Hair feature analysis method and system
Through artificial intelligence model and R-CNN technology, combined with global features and framework-shielding enhancement module, the accuracy and consistency of traditional hair analysis methods are solved, and more accurate hair follicle detection and width calculation are achieved, which improves hair loss diagnosis and the therapeutic effect.
Patent Information
- Application Number
- CN202310614302.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-29
- Publication Date
- 2024-11-29
AI Technical Summary
Traditional hair analysis methods rely on manual examinations, which are time-consuming, labor-intensive and prone to human errors, lack accuracy and repetition, and are difficult to provide consistent hair follicle detection and width calculation results, affecting the diagnosis and treatment effects of hair loss.
Using artificial intelligence models, pre-processing by capturing high-magnification and low-magnification images of scalp areas, and using R-CNN model or its variants for detection and calculation, combined with the global feature enhancement module and framework-shielding enhancement module, Improve the accuracy and consistency of detection and calculations.
Achieve more accurate and consistent hair follicle detection and width calculations, improving the effectiveness of hair loss diagnosis and treatment, and providing personalized hair care advice and health assessment.
Smart Images

Figure CN119048726A_ABST
Abstract
Description
Technical Field
[0001] The present invention is directed to a hair analysis system and method, and more particularly to a hair analysis system and method that uses an artificial intelligence model to analyze hair characteristics, such as hair diameter, fine hair density ratio, and total hair density ratio. Background Art
[0002] Hair loss is a common problem that affects millions of people worldwide. Analyzing hair characteristics, such as hair diameter, fine hair density ratio, and total hair density ratio, is essential for diagnosing hair loss, assessing its severity, and determining appropriate treatment options. Traditional hair analysis methods rely on manual inspection, which can be time-consuming, laborious, and prone to human error. Furthermore, these methods may lack accuracy and repeatability as results may vary depending on the experience and skills of the examiner. In addition, recent advances in computer vision and artificial intelligence have enabled the development of automated hair analysis systems. These technologies typically utilize machine learning algorithms to process and analyze images of the scalp area, detect hair follicles, and calculate hair width.
[0003] Therefore, how to use artificial intelligence models to meet the specific requirements of hair analysis tasks to provide more accurate and consistent results in detecting hair follicles and calculating hair width, and ultimately help to more effectively diagnose and treat hair loss problems, is something that is worth considering for those with general knowledge in this field. Summary of the invention
[0004] The object of the present invention is to provide a method and system for analyzing hair features using an artificial intelligence model to meet the specific requirements of hair analysis tasks to provide more accurate and consistent results in detecting hair follicles and calculating hair width.
[0005] Based on the above and other purposes, the present invention provides a hair feature analysis method, comprising the following steps: firstly capturing a high magnification image and a low magnification image of a scalp area, and then preprocessing the two images. Next, the preprocessed high magnification image and low magnification image are input into an artificial intelligence model, and the hair follicles in the scalp area are detected and the hair width is calculated. Afterwards, at least one hair feature can be calculated through the high magnification image and the low magnification image analyzed by the artificial intelligence model. Among them, the artificial intelligence model can adopt an R-CNN model or a variant of the R-CNN model.
[0006] Based on the above and other purposes, the present invention also provides a hair feature analysis system, which includes: an image capture device, a preprocessing module, an artificial intelligence model, and an analysis module. The image capture device is used to capture a high magnification image and a low magnification image of the scalp area, and the preprocessing module is used to preprocess these images. The artificial intelligence model can simultaneously detect hair follicles and calculate hair width, and the model can be an R-CNN model or a variant of the R-CNN model. In addition, the analysis module is used to receive the image analyzed by the artificial intelligence model and calculate at least one hair feature.
