Acupuncture point positioning method and system based on image enhancement

Abstract

The invention discloses an acupoint positioning method and system based on image enhancement, and belongs to the technical field of image processing, and the method comprises the steps of human body acupoint image collection, human body acupoint image enhancement, acupoint recognition model construction and acupoint positioning. According to the scheme, through parameter optimization of the gamma value and the fractional order, according to the difference between the oscillation coefficient and the individual position, random disturbance is introduced for updating, the parameter combination is continuously optimized in combination with the evolution rate, and gamma transformation and fractional order enhancement are performed by using the optimal parameter combination, so that the image quality is improved; the activation and recombination unit performs intelligent screening on features through channel gating activation and then performs splitting and recombination on a feature map to obtain a composite feature map, and the bitype convolution optimization unit obtains a core convolution feature map and a detail convolution feature map by splitting the composite feature map and combining group convolution with point-by-point convolution. And weighting is carried out through global average pooling and a soft attention mechanism, so that the accuracy and the stability of acupoint recognition are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and specifically refers to an acupoint positioning method and system based on image enhancement. Background Art

[0002] Acupoint positioning method is a method that uses image processing and deep learning technology to analyze and locate images of human acupoints, extract feature information from human acupoint images, automatically identify and calibrate acupoints, reduce manual intervention, and improve positioning accuracy and automation level. However, the existing acupoint positioning methods have the problems of poor quality of human acupoint images, unclear display of acupoint-related features, and difficulty in accurately identifying and analyzing acupoint features, resulting in a high misidentification rate of acupoint positioning; the existing acupoint positioning methods have the problem that human acupoint image features have multi-scale and complex local details, and features of different scales are often expressed in different ways in the image. It is difficult for existing methods to accurately extract details and make effective feature fusion at different scales, resulting in difficulty in accurately locating acupoint positions. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an acupoint positioning method and system based on image enhancement. In view of the problems in the existing acupoint positioning methods, such as poor quality of human acupoint images, unclear display of acupoint-related features, difficulty in accurately identifying and analyzing acupoint features, and a high misrecognition rate of acupoint positioning, this scheme performs parameter optimization on gamma value and fractional order, updates individual positions according to the difference between oscillation coefficient and individual positions, introduces random disturbance, continuously updates the number according to the evolution rate, finds the optimal parameter combination, and uses gamma function and fractional-order differential algorithm to perform gamma transformation and fractional-order enhancement on human acupoint images based on the optimal parameter combination to obtain human acupoint enhanced images, which more clearly display acupoint-related features, provide reliable input for subsequent acupoint positioning, and are beneficial to the identification and analysis of acupoint features, thereby improving the accuracy of acupoint positioning. In view of the problems in the existing acupoint positioning methods, The features of human acupoint images have multi-scale and complex local details, and features of different scales are often expressed in different ways in the image. It is difficult for existing methods to accurately extract details and make effective feature fusion at different scales, which leads to the problem of difficulty in accurately locating the acupoint positions. This solution designs intelligent feature convolution to replace the standard convolution in the backbone network. The activation reorganization unit intelligently screens the features through channel gating activation, and then splits and reorganizes the feature map to obtain a composite feature map. The dual-type convolution optimization unit splits the composite feature map and combines group convolution with point-by-point convolution to obtain the core convolution feature map and the detail convolution feature map, which are then weighted through global average pooling and soft attention mechanism to obtain a feature fusion map, which reduces redundant information and optimizes the fusion of multi-scale features, effectively extracts subtle and key acupoint features from complex human acupoint images, and significantly improves the accuracy and stability of acupoint recognition.

[0004] The technical solution adopted by the present invention is as follows: The acupoint positioning method based on image enhancement provided by the present invention comprises the following steps:

[0005] Step S1: human body acupuncture point image acquisition;

[0006] Step S2: human body acupoint image enhancement;

[0007] Step S3: constructing an acupoint recognition model;

[0008] Step S4: Acupoint positioning.

[0009] Furthermore, in step S1, the human acupoint image acquisition is to acquire human acupoint images that have been annotated with acupoints.

[0010] Further, in step S2, the human body acupoint image enhancement performs parameter optimization on the gamma value γ of the gamma function and the fractional order α of the fractional order differential algorithm to find the optimal parameter combination (γ, α), and based on the optimal parameter combination (γ, α), the gamma function and the fractional order differential algorithm are used to perform gamma transformation and fractional order enhancement on the human body acupoint image to obtain the human body acupoint enhanced image, and a training data set and a test data set are constructed based on the human body acupoint enhanced image; the parameter optimization specifically includes the following steps:

[0011] Step S21: initial individual position; establish a parameter search space for the gamma value γ of the gamma function and the fractional order α of the fractional differential algorithm, and pre-set the number of initial individual positions H 0 , the minimum number of individual positions H min , maximum search times T and fitness threshold f th , randomly initialize H in the parameter search space 0 Individual positions, each individual position represents a set of parameter combinations, and the information entropy, average gradient and variance weighted combination of the human acupoint enhanced image obtained based on the parameter combination is used as the fitness value of the corresponding individual position;

[0012] Step S22: Update individual positions; update each individual position according to the difference between the oscillation coefficient and the individual position, and introduce random disturbance; the formula used is as follows:

[0013] ;

[0014] ;

[0015] In the formula, and are the hth individual positions in the t+1th and tth searches, respectively, Q t is the oscillation coefficient at the t-th search, h is the individual position index, t is the search number index, is an individual position randomly selected during the t-th search, r 1 and r 2 are two random numbers that do not interfere with each other. sin(·) and round(·) are the sine function and rounding function respectively. randn is a random number that follows a normal distribution.

