance feature fusion. Specifically, a skip connection layer is added after each block layer to connect to the last block layer, and then through adaptive reduction Sampling enriches local features and global features (local features include small features, such as red blood streaks, and global features include large-area features, such as oil), which is conducive to subsequent extraction or classification of features at different scales;

(4) The improved MobileNet deep network model of the present invention adds a 1*1 convolution layer at the end. The addition of a 1*1 convolution layer makes the model pay more attention to classification information and further accelerates the convergence speed;

(5) The present invention inputs the categories and confidence levels corresponding to scalp and hair attributes into the constructed score mapping function to map the scores, so that users can intuitively feel the state of their own scalp based on the scores based on the scores.

The above description is only an overview of the technical solutions of the present invention. In order to understand the technical means of the present invention more clearly, so that they can be implemented according to the contents of the description, and in order to make the above and other objects, features and advantages of the present invention more obvious and easy to understand. Understand, the specific embodiments of the present invention are listed below.

From the following detailed description of specific embodiments of the present invention in conjunction with the accompanying drawings, those skilled in the art will further understand the above and other objects, advantages and features of the present invention.

Description of drawings

Figure 1 is a flow chart of a scalp hair detection method according to an embodiment of the present invention;

Figure 2 is an example diagram of the MobileNet deep network model in the prior art;

Figure 3 is an example diagram of an improved MobileNet deep network model according to an embodiment of the present invention;

Figure 4 is an example diagram of the block layer according to the embodiment of the present invention;

Figure 5 is a hierarchical table of the MobileNet deep network model in the prior art;

Figure 6 is a hierarchical table of the improved MobileNet deep network model according to the embodiment of the present invention;

Figure 7 is a comparison diagram of model loss between MobileNet in the prior art and the improved MobileNet according to the embodiment of the present invention;

Figure 8 is a comparison diagram of model accuracy between MobileNet in the prior art and the improved MobileNet according to the embodiment of the present invention;

Figure 9 is a detailed flow chart of a method for detecting red blood filament attributes on the scalp according to an embodiment of the present invention;

Figure 10 is a structural block diagram of a scalp hair detection system according to an embodiment of the present invention;

Figure 11 is a frame diagram of a scalp hair detection device according to an embodiment of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention; obviously, the described embodiments are only some of the embodiments of the present invention, not all of them. Based on The embodiments of the present invention and all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

In the description of the present invention, it should be noted that the terms "comprising", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article or device including a series of elements not only includes Those elements, but also other elements not expressly listed or inherent in such process, method, article or equipment. Without further limitation, an element defined by the statement "comprises a..." does not exclude the presence of additional identical elements in a process, method, article, or apparatus that includes the stated element.

In the description of the present invention, it should be noted that the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

In the description of the present invention, it should be noted that, unless otherwise clearly stated and limited, the terms "installed", "provided with", "set/connected", "connected", etc., should be understood in a broad sense, such as " "Connection" can be a fixed connection, a detachable connection, or an integral connection. It can be a mechanical connection or an electrical connection. It can be a direct connection or an indirect connection through an intermediate medium. It can be internal to two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood in specific situations.

In the description of the present invention, it should be noted that, unless otherwise expressly stated and limited, step identification S101, S102, S103, etc. are only used for convenience of expression and do not indicate the execution order. The corresponding execution order can be adjusted.

As shown in Figure 1, a scalp hair detection method of the present invention includes:

S101, obtain different scalp and hair images;

S102, label and classify scalp and hair images according to scalp and hair attributes to form a classification data set based on scalp and hair attributes;

S103, input the labeled classification data set images into the improved MobileNet deep network model for training, and obtain the trained deep network model based on scalp and hair attributes;

S104. Input the scalp and hair image to be tested into the trained deep network model to obtain detection results corresponding to scalp and hair attributes; the detection results include categories and confidence levels corresponding to the categories.

Obtaining different scalp and hair images, specifically including acquiring scalp and hair images of different light sources, different angles, different ages, different genders, etc., by selecting different scalp and hair images for subsequent analysis of the improved MobileNet deep network model (hereinafter referred to as MobileHairNet deep network) Model) is trained so that the improved MobileNet deep network model can detect the categories and confidence levels of scalp and hair images under different circumstances, expanding the scope of adaptation.