[0007] As in the above-mentioned hair feature analysis method or system, the artificial intelligence model is an HTC enhanced model, which is modified from the mixed task cascade model and has the following characteristics: the model replaces the semantic segmentation module in the mixed task cascade model with a global feature enhancement module, and connects the refined semantic features of the global feature enhancement module with multiple frame branches. In addition, the model also introduces a frame-shielding enhancement module, which is connected to the above-mentioned frame branch and at least one shielding branch.
[0008] As in the above-mentioned hair feature analysis method or system, the HTC enhanced model has multiple features. Among them, the global feature enhancement module includes multiple convolutional layers with different kernel sizes to extract multi-scale features. In one embodiment, the HTC enhanced model is provided with the frame branch and the shield branch at each stage, and the number of the frame-shield enhancement modules is multiple, and each frame-shield enhancement module is connected between the frame branch and the shield branch at each stage. In addition, in another embodiment, the HTC enhanced model is provided with a frame branch at each stage but only one shield branch, and the frame-shield enhancement module is connected between the last frame branch and the shield branch.
[0009] In summary, by introducing the R-CNN model or its variants, the hair analysis method and system of the present invention can more accurately and consistently detect hair follicles and calculate hair width, which helps to diagnose and treat hair loss problems more effectively.
[0010] In order to make the above-mentioned purposes, features and advantages of the present invention more clearly understood, the following will be described in detail with reference to the embodiments and the accompanying drawings. It should be noted that the components in the accompanying drawings are only for illustration and are not drawn according to the actual proportions of the components. BRIEF DESCRIPTION OF THE DRAWINGS
[0011] Various embodiments will be described below with reference to the accompanying drawings, which are intended to be illustrative and not intended to be limiting in any way, wherein like reference numerals represent like components, and wherein:
[0012] Figure 1A first embodiment of a hair characteristic analysis system of the present invention is shown.
[0013] Figure 2 A second embodiment of the hair feature analysis system of the present invention is shown.
[0014] Figure 3 Depicted is a flow chart of the steps of capturing high-magnification and low-magnification images of the scalp area and pre-processing.
[0015] Figure 4 A schematic diagram of the mixed-task cascade model is shown.
[0016] Figure 5 A schematic diagram of the HTC enhancement model of the present invention is depicted.
[0017] Figure 6 A schematic diagram of the global feature enhancement module is shown.
[0018] Figure 7 Shown is a schematic diagram of a frame-shielding enhancement module.
[0019] Figure 8 Depicted is a schematic diagram of another embodiment of the HTC enhancement model.
[0020] Fig. 9 A flow chart of the hair feature analysis method of the present invention is shown.
[0021] Figures 10 to 12 Displays the real-life image output by the hair feature analysis system in actual operation.
[0022] Fig.13 Shown is a visualized hair analysis result. DETAILED DESCRIPTION
[0023] The present invention is best understood with reference to the detailed description and accompanying drawings set forth herein. Various embodiments will be discussed below with reference to the accompanying drawings. However, it will be readily appreciated by those skilled in the art that the detailed description given herein with respect to the accompanying drawings is for illustrative purposes only, as these methods and systems may extend beyond the described embodiments. For example, the teachings given and the needs of a particular application may produce a variety of optional and suitable methods to implement the functions of any of the details described herein. Therefore, any method may extend beyond the specific implementation selection range in the following embodiments described and shown.
[0024] See also Figure 1 , Figure 1The first embodiment of a hair feature analysis system of the present invention is shown. The hair feature analysis system 100 includes an image capture device 110, a preprocessing module 120, an artificial intelligence model 130, and an analysis module 140. The image capture device 110 is responsible for capturing a high magnification image and a low magnification image of a scalp area. In this embodiment, the image capture device 110 can capture a high magnification image and a low magnification image of a scalp area at the same time. By capturing high magnification and low magnification images at the same time, complementary information of two different images can be obtained, which helps to improve the accuracy and reliability of hair width and density measurement. These images provide detailed information of hair follicles and hair width, which is essential for analyzing hair features. The preprocessing module 120 is responsible for preprocessing the captured high magnification and low magnification images, including noise reduction, resizing, normalization or other image enhancement techniques. In addition, the preprocessing module 120 can also evaluate the quality of the image. If it is determined that the image quality is not good, the image capture device 110 can be ordered to recapture the image or the operator of the hair feature analysis system 100 can be notified to recapture the image.