[0016] Step S23: Search times detection; if the search times , then go to step S24; otherwise, go to step S25;

[0017] Step S24: Update quantity; including the following steps:

[0018] Step S241: Calculate the evolution rate; the formula used is as follows:

[0019] ;

[0020] Where, L t+1 is the evolution rate at the t+1th search, and are the global optimal positions at the oth and o-1th searches, respectively. and They are and The fitness value of , o is the search number index, is the round-up symbol, is a smoothing term;

[0021] Step S242: Evolution rate detection; if the evolution rate L t+1 Less than or equal to 10 -6 , then go to step S243; otherwise, go to step S25;

[0022] Step S243: Update the number of individual positions; calculate the new number of individual positions H based on the fitness value of the global optimal position t+1 , if H t+1 Less than or equal to H t , then all individual positions are arranged in ascending order according to the fitness value, and the first H t+1 Individual positions, discard the remaining individual positions; otherwise, keep the original H t individual positions and initialize them randomly again in the parameter search space individual positions; the formula used is as follows:

[0023] ;

[0024] In the formula, H t , H t+1 and H 0.5T are the number of individual positions during the t-th, t+1-th and 0.5T-th searches, respectively. is the global optimal position at the t+1th search, yes The fitness value of It is the minimum fitness value since the t+1 search;

[0025] Step S25: Determine the optimal parameter combination; update the fitness value of the individual position, and when the fitness value of the global optimal position is less than the fitness threshold f th When , the parameter combination represented by the global optimal position is the optimal parameter combination (γ, α), and the parameter optimization is completed; otherwise, if the maximum number of searches is reached, return to step S21 to re-initialize the individual positions; otherwise, the number of searches is increased by 1 and return to step S22 to continue searching.

[0026] Further, in step S3, the acupoint recognition model is constructed using the YOLOv8 network architecture to complete the construction of the acupoint recognition model; the YOLOv8 network architecture includes a backbone network, a neck network and a detection head; the backbone network is responsible for extracting feature maps of different scales from the human acupoint enhanced image of the training data set through convolution operations; the neck network is responsible for further fusing and processing the multi-scale feature maps extracted by the backbone network; the detection head is responsible for acupoint recognition according to the feature maps output by the neck network; in the construction of the acupoint recognition model, a smart feature convolution is designed to replace the standard convolution in the backbone network, and the smart feature convolution is composed of an activation recombination unit and a dual-type convolution optimization unit, and specifically includes the following steps:

[0027] Step S31: activating the recombination unit; comprising the following steps:

[0028] Step S311: group normalization: performing group normalization on the input feature map I;

[0029] Step S312: Calculate weights; normalize the scaling factor of each channel to obtain the weight corresponding to each channel;

[0030] Step S313: channel gating activation; the weighted feature map is mapped to the interval (0, 1) through the Sigmoid function, and the threshold W is pre-set th Perform gate control, for the th The channel of , set its weight to 1 to obtain the activation weight W 1 , for values ​​less than the threshold W th The channel of the , set its weight to 0 to obtain the suppression weight W 2 ; The formula used is as follows:

[0031] ;

[0032] Where Gate (·) and Sigmoid (·) are the gate function and activation function respectively, W a is the weight, a is the index, W 1 and W 2 are activation weight and inhibition weight, W 1 =1,W 2 =0, is a set of weights, It is the feature map after group normalization;

[0033] Step S314: Split the input feature map; multiply I by W 1 and W 2 , thereby splitting I into feature map activation components and feature map suppression component ;

[0034] Step S315: Split and reassemble; for acupoint recognition, it is possible to extract diverse features from different channels and details, and improve the interactivity between different features through splitting and reassembling, thereby capturing the complex morphological information of the acupoints; and They are evenly split along the channels, from Evenly split into two sub-feature maps of equal size and ,from Evenly split into two sub-feature maps of equal size and , cross-combining the evenly split sub-feature maps to obtain two new feature maps I ω1 and I ω2 , and then I ω1 and I ω2 Connect and get the composite feature graph I ω ;

[0035] Step S32: dual-type convolution optimization unit; comprising the following steps:

[0036] Step S321: Split the composite feature graph; ω according to and The channels are split to obtain the core feature map I upper And detail feature map I below , and then apply 1×1 convolution to compress I upper and I below The number of channels is , so as to obtain the core feature map after convolution compression. and detail feature map ;in, is the control parameter, U is the number of channels in I;

[0037] Step S322: dual-type convolution combination; for , first perform group convolution separately And point-wise convolution , and then and Add together to get the core convolution feature map O 1 ;for , first perform point-by-point convolution , and then and Connect to get the detail convolution feature map O 2 ;

[0038] Step S323: feature reconstruction and optimization; using global average pooling to process O 1 and O2 , and obtain the core global feature B 1 and detail global features B 2 , and then generate the core weight factor through the soft attention mechanism and detail weight factor , then use and For O 1 and O 2 Perform weighted combination to obtain the feature fusion graph O.

