Intelligent massage method and system and computer readable storage medium

Abstract

The invention discloses an intelligent massage method and system and a computer readable storage medium, which are applied to the field of massage health care, and comprise the following steps: obtaining a target heat distribution diagram of a target object; identifying and analyzing the target heat distribution diagram to obtain a lesion identification result of the target object; obtaining depth image information corresponding to the lesion part based on the lesion recognition result; the diseased region is positioned based on the lesion recognition result and the depth image information, and a massage component is controlled to massage the diseased region according to the positioning result. The target heat distribution diagram is recognized and analyzed to obtain the lesion recognition result, the depth image information corresponding to the lesion part is obtained according to the lesion recognition result, then the lesion part is positioned and massaged according to the lesion recognition result and the depth image information, the recognition and positioning precision can be improved, and the massage effect is improved. Manual operation is reduced, and massage efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of massage health care, and in particular to an intelligent massage method, system and computer-readable storage medium. Background Art

[0002] With the accelerating pace of life and increasing work pressure, sub-health issues are becoming increasingly serious. To alleviate physical fatigue and discomfort, smart massagers have emerged. After several iterations, the current mainstream smart massagers mainly use built-in pressure or temperature sensors to monitor the user's body shape and posture in real time to adjust the massage intensity and provide a personalized massage experience. However, due to the differences in the bodies of different users, such as some users with localized vascular and tendon blockages or inflammation, current massagers only provide general kneading for neck massage, unable to accurately locate the affected area and massage to relieve the blockage or inflammation.

[0003] Therefore, how to provide a massage that can perform personalized massage based on the user's physical differences to relieve local blood vessel and tendon blockage or inflammation is a technical problem that technical personnel in this field urgently need to solve. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide an intelligent massage method, system and computer-readable storage medium, which solves the problem that the massage machine in the prior art only performs overall kneading and cannot accurately locate the diseased area and massage to relieve congestion or inflammation.

[0005] To solve the above technical problems, the present invention provides an intelligent massage method, comprising:

[0006] Obtaining a target heat distribution map of a target object;

[0007] Performing identification and analysis on the target heat distribution map to obtain a lesion identification result of the target object;

[0008] Acquiring depth image information corresponding to the lesion site based on the lesion recognition result;

[0009] The lesion site is located based on the lesion recognition result and the depth image information, and a massage component is controlled to massage the lesion site according to the positioning result.

[0010] Optionally, the recognition analysis of the target heat distribution map is implemented based on an image AI recognition model, and the image AI recognition model includes a recognition model and a classification model. Then, the recognition analysis of the target heat distribution map to obtain the lesion recognition result of the target object includes:

[0011] Analyzing the target heat distribution map using a recognition model trained based on a deep learning algorithm to identify the planar position information of the lesion site;

[0012] Using a classification model obtained by training a convolutional neural network based on a residual network architecture, the lesion site is classified to determine the lesion category;

[0013] Accordingly, obtaining depth image information corresponding to the lesion site based on the lesion recognition result includes:

[0014] Based on the planar position information of the lesion part, acquiring the depth image information corresponding to the lesion part;

[0015] Accordingly, locating the lesion site based on the lesion recognition result and the depth image information, and controlling the massage component to massage the lesion site according to the positioning result, includes:

[0016] The lesion part is positioned based on the planar position information of the lesion part and the depth image information, and a massage component is controlled to massage the lesion part according to the positioning result.

[0017] Optionally, locating the lesion part based on the planar position information of the lesion part and the depth image information, and controlling a massage component to massage the lesion part according to the positioning result, includes:

[0018] determining a relative position between the lesion site and the massage component based on the planar position information of the lesion site and the depth image information;

[0019] determining a corresponding massage strategy based on the lesion category;

[0020] The massage component is controlled to move to the lesion site based on the relative position to execute the massage strategy.

[0021] Optionally, controlling the massage component to move to the lesion site to execute the massage strategy based on the relative position includes:

[0022] The massage component is controlled to move to the center of the lesion site based on the relative position to execute the massage strategy.

[0023] Optionally, the classification model obtained by training a convolutional neural network based on a residual network architecture is used to classify the lesion site and determine the lesion category, including:

[0024] Inputting the identified lesion image into the classification model, and outputting features of the lesion image through the convolution operation of each convolutional layer; wherein, if the current convolutional layer has a previous convolutional layer, the features output by the current convolutional layer are the sum of the output item of the current convolutional layer through the convolution operation and the input item of the previous convolutional layer;

[0025] The features of the lesion area image are input into a fully connected layer, and the lesion category is determined according to the lesion category probability output by the fully connected layer.

[0026] Optionally, before identifying the planar position information of the lesion site, the method further includes:

[0027] Acquiring target morphological features of the target object;

[0028] The method of analyzing the target heat distribution map using a recognition model trained based on a deep learning algorithm to identify the planar position information of the lesion site includes:

[0029] Analyzing the target heat distribution map to extract features of abnormal temperature areas;

[0030] Combining the temperature anomaly region characteristics with the target morphological characteristics to obtain a comprehensive feature vector;

[0031] The comprehensive feature vector is input into the recognition model trained based on the deep learning algorithm to identify the planar position information of the lesion site.

[0032] Optionally, the process of training the recognition model based on a deep learning algorithm includes:

[0033] Collecting multiple target heat distribution maps, and dividing the collected target heat distribution maps into a training set and a validation set; the target heat distribution maps in the training set are marked with lesion sites;

[0034] Using the training set to train the recognition model, and using the trained recognition model to recognize the validation set to generate a recognition result;

[0035] An index for comprehensively measuring the accuracy and recall rate is calculated based on the recognition result, and whether the recognition model is trained is determined based on the index for comprehensively measuring the accuracy and recall rate.

[0036] Optionally, the use of a recognition model trained based on a deep learning algorithm to analyze the target heat distribution map and identify the planar position information of the lesion site includes:

[0037] Analyzing and processing the target heat distribution map using the recognition model to mark the lesion site;

[0038] The YOLOv8 target detection algorithm is used to identify and extract the planar position information of the lesion site.

[0039] The present invention also provides an intelligent massage system, comprising:

[0040] Target heat distribution acquisition component, massage component, depth image information acquisition component, memory and processor;

[0041] The target heat distribution acquisition component and the depth image information acquisition component are both in communication with the processor, and are used to send the acquired target heat distribution map and depth image information of the target object to the processor respectively; the processor is in communication with the massage component, and is used to control the massage component to massage the lesion area;

[0042] The memory is used to store a computer program, and the processor is used to execute the computer program to implement the steps of the above-mentioned intelligent massage method.

[0043] Optionally, the depth image information acquisition component is a depth camera;

[0044] The target heat distribution acquisition component is an infrared thermal imaging camera, and the processor includes an intelligent recognition component and a main control component;

[0045] The intelligent recognition component is respectively connected to the depth camera and the infrared thermal imaging camera for identifying the planar position information of the lesion site according to the acquired target heat distribution map, and determining the corresponding lesion category and the corresponding massage strategy;

[0046] The main control component is respectively connected to the intelligent recognition component and the massage component for communication, and is used to receive the corresponding massage strategy determined by the intelligent recognition component, and control the massage component to move to a corresponding position to execute the massage strategy.

[0047] The present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the intelligent massage method described above are implemented.

[0048] It can be seen that the intelligent massage method provided by the present invention includes obtaining a target heat distribution map of a target object; identifying and analyzing the target heat distribution map to obtain a lesion recognition result of the target object; obtaining depth image information corresponding to the lesion site based on the lesion recognition result; locating the lesion site based on the lesion recognition result and the depth image information, and controlling the massage component to massage the lesion site according to the positioning result. The present invention obtains a lesion recognition result by identifying and analyzing the target heat distribution map, and obtains depth image information corresponding to the lesion site using the lesion recognition result, and then locates the lesion site using the above-mentioned lesion recognition result and depth image information, and controls the massage component to massage the lesion site, which can improve the accuracy of recognition and positioning, reduce manual operations, and improve massage efficiency. In addition, the present invention also provides an intelligent massage system and a computer-readable storage medium, which also have the above-mentioned beneficial effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0050] Figure 1 A flow chart of an intelligent massage method provided by an embodiment of the present invention;

[0051] Figure 2 This is an example of back infrared thermal imaging provided by an embodiment of the present invention;

[0052] Figure 3 Another example of back infrared thermal imaging provided by an embodiment of the present invention;

[0053] Figure 4 A schematic structural diagram of an intelligent massage system provided by an embodiment of the present invention;

[0054] Figure 5 A schematic structural diagram of another intelligent massage system provided by an embodiment of the present invention;

[0055] Figure 6 A schematic diagram of the structural connection of an intelligent massage system provided by an embodiment of the present invention;

[0056] The following are the descriptions of the reference numerals:

[0057] 10-target heat distribution acquisition component, 11-infrared thermal imaging camera, 20-massage component, 30-memory, 40-processor, 41-intelligent recognition component, 42-main control component, 50-communication interface, 51-communication bus, 60-depth camera, 70-intelligent robotic arm. DETAILED DESCRIPTION

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0059] With the accelerated pace of life and increased work pressure, sub-health issues are becoming increasingly serious. To alleviate physical fatigue and discomfort, smart massagers have emerged. After several updates and iterations, the current mainstream smart massagers mainly use built-in pressure sensors or temperature sensors to monitor the user's body shape and posture in real time to adjust the massage intensity and provide a personalized massage experience. However, due to differences in the bodies of different users, for example, some users may have localized blood vessel blockages or inflammation. Current massagers use infrared cameras to capture infrared radiation emitted by the human body, generate images corresponding to the body's thermal state, and display them on the control panel. This allows the operator to focus massage on areas with higher heat distribution based on the corresponding thermal state images for different users. Therefore, existing massagers simply incorporate infrared cameras, and the massage of the affected area is still a general kneading, which cannot accurately locate the affected area and massage it to relieve blockage or inflammation.

[0060] The present invention, however, identifies and analyzes the target heat distribution map to obtain a lesion recognition result, uses the lesion recognition result to obtain depth image information corresponding to the lesion site, and then uses the lesion recognition result and depth image information to locate the lesion site and control the massage component to massage the lesion site. This improves the accuracy of recognition and positioning, reduces manual operation, and improves massage efficiency. Furthermore, the present invention provides an intelligent massage system and computer-readable storage medium, which also have the aforementioned beneficial effects.

[0061] Please refer to Figure 1 , Figure 1 This is a flow chart of an intelligent massage method provided by an embodiment of the present invention. The method may include:

[0062] S101: Acquire a target heat distribution map of a target object.

[0063] The execution subject of this embodiment is an intelligent massage system. The target object in this embodiment is generally a human body, and can be specifically applied to human back massage, so the target heat distribution obtained is Figure 1 It is generally a heat distribution map of the human back, but this embodiment is not limited to this. The target heat distribution acquisition component and image AI recognition model provided in this embodiment can identify and guide the precise positioning of the lesion site to achieve rapid identification of the lesion site and provide targeted massage.

[0064] This embodiment does not limit the specific method of obtaining the target heat distribution map of the target object, as long as the target heat distribution map of the target object can be accurately obtained. For example, the target heat distribution map can be obtained using an infrared thermal imaging camera, or using a thermal radiometer, or using a microwave thermal imager, a thermistor array, a pyroelectric sensor, a thermocouple array, etc. to obtain the target heat distribution map. It should be further explained that the use of an infrared thermal imaging camera to obtain the target heat distribution map can not only ensure the accuracy of the target heat distribution map, but also avoid contact with the target, and avoid interference and discomfort to the target. This embodiment also does not limit the startup timing of the program for obtaining the target heat distribution map. For example, it can be obtained in real time after startup, or in order to reduce operating costs, the target heat distribution map can be obtained only once after startup for subsequent processing.

[0065] S102: Identify and analyze the target heat distribution map to obtain a lesion identification result of the target object.

[0066] It should be noted that for the same target heat distribution map, multiple lesion sites can be identified. When executing the massage strategy, multiple massage components can be controlled to massage multiple lesion sites at the same time, or one massage component can be controlled to massage multiple lesion sites in sequence. When controlling one massage component to massage multiple lesion sites in sequence, there is a priority relationship in massaging the multiple lesion sites. In this embodiment, a priority can be set for each lesion site so that each lesion site can be massaged in an orderly manner. This embodiment does not limit the specific method of setting the priority of the lesion site, as long as it can ensure that the massage component performs the massage operation in an orderly manner. For example, the closer the lesion site is to the initial position of the massage component, the higher its priority; or the lesion site with a higher degree of severity can be set to have a higher priority; or it can be set to comprehensively consider the lesion site with a higher degree of severity and the lesion site with a higher priority.

[0067] Furthermore, in order to ensure the convenience of identifying the lesion of the target object and obtain the lesion location information, the recognition analysis of the above-mentioned target heat distribution map can be set to be implemented based on the image AI recognition model. The image AI recognition model includes a recognition model and a classification model. The target heat distribution map is recognized and analyzed to obtain the lesion recognition result of the target object, which may include:

[0068] Using the recognition model trained based on the deep learning algorithm, the target heat distribution map is analyzed to identify the planar position information of the lesion site;

[0069] The classification model obtained by training the convolutional neural network based on the residual network architecture is used to classify the lesion site and determine the lesion category;

[0070] Accordingly, obtaining depth image information corresponding to the lesion site based on the lesion recognition result may include:

[0071] Based on the plane position information of the lesion part, obtaining the depth image information corresponding to the lesion part;

[0072] Accordingly, locating the lesion site based on the lesion recognition result and the depth image information, and controlling the massage component to massage the lesion site according to the positioning result may include:

[0073] The lesion part is located based on the plane position information of the lesion part and the depth image information, and the massage component is controlled to massage the lesion part according to the positioning result.

[0074] It should be noted that in this embodiment, a deep learning algorithm is used to train an image AI recognition model using the collected training set and validation set. After the training is completed, the image AI recognition model can be used to analyze the acquired target heat distribution map and identify the information of the lesion site in the target heat distribution map to achieve automatic recognition and extraction of the lesion site. In this embodiment, a recognition model trained based on a deep learning algorithm is used to obtain the planar position information of the lesion site of the target object, and a classification model trained based on a convolutional neural network architecture is used to determine the lesion category and the depth image information corresponding to the lesion site. By combining the above-mentioned planar position information of the lesion site and the above-mentioned depth image information, the three-dimensional position positioning coordinates of the lesion site can be determined. The lesion category of the lesion site determined by the classification model can be used to obtain the corresponding massage strategy. Therefore, by combining the above-mentioned three-dimensional position positioning coordinates of the lesion site and the corresponding massage strategy, the step of controlling the massage component to massage the lesion site of the target object can be completed. In this embodiment, in order to ensure that the lesion part of the target object is automatically massaged, a convolutional neural network based on a residual network architecture is further used to train a classification model, and the trained classification model is used to classify the lesion part extracted by the above identification to determine the lesion category of the lesion part. From obtaining the heat distribution map of the target to identifying the lesion category of the lesion part, no human intervention is required in the entire process, and automatic identification and determination of the lesion category can be achieved.

[0075] Furthermore, in order to improve the processing performance and stability of the classification model, the classification model obtained by training the convolutional neural network based on the residual network architecture can classify the lesion site and determine the lesion category, which may include the following steps:

[0076] Step S11: Input the identified lesion image into the classification model, and output the features of the lesion image through the convolution operation of each convolution layer; wherein, if the current convolution layer has a previous convolution layer, the features output by the current convolution layer are the sum of the output item of the current convolution layer through the convolution operation and the input item of the previous convolution layer;

[0077] Step S12: Input the features of the lesion area image into the fully connected layer, and determine the lesion category based on the lesion category probability output by the fully connected layer.

[0078] It should be noted that in this embodiment, when the current convolution layer has a previous convolution layer, the feature output by the current convolution layer is set to be the sum of the output item of the current convolution layer through the convolution operation and the input item of the previous convolution layer, allowing the model to reuse features between different layers and improve the efficiency of feature utilization. When the current convolution layer is the initial convolution layer, the feature output by the initial convolution layer is only the output item of the initial convolution layer through the convolution operation. In addition, in this embodiment, the feature of the image of the lesion site finally output by the convolution layer is input to the fully connected layer. At this time, the lesion category corresponding to the maximum probability can be selected from the lesion category probabilities output by the fully connected layer as the final lesion category.

[0079] Furthermore, in order to ensure that the lesion category is determined successfully based on the extracted image of the lesion site, it can be set that if the current convolution layer is the initial convolution layer, the formula is substituted into:

[0080] , get the features of the current convolutional layer output; among them, is the input item of the current convolutional layer, is the convolution kernel, is the bias term, The feature output by the current convolutional layer;

[0081] If the current convolutional layer has a previous convolutional layer, substitute the formula:

[0082] , get the features of the current convolutional layer output; among them, is the output item of the current convolution layer through the convolution operation, is the input of the previous convolutional layer, It is the feature output by the current convolutional layer;

[0083] The features of the lesion image are input into the fully connected layer. The lesion category is determined based on the lesion category probability output by the fully connected layer. The following steps may be performed:

[0084] Substitute the features of the lesion image into the fully connected layer classification formula:

[0085] , and obtain the probability of lesion category; where, is the weight matrix of the fully connected layer, is the bias term, is the feature of the image of the lesion, is the category probability after Softmax conversion. Softmax conversion is a function that converts the network output into a probability distribution;

[0086] Substitute the lesion category probability into the category determination formula:

[0087] , determine the lesion category; among them, is the category probability after Softmax conversion, The confirmed lesion category.

[0088] It should be noted that, in this embodiment, if the current convolution layer is the initial convolution layer, the image of the lesion site is input into the formula In this case, the image of the lesion site identified and extracted is used as the input of the current convolutional layