Method and intelligent system for human body multi-granularity analysis

Abstract

The invention relates to the technical field of image recognition, in particular to a method for human body multi-granularity analysis and an intelligent system.The method for human body multi-granularity analysis comprises the following steps that S1, the intelligent system is initialized, and image information containing human body parts is obtained; s2, inputting image information as an original picture, extracting human body region features, and extracting corresponding category channels and category segmentation regions by adopting a segmentation model according to the image information; and S3, obtaining the extracted category channel and the category segmentation region, outputting human body analysis results of different granularities according to a preset segmentation demand, and completing region coordinate output. According to the method, the corresponding category channel and the category segmentation region can be extracted by adopting the segmentation model based on the deep learning algorithm according to the image information, and human body features of different granularities such as the human body region, the human body part, the skin region and the skin lesion region in the image information can be analyzed according to requirements.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular to a method and an intelligent system for multi-granularity analysis of a human body. Background Art

[0002] In many scenarios of real applications, it is necessary to perform intelligent analysis of the human body structure, such as separating the human body area from the background area, analyzing the human body area according to different parts, or only extracting the skin area to provide intelligent assistance for subsequent operations, such as human appearance transfer, pedestrian re-identification, fashion synthesis, etc. Even in the medical field, more attention will be paid to the skin lesion area and thus it will be focused on. At present, the solution for intelligent analysis of human body structure mainly adopts single-granularity analysis, that is, it can only realize human body area classification or human body part analysis. There is no universal method that can perform multiple different granularities of analysis on the human body at one time. The application of this single function will lead to problems such as poor versatility and low application efficiency. Summary of the invention

[0003] The technical problem to be solved by the present invention is as follows: the existing human body structure analysis method mainly adopts single-granularity analysis, which can only realize human body region classification or human body part analysis, and cannot realize multiple different granularity analyses of the human body at one time. The application of this single function will lead to problems such as poor versatility and low application efficiency.

[0004] In order to solve the above technical problems, the first aspect of the present invention adopts the following technical solution: a method for multi-particle analysis of human body, comprising the following steps: S1: Initialize the intelligent system and obtain image information containing human body parts; S2: Input image information as the original image and extract the human body region features. According to the image information, the segmentation model is used to extract the corresponding category channel and category segmentation region; S3: Obtain the extracted category channels and category segmentation areas, output human body analysis results of different granularities according to the preset segmentation requirements, and complete the area coordinate output.

[0005] When the present invention is working, it can extract corresponding category channels and category segmentation areas according to image information using a segmentation model based on a deep learning algorithm, and can parse human features of different granularities such as human body areas, human body parts, skin areas and skin lesion areas in the image information according to needs, which greatly simplifies the workflow of human body analysis work, and can ensure the accuracy of analysis results of various granularities, with high flexibility and high analysis efficiency.

[0006] Preferably, in step S3, when obtaining the extracted category channels and category segmentation regions, outputting human body analysis results of different granularities according to preset segmentation requirements and completing the output of region coordinates, the following steps are adopted: A1: Determine the segmentation requirement and select at least one corresponding category channel according to the segmentation requirement; A2: Perform grayscale visualization conversion according to the category segmentation area corresponding to the selected category channel to obtain a grayscale image; A3: After Gaussian blurring the acquired grayscale image, the denoised grayscale image is subjected to binary threshold processing to highlight and clarify the edges of the category segmentation area; A4: Keep the selected category segmentation area as the foreground and remove the background area; A5: Extract the coordinates of the edge of the category segmentation area that serves as the foreground and output the foreground coordinates.

[0007] When the present invention is working, it can output human body analysis results of various granularities, and can dynamically adjust the number of category channels according to segmentation requirements. It has a high degree of customization, is applicable to various application fields, has good versatility, and greatly reduces the difficulty of development and application.

