Hand acupoint detection method and system based on regional correlation coefficient learning attention

Abstract

The application relates to the technical field of image processing, in particular to a hand acupoint detection method and system based on regional correlation coefficient learning attention, which comprises the following steps: acquiring a hand image, performing feature extraction on the hand image to obtain a first feature map; and processing the first feature map by using a regional correlation coefficient learning attention network to obtain a third feature map; wherein the regional correlation coefficient learning attention network comprises a self-adaptive block perception attention module and a self-adaptive correlation coefficient channel attention module, the self-adaptive block perception attention module is used for detecting fine-grained acupoints in the first feature map to obtain a second feature map; the self-adaptive correlation coefficient channel attention module is used for modeling acupoint features of different skin colors in the second feature map to obtain the third feature map; and the labels and coordinate points of the hand acupoints are predicted based on the third feature map; the application can detect fine-grained acupoints and locate acupoints under different skin colors, thereby improving the effectiveness of hand acupoint detection.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a hand acupoint detection method and system based on regional correlation coefficient learning attention. Background Art

[0002] Chinese acupuncture is a unique form of healing in Traditional Chinese Medicine that stimulates acupuncture points to regulate the flow of qi (energy) and blood to treat a variety of conditions, such as headaches, insomnia, and facial paralysis.

[0003] Traditional Chinese acupuncture (CA) plays an important role in disease treatment, prevention, and healthcare. However, due to its complexity, detecting acupoints on the hand from the body surface is difficult. This is primarily due to: 1) the acupoint area closely resembles surrounding areas; and 2) the skin tone and depth of the hand vary. With the advancement of computer hardware technology, deep neural networks have been widely applied in various fields. Due to their powerful feature recognition capabilities, deep neural networks have also found application in Chinese acupuncture.

[0004] Therefore, it is necessary to design computer-assisted tools to train acupuncturists or assist doctors in performing acupuncture, and combine deep neural networks to solve the above technical problems, which can detect fine-grained acupoints and locate acupoints under different skin colors, thereby improving the effectiveness of hand acupoint detection. Summary of the Invention

[0005] The purpose of the present invention is to provide a hand acupoint detection method and system based on regional correlation coefficient learning attention, which can detect fine-grained acupoints and locate acupoints under different skin colors, thereby improving the effectiveness of hand acupoint detection.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a method for detecting hand acupoints based on regional correlation coefficient learning attention, the method comprising the following steps:

[0008] Acquire a hand image, perform feature extraction on the hand image, and obtain a first feature map;

[0009] The first feature map is processed using a regional correlation coefficient learning attention network to obtain a third feature map; wherein the regional correlation coefficient learning attention network includes an adaptive block-aware attention module and an adaptive correlation coefficient channel attention module, and the adaptive block-aware attention module is used to detect fine-grained acupoints in the first feature map to obtain a second feature map; and the adaptive correlation coefficient channel attention module is used to model acupoint features of different skin colors in the second feature map to obtain a third feature map;

[0010] The labels and coordinates of the hand acupuncture points are predicted based on the third feature map.

[0011] Preferably, the step of extracting features from the hand image to obtain a first feature map includes:

[0012] A first feature map is extracted from the hand image using head convolution.

[0013] Preferably, detecting fine-grained acupuncture points in the first feature map by the adaptive block-aware attention module to obtain a second feature map comprises:

[0014] Dividing the first feature map into multiple patches evenly, and connecting the multiple patches to generate a patch feature map;

[0015] The patch feature map is processed by using an average standard and then input into a patch attention module, and an attention weight is output;

[0016] The attention weights are processed by adaptive masking to obtain the second feature map.

[0017] Preferably, the expression of the attention weight is:

[0018] K=Softmax{MLP{Norm[MLP(J)]}};

[0019] Among them, K represents the attention weight, Softmax{·} represents the Softmax activation function processing, MLP{·} represents the multi-layer perceptron processing, and Norm{·} represents the normalization processing.

[0020] Preferably, the processing of the attention weights by the adaptive mask to obtain the second feature map includes:

[0021] After normalizing the attention weights, they are multiplied by the learnable parameters to obtain the mask map.

[0022] The attention weight is multiplied by the mask map to obtain an attention feature map, which is reshaped into the second feature map.