[0039] Furthermore, in step S4, the acupoint positioning is to collect images of human acupoints to be positioned and perform image enhancement to obtain enhanced images of human acupoints to be positioned, and then input the enhanced images of human acupoints to be positioned into the acupoint recognition model for processing, identify the areas where the acupoints are located, and mark them, output the marked human acupoint images, and complete the acupoint positioning.

[0040] The acupoint positioning system based on image enhancement provided by the present invention comprises a human acupoint image acquisition module, a human acupoint image enhancement module, an acupoint recognition model building module and an acupoint positioning module;

[0041] The human body acupoint image acquisition module acquires human body acupoint images that have been annotated with acupoints, and sends the data to the human body acupoint image enhancement module;

[0042] The human acupuncture point image enhancement module performs parameter optimization on the gamma value and the fractional order, updates the individual position according to the difference between the oscillation coefficient and the individual position, introduces random disturbance, continuously updates the number according to the evolution rate, finds the optimal parameter combination, and uses the gamma function and the fractional order differential algorithm to perform gamma transformation and fractional order enhancement on the human acupuncture point image based on the optimal parameter combination to obtain the human acupuncture point enhanced image, and sends the data to the acupuncture point recognition model construction module;

[0043] The acupoint recognition model construction module designs intelligent feature convolution to replace the standard convolution in the backbone network. The activation reorganization unit intelligently screens the features through channel gating activation, and then splits and reorganizes the feature map to obtain a composite feature map. The dual-type convolution optimization unit splits the composite feature map and combines the group convolution and the point-by-point convolution to obtain a core convolution feature map and a detail convolution feature map, and then weights them through global average pooling and soft attention mechanism to obtain a feature fusion map, and sends the data to the acupoint positioning module.

[0044] The acupoint positioning module collects the human acupoint image to be located and performs image enhancement, and then inputs it into the acupoint recognition model for processing, identifies the area where the acupoint is located, and marks it, and outputs the marked human acupoint image.

[0045] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0046] (1) In view of the problems in existing acupoint positioning methods, such as poor quality of human acupoint images, unclear display of acupoint-related features, difficulty in accurately identifying and analyzing acupoint features, and high misrecognition rate of acupoint positioning, this scheme optimizes the gamma value and fractional order parameters. According to the difference between the oscillation coefficient and the individual position, random perturbations are introduced to update the individual positions. The number is continuously updated according to the evolution rate to find the optimal parameter combination. Based on the optimal parameter combination, the gamma function and fractional order differential algorithm are used to perform gamma transformation and fractional order enhancement on the human acupoint images to obtain enhanced images of human acupoints. Then, training data sets and test data sets are constructed, which improves the quality of human acupoint images, displays acupoint-related features more clearly, and provides reliable input for subsequent acupoint positioning, which is conducive to the recognition and analysis of acupoint features, reduces the misrecognition rate, and thus improves the accuracy of acupoint positioning.

[0047] (2) In view of the existing acupoint positioning methods, the features of human acupoint images have multi-scale and complex local details, and features of different scales are often expressed in different ways in the image. It is difficult for existing methods to accurately extract details and make effective feature fusion at different scales, which leads to the problem of difficulty in accurately locating the acupoint position. This scheme designs intelligent feature convolution to replace the standard convolution in the backbone network. The activation reorganization unit intelligently screens the features through channel gating activation, and then splits and reorganizes the feature map to obtain a composite feature map. The dual-type convolution optimization unit splits the composite feature map and combines group convolution and point-by-point convolution to obtain the core convolution feature map and the detail convolution feature map, which are then weighted through global average pooling and soft attention mechanism to obtain the feature fusion map, which realizes the effective extraction of details and overall information, thereby improving the perception ability of complex acupoint morphology, reducing redundant information and optimizing the fusion of multi-scale features, ensuring the accurate capture of features of different scales, and effectively extracting subtle and key acupoint features from complex human acupoint images, significantly improving the accuracy and stability of acupoint recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A schematic diagram of the flow of the acupoint positioning method based on image enhancement provided by the present invention;

[0049] Figure 2 A schematic diagram of an acupoint positioning system based on image enhancement provided by the present invention;

[0050] Figure 3 is a schematic flow chart of step S2;

[0051] Figure 4 It is a schematic diagram of the process of step S3.