[0008] Preferably, the method further includes the step of further expanding the multi-granularity parsing capability of the segmentation model, specifically using the following steps: B1: Collect relevant data that need to be expanded in granularity, and annotate them to obtain a new granularity dataset; B2: Obtain the existing data set of the segmentation model, perform data annotation fusion on the existing data set and the new granularity data set, and obtain a fused data set with the new granularity features added; B3: The segmentation model is transferred and trained by fusing the dataset, and the segmentation model automatically extracts and learns new granular features; B4: After the training and learning is completed, the segmentation model is verified for model performance. When the model performance verification is qualified, the segmentation model is saved to complete the expansion of the segmentation model's ability to analyze the human body at more granularity.

[0009] Preferably, in step B3, when the segmentation model is transferred and trained by fusing the data set and the segmentation model automatically extracts and learns new granularity features, the following steps are adopted: C1: Divide the fused data set into a training set and a validation set, perform data augmentation, and perform standardization or normalization; C2: Use a deep learning framework to define the model structure, select an appropriate loss function based on the preset task objectives, configure the optimizer to optimize the hyperparameters of the segmentation model, determine the evaluation criteria, and define the evaluation indicators that are monitored first during training and validation; C3: Load the data in the training set in batches and participate in training. Repeat the process of forward propagation, loss calculation, backpropagation and parameter update until all epochs are completed and go to step B4.

[0010] Preferably, in step B4, the performance of the segmentation model after training and learning is verified. When the model performance verification is qualified, the segmentation model is saved. When the segmentation model's ability to parse the human body at more granularity is expanded, the following steps are adopted to divide part of the data in the fusion data set into a verification set, evaluate the model performance of the trained segmentation model on the verification set, monitor the evaluation indicators, and when the model performance verification is qualified, the segmentation model is saved, completing the expansion of the segmentation model's ability to parse the human body at more granularity.

[0011] When the present invention is working, it can expand the human body analysis capabilities with more granularity according to the customer's customized needs, has a high degree of customization, is convenient to apply to various related application fields, is easy to adjust, can effectively reduce the difficulty of adjustment, and has high application efficiency.

[0012] Preferably, in step B1, when collecting relevant data whose granularity needs to be expanded and annotating to obtain a new granularity data set, the following steps are adopted to collect pictures containing lesions of different skin diseases, and obtain a skin lesion data set after polygonal annotation of the skin lesion area; In step B2, when obtaining an existing data set of the segmentation model, annotating and fusing the existing data set with the new granularity data set to obtain a fused data set with new granularity features, the following steps are adopted to obtain an existing data set of the segmentation model, annotating and fusing the existing data set with the skin lesion data set to obtain a fused data set with skin lesion features.

[0013] Preferably, the step S3 further includes the following steps: obtaining coordinate information of human body parts and skin lesion areas to perform auxiliary analysis of disease severity, and outputting the skin lesion severity level by calculating the skin lesion ratio.

[0014] Preferably, step S3 further includes the following steps: acquiring features of human body parts, skin areas and skin lesion areas for auxiliary disease diagnosis analysis, and outputting the probabilities of several skin diseases in the skin lesion areas by combining features learned by the model.

[0015] Preferably, step S3 also includes the following steps: obtaining features of human body parts, skin areas and skin lesion areas for auxiliary disease diagnosis analysis, outputting the probabilities of several skin diseases in the skin lesion area through the feature combination learned by the model, obtaining the coordinate information of human body parts and skin lesion areas for auxiliary disease severity analysis, and outputting the skin lesion severity level by calculating the skin lesion ratio.

[0016] When the present invention is working, the segmentation model completed by training the fusion data set with skin lesion features can realize the segmentation function of the skin lesion area, and can output the skin lesion severity level and the probability of several skin diseases in the skin lesion area through disease auxiliary diagnosis analysis and disease severity auxiliary analysis, with good auxiliary effect.