[0023] Preferably, the step of modeling the acupoint features of different skin colors in the second feature map by the adaptive correlation coefficient channel attention module to obtain a third feature map comprises:

[0024] Inputting the second feature map into a channel-related selector to generate a channel selection mask and an inverse mask;

[0025] Multiply the second feature map by the mask to obtain a valid feature map; multiply the second feature map by the inverse mask to obtain a redundant feature map;

[0026] After multiplying the learnable factor with the second feature map, the third feature map is obtained by adding the learnable factor to the effective feature map and the redundant feature map.

[0027] Preferably, the inputting the second feature map into the channel correlation selector to generate the channel-selected mask and the inverse mask comprises:

[0028] The second feature map is sequentially subjected to the pooling processing, the convolution processing and the activation processing to generate the channel-selected mask and the inverse mask, respectively.

[0029] Preferably, the predicting the position and the label of the hand acupoint based on the third feature map comprises:

[0030] The third feature map is subjected to the inverted bottleneck processing and the deconvolution processing to obtain the coordinate point position and the label of the hand acupoint.

[0031] In a second aspect, an embodiment of the present application provides a hand acupoint detection system based on region correlation coefficient learning attention, which comprises:

[0032] At least one processor;

[0033] At least one memory for storing at least one program;

[0034] When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of the above.

[0035] The present application has the following beneficial effects: the adaptive block perception attention module can capture detailed information at multiple scales, and is designed to detect fine-grained acupoints; the adaptive correlation coefficient channel attention module can model channel correlation and is designed to locate hand acupoints on various skin colors; the present application can detect fine-grained acupoints and locate acupoints on different skin colors. The present application can detect fine-grained acupoints and locate acupoints on different skin colors, thereby improving the effectiveness of hand acupoint detection. BRIEF DESCRIPTION OF DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0037] Figure 1 is a flowchart of a hand acupoint detection method based on region correlation coefficient learning attention in an embodiment of the present application;

[0038] Figure 2is a framework diagram of the regional correlation coefficient learning attention network in an embodiment of the present invention;

[0039] Figure 3 a is a schematic diagram comparing the appearance of the acupoint area and the surrounding area in an embodiment of the present invention; b is another schematic diagram comparing the appearance of the acupoint area and the surrounding area in an embodiment of the present invention; c is a schematic diagram comparing the skin color of the acupoint area and the surrounding area in an embodiment of the present invention; d is another schematic diagram comparing the skin color of the acupoint area and the surrounding area in an embodiment of the present invention;

[0040] Figure 4 a is a schematic diagram of 11 acupoints selected from the palm in an embodiment of the present invention; b is a schematic diagram of 7 acupoints selected from the back of the hand in an embodiment of the present invention; c is a schematic diagram of the reference distance of the palm in an embodiment of the present invention; d is a schematic diagram of the reference distance of the back of the hand in an embodiment of the present invention;

[0041] Figure 5 2 is a schematic diagram of the structure of a hand acupoint detection system based on regional correlation coefficient learning attention in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The following will be combined with the embodiments and drawings to clearly and completely describe the concept, specific structure and technical effects of the present invention so as to fully understand the purpose, scheme and effect of the present invention. It should be noted that the embodiments and features in the embodiments of the present invention can be combined with each other unless there is a conflict.

[0043] In the related art, a new method for acupoint localization that combines RGB and depth images to guide robot-controlled acupuncture is disclosed. However, both RGB and depth images are susceptible to changes in skin color, which can reduce image quality and reduce the accuracy of acupoint localization. In addition, a hand-based acupoint localization method that combines a dual attention mechanism with a cascade network model is disclosed, ensuring real-time performance and robustness under complex conditions, including lighting changes. However, changes in skin texture and color often cause the acupoint area to be very similar to the surrounding area, making acupoint localization on the hand difficult.

[0044] A public image-based deep learning method for facial acupoint detection uses a high-resolution network for feature extraction. However, variations in light intensity still affect image quality and feature extraction.

[0045] Two published network-based methods use computer vision to automatically identify acupuncture points on the face and hands in real time, improving the accuracy and accessibility of acupuncture. However, these methods show limited robustness in distinguishing similar acupuncture points.