Label and classify scalp and hair images based on scalp and hair attributes. Specifically, you can ask a professional doctor to mark it. Different scalp and hair attributes can be divided into different categories, or they can all be divided into three categories, such as mild, moderate and moderate. Professional annotation will help the subsequent MobileHairNet deep network model to continuously update its own network structure feature parameters during the training process, and adjust the network structure feature parameters to the optimal state.

In the present invention, the scalp hair attributes include at least one of hair thickness, hair damage degree, hair oil, scalp cuticle, scalp red blood filaments and hair follicle subcutaneous oil; each scalp hair attribute corresponds to a MobileHairNet deep network model. The structure of the MobileHairNet deep network model corresponding to each scalp hair attribute is the same, but the network structure characteristic parameters may be different.

It should be noted that since the characteristics of each scalp and hair attribute are different, a variable magnification lens can be used when collecting images of different scalp and hair attributes. For example, 50x, 100x and 200x optical lenses are used to magnify the hair and scalp respectively to observe the characteristics of the scalp and hair. The 50x lens can make it easier to observe redness and other conditions on the scalp, and the 100x lens can make it easier to observe the cuticle of the scalp, scalp oil, and hair follicle hair loss. 200x lens for observing hair thickness, Hair damage, etc.

In addition, this embodiment uses three-spectrum recognition technology to identify the characteristics of scalp and hair images. When collecting scalp and hair images, scalp and hair images can be collected under different light sources. Corresponding to the characteristics of red blood streaks in the scalp and subcutaneous oil in hair follicles, it is difficult to distinguish these characteristics with the naked eye under traditional white light. With the assistance of polarized light, the specular reflection of natural light can be eliminated, making it easier to observe the characteristics of red blood filaments under the surface of the skin. UV light sources with wavelengths between 280nm and 400nm are easily reflected by the subcutaneous oil of hair follicles, resulting in the production of bright red light. Specifically, which magnification lens and which kind of light are more suitable for extracting which scalp and hair attributes can be obtained after experimental testing, and then the images corresponding to the scalp and hair attributes are obtained under the corresponding magnification lens and light source as test images for training.

It should be noted that the collection of scalp and hair images can be implemented on a device that performs a scalp and hair detection method, or can be collected on other devices and then sent to a device that performs a scalp and hair detection method. The specific settings are set according to needs. This implementation Examples are not limited.

In this embodiment, the MobileHairNet deep network model includes: a first convolution layer, several block layers, a pooling layer, a second convolution layer, and a third convolution layer; each block layer includes: a fourth convolution layer. , deep convolution layer and fifth convolution layer; after each block layer, there is a skip connection layer connected to the last block layer.

Further, the improved MobileNet deep network model also includes several first activation function layers, and each convolutional layer is connected to a first activation function layer to perform feature information on the scalp and hair images extracted by the convolutional layer. Nonlinear operations.

The first convolution layer, the second convolution layer, the third convolution layer, the fourth convolution layer and the fifth convolution layer here are conv layers that do not include an activation function, and the depth convolution layer does not include an activation function. dwconv layer. There is a first activation function layer connected behind the first convolution layer, the second convolution layer, the third convolution layer, the fourth convolution layer and the fifth convolution layer respectively. There is also a first activation function layer connected behind the dwconv layer. function layer. In other embodiments, the conv layer and the first activation function layer may also be collectively referred to as the convolution layer, and the dwconv layer and the first activation function layer may be collectively referred to as the depth convolution layer, which is not specifically limited in this embodiment.

Specifically, the first activation function layer includes a ReLU layer.

The end of the MobileHairNet deep network model also includes a fully connected layer. The fully connected layer outputs three 1*1 channel images, and the second activation function layer connected to the fully connected layer activates and outputs the confidence of each category. .

The second activation function layer includes a Softmax layer.

Using the ReLU function to introduce nonlinear factors enables the MobileHairNet deep network model to fit nonlinear functions. It suppresses uninterested feature areas and focuses on important features. The formula of the ReLU function is as follows: x is the value of each pixel in the picture.

The formula of the Softmax function is as follows:

in, is the channel value output by the i-th category, n is the number of categories, such as mild, moderate and severe, including three categories.

Finally, according to the confidence of each category in the output, the category with the maximum confidence is taken out and used as the category of the current image output, for example, through softmax output [0.5, 0.2, 0.3]. Then the probability that the information mapped in the three corresponding categories is mild is 50%, the probability of moderate is 20%, and the probability of severe is 20%. Finally, the maximum probability is taken as the current category. In order to verify whether the MobileHairNet deep network model has learned the expected attention features, the test set is used for verification. Use Cross Entropy Loss to calculate the loss of the model, as follows:

Among them, H(y,p) represents the model loss; y represents the true value of the image label in the test set; p represents the predicted value of the output label after being sent to the model; N represents the number of images in the test set; M represents the number of categories; c represents Current output category; y _ic represents the true value of the c-th classification of the i sample; p _ic represents the prediction of the output after the c-th classification of the i-th sample is fed into the model

When the model undergoes multiple rounds of training on the training set data, the loss of the training set and the loss of the test set continue to decrease, indicating that the model can converge and use the accuracy function Precision to calculate the accuracy of this round of weights in the test set, as follows:

Conduct multiple rounds of training and save the optimal model. At the beginning of training, you can set a preset number of rounds for the network Conduct training. A comprehensive evaluation is conducted based on the accuracy Precision and model loss function Cross Entropy Loss obtained in each round. Save a model with a higher model test set accuracy and a lower training set model loss to ensure that the model has a higher prediction accuracy and feature learning ability.

Specifically, see Figure 2 which shows an example diagram of the MobileNet deep network model in the prior art. Figure 3 shows an example diagram of the MobileHairNet deep network model according to the embodiment of the present invention. Figure 4 shows the block layer of the embodiment of the present invention. sample graph. Referring to FIG. 5 , a hierarchical table of the MobileNet deep network model in the prior art is shown, and FIG. 6 is a hierarchical table of the MobileHairNet deep network model according to an embodiment of the present invention. As can be seen from Figures 2 to 6, the MobileHairNet deep network model of this embodiment has the following characteristics and beneficial effects compared with the existing MobileNet deep network model:

(1) Reduce the number of block layers in Mobilenet. Since the features are relatively obvious, there is no need to perform too many block layers to extract information. Therefore, the MobileHairNet deep network model in this embodiment removes the middle 14×14×96 and 7×7×160 input block layers, thereby reducing the amount of calculation. , speed up the operation.

(2) Add a skip connection layer to enhance feature fusion. A skip connection layer is added after different block layers, connected to the last block layer, and then through adaptive downsampling (referring to scaling, reducing by a certain ratio, otherwise it cannot be superimposed to the last layer) to make the local features and global features Become rich, which is conducive to subsequent extraction or classification of features of different scales (local features refer to small features, such as red blood streaks on the scalp, and global features refer to large-area features such as oil, similar to different sizes of features) Learn from pictures, and you can quickly distinguish similar pictures of different sizes next time).

Through experiments, it can be seen that in the same batch of data sets, the accuracy of the model connected by the skip connection layer will be 5.8% higher than that of the model without the skip connection layer. Under the same number of trainings, the average loss will be 17% smaller than the model without skip layer connection, the convergence speed will be faster, and the final model will be reached 25 rounds earlier (the accuracy of subsequent iteration epochs is due to overfitting of the model and the accuracy has not improved. The optimal number of iterations is determined by the minimum value of the loss value or the maximum value of ACC (accuracy)). The model loss comparison diagram between MobileNet in the prior art and the improved MobileNet according to the embodiment of the present invention is shown in Figure 7. The model accuracy comparison diagram between MobileNet in the prior art and the improved MobileNet according to the embodiment of the present invention is shown in Figure 8.

(3) Add a terminal 1*1 convolution layer to speed up convergence. Since the skip connection layer is connected to the last layer of blocks, after the maxpooling pooling layer, plus the previous optimization solution (reducing the number of block layers in Mobilenet), compared with the 1280 layers of the original Mobilenet, the pooling layer of the present invention is from The number of layers is reduced to 171 (3+32+16+24+32+64), further compressing the amount of calculation, and finally a 1*1 convolution layer is added to make The model pays more attention to classification information and further accelerates the convergence speed.

Under these three strategies, the MobileHairNet deep network model of the present invention is only 2.4M in size and has an inference speed of 5ms. Compared with the SqueezeNet model, which is 4.8M, it is smaller and the inference speed is 35ms faster.

In this embodiment, after obtaining the detection results corresponding to scalp hair attributes, a scalp hair detection method further includes: inputting the category and confidence level into the constructed score mapping function to obtain a score corresponding to the category and confidence level.

The score mapping function is as follows:

During specific implementation, after obtaining the category cls with the highest confidence and the confidence x, the category cls with the highest confidence and the confidence Perform score mapping. Allow users to intuitively feel the condition of their hair and scalp through scores, such as mapping mild to 40-60 points, moderate to 60-80 points, and severe to 80-100 points based on confidence. .

As shown in Figure 10, a scalp hair detection system includes:

Image acquisition module 1001, used to acquire different scalp and hair images;

The classification data set annotation module 1002 is used to annotate and classify scalp and hair images according to scalp and hair attributes to form a classification data set based on scalp and hair attributes;

The deep network model training module 1003 is used to input the labeled classification data set images into the improved MobileNet deep network model for training, and obtain the trained deep network model based on scalp and hair attributes;

The detection result output module 1004 is used to input the scalp and hair image to be tested into the trained deep network model to obtain detection results corresponding to scalp and hair attributes; the detection results include categories and confidence levels corresponding to the categories.

In this embodiment, the image acquisition module 1001, classification data set annotation module 1002, deep network For further functional description of modules such as the network model training module 1003 and the detection result output module 1004, see the aforementioned scalp hair detection method for details.

Referring to Figure 11, a scalp hair detection device 110 includes a memory 1101, a processor 1102, and a computer program 1103 stored in the memory 1101 and executable on the processor 1102. When the processor 1102 executes the program Implement the scalp hair detection method.

In an embodiment, the computer program 1103 may be divided into one or more modules/units, and the one or more modules/units are stored in the memory 1101 and executed by the processor 1102 to Complete this application. The one or more modules/units may be a series of computer program 1103 instruction segments capable of completing specific functions. The instruction segments are used to describe the execution process of the computer program 1103 in the scalp hair detection device 110 .

The scalp and hair detection device 110 may be a proprietary scalp and hair detection instrument, or may be a computing device such as a mobile phone, desktop computer, notebook, PDA, cloud server, etc. The scalp hair detection device 110 may include, but is not limited to, a processor 1102 and a memory 1101. Those skilled in the art can understand that FIG. 11 is only an example of the scalp and hair detection device 110 and does not constitute a limitation on the scalp and hair detection device 110. It may include more or fewer components than shown in the figure, or combine certain components. Or different components, for example, the scalp hair detection device 110 may also include input and output devices, network access devices, buses, etc.

The so-called processor 1102 can be a central processing unit (Central Processing Unit, CPU), or other general-purpose processor 1102, a digital signal processor 1102 (Digital Signal Processor, DSP), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC ), off-the-shelf programmable gate array (Field-Programmable GateArray, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor 1102 may be a microprocessor 1102 or the processor 1102 may be any conventional processor 1102 or the like.

The memory 1101 may be an internal storage unit of the scalp hair detection device 110, such as a hard disk or memory of the scalp hair detection device 110. The memory 1101 may also be an external storage device of the scalp and hair detection device 110, such as a plug-in hard drive, a smart media card (SMC), or a secure digital device equipped on the scalp and hair detection device 110. Digital, SD) card, Flash Card, etc. Further, the memory 1101 may also include both an internal storage unit of the scalp hair detection device 110 and an external storage device. The memory 1101 uses For storing the computer program 1103 and other programs and data required by the terminal device. The memory 1101 can also be used to temporarily store data that has been output or is to be output.

The above are only preferred specific embodiments of the present invention; however, the protection scope of the present invention is not limited thereto. Any person familiar with the technical field who is familiar with the technical field shall make equivalent substitutions or changes based on the technical solutions and improvement concepts of the present invention within the technical scope disclosed in the present invention, and shall be covered by the protection scope of the present invention.

Industrial applicability

The present invention is a scalp hair detection method that uses an improved MobileNet deep network model to detect scalp and hair attributes in scalp and hair images, and finally outputs the category and confidence level corresponding to the scalp and hair attributes, which improves the calculation speed and makes terminal processing more convenient. Side deployment, with good industrial practicality.