[0017] To solve the above technical problems, the second aspect of the present invention adopts the following technical solution: an intelligent system for multi-granularity analysis of the human body, applying a method for multi-granularity analysis of the human body as described above, the intelligent system includes a processor and a memory, the memory stores at least one instruction, at least one program, code set or instruction set, the instruction, program, code set or instruction set is loaded by the processor and executed by a method for multi-granularity analysis of the human body as described above.

[0018] The beneficial technical effects of the present invention include: 1. The present invention can extract corresponding category channels and category segmentation areas according to image information using a segmentation model based on a deep learning algorithm, and can parse human features of different granularities such as human body areas, human body parts, skin areas, and skin lesion areas in the image information according to needs, which greatly simplifies the workflow of human body analysis work, and can ensure the accuracy of analysis results of various granularities, with high flexibility and high analysis efficiency.

[0019] 2. The present invention can output human body analysis results of various granularities, and can dynamically adjust the number of category channels according to segmentation requirements. It has a high degree of customization, is applicable to various application fields, has good versatility, and greatly reduces the difficulty of development and application.

[0020] 3. The present invention can expand the human body analysis capabilities with more granularity according to customer customization needs, has a high degree of customization, is convenient for application in various related application fields, is easy to adjust, and can effectively reduce the difficulty of adjustment, with high application efficiency.

[0021] 4. The present invention can realize the segmentation function of the skin lesion area by training the segmentation model with the fusion data set that adds skin lesion features, and can output the skin lesion severity level and the probability of several skin diseases in the skin lesion area through disease auxiliary diagnosis analysis and disease severity auxiliary analysis, with good auxiliary effect.

[0022] Other features and advantages of the present invention will be disclosed in detail in the following specific embodiments and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The present invention will be further described below in conjunction with the accompanying drawings: Figure 1 is a workflow diagram of a method for multi-granularity analysis of the human body; Figure 2 is a workflow diagram of step S3 in a method for multi-granularity analysis of human body; Figure 3 The segmentation map output by multi-granularity analysis of the whole body parts; Figure 4 This is the segmentation map of the back area when multi-granularity analysis is performed on the whole body. Figure 5 Segmentation of the left calf posterior region during multi-granularity analysis of whole body parts; Figure 6 It is the segmentation map output by multi-granularity analysis of local parts; Figure 7 It is the segmentation map of the back area during multi-granularity analysis of local parts; Figure 8 Workflow diagram to further expand the multi-granularity parsing capability of the segmentation model; Fig. 9 A workflow diagram for outputting the severity level of skin lesions in the skin lesion area; Fig.10 A workflow diagram for outputting the probability of skin diseases in skin lesion areas; Fig.11 A workflow diagram for simultaneously outputting the probability of skin diseases and the severity of skin lesions in the skin lesion area; Fig.12 This is the segmentation map output by multi-granularity analysis of the skin lesion area. DETAILED DESCRIPTION

[0024] The technical solutions of the embodiments of the present invention are explained and described below in conjunction with the drawings of the embodiments of the present invention, but the following embodiments are only preferred embodiments of the present invention, not all. Based on the embodiments in the implementation mode, other embodiments obtained by those skilled in the art without creative work are all within the protection scope of the present invention.

[0025] In the following description, terms such as "inside", "outside", "up", "down", "left", "right", etc. that indicate directions or positional relationships are only used to facilitate the description of the embodiments and simplify the description, and do not indicate or imply that the referred device or element must have a specific direction, be constructed and operate in a specific direction. Therefore, it should not be understood as a limitation of the present invention. Embodiment 1:

[0026] See also Figure 1 This embodiment discloses a method for multi-granularity analysis of human body, comprising the following steps: S1: Initialize the intelligent system and obtain image information containing human body parts; S2: Input image information as the original image and extract the human body region features. According to the image information, the segmentation model is used to extract the corresponding category channel and category segmentation region; S3: Obtain the extracted category channels and category segmentation areas, output human body analysis results of different granularities according to the preset segmentation requirements, and complete the area coordinate output.

[0027] When this embodiment is working, it can extract corresponding category channels and category segmentation areas according to image information using a segmentation model based on a deep learning algorithm, and can parse human features of different granularities such as human body areas, human body parts, skin areas, and skin lesion areas in the image information as needed, which greatly simplifies the workflow of human body analysis work, and can ensure the accuracy of analysis results of various granularities, with high flexibility and high analysis efficiency.

[0028] See also Figure 2 In the specific implementation, in the step S3, the extracted category channels and category segmentation regions are obtained, and the human body analysis results of different granularities are output according to the preset segmentation requirements and the regional coordinate output is completed, and the following steps are adopted: A1: Determine the segmentation requirement and select at least one corresponding category channel according to the segmentation requirement; A2: Perform grayscale visualization conversion according to the category segmentation area corresponding to the selected category channel to obtain a grayscale image; A3: After Gaussian blurring the acquired grayscale image, the denoised grayscale image is subjected to binary threshold processing to highlight and clarify the edges of the category segmentation area; A4: Keep the selected category segmentation area as the foreground and remove the background area; A5: Extract the coordinates of the edge of the category segmentation area that serves as the foreground and output the foreground coordinates.

[0029] When this embodiment is working, it can output human body analysis results of various granularities, and can dynamically adjust the number of category channels according to segmentation requirements. It has a high degree of customization, is suitable for various application fields, has good versatility, and greatly reduces the difficulty of development and application.

[0030] While working, see Figures 3 to 7 , obtain image information containing human body parts, input image information as the original image and extract human body region features, and use the segmentation model to extract the corresponding category channels and category segmentation regions based on the image information. For the classification, detection and segmentation results of whole body parts, please refer to Figure 3 , the background area has been completely removed, among which, Figure 3The yellow area on the upper left is the back of the left arm, the pink area on the upper middle is the back, the yellow area on the upper right is the back of the right arm, the yellow area on the lower left is the back of the left thigh, the purple area on the lower left is the back of the left calf, the purple area on the lower right is the back of the right thigh, and the yellow area on the lower right is the back of the right calf. The human body part segmentation is completed. Please refer to Figure 4 and Figure 5 , the evaluation area can be selected by clicking, for further local human body analysis, e.g. Figure 6 When analyzing local parts, more detailed regional segmentation can be performed, and the evaluation parts can also be selected at this time.

[0031] In specific implementation, the segmentation model includes: Backbone, responsible for extracting multi-scale features of the input image, generally ResNet or SwinTransformer can be used; Transformer encoder——Responsible for feature enhancement of the features extracted by Backbone and their corresponding positional embeddings, i.e. position encoding, and initializing the learnable parameters for the subsequent Transformer decoder; Transformer decoder - responsible for merging the learnable parameters output by the Transformer encoder and updating them layer by layer; Prediction head——responsible for outputting classification results, namely Classes, category channels, rectangular boxes, namely Boxes, object target positioning, segmentation areas, namely Masks, object segmentation, and combining the three information to give the segmentation area of ​​​​a specific category of objects. Embodiment 2:

[0032] See also Figure 8 This embodiment provides a method for multi-granularity analysis of the human body. The similarities with other embodiments are not repeated here, and the differences are described in detail below.

[0033] See also Figure 8 In this embodiment, the step of further expanding the multi-granularity parsing capability of the segmentation model is also included, specifically using the following steps: B1: Collect relevant data that need to be expanded in granularity, and obtain a new granularity data set after annotation. In specific implementation, the annotation information should be polygon annotations of different granularity areas and corresponding category names. It can be done manually or pre-annotated through an existing segmentation model and then manually reviewed, which can reduce the investment in labor costs to a certain extent; B2: Obtain an existing data set of the segmentation model, fuse the existing data set with the new granularity data set for data annotation, and obtain a fused data set with new granularity features added. In specific implementation, when fusing the existing data set with the new granularity data set for data annotation, the following steps are adopted to put the images of the existing data set and the images of the new granularity data set together, and fuse the related annotation files into one annotation file; B3: The segmentation model is transferred and trained by fusing the dataset, and the segmentation model automatically extracts and learns new granular features; B4: After the training and learning is completed, the segmentation model is verified for model performance. When the model performance verification is qualified, the segmentation model is saved to complete the expansion of the segmentation model's ability to analyze the human body at more granularity.