[0046] In summary, despite the promising performance of these deep learning-based models, challenges remain in Chinese acupuncture. First, acupoint areas closely resemble the surrounding skin, making it difficult to distinguish them based on appearance alone. Subtle differences are often difficult to detect, leading to potential misidentification, especially in areas of low contrast. Second, variations in skin color can complicate the observation or palpation of specific landmarks on hand acupoints. For example, darker skin tones can make it more difficult to discern details or shallow depressions, thus increasing the challenge of acupoint localization.

[0047] To solve the above technical problems, the present invention proposes a Region Correlation Learning Attention Network (RCLA-Net) for Chinese acupuncture hand acupoint detection, which includes an Adaptive Patch Aware Attention (APAA) module and an Adaptive Correlation Channel Attention (ACCA) module. The APAA module captures the feature information of small patches to detect fine-grained acupoints, and the ACCA module locates acupoints under different skin colors through channel correlation modeling, thereby improving the effectiveness of hand acupoint detection.

[0048] See Figure 1 The present invention provides a hand acupoint detection method based on regional correlation coefficient learning attention, the method comprising the following steps:

[0049] S100, acquiring a hand image, performing feature extraction on the hand image, and obtaining a first feature map;

[0050] S200, processing the first feature map using a regional correlation coefficient learning attention network to obtain a third feature map; wherein the regional correlation coefficient learning attention network includes an adaptive block-aware attention module and an adaptive correlation coefficient channel attention module, detecting fine-grained acupoints in the first feature map using the adaptive block-aware attention module to obtain a second feature map; and modeling acupoint features of different skin colors in the second feature map using the adaptive correlation coefficient channel attention module to obtain a third feature map;

[0051] S300: Predict labels and coordinates of hand acupoints based on the third feature map.

[0052] The present invention uses an adaptive block perception attention module to capture detailed information at multiple scales and is designed to detect fine-grained acupoints. The adaptive correlation coefficient channel attention module can model channel correlation and be designed to locate hand acupoints on various skin tones. The present invention can detect fine-grained acupoints and locate acupoints under different skin tones, thereby improving the effectiveness of hand acupoint detection.

[0053] In some improved embodiments, in S100, extracting features from the hand image to obtain a first feature map includes:

[0054] A first feature map is extracted from the hand image using head convolution.

[0055] In some improved embodiments, in S200, detecting fine-grained acupuncture points in the first feature map by the adaptive block-aware attention module to obtain a second feature map includes:

[0056] Dividing the first feature map into multiple patches evenly, and connecting the multiple patches to generate a patch feature map;

[0057] The patch feature map is processed by using an average standard and then input into a patch attention module, and an attention weight is output;

[0058] The attention weights are processed by adaptive masking to obtain the second feature map.

[0059] In some improved embodiments, the expression of the attention weight is:

[0060] K=Softmax{MLP{Norm[MLP(J)]}};

[0061] Among them, K represents the attention weight, Softmax{·} represents the Softmax activation function processing, MLP{·} represents the multi-layer perceptron processing, and Norm{·} represents the normalization processing.

[0062] In some improved embodiments, processing the attention weights by adaptive masking to obtain the second feature map includes:

[0063] After normalizing the attention weights, they are multiplied by the learnable parameters to obtain the mask map.

[0064] The attention weight is multiplied by the mask map to obtain an attention feature map, which is reshaped into the second feature map.

[0065] like Figure 2As shown in Figure 2, the Regional Correlation Learning Attention Network (RCLA-Net) consists of two key components: the Adaptive Patch-Aware Attention (APAA) module and the Adaptive Correlation Channel Attention (ACCA) module. A detailed description of each module is given below.

[0066] The processing of the adaptive block-aware attention module for fine acupoint recognition is as follows:

[0067] In traditional Chinese acupuncture, acupoint areas are very similar in appearance to the surrounding areas on the hand surface. First, both areas exhibit consistent skin texture, without distinct boundaries that distinguish natural skin patterns. For example, the texture surrounding an acupoint on the hand naturally extends into the acupoint itself. Furthermore, surface landmarks, such as muscle contours and joint locations, are similarly aligned in both the acupoint area and the surrounding area, making them visually difficult to distinguish. Figure 3 Based on these observations, we design an adaptive patch-aware attention (APAA) module to perform acupoint recognition by capturing fine-grained features on small patches, as shown in a and b in Fig. Figure 2 The top module is shown.