[0052] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0054] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0055] Example 1, see Figure 1 The present invention provides an acupoint positioning method based on image enhancement, which comprises the following steps:

[0056] Step S1: collecting human acupoint images; collecting human acupoint images marked with acupoints;

[0057] Step S2: human acupuncture point image enhancement; performing parameter optimization on the gamma value and the fractional order, updating the individual position according to the difference between the oscillation coefficient and the individual position, and introducing random disturbance, continuously updating the number according to the evolution rate, finding the optimal parameter combination, and performing gamma transformation and fractional order enhancement on the human acupuncture point image based on the optimal parameter combination using the gamma function and the fractional order differential algorithm to obtain the human acupuncture point enhanced image;

[0058] Step S3: Construct an acupoint recognition model; design intelligent feature convolution to replace the standard convolution in the backbone network, activate the reorganization unit to intelligently screen the features through channel gating activation, and then split and reorganize the feature map to obtain a composite feature map. The dual-type convolution optimization unit splits the composite feature map and combines the group convolution and point-by-point convolution to obtain the core convolution feature map and the detail convolution feature map, and then weights them through global average pooling and soft attention mechanism to obtain a feature fusion map;

[0059] Step S4: acupoint positioning; collect the human acupoint image to be positioned and perform image enhancement, then input it into the acupoint recognition model for processing, identify the area where the acupoint is located, mark it, and output the marked human acupoint image.

[0060] Example 2, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S2, the human acupoint image is enhanced; the quality of the human acupoint image used for acupoint positioning research is improved, and the enhanced human acupoint image can more clearly display the acupoint-related features, provide better quality human acupoint image data for acupoint positioning, and make it more conducive to the recognition and analysis of acupoint features; the gamma value γ of the gamma function and the fractional order α of the fractional order differential algorithm are optimized to find the optimal parameter combination (γ, α), and the gamma function and the fractional order differential algorithm are used based on the optimal parameter combination (γ, α) to perform gamma transformation and fractional order enhancement on the human acupoint image, optimize the image contrast, brightness and texture details, thereby improving the image quality, obtain the human acupoint enhanced image, and construct a training data set and a test data set based on the human acupoint enhanced image; the parameter optimization specifically includes the following steps:

[0061] Step S21: initial individual position; determine the comprehensive index for measuring the quality of human acupoint enhanced image, so as to evaluate the effect of image enhancement under different parameter combinations, so as to find the optimal parameter combination for enhancing the image. High-quality images are helpful to more accurately identify acupoint features; establish a parameter search space for the gamma value γ of the gamma function and the fractional order α of the fractional differential algorithm, and pre-set the number of initial individual positions H 0 , the minimum number of individual positions H min , maximum search times T and fitness threshold f th , randomly initialize H in the parameter search space 0 Individual positions are used, and each individual position represents a set of parameter combinations. The information entropy, average gradient, and variance weighted combination of the human acupoint enhanced image obtained based on the parameter combination is used as the fitness value of the corresponding individual position. The formula used is as follows:

[0062] ;

[0063] ;

[0064] ;

[0065] ;

[0066] Where f is the fitness value, S, R and D are information entropy, average gradient and variance respectively, i is the gray level index, p iis the probability density function of the i-th gray level, M and N are the sizes of the human acupoint image in the vertical and horizontal directions, m and n are the position indexes in the vertical and horizontal directions, respectively. The image of acupuncture points on the human body The pixel value at The image of acupuncture points on the human body The gradient at , ε 1 , ε 2 and ε 3 They are the weight coefficients of information entropy, average gradient and variance respectively;

[0067] Step S22: Update individual positions; search for the optimal parameter combination more effectively in the parameter search space, while increasing the diversity of the search to avoid falling into the local optimum, and ensure that the image enhancement parameters most suitable for acupoint positioning are found; update each individual position according to the difference between the oscillation coefficient and the individual position, and introduce random perturbations to enhance the diversity and exploration of the search process; the formula used is as follows:

[0068] ;

[0069] ;

[0070] In the formula, and are the hth individual positions in the t+1th and tth searches, respectively, Q t is the oscillation coefficient at the t-th search, h is the individual position index, t is the search number index, is an individual position randomly selected during the t-th search, r 1 and r 2 are two random numbers in the range (0, 1) that do not interfere with each other. sin(·) and round(·) are the sine function and rounding function respectively. randn is a random number that follows a normal distribution.

[0071] Step S23: Search times detection; This avoids the waste of resources caused by unlimited search, and also prevents premature stopping of the search and missing the optimal parameter combination, thereby improving the efficiency and accuracy of the optimization of the entire human acupoint image enhancement parameter; if the search times are , then go to step S24; otherwise, go to step S25;

[0072] Step S24: updating the number; dynamically adjusting the number of individual positions according to the actual search situation can more effectively utilize computing resources and improve the probability of finding the optimal parameter combination, thereby providing more suitable image enhancement parameters for acupoint positioning; comprising the following steps:

[0073] Step S241: Calculate the evolution rate; quantify the degree of evolution in the search process, provide a basis for judging whether the search is stagnant, and thus determine whether the number of individual positions needs to be adjusted in the process of optimizing the image enhancement parameters; the formula used is as follows:

[0074] ;

[0075] Where, L t+1 is the evolution rate at the t+1th search, and are the global optimal positions at the oth and o-1th searches respectively. The global optimal position is the individual position with the smallest fitness value. and They are and The fitness value of , o is the search number index, is the round-up symbol, is the smoothing term, Used to avoid the denominator being 0;

[0076] Step S242: Evolution rate detection; if the evolution rate L t+1 Less than or equal to 10 -6 , indicating that the search process has come to a standstill, then go to step S243; otherwise, go to step S25;