[0034] In the specific implementation, in step B3, the segmentation model is transferred and trained by fusing the data set, and the segmentation model automatically extracts and learns new granular features, and the following steps are adopted: C1: Divide the fused data set into a training set and a validation set, perform data augmentation, and perform standardization or normalization; C2: Use a deep learning framework to define the model structure, select an appropriate loss function based on the preset task objectives, configure the optimizer to optimize the hyperparameters of the segmentation model, determine the evaluation criteria, and define the evaluation indicators that are monitored first during training and validation; C3: Load the data in the training set in batches and participate in training. Repeat the process of forward propagation, loss calculation, backpropagation and parameter update until all epochs are completed and go to step B4.

[0035] Preferably, in step B4, the performance of the segmentation model after training and learning is verified. When the model performance verification is qualified, the segmentation model is saved. When the segmentation model's ability to parse the human body at more granularity is expanded, the following steps are adopted to divide part of the data in the fusion data set into a verification set, evaluate the model performance of the trained segmentation model on the verification set, monitor the evaluation indicators, and when the model performance verification is qualified, the segmentation model is saved to complete the expansion of the segmentation model's ability to parse the human body at more granularity.

[0036] When this embodiment is working, it can expand the human body analysis capabilities with more granularity according to customer customization needs, has a high degree of customization, is convenient for application in various related application fields, is easy to adjust, and can effectively reduce the difficulty of adjustment, with high application efficiency. Embodiment three:

[0037] This embodiment provides a method for multi-granularity analysis of the human body. The similarities with other embodiments are not repeated here, and the differences are described in detail below.

[0038] In this embodiment, in step B1, when collecting relevant data that need to expand the granularity and annotating them to obtain a new granularity data set, the following steps are adopted to collect pictures containing lesions of different skin diseases, and obtain a skin lesion data set after polygonal annotation of the skin lesion area; In step B2, when obtaining an existing data set of the segmentation model, annotating and fusing the existing data set with the new granularity data set to obtain a fused data set with new granularity features, the following steps are adopted to obtain an existing data set of the segmentation model, annotating and fusing the existing data set with the skin lesion data set to obtain a fused data set with skin lesion features.

[0039] See also Fig. 9 In the specific implementation, in the step S3, the following steps are also included: obtaining the coordinate information of the human body part and the skin lesion area to perform auxiliary analysis of the severity of the disease, and outputting the skin lesion severity level by calculating the skin lesion ratio.

[0040] See also Fig.10 Preferably, in the step S3, the following steps are also included: obtaining the characteristics of the human body part, skin area and skin lesion area for auxiliary disease diagnosis analysis, and outputting the probabilities of several skin diseases in the skin lesion area by combining the features learned by the model.

[0041] As a further improvement of this embodiment, in step S3, the features of the human body part, the skin area and the skin lesion area are obtained for disease-assisted diagnosis analysis, and the probabilities of several skin diseases in the skin lesion area are output by combining the features learned by the model, and the following steps are adopted: D1: Acquire the features of human body parts, skin areas and skin lesion areas, determine the size of the recognition window according to a preset size, divide the skin lesion area to be recognized according to a preset step length to obtain several recognition windows, and add a mark to each recognition window according to the features of its corresponding human body part or skin area; D2: Extract and identify features of each recognition window through the segmentation model to obtain the probability of skin disease in each recognition window; D3: setting the recognition window whose skin disease probability is higher than the preset determination threshold as the recognition base point, and sequentially and gradually fusing several recognition windows whose skin disease probability meets the currently selected similarity classification level according to the preset similarity classification level; D4: Change the size of the recognition window and divide the fused skin lesion area to obtain several recognition windows. Use the segmentation model to extract and recognize the features of the recognition windows located at the edge, obtain the skin disease probability of each recognition window, and filter out the recognition windows with skin disease probabilities lower than the preset filtering threshold. When all the recognition windows similar to all skin diseases have been determined, calculate and output several skin disease probabilities for the skin lesion area.

[0042] Preferably, in step D2, the following step is further included: when the skin disease probabilities of all recognition windows are less than a preset determination threshold, the size and step length of the recognition window are changed, and the process proceeds to step D1, wherein the change of the recognition window size adopts the following formula: ; in: To change the size of the recognition window, is the size of the recognition window before the change, s is the step size before the change, m is the preset adjustment rate parameter, which can be set in the range of 2 to 3 in the specific implementation, and can be dynamically adjusted according to the number of changes, so as to speed up the recognition speed of effective content, H is the skin disease probability set of several recognition windows, is the mode of the probability of skin diseases in set H.

[0043] In step D4, the size of the identification window is changed and the fused skin lesion area is divided to obtain a plurality of identification windows. The identification windows located at the edge are subjected to feature extraction and identification through the segmentation model to obtain the skin disease probability of each identification window. When the identification windows with skin disease probabilities lower than a preset filtering threshold are filtered out, the following steps are adopted: E1: Determine the recognition base point, divide the fused skin lesion area at least once inward from the edge position of the fused skin lesion area according to the changed recognition window size and step length to obtain several recognition windows, extract and recognize features of each recognition window through the segmentation model, and obtain the skin disease probability of each recognition window; E2: Filter out the recognition windows whose skin disease probabilities are lower than the preset filtering threshold from the skin lesion area and go to step E1. When the skin disease probabilities of all recognition windows are higher than the preset determination threshold, end the recognition operation and output the coordinate information of the skin lesion area and the skin disease type.

[0044] In the step E1, the identification base point is determined, and the fused skin lesion area is divided inward at least once from the edge position of the fused skin lesion area according to the changed identification window size and step length to obtain a number of identification windows. The segmentation model is used to extract and identify features of each identification window. When the skin disease probability of each identification window is obtained, the following steps are adopted: from the edge position of the fused skin lesion area to the direction close to the identification base point according to the changed identification window size and step length, the segmentation model is used to extract and identify the features of the identification window to obtain the skin disease probability of the identification window. When the skin disease probability of the identification window is lower than the preset filtering threshold, the identification window is continued to be obtained in the direction close to the identification base point according to the changed identification window size and step length. The segmentation model is used to extract and identify the features of the identification window to obtain the skin disease probability of the identification window. This is repeated in sequence until the skin disease probability of the current identification window is higher than the preset determination threshold. This is repeated in sequence to complete the redetermination of the edge position of all fused skin lesion areas.

[0045] In step E1, the change in the recognition window size is calculated using the following formula: ; in: To change the size of the recognition window, is the size of the recognition window before changing, n is the preset adjustment rate parameter, is the probability of the skin disease corresponding to the current recognition base point in the recognition window, L is the set of all skin disease probabilities in the current recognition window except the skin disease corresponding to the current recognition base point, To take the maximum value of the probability of skin diseases in set L.

[0046] Of course, in addition to dividing the edge of the lesion area of ​​a specific skin disease by corrosion, the edge of the lesion area of ​​a specific skin disease can also be determined by expansion, that is, by obtaining other skin diseases except the skin disease corresponding to the identification base point in the identification window of the edge position of the lesion area after fusion, and dividing the identification window from the inside to the outside through other skin diseases, filtering the identification windows with probabilities of other skin diseases higher than the preset filtering threshold, and redetermining the edge position of the lesion area can also be achieved. Although the accuracy will be reduced, it can cover a larger lesion area, which has certain advantages for the division of lesion areas with lower complexity.