[0068] The expression of the adaptive patch-aware attention module is defined as:

[0069] APPA(I)=AM[PA(I)];

[0070] Among them, PA and AM represent patch attention and adaptive mask respectively, which can be described as follows:

[0071] First, the first feature map I is evenly divided into small patches, which are then concatenated to generate the patch feature map J, which is then processed with the averaging criterion.

[0072] like Figure 2 The adaptive patch-aware attention module shown at the top, the first feature map is divided into small patches, allowing attention to focus on local details at different locations. This enables the adaptive patch-aware attention module to analyze each local region in a fine-grained manner, further distinguishing the hand acupoint area from similar surrounding areas.

[0073] Then, the patch feature map J is input into the multi-layer perceptron (MLP) to produce the attention weight K, which is defined as:

[0074] PA(J)=Softmax{MLP{Norm[MLP(J)]}};

[0075] Among them, PA(J) represents the patch attention processing of the patch feature map J, and the attention weight K=PA(J).

[0076] Next, the attention weight K is processed by adaptive masking (AM). Specifically, the attention weight K is normalized and multiplied by the learnable parameter α to obtain the mask map M.

[0077] Finally, the attention weight K is multiplied by the mask map M to obtain the attention feature map L, which is then reshaped into the second feature map X.

[0078] In some improved embodiments, in S200, the step of modeling the acupoint features of different skin colors in the second feature map using the adaptive correlation coefficient channel attention module to obtain a third feature map includes:

[0079] Inputting the second feature map into a channel-related selector to generate a channel selection mask and an inverse mask;

[0080] Multiply the second feature map by the mask to obtain a valid feature map; multiply the second feature map by the inverse mask to obtain a redundant feature map;

[0081] After multiplying the learnable factor by the second feature map, it is added to the effective feature map and the redundant feature map to obtain the third feature map.

[0082] In some improved embodiments, inputting the second feature map into a channel-related selector to generate a channel selection mask and an inverse mask includes:

[0083] The second feature map is sequentially subjected to pooling, convolution, and activation processing to generate a channel selection mask and an inverse mask, respectively.

[0084] Specifically, the convolution kernel size of the convolution processing is 1×1; the activation function of the activation processing is the ReLU function.

[0085] The following is a detailed introduction to the processing of the adaptive correlation coefficient channel attention module for skin color modeling:

[0086] In acupuncture, skin color plays a crucial role in determining the clarity, color accuracy, depth, and three-dimensional effect of acupoint imaging on the body surface. Specifically, excessive skin color will obscure details, while light skin color will cause changes in imaging, such as Figure 3 As shown in c and d in Figure 2, accurate skin color modeling is crucial for optimal visualization and accurate assessment of acupoint areas. To address this problem, the present invention designs an adaptive correlation coefficient channel attention (ACCA) module to model acupoint features under different skin colors by modeling channel correlation, as shown in Figure 2. Figure 2 As shown in the adaptive correlation coefficient channel attention module at the bottom, the expression of the adaptive correlation coefficient channel attention module is defined as:

[0087] ;

[0088] Where H(·) is the channel-dependent selector, X is the second feature map, β is the learnable factor, and Z is the third feature map.

[0089] The processing process of the adaptive correlation coefficient channel attention module on the second feature map X is described as follows:

[0090] First, the second feature map X is input into the channel-dependent selector H(·), which includes pooling, convolution and ReLU blocks to generate the channel selection mask H(X) and the inverse mask 1-H(X).

[0091] Next, multiply the second feature map X by the mask H(X) and the inverse mask 1-H(X) respectively to obtain the effective feature map Y and the redundant feature map Yˊ, which are expressed as:

[0092] ;

[0093] Finally, the learnable factor β is multiplied by the second feature map X and then added to the valid feature map Y and the redundant feature map Yˊ to obtain the third feature map Z.

[0094] Based on the above observations, the adaptive correlation coefficient channel attention module allows the regional correlation coefficient learning attention network to focus on effective features while suppressing redundant information, thereby improving the overall performance of acupoint localization for different skin colors.

[0095] In some improved embodiments, in S300, predicting the positions and labels of hand acupuncture points based on the third feature map includes:

[0096] The third feature map is subjected to inverted bottleneck processing and deconvolution processing to obtain coordinate point positions and labels of the hand acupuncture points.

[0097] In order to verify the effectiveness of the proposed regional correlation coefficient learning attention network for hand acupoint detection, this paper conducts extensive experiments on the hand acupoint dataset.

[0098] A. Hand acupoint dataset:

[0099] The 11K Hand Subset (11KHS) dataset is used. This dataset consists of 3860 palm hand images and 4175 back hand images. The resolution of each image is 1600×1200 and was collected from subjects aged 18-75. Specifically, 11 acupoints used in traditional Chinese medicine acupuncture were selected from the palm of the hand, namely: Sifeng acupoint (extraordinary meridian points EX-UE10, a total of four points), Shixuan acupoint (extraordinary meridian points EX-UE11, a total of four points), Shaofu acupoint (heart meridian HT8), Laogong acupoint (pericardium meridian PC8), and Zhongchong acupoint (pericardium meridian PC9). Figure 4In addition, 7 acupoints were selected from the back of the hand for acupuncture, namely: Hegu (Large Intestine Meridian LI4), Zhongzhu (Triple Burner Meridian SJ3), Yemen (Triple Burner Meridian SJ2), Shaoze (Small Intestine Meridian SI1), Shaochong (Heart Meridian HT9), Guanchong (Triple Burner Meridian SJ1), and Shangyang (Large Intestine Meridian LI1), as shown in Figure 1. Figure 4 As shown in b.

[0100] B. Evaluation indicators:

[0101] In order to evaluate the performance of the regional correlation coefficient learning attention network, the present invention regards the acupoint detection rate (ADR) as an evaluation indicator:

[0102] (5);

[0103] Where N is the total number of acupoints on the hand, and C is the number of correctly predicted acupoints. When the error E is less than the threshold λ (which is set to 0.04, 0.06, 0.10, and 0.12 in this paper), the predicted acupoint is considered to be correct, and E is defined as:

[0104] (6);

[0105] in, and are the j-th predicted acupoint and the true label respectively.

[0106] In addition, in order to eliminate the influence of image size and hand size on the results, the distance between two fixed points is used as the reference value for normalization. Specifically, for palm images, the Euclidean distance between Laogong and Shaofu points is set as the normalization factor. For back of hand images, the Euclidean distance between Yimen and Zhongzhu points is used as the normalization factor, as shown in the following example: Figure 4 As shown in c and d.

[0107] Comparison with hand acupoint detection methods on the 11KHS dataset;

[0108] In this paper, RCLA-Net (Regional Correlation Coefficient Learning Attention Network) is compared with five hand acupoint detection methods, including SimpleNet, EfficientHRNet, HigherHRNet, LECANet and LitePose, on the 11KHS dataset. The quantitative results are shown in Table 1.

[0109] Table 1: Quantitative results on the 11KHS dataset

[0110]

[0111] As can be seen from Table 1, the performance of the proposed regional correlation coefficient learning attention network (RCLA-Net) is better than that of other methods, and the ADRs on the 11KHS dataset are 31.24%, 49.47%, 69.28% and 76.11% when the error threshold λ is set to 0.04, 0.06, 0.10 and 0.12, respectively. In addition, the proposed regional correlation coefficient learning attention network improves the second best method LitePose. The results show that in hand acupoint detection, the adaptive patch correlation attention mechanism is more effective than multi-branch information fusion. In addition, RCLA-Net is significantly better than SimpleNet because SimpleNet is difficult to effectively learn fine-grained local characteristics. It can be seen that the experiments on the benchmark dataset prove that the proposed regional correlation coefficient learning attention network can improve the effectiveness of Chinese acupuncture hand acupoint detection.

[0112] Corresponding to the method of Figure 1 , with reference to Figure 5 , the embodiment of the present application provides a hand acupoint detection system based on regional correlation coefficient learning attention, comprising:

[0113] at least one processor;

[0114] at least one memory for storing at least one program;

[0115] When the at least one program is executed by the at least one processor, the at least one processor implements the above-mentioned method.