[0077] Step S243: Update the number of individual positions; adjust the number of individual positions according to the search status to optimize the search process and increase the probability of finding the optimal parameter combination suitable for acupoint positioning image enhancement; calculate the new number of individual positions H based on the fitness value of the global optimal position t+1 , if H t+1 Less than or equal to H t , then all individual positions are arranged in ascending order according to the fitness value, and the first H t+1 Individual positions are selected, and the remaining individual positions are discarded, so that the search is concentrated near the global optimal position, and a more refined local search is performed; otherwise, the original H is retained. t individual positions and initialize them randomly again in the parameter search space Individual positions are added to expand the search range and increase the chance of finding a better solution; the formula used is as follows:

[0078] ;

[0079] In the formula, H t , H t+1 and H 0.5T are the number of individual positions during the t-th, t+1-th and 0.5T-th searches, respectively. is the global optimal position at the t+1th search, yes The fitness value of It is the minimum fitness value since the t+1 search;

[0080] Step S25: Determine the optimal parameter combination; update the fitness value of the individual position, and when the fitness value of the global optimal position is less than the fitness threshold f th When , the parameter combination represented by the global optimal position is the optimal parameter combination (γ, α), and the parameter optimization is completed; otherwise, if the maximum number of searches is reached, return to step S21 to re-initialize the individual positions; otherwise, the number of searches is increased by 1 and return to step S22 to continue searching.

[0081] By performing the above operations, in order to solve the problems in the existing acupoint positioning methods, such as poor quality of human acupoint images, unclear display of acupoint-related features, difficulty in accurately identifying and analyzing acupoint features, and high misrecognition rate of acupoint positioning, this scheme performs parameter optimization on gamma value and fractional order, updates individual positions according to the difference between the oscillation coefficient and the individual position, introduces random disturbance, continuously updates the number according to the evolution rate, finds the optimal parameter combination, and uses the gamma function and fractional-order differential algorithm to perform gamma transformation and fractional-order enhancement on the human acupoint image based on the optimal parameter combination to obtain the human acupoint enhanced image, and then constructs the training data set and the test data set, thereby improving the quality of the human acupoint image, displaying acupoint-related features more clearly, providing reliable input for subsequent acupoint positioning, facilitating the recognition and analysis of acupoint features, reducing the misrecognition rate, and thus improving the accuracy of acupoint positioning.

[0082] Example 3, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S3, an acupoint recognition model is constructed, and the YOLOv8 network architecture is used to complete the construction of the acupoint recognition model; the YOLOv8 network architecture includes a backbone network, a neck network and a detection head; the backbone network is responsible for extracting feature maps of different scales from the human acupoint enhanced image of the training data set through convolution operations; the neck network is responsible for further fusing and processing the multi-scale feature maps extracted by the backbone network to enhance the expression ability of the features; the detection head is responsible for acupoint recognition according to the feature maps output by the neck network; in the construction of the acupoint recognition model, a smart feature convolution is designed to replace the standard convolution in the backbone network. The smart feature convolution is composed of an activation recombination unit and a dual-type convolution optimization unit, and specifically includes the following steps:

[0083] Step S31: activating the recombination unit; in acupoint recognition, the key points and detail information of the human body are usually more complex, requiring more accurate feature extraction, activating the recombination unit helps eliminate data noise, while retaining important features, effectively improving the recognition ability of features of different scales and importance, avoiding some irrelevant or interfering features from affecting the recognition accuracy, and for acupoint recognition, focusing on the key point positions in the human feature map, improving the accuracy of acupoint recognition; comprising the following steps:

[0084] Step S311: group normalization: perform group normalization on the input feature map I; the formula used is as follows:

[0085] ;

[0086] Where μ and σ are the mean and standard deviation of I, respectively, φ and β are the scaling factor and offset factor, respectively. is the feature map after group normalization. is the smoothing term, Used to avoid the denominator being 0;

[0087] Step S312: Calculate the weight; normalize the scaling factor of each channel to obtain the weight corresponding to each channel; the formula used is as follows:

[0088] ;

[0089] Where U is the number of channels in I, ω u is the weight corresponding to the u-th channel, u and v are the channel indices, φ u and φ v are the scaling factors for the uth and vth channels respectively, is a set of weights;

[0090] Step S313: channel gating activation; for acupoint recognition, structures such as joints, muscles or bones of the human body often contain differentiated feature information. Through channel gating activation, features with high information content can be intelligently retained while suppressing irrelevant or redundant features, thereby improving the stability and accuracy of acupoint recognition; the weighted feature map is mapped to the interval (0, 1) through the Sigmoid function, and the threshold W is pre-set th Perform gate control, for the th The channel of , set its weight to 1 to obtain the activation weight W 1 , for values ​​less than the threshold W th The channel of the , set its weight to 0 to obtain the suppression weight W 2 ; The formula used is as follows:

[0091] ;

[0092] Where Gate (·) and Sigmoid (·) are the gate function and activation function respectively, W a is the weight, a is the index, W 1 and W 2 are activation weight and inhibition weight, W 1 =1,W 2 =0;