[0025] The artificial intelligence model 130 is responsible for detecting hair follicles and calculating hair width from both the preprocessed high magnification and low magnification images. This artificial intelligence model 130 can be an R-CNN model or other variants of the R-CNN model. What is a variant of the R-CNN model will be introduced in more detail later. In addition, the analysis module 140 is responsible for analyzing the image and calculating the hair features based on the output of the artificial intelligence model 130. Using the information provided by the artificial intelligence model 130, the analysis module 140 can perform further analysis, calculate the hair features of the scalp area, and provide personalized recommendations for hair care to understand the health of the hair and discover potential problems related to the hair. This collaboration between the artificial intelligence model 130 and the analysis module 140 ensures a comprehensive and accurate assessment of the health and characteristics of the hair.
[0026] Next, please refer to Figure 2 , Figure 2A second embodiment of the hair feature analysis system of the present invention is illustrated. In addition to the image capture device 110, preprocessing module 120, artificial intelligence model 130, and analysis module 140 described above, the hair feature analysis system 100' of the second embodiment further includes other components, such as a display device 150 for displaying the analyzed image and calculated hair features to the user, a storage module 160 for storing the analyzed image and calculated hair features in a database, and a report generation module 170 for generating a report based on the calculated hair features. In the second embodiment, the report generation module 170 is a comprehensive report generated based on the calculated hair features. The report may contain various types of information, such as charts, statistical data summaries, and written explanations of the data, and the report may also be shared with others, printed, or saved in a digital format for further analysis or record keeping.
[0027] Please refer to Figure 3 , Figure 3 Depicted is a flow chart of the steps of capturing high magnification and low magnification images of a scalp region and preprocessing. In step S210, the capture device 110 captures high magnification and low magnification images of the scalp region. The high magnification images provide detailed information about the hair follicles, while the low magnification images provide an overall view of hair density and distribution, which facilitates subsequent processing and analysis. In step S220, the preprocessing module 120 preprocesses the captured high magnification and low magnification images, such as noise removal, resizing, and normalization, to improve image quality and facilitate subsequent analysis. In step S230, the preprocessing module 120 evaluates the quality of the preprocessed images. If the image quality is not good enough for further processing, the system proceeds to step S240; otherwise, the process proceeds to step S250.
[0028] In step S240, the preprocessing module 120 commands the image capture device 110 to automatically retake the image or notifies the operator of the hair feature analysis system 100 to manually retake the image. When a new image is taken, the process returns to step S220 for preprocessing. In step S250, the preprocessed high magnification and low magnification images are transmitted to the artificial intelligence model 130. After that, the process continues to perform hair feature analysis, which will be described in the following paragraphs.
[0029] Please refer to Figure 4 , Figure 4A schematic diagram of a hybrid task cascade model is depicted, and the hybrid task cascade model (Hybrid Task Cascade, HTC) is hereinafter referred to as the HTC model. The HTC model 300 is a powerful deep learning model that performs well in object detection and segmentation tasks. The HTC model 300 includes multiple stages, each of which includes a framework branch 310 for bounding box detection and a shielding branch 320 for instance segmentation. The HTC model 300 also includes a semantic segmentation module 330 for processing the entire image, which is constructed using a fully convolutional model in this embodiment. The semantic segmentation module 330 is constructed based on the output of the feature pyramid network 340. Then, the region proposal network 350 generates a set of region proposals. In order to fuse multi-level features, the semantic segmentation model 330 combines mid-level features with higher-level features containing global information and lower-level features containing local information. In the HTC model 300, the framework branch 310 and the shielding branch 320 of each stage take the features extracted from the backbone network (i.e., the region proposal network 350) as input, and also use the semantic features of the semantic segmentation module 330. This semantic feature is fused with the features extracted by the ROI pooling operation or the ROI alignment operation, which enhances the performance of the HTC model 130 in object detection and segmentation tasks.