[0047] See also Fig.11As a further improvement of this embodiment, in step S3, the following steps are also included: obtaining the features of human body parts, skin areas and skin lesion areas for auxiliary disease diagnosis analysis, outputting the probabilities of several skin diseases in the skin lesion area through the feature combination learned by the model, obtaining the coordinate information of the human body parts and the skin lesion area for auxiliary disease severity analysis, and outputting the skin lesion severity level by calculating the skin lesion ratio, which can further improve the analysis accuracy and is suitable for high-difficulty scenarios with complex skin lesion areas and fuzzy disease probabilities, and has a good application effect.

[0048] When this embodiment is working, the segmentation model trained by the fusion data set with skin lesion features added can realize the segmentation function of the skin lesion area, and can output the skin lesion severity level and the probabilities of several skin diseases in the skin lesion area through disease auxiliary diagnosis analysis and disease severity auxiliary analysis, with good auxiliary effect.

[0049] While working, see Fig.12 , which segmented the background area, i.e., the areas marked with numbers 1, 2, 3, and 4, the foreground area, i.e., the leg area marked with number 5, and the skin lesion area, i.e., the areas marked with numbers 6, 7, 8, 9, and 10. Each area is covered by a mask of a different color to facilitate identification and subsequent analysis. Embodiment 4:

[0050] The present embodiment provides an intelligent system for multi-granularity analysis of the human body, which applies a method for multi-granularity analysis of the human body as in the above embodiment. The intelligent system includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the instruction, program, code set or instruction set is loaded by the processor and executed by a method for multi-granularity analysis of the human body as in the above embodiment.

[0051] The beneficial technical effects of this embodiment include: the present invention can extract corresponding category channels and category segmentation areas according to image information using a segmentation model based on a deep learning algorithm, and can parse human features of different granularities such as human body areas, human body parts, skin areas and skin lesion areas in the image information as needed, which greatly simplifies the workflow of human body analysis work, and can ensure the accuracy of analysis results of various granularities, with high flexibility and high analysis efficiency.

[0052] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Those skilled in the art should understand that the present invention includes but is not limited to the contents described in the drawings and the above specific embodiments. Any modification that does not deviate from the functional and structural principles of the present invention will be included in the scope of the claims.

Claims

1. A method for multi-granularity analysis of human body, characterized in that: The following steps are involved: S1: Initialize the intelligent system and obtain image information containing human body parts; S2: Input image information as the original image and extract the human body region features. According to the image information, the segmentation model is used to extract the corresponding category channel and category segmentation region; S3: Obtain the extracted category channels and category segmentation areas, output human body analysis results of different granularities according to the preset segmentation requirements, and complete the area coordinate output.

2. A method for multi-granularity analysis of human body according to claim 1, characterized in that: In step S3, the extracted category channels and category segmentation regions are obtained, and human body analysis results of different granularities are output according to preset segmentation requirements and the region coordinate output is completed, and the following steps are adopted: A1: Determine the segmentation requirement and select at least one corresponding category channel according to the segmentation requirement; A2: Perform grayscale visualization conversion according to the category segmentation area corresponding to the selected category channel to obtain a grayscale image; A3: After Gaussian blurring the acquired grayscale image, the denoised grayscale image is subjected to binary threshold processing to highlight and clarify the edges of the category segmentation area; A4: Keep the selected category segmentation area as the foreground and remove the background area; A5: Extract the coordinates of the edge of the category segmentation area that serves as the foreground and output the foreground coordinates.

3. A method for multi-granularity analysis of human body according to claim 1, characterized in that: The method also includes steps to further expand the multi-granularity parsing capability of the segmentation model, specifically the following steps: B1: Collect relevant data that need to be expanded in granularity, and annotate them to obtain a new granularity dataset; B2: Obtain the existing data set of the segmentation model, perform data annotation fusion on the existing data set and the new granularity data set, and obtain a fused data set with the new granularity features added; B3: The segmentation model is transferred and trained by fusing the dataset, and the segmentation model automatically extracts and learns new granular features; B4: After the training and learning is completed, the segmentation model is verified for model performance. When the model performance verification is qualified, the segmentation model is saved to complete the expansion of the segmentation model's ability to analyze the human body at more granularity.

4. A method for multi-granularity analysis of human body according to claim 3, characterized in that: In step B3, the segmentation model is transferred and trained by fusing the data set, and the segmentation model automatically extracts and learns new granular features, and the following steps are adopted: C1: Divide the fused data set into a training set and a validation set, perform data augmentation, and perform standardization or normalization; C2: Use a deep learning framework to define the model structure, select an appropriate loss function based on the preset task objectives, configure the optimizer to optimize the hyperparameters of the segmentation model, determine the evaluation criteria, and define the evaluation indicators that are monitored first during training and validation; C3: Load the data in the training set in batches and participate in training, repeat the process of forward propagation, loss calculation, backpropagation and parameter update until all epochs are completed and go to step B4.

5. A method for multi-granularity analysis of human body according to claim 3, characterized in that: In step B4, the segmentation model after training and learning is completed is subjected to model performance verification. When the model performance verification is qualified, the segmentation model is saved. When the segmentation model's ability to parse the human body at more granularity is expanded, the following steps are adopted to divide part of the data in the fusion data set into a verification set, and the model performance of the trained segmentation model is evaluated on the verification set, and the evaluation indicators are monitored. When the model performance verification is qualified, the segmentation model is saved, and the segmentation model's ability to parse the human body at more granularity is expanded.

6. A method for multi-granularity analysis of human body according to claim 3, characterized in that: In step B1, when collecting relevant data that need to be expanded in granularity and annotating them to obtain a new granularity data set, the following steps are adopted to collect pictures containing lesions of different skin diseases, and obtain a skin lesion data set after polygonal annotation of the skin lesion area; In step B2, when obtaining an existing data set of the segmentation model, annotating and fusing the existing data set with the new granularity data set to obtain a fused data set with new granularity features, the following steps are adopted to obtain an existing data set of the segmentation model, annotating and fusing the existing data set with the skin lesion data set to obtain a fused data set with skin lesion features.

7. A method for multi-granularity analysis of human body according to claim 6, characterized in that: In the step S3, the following steps are also included: obtaining the coordinate information of the human body parts and the skin lesion areas to perform auxiliary analysis of the severity of the disease, and outputting the severity level of the skin lesions by calculating the skin lesion ratio.

8. A method for multi-granularity analysis of human body according to claim 6, characterized in that: In the step S3, the following steps are also included: obtaining the features of the human body parts, skin areas and skin lesion areas for disease-assisted diagnosis analysis, and outputting the probabilities of several skin diseases in the skin lesion areas through the feature combination learned by the model.

9. A method for multi-granularity analysis of human body according to claim 6, characterized in that: In the step S3, the following steps are also included: obtaining the features of the human body parts, skin areas and skin lesion areas for auxiliary disease diagnosis analysis, outputting the probabilities of several skin diseases in the skin lesion area through the feature combination learned by the model, obtaining the coordinate information of the human body parts and skin lesion areas for auxiliary analysis of the severity of the disease, and outputting the severity level of the skin lesions by calculating the skin lesion ratio.

10. An intelligent system for multi-granularity analysis of human body, using a method for multi-granularity analysis of human body as claimed in any one of claims 1 to 9, characterized in that: The intelligent system includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the instruction, program, code set or instruction set is loaded and executed by the processor according to a method for multi-granularity analysis of the human body as described in any one of claims 1 to 9.

Images