[0116] It can be seen that the contents in the above-mentioned method embodiments are all applicable to the present system embodiment, the present system embodiment specifically implements the same functions as the above-mentioned method embodiments, and achieves the same beneficial effects as the above-mentioned method embodiments.

[0117] In addition, the embodiment of the present application also discloses a computer program product or a computer program, which is stored in a computer readable storage medium. The processor of the computer device can read the computer program from the computer readable storage medium, and the processor executes the computer program, so that the computer device executes the above-mentioned method. Similarly, the contents in the above-mentioned method embodiments are all applicable to the present storage medium embodiment, the present storage medium embodiment specifically implements the same functions as the above-mentioned method embodiments, and achieves the same beneficial effects as the above-mentioned method embodiments.

[0118] Those skilled in the art will appreciate that all or some of the methods and systems disclosed above can be implemented as software, firmware, hardware, or any suitable combination thereof. Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on computer-readable media, which may include computer storage media (or non-transitory media) and communication media (or transitory media). As is well known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVDs) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. Furthermore, as is well known to those skilled in the art, communication media typically embodies computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0119] The above is a specific description of the preferred implementation of the present disclosure, but the present disclosure is not limited to the above-mentioned implementation mode. Technical personnel familiar with the art can also make various equivalent modifications or substitutions without violating the spirit of the present disclosure. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present disclosure.

Claims

1. A hand acupoint detection method based on regional correlation coefficient learning attention, characterized in that: The method comprises the following steps: Acquire a hand image, perform feature extraction on the hand image, and obtain a first feature map; The first feature map is processed using a regional correlation coefficient learning attention network to obtain a third feature map; wherein the regional correlation coefficient learning attention network includes an adaptive block-aware attention module and an adaptive correlation coefficient channel attention module, and the adaptive block-aware attention module is used to detect fine-grained acupoints in the first feature map to obtain a second feature map; and the adaptive correlation coefficient channel attention module is used to model acupoint features of different skin colors in the second feature map to obtain a third feature map; Predicting labels and coordinates of hand acupuncture points based on the third feature map; The detecting fine-grained acupuncture points in the first feature map by the adaptive block-aware attention module to obtain a second feature map comprises: Dividing the first feature map into multiple patches evenly, and connecting the multiple patches to generate a patch feature map; The patch feature map is processed by using an average standard and then input into a patch attention module, and an attention weight is output; The attention weights are processed by adaptive mask to obtain the second feature map; The method of modeling the acupoint features of different skin colors in the second feature map by the adaptive correlation coefficient channel attention module to obtain a third feature map includes: Inputting the second feature map into a channel-related selector to generate a channel selection mask and an inverse mask; Multiply the second feature map by the mask to obtain a valid feature map; multiply the second feature map by the inverse mask to obtain a redundant feature map; multiply the learnable factor by the second feature map, add the valid feature map and the redundant feature map to obtain the third feature map.

2. The method according to claim 1, characterized in that The step of extracting features from the hand image to obtain a first feature map includes: A first feature map is extracted from the hand image using head convolution.

3. The method according to claim 1, characterized in that The expression of the attention weight is: K=Softmax{MLP{Norm[MLP(J)]}}; Among them, K represents the attention weight, Softmax{·} represents the Softmax activation function processing, MLP{·} represents the multi-layer perceptron processing, and Norm{·} represents the normalization processing.

4. The method according to claim 1, wherein The attention weights are processed by the adaptive mask to obtain a second feature map, including: After normalizing the attention weights, they are multiplied by the learnable parameters to obtain the mask map. The attention weight is multiplied by the mask map to obtain an attention feature map, which is reshaped into the second feature map.

5. The method according to claim 1, wherein The step of inputting the second feature map into a channel-related selector to generate a channel selection mask and an inverse mask includes: The second feature map is sequentially subjected to pooling, convolution, and activation processing to generate a channel selection mask and an inverse mask, respectively.

6. The method according to claim 1, characterized in that The predicting the positions and labels of the hand acupuncture points based on the third feature map includes: The third feature map is subjected to inverted bottleneck processing and deconvolution processing to obtain coordinate point positions and labels of the hand acupuncture points.

7. A hand acupoint detection system based on regional correlation coefficient learning attention, characterized in that The system comprises: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of claims 1 to 6.

Images