[0030] In summary, the HTC model 300 extracts multi-scale features from the input image using a feature pyramid network 340. Then, the region proposal network 350 generates a set of region proposals. Each stage refines the region proposal, outputs the final bounding box coordinates of the detected object, and predicts a binary mask for each detected object in the refined bounding box. The semantic segmentation branch 330 provides additional spatial context to improve the detection and segmentation of objects in a cluttered background. The output of the HTC model 300 includes multiple detected hair follicles and multiple hairs, which are then used by the analysis module 140 to calculate various hair features, such as average hair diameter, fine hair density ratio, and total hair density ratio. For a more detailed description of the hybrid task cascade model, please refer to the paper "Hybrid Task Cascade for Instance Segmentation" (https: / / arxiv.org / abs / 1901.07518).
[0031] Please refer to Figure 5 , Figure 5 The schematic diagram of the HTC enhanced model of the present invention is shown. In the HTC enhanced model 400, the original HTC model 300 ( Figure 4 ) is replaced with a global feature enhancement module 430. In addition, the HTC enhancement model 400 further includes at least one box-maskenhance module 460. The box-maskenhance module creates a stronger connection mode, strengthens the relationship between the frame branch 310 and the mask branch 320, and thus potentially improves the result of instance segmentation. Therefore, the HTC enhancement model 400 is more suitable for the simultaneous detection of hair follicles and calculation of hair width required in the hair feature analysis system 100. That is, the HTC enhancement model 400 maintains the advantages of the original HTC model 300, but introduces technical features that are added or modified for hair analysis. That is, by replacing the semantic segmentation module 330 with the global feature enhancement module 430 and adding the box-maskenhance module 460, the HTC enhancement model 400 can effectively handle the unique challenges associated with hair follicle detection and hair strand segmentation, thereby ultimately bringing improved hair analysis results.
[0032] Please refer to Figure 6 , Figure 6A schematic diagram of a global feature enhancement module is depicted. The global feature enhancement module 430 is to enhance the global features extracted from the feature pyramid network 340 and integrate them into the HTC enhancement model 400, thereby improving the performance of the HTC enhancement model 400. In this embodiment, the output of the feature pyramid network 340 is first connected to a 13x13 convolution layer 432. This 13x13 convolution layer 432 is responsible for capturing complex and detailed global features from the feature pyramid network 340. The output of this 13x13 convolution layer 432 is connected to a 3x3 convolution layer 434, which is used to extract and refine local features from the input (i.e., the output of the 13x13 convolution layer 432) and enhance the representation of global features. Subsequently, the output of the 3x3 convolution layer 434 is connected to a 1x1 convolution layer 436. This 1x1 convolution layer 436 acts as a bottleneck, reducing the dimension of the features and compressing information while retaining key features. The output of this 1x1 convolutional layer 436 is connected to a global average pooling layer 438. The global average pooling layer 438 calculates the average of the input feature map to generate a compact representation that captures global context information. The generated feature map is then output and used to improve the performance of the HTC enhanced model 400 in detecting hair follicles and calculating hair width. By incorporating the global feature enhancement module 430 into the HTC enhanced model 400, the hair feature analysis system 100 can better analyze hair features and provide valuable insights into hair health and personalized hair care recommendations.
[0033] Please refer to Figure 7 , Figure 7 Shown is a schematic diagram of a frame-shield enhancement module. The frame-shield enhancement module 460 is designed to create a stronger connection, combining the frame branch 310 and the shield branch 320, thereby potentially improving the quality of the instance segmentation results. In this embodiment, the frame-shield enhancement module 460 includes a fully connected layer 462, a reshape layer 464, and a 13x13 convolution layer 466. The fully connected layer 462 is connected to one of the output layers of the frame branch 310, and the output dimension is, for example, Nx4096, where N represents the number of detected bounding boxes. The function of the fully connected layer 462 is to capture features from the frame branch 310, which can improve the prediction effect of the shield branch 320.
[0034] Afterwards, the fully connected layer 462 output tensor is converted from Nx4096 to Nx256*14*14 via the reshape layer 464. This operation enables the tensor to match the size required by the following 13x13 convolution layer 466. The reshape layer 464 connects the reshaped output to the 13x13 convolution layer 466, which in turn connects to the mask branch 320. By connecting the features of the frame branch 310 to the mask branch 320 in this way, the frame-mask enhancement module 460 can make the prediction of the mask branch 320 more accurate.
[0035] The following will introduce two configuration embodiments of the frame-shielding enhancement module. In one embodiment, Figure 5 As shown, a frame-shield enhancement module 460 is placed between the frame branch 310 and the shield branch 320 of each stage of the HTC enhancement model 400. This design helps to propagate more accurate bounding box information into the shield prediction process at each stage. By incorporating the frame-shield enhancement module 460, the HTC enhancement model 400 is able to generate more accurate shield predictions and achieve better alignment between bounding boxes and shields.
[0036] Next, please refer to Figure 8 , which is a schematic diagram of another embodiment of the HTC enhancement model. The shield branch 320' is placed after the last frame branch 310, and a frame-shield enhancement module 460 is used to connect them. The focus of this design is to improve the shield prediction in the last stage of the HTC enhancement model 400', so that the most refined bounding box information can be used. By connecting the last frame branch 310 with the shield branch 320' using the frame-shield enhancement module 460, the HTC enhancement model 400' may generate a more accurate instance mask that is better aligned with the detected bounding box.
[0037] The following is a comprehensive description of the hair feature analysis method of the present invention. Fig. 9 , Fig. 91 shows a flow chart of the hair feature analysis method of the present invention. The hair feature analysis method of the present invention includes several steps. First, in step S610, a high magnification image and a low magnification image of a scalp area are captured. Then, in step S620, the captured images are pre-processed by the pre-processing module 120, including various techniques such as resizing, normalization and augmentation, to prepare input images for use by the artificial intelligence model 130. Next, in step S630, the pre-processed images are input to the artificial intelligence model 130, and the artificial intelligence model 130 uses the HTC enhancement model 400 to simultaneously perform hair follicle detection and calculate hair width. Next, in step S640, the analysis module 140 post-processes the output of the artificial intelligence model 130. Then, in step S650, the results of the analysis module 140 are output.
[0038] Here, we will introduce step S640 and step S650 in more detail, and first introduce step S640. After the artificial intelligence model 130 simultaneously detects multiple hair follicles and multiple hair widths in the scalp area from the pre-processed image, the analysis module 140 receives the information output by the artificial intelligence model 130. The analysis module 140 segments the hair based on the information, and the middle segment of each hair is selected as the basis for width calculation. Then, by analyzing the selected hair segment, the number of pixels corresponding to the hair width is determined. After that, the analysis module 140 converts the pixel count representing the hair width into a real-world measurement scale (e.g., microns) to provide an accurate hair width. In addition, the analysis module 140 also calculates the fine hair density ratio, that is, the ratio of fine hair (i.e., hair with a diameter less than a certain threshold) to the total number of hair. This indicator can help evaluate hair strength and identify possible hair thinning or breakage problems. In addition, the analysis module 140 also calculates the total hair density ratio, that is, the ratio of the total number of hair follicles to the area in the examined scalp area. This parameter helps identify the degree of hair loss or thinning.
[0039] Once the analysis module 140 calculates these hair features, the hair feature analysis system 100 proceeds to step S650 to present the results to the user. In step S650, the display device 150 displays the analyzed image and the calculated head features, providing the user with a visual representation of the hair analysis results (e.g., Fig.13). In addition, the storage module 160 stores the analyzed images and calculated hair features in a database for easy retrieval and comparison of the results at any time. This function can be used to track the progress of hair health or evaluate the effectiveness of hair care treatments. In addition, the report generation module 170 generates a report based on the calculated hair features. This report may include a summary of the hair analysis results, personalized hair care recommendations, and any other information related to the user's hair health. The report can be printed, emailed, or otherwise shared with the user for reference and discussion.
[0040] See also Figures 10 to 12 , Figures 10 to 12 The real-life images output by the hair feature analysis system in actual operation are shown. These real-life images show that the artificial intelligence model 130 of the present invention demonstrates its ability to accurately identify hair follicles and hair, which can serve as empirical evidence of the effectiveness of the artificial intelligence model 130 in detecting and analyzing hair-related features. It should be noted that Figures 10 to 12 The real scene image is the output result when the artificial intelligence model 130 includes the HTC enhancement model 400. Fig.10 In the example, the artificial intelligence model 130 is able to effectively distinguish individual hair follicles from the surrounding skin and hair strands, highlighting its accuracy in detecting the location of hair follicles. This ability to accurately detect hair follicles enables the hair feature analysis system 100 to provide reliable hair density and distribution measurements.
[0041] Fig.11 The ability of the artificial intelligence model 130 to identify individual hairs is demonstrated. When individual hairs are effectively identified, the width of each hair can be accurately calculated, which can provide the user with valuable information about hair health, such as average hair diameter, fine hair density ratio, and total hair density ratio. Fig.12 The artificial intelligence model 130 is shown to successfully segment individual hairs and hair follicles, and the effectiveness of the HTC enhanced model 400 in object detection and instance segmentation tasks is also confirmed. This segmentation enables the hair feature analysis system 100 to simultaneously detect hair follicles and calculate hair width, thereby improving the analysis results. Figures 10 to 12 The real-life images shown provide a strong visual demonstration of the ability of the artificial intelligence model 130 to accurately detect and analyze hair follicles and hair, while demonstrating the practicality and effectiveness of the hair feature analysis system 100 of the present invention in a real environment.
[0042] In the above-mentioned embodiment, the artificial intelligence model 130 is implemented using the HTC model and the HTC enhanced model, but the artificial intelligence model 130 may also use the R-CNN model. When the artificial intelligence model 130 is an R-CNN model, the first step will be to generate potential bounding boxes or region proposals for the hair follicles in the input image. A selective search algorithm or other region proposal algorithm may be used to determine regions of interest based on color, texture, and size. Next, for each region of interest, the R-CNN model will apply a pre-trained convolutional neural network (CNN) to extract meaningful features, which will help identify the patterns and features of hair follicles and hair, enabling the model to accurately identify and segment them. After feature extraction, a classifier is applied to each region of interest to determine whether it contains hair follicles. The classifier is, for example, a support vector machine, which uses the features extracted in the previous step to make decisions. Next, bounding box regression is performed in parallel with the classification step, which will optimize the coordinates of the proposed bounding box to better select the hair follicles and improve the accuracy of hair follicle detection and segmentation. Afterwards, the R-CNN model can pass the output data to the analysis model 140 for subsequent processing.
[0043] In addition, the artificial intelligence model 130 may also use variants of other R-CNN models, such as a cascade R-CNN model or a masked R-CNN model. The cascade R-CNN model can refer to the paper "Cascade R-CNN: High Quality Object Detection and Instance Segmentation" (https: / / arxiv.org / abs / 1906.09756), and the masked R-CNN model can refer to the paper "Mask R-CNN" (https: / / arxiv.org / abs / 1703.06870).
[0044] The above embodiments are merely examples for the convenience of explanation. Although any modifications are made by those skilled in the art, they will not deviate from the scope of protection as set forth in the claims.
Claims
1. A hair feature analysis method, characterized in that: include: capturing a high magnification image and a low magnification image of a scalp region; Preprocessing the high magnification image and the low magnification image; Inputting the pre-processed high magnification image and the low magnification image into an artificial intelligence model to simultaneously detect a plurality of hair follicles and a plurality of hair widths in the scalp area; Calculating at least one hair feature by analyzing the high magnification image and the low magnification image through an artificial intelligence model; Among them, the artificial intelligence model includes the R-CNN model or a variant of the R-CNN model.
2. The hair feature analysis method according to claim 1, wherein the artificial intelligence model is a HTC enhanced model, the HTC enhanced model is modified from a hybrid task cascade model, and the HTC enhanced model has the following characteristics: A semantic segmentation module in the mixed task cascade model is replaced with a global feature enhancement module; A refined semantic feature from the global feature enhancement module is connected to a plurality of frame branches; and at least one frame-shielding enhancement module is introduced to connect the frame branch and at least one shielding branch.
3. The hair feature analysis method according to claim 2, characterized in that: The global feature enhancement module includes multiple convolutional layers with different kernel sizes to extract multi-scale features.
4. The hair feature analysis method according to claim 2, characterized in that: The HTC enhanced model is provided with the frame branch and the shielding branch in each stage, and the number of the frame-shielding enhanced modules is multiple, and each frame-shielding enhanced module is connected between the frame branch and the shielding branch in each stage.
5. The hair feature analysis method according to claim 2, characterized in that: The HTC enhanced model is provided with the frame branch but only with one shielding branch at each stage, the frame-shielding enhancement module is arranged between the last frame branch and the shielding branch, and the frame-shielding enhancement module connects the last frame branch and the shielding branch.
6. The hair feature analysis method according to claim 1, characterized in that: The hair characteristics include average hair diameter, fine hair density ratio, or total hair density ratio.
7. The hair feature analysis method according to claim 1, characterized in that: The artificial intelligence model includes a mixed task cascade model.
8. A hair feature analysis system, characterized in that: include: an image capture device for capturing a high magnification image and a low magnification image of a scalp region; A pre-processing module, used for pre-processing the captured high magnification image and the low magnification image; An artificial intelligence model for simultaneously detecting hair follicles and calculating hair width, wherein the artificial intelligence model comprises an R-CNN model or a variant of the R-CNN model; An analysis module is used to receive the image analyzed by the artificial intelligence model and calculate at least one hair feature.
9. The hair feature analysis system according to claim 8, characterized in that: The artificial intelligence model is an HTC enhanced model, which is modified from the mixed task cascade model and has the following characteristics: A semantic segmentation module in the mixed task cascade model is replaced with a global feature enhancement module; A refined semantic feature from the global feature enhancement module is connected to a plurality of frame branches; a frame-shielding enhancement module is introduced to connect the frame branch and at least one shielding branch.
10. The hair feature analysis system according to claim 9, characterized in that: The global feature enhancement module includes multiple convolutional layers with different kernel sizes to extract multi-scale features.
11. The hair feature analysis system according to claim 9, characterized in that: The HTC enhanced model is provided with the frame branch and the shielding branch in each stage, and the number of the frame-shielding enhanced modules is multiple, and each frame-shielding enhanced module is connected between the frame branch and the shielding branch in each stage.
12. The hair feature analysis system according to claim 9, characterized in that: The HTC enhanced model is provided with the frame branch but only with one shielding branch at each stage, the frame-shielding enhancement module is arranged between the last frame branch and the shielding branch, and the frame-shielding enhancement module connects the last frame branch and the shielding branch.
13. The hair feature analysis system according to claim 8, characterized in that: The artificial intelligence model includes a mixed task cascade model.
Citation Information
Cited By
Intelligent scalp health assessment system based on multi-modal imaging and data fusion
CN120565124A
Intelligent evaluation system for scalp health based on multi-modal imaging and data fusion
CN120565124B