[0093] Step S314: Split the input feature map; multiply I by W 1 and W 2 , thereby splitting I into feature map activation components and feature map suppression component ; The formula used is as follows:

[0094] ;

[0095] ;

[0096] In the formula, is the element-wise multiplication operator;

[0097] Step S315: Split and reassemble; for acupoint recognition, it is possible to extract diverse features from different channels and details, and improve the interactivity between different features through splitting and reassembling, thereby capturing the complex morphological information of the acupoints; and They are evenly split along the channels, from Evenly split into two sub-feature maps of equal size and ,from Evenly split into two sub-feature maps of equal size and , cross-combining the evenly split sub-feature maps to obtain two new feature maps I ω1 and I ω2 , and then I ω1 and I ω2 Connect and get the composite feature graph I ω ; The formula used is as follows:

[0098] ;

[0099] ;

[0100] ;

[0101] ;

[0102] ;

[0103] In the formula, is a uniform splitting operation along the channel, is the element summation operator, ∪ is the concatenation operator;

[0104] Step S32: dual-type convolution optimization unit; combined with dual-type convolution, the overall information and subtle changes can be fully extracted to optimize the recognition effect of acupoints. When processing complex human acupoint images, learning the detail level can effectively improve the accuracy of acupoint recognition; including the following steps:

[0105] Step S321: Split the composite feature graph; ω according to and The channels are split to obtain the core feature map I upper And detail feature map I below , and then apply 1×1 convolution to compress I upper and I below The number of channels is reduced to reduce channel redundancy, thereby obtaining the core feature map after convolution compression. and detail feature map ;in, is the control parameter, ;

[0106] Step S322: dual-type convolution combination; acupoint recognition is a task that requires very high position accuracy. Through this dual-type convolution optimization, the error can be minimized when processing features of different scales to ensure accurate recognition and positioning of acupoints; , first perform group convolution separately And point-wise convolution , and then and Add together to get the core convolution feature map O 1 ;for , first perform point-by-point convolution , and then and Connect to get the detail convolution feature map O 2 ; The formula used is as follows:

[0107] ;

[0108] ;

[0109] Among them, GWC (·) and PWC (·) are group convolution and point-wise convolution, respectively;

[0110] Step S323: feature reconstruction and optimization; the importance of each feature map can be automatically adjusted, and it is helpful to focus on important acupoint features according to actual needs, so that intelligent adjustments can be made according to the characteristics of the input image in different situations to avoid over-reliance on a single feature, which not only helps to improve the perception of complex acupoint features, but also reduces the interference of irrelevant features through a weighted mechanism, providing more robust acupoint recognition performance; global average pooling is used to process O 1 and O 2 , get the core global feature B 1 and detail global features B 2 , and then generate the core weight factor through the soft attention mechanism and detail weight factor , then use and For O 1 and O 2 Perform weighted combination to obtain the feature fusion graph O; the formula used is as follows:

[0111] ;

[0112] ;

[0113] ;

[0114] Among them, AvgPooling (·) is the global average pooling function, O g , B g and They are convolutional feature map, global feature and weight factor respectively, and g is the index.

[0115] By performing the above operations, the existing acupoint positioning methods have the problem that the human acupoint image features have multi-scale and complex local details, and the features of different scales are often expressed in different ways in the image. It is difficult for the existing methods to accurately extract the details and make effective feature fusion at different scales, which leads to the problem that it is difficult to accurately locate the acupoint position. This solution designs intelligent feature convolution to replace the standard convolution in the backbone network. The activation reorganization unit intelligently screens the features through channel gating activation, and then splits and reorganizes the feature map to obtain a composite feature map. The dual-type convolution optimization unit splits the composite feature map and combines the group convolution and point-by-point convolution to obtain the core convolution feature map and the detail convolution feature map, and then weights them through global average pooling and soft attention mechanism to obtain the feature fusion map, which realizes the effective extraction of details and overall information, thereby improving the perception of complex acupoint morphology, reducing redundant information and optimizing the fusion of multi-scale features, ensuring the accurate capture of features of different scales, and effectively extracting subtle and key acupoint features from complex human acupoint images, significantly improving the accuracy and stability of acupoint recognition.

[0116] Example 4, see Figure 1 This embodiment is based on the above embodiment. In step S4, acupoint positioning is to collect images of human acupoints to be positioned and perform image enhancement to obtain enhanced images of human acupoints to be positioned, and then input the enhanced images of human acupoints to be positioned into the acupoint recognition model for processing, identify the areas where the acupoints are located, and mark them, output the marked human acupoint images, and complete acupoint positioning.

[0117] Example 5, see Figure 2 , this embodiment is based on the above embodiment, and the acupoint positioning system based on image enhancement provided by the present invention includes a human acupoint image acquisition module, a human acupoint image enhancement module, an acupoint recognition model building module and an acupoint positioning module;

[0118] The human body acupoint image acquisition module acquires human body acupoint images that have been annotated with acupoints, and sends the data to the human body acupoint image enhancement module;

[0119] The human acupuncture point image enhancement module performs parameter optimization on the gamma value and the fractional order, updates the individual position according to the difference between the oscillation coefficient and the individual position, introduces random disturbance, continuously updates the number according to the evolution rate, finds the optimal parameter combination, and uses the gamma function and the fractional order differential algorithm to perform gamma transformation and fractional order enhancement on the human acupuncture point image based on the optimal parameter combination to obtain the human acupuncture point enhanced image, and sends the data to the acupuncture point recognition model construction module;

[0120] The acupoint recognition model construction module designs intelligent feature convolution to replace the standard convolution in the backbone network. The activation reorganization unit intelligently screens the features through channel gating activation, and then splits and reorganizes the feature map to obtain a composite feature map. The dual-type convolution optimization unit splits the composite feature map and combines the group convolution and the point-by-point convolution to obtain a core convolution feature map and a detail convolution feature map, and then weights them through global average pooling and soft attention mechanism to obtain a feature fusion map, and sends the data to the acupoint positioning module.

[0121] The acupoint positioning module collects the human acupoint image to be located and performs image enhancement, and then inputs it into the acupoint recognition model for processing, identifies the area where the acupoint is located, and marks it, and outputs the marked human acupoint image.

[0122] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0123] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0124] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. Acupoint positioning method based on image enhancement, characterized in that: The method comprises the following steps: Step S1: human acupoint image acquisition: acquiring human acupoint images with acupoint annotations; Step S2: human body acupuncture point image enhancement; Step S3: constructing an acupoint recognition model; Step S4: acupoint positioning; collecting the human acupoint image to be positioned and performing image enhancement, then inputting it into the acupoint recognition model for processing, identifying the area where the acupoint is located, and marking it, and outputting the marked human acupoint image; Step S2 includes step S22: updating individual positions; updating each individual position according to the difference between the oscillation coefficient and the individual position, and introducing random disturbance; the formula used is as follows: ； ； In the formula, and are the hth individual positions in the t+1th and tth searches, respectively, Q t is the oscillation coefficient at the t-th search, h is the individual position index, t is the search number index, T is the maximum search number, is an individual position randomly selected during the t-th search, r1 and r2 are two random numbers that do not interfere with each other, sin(·) and round(·) are the sine function and rounding function respectively, and randn is a random number that obeys the normal distribution.

2. The acupoint localization method based on image enhancement according to claim 1, characterized in that: In step S2, the human acupoint image enhancement performs parameter optimization on the gamma value γ of the gamma function and the fractional order α of the fractional order differential algorithm to find the optimal parameter combination (γ, α), and based on the optimal parameter combination (γ, α), the gamma function and the fractional order differential algorithm are used to perform gamma transformation and fractional order enhancement on the human acupoint image to obtain the human acupoint enhanced image, and a training data set and a test data set are constructed based on the human acupoint enhanced image; the parameter optimization specifically includes the following steps: Step S21: initial individual position; establish a parameter search space for the gamma value γ of the gamma function and the fractional order α of the fractional differential algorithm, and pre-set the initial individual position number H0, the minimum individual position number H min , maximum search times T and fitness threshold f th , randomly initialize H in the parameter search space 0 Individual positions, each individual position represents a set of parameter combinations, and the information entropy, average gradient and variance weighted combination of the human acupoint enhanced image obtained based on the parameter combination is used as the fitness value of the corresponding individual position; Step S22: Update individual position; Step S23: Search times detection; if the search times , then go to step S24; otherwise, go to step S25; where t is the search number index; Step S24: Update quantity; Step S25: Determine the optimal parameter combination; update the fitness value of the individual position, and when the fitness value of the global optimal position is less than the fitness threshold f th When , the parameter combination represented by the global optimal position is the optimal parameter combination (γ, α), and the parameter optimization is completed; otherwise, if the maximum number of searches is reached, return to step S21 to re-initialize the individual positions; otherwise, the number of searches is increased by 1 and return to step S22 to continue searching.

3. The acupoint localization method based on image enhancement according to claim 2, characterized in that: In step S24, the updating quantity specifically includes the following steps: Step S241: Calculate the evolution rate; the formula used is as follows: ； Where, L t+1 is the evolution rate at the t+1th search, and are the global optimal positions at the oth and o-1th searches, respectively. and They are and The fitness value of , o is the search number index, is the round-up symbol, is a smoothing term; Step S242: Evolution rate detection; if the evolution rate L t+1 Less than or equal to 10 -6 , then go to step S243; otherwise, go to step S25; Step S243: Update the number of individual positions; calculate the new number of individual positions H based on the fitness value of the global optimal position t +1 , if H t+1 Less than or equal to H t , then all individual positions are arranged in ascending order according to the fitness value, and the first H t+1 Individual positions, discard the remaining individual positions; otherwise, keep the original H t individual positions and initialize them randomly again in the parameter search space individual positions; the formula used is as follows: ； In the formula, H t , H t+1 and H 0.5T are the number of individual positions during the t-th, t+1-th and 0.5T-th searches, respectively. is the global optimal position at the t+1th search, yes The fitness value of It is the minimum fitness value since the t+1 search.

4. The acupoint locating method based on image enhancement according to claim 1, characterized in that: In step S3, the acupoint recognition model is constructed using the YOLOv8 network architecture to complete the construction of the acupoint recognition model; the YOLOv8 network architecture includes a backbone network, a neck network and a detection head; the backbone network is responsible for extracting feature maps of different scales from the human acupoint enhanced image of the training data set through convolution operations; the neck network is responsible for further fusing and processing the multi-scale feature maps extracted by the backbone network; the detection head is responsible for acupoint recognition according to the feature maps output by the neck network; in the construction of the acupoint recognition model, a smart feature convolution is designed to replace the standard convolution in the backbone network, and the smart feature convolution is composed of an activation recombination unit and a dual-type convolution optimization unit, and specifically includes the following steps: Step S31: activating the recombination unit; comprising the following steps: Step S311: group normalization: performing group normalization on the input feature map I; Step S312: Calculate weights; normalize the scaling factor of each channel to obtain the weight corresponding to each channel; Step S313: channel gating activation; the weighted feature map is mapped to the interval (0, 1) through the Sigmoid function, and the threshold W is pre-set th Perform gate control, for the th The channel is set to 1 to obtain the activation weight W1. th The channel of , set its weight to 0 to obtain the suppression weight W2; the formula used is as follows: ； Where Gate (·) and Sigmoid (·) are the gate function and activation function respectively, W a is the weight, a is the index, W1 and W2 are activation weight and inhibition weight respectively, W1=1, W2=0, is a set of weights, It is the feature map after group normalization; Step S314: Split the input feature map; multiply I by W1 and W2 respectively, so as to split I into feature map activation components and feature map suppression component ; Step S315: Split and reassemble; for acupoint recognition, it is possible to extract diverse features from different channels and details, and improve the interactivity between different features through splitting and reassembling, thereby capturing the complex morphological information of the acupoints; and They are evenly split along the channels, from Evenly split into two sub-feature maps of equal size and ,from Evenly split into two sub-feature maps of equal size and , cross-combining the evenly split sub-feature maps to obtain two new feature maps I ω1 and I ω2 , and then I ω1 and I ω2 Connect and get the composite feature graph I ω ; Step S32: Dual-type convolution optimization unit.

5. The acupoint positioning method based on image enhancement according to claim 4, characterized in that: In step S32, the dual-type convolution optimization unit specifically includes the following steps: Step S321: Split the composite feature graph; ω according to and The channels are split to obtain the core feature map I upper And detail feature map I below , and then apply 1×1 convolution to compress I upper and I below The number of channels is , so as to obtain the core feature map after convolution compression. and detail feature map ;in, is the control parameter, U is the number of channels in I; Step S322: dual-type convolution combination; for , first perform group convolution separately And point-wise convolution , and then and Add together to get the core convolution feature map O1; for , first perform point-by-point convolution , and then and Connect to obtain the detail convolution feature map O2; Step S323: Feature reconstruction and optimization; use global average pooling to process O1 and O2 respectively to obtain the core global feature B1 and the detail global feature B2, and then generate the core weight factor through the soft attention mechanism and detail weight factor , then use and Perform weighted combination on O1 and O2 respectively to obtain the feature fusion graph O.

6. The acupoint locating method based on image enhancement according to claim 1, characterized in that: In step S4, the acupoint positioning is to collect the image of the human acupoint to be located and perform image enhancement to obtain the enhanced image of the human acupoint to be located, and then input the enhanced image of the human acupoint to be located into the acupoint recognition model for processing, identify the area where the acupoint is located, and mark it, output the marked human acupoint image, and complete the acupoint positioning.

7. An acupoint locating system based on image enhancement, used to implement the acupoint locating method based on image enhancement as claimed in any one of claims 1 to 6, characterized in that: It includes a human acupoint image acquisition module, a human acupoint image enhancement module, an acupoint recognition model building module and an acupoint positioning module; The human body acupoint image acquisition module acquires human body acupoint images that have been annotated with acupoints, and sends the data to the human body acupoint image enhancement module; The human acupuncture point image enhancement module performs parameter optimization on the gamma value and the fractional order, updates the individual position according to the difference between the oscillation coefficient and the individual position, introduces random disturbance, continuously updates the number according to the evolution rate, finds the optimal parameter combination, and uses the gamma function and the fractional order differential algorithm to perform gamma transformation and fractional order enhancement on the human acupuncture point image based on the optimal parameter combination to obtain the human acupuncture point enhanced image, and sends the data to the acupuncture point recognition model construction module; The acupoint recognition model construction module designs intelligent feature convolution to replace the standard convolution in the backbone network. The activation reorganization unit intelligently screens the features through channel gating activation, and then splits and reorganizes the feature map to obtain a composite feature map. The dual-type convolution optimization unit splits the composite feature map and combines the group convolution and the point-by-point convolution to obtain a core convolution feature map and a detail convolution feature map, and then weights them through global average pooling and soft attention mechanism to obtain a feature fusion map, and sends the data to the acupoint positioning module. The acupoint positioning module collects the human acupoint image to be located and performs image enhancement, and then inputs it into the acupoint recognition model for processing, identifies the area where the acupoint is located, and marks it, and outputs the marked human acupoint image.

Images