Claims

A scalp hair detection method, characterized by including:

Acquire different scalp and hair images;

According to scalp and hair attributes, scalp and hair images are annotated and classified to form a classification data set based on scalp and hair attributes;

Input the labeled classification data set images into the improved MobileNet deep network model for training, and obtain the trained deep network model based on scalp and hair attributes;

Input the scalp and hair image to be tested into the trained deep network model to obtain detection results corresponding to scalp and hair attributes; the detection results include categories and confidence levels corresponding to the categories.
The scalp hair detection method according to claim 1, characterized in that the improved MobileNet deep network model includes in sequence: a first convolution layer, several block layers, a pooling layer, a second convolution layer and a third convolution layer. Accumulation layer; each block layer includes: the fourth convolution layer, the depth convolution layer and the fifth convolution layer; each block layer is followed by a skip connection layer connected to the last block layer.
The scalp hair detection method according to claim 2, characterized in that the improved MobileNet deep network model also includes a plurality of first activation function layers, and each convolution layer is connected to a first activation function layer respectively to detect The feature information of the scalp and hair image extracted by the convolutional layer is subjected to nonlinear operations.
The scalp hair detection method according to claim 3, wherein the first activation function layer includes a ReLU layer.
The scalp hair detection method according to claim 2, characterized in that the end of the improved MobileNet deep network model includes a fully connected layer, and the fully connected layer outputs three 1*1 channel images, which are connected to the fully connected layer. The second activation function layer connected by the layers activates and outputs the confidence of each classification.
The scalp hair detection method according to claim 5, wherein the second activation function layer includes a Softmax layer.
The scalp hair detection method according to claim 1, characterized in that the loss calculation function of the improved MobileNet deep network model is as follows:

Among them, H(y,p) represents the model loss; y represents the true value of the image label of the test set; p represents the input The predicted value of the output label after the model; N represents the number of pictures in the test set; M represents the number of categories; c represents the current output category; y ic represents the c-th classification true value of the i sample; p ic represents the c-th classification true value of the i-th sample The predicted values output after c categories are fed into the model.
The scalp hair detection method according to claim 1, further comprising an accuracy calculation function as follows:

Among them, Precision represents the accuracy of this round of weights in this test set; TP represents the number of correct judgments; FP represents the number of incorrect judgments.
The scalp hair detection method according to claim 1, characterized in that after obtaining the detection results corresponding to scalp hair attributes, it further includes: inputting the category and confidence into the constructed score mapping function to obtain a score corresponding to the category and confidence. .
The scalp hair detection method according to claim 9, characterized in that the score mapping function is as follows:

Among them, x is the confidence of the detection result output; cls is the category of the detection result output; sigmoid(x) represents the mapping intermediate function; f(x, cls) represents the mapping function; f represents the score corresponding to the confidence.
The scalp hair detection method according to claim 1, wherein the scalp hair attributes include at least one of hair thickness, hair damage degree, hair oil, scalp cuticle, scalp red blood cells and hair follicle subcutaneous oil; Each scalp and hair attribute corresponds to an improved MobileNet deep network model.
The scalp and hair detection method according to claim 11, characterized in that when obtaining the scalp and hair image, the scalp and hair images can be collected under different light sources according to the attributes of the scalp and hair, and the scalp and hair images are collected through three-spectrum recognition technology. characteristics to identify.
A scalp hair detection system, which is characterized by including:

Image acquisition module, used to acquire different scalp and hair images;

Classification data set annotation module, used to annotate scalp and hair images according to scalp and hair attributes Classification to form a classification data set based on scalp and hair attributes;

The deep network model training module is used to input the labeled classification data set images into the improved MobileNet deep network model for training, and obtain the trained deep network model based on scalp and hair attributes;

The detection result output module is used to input the scalp and hair image to be tested into the trained deep network model to obtain detection results corresponding to scalp and hair attributes; the detection results include categories and confidence levels corresponding to the categories.
The scalp and hair detection system according to claim 13, wherein the scalp and hair attributes include at least one of hair thickness, hair damage degree, hair oil, scalp cuticle, scalp red blood cells and hair follicle subcutaneous oil; Each scalp and hair attribute corresponds to an improved MobileNet deep network model.
The scalp and hair detection system according to claim 14, characterized in that when the image acquisition module acquires a scalp and hair image, the scalp and hair images can be collected under different light sources according to the attributes of the scalp and hair, and the three-spectrum recognition technology is used to collect the scalp and hair images. Features of scalp and hair images are identified.
A scalp and hair detection device, characterized in that it includes a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements any of claims 1 to 12. The scalp hair detection method described in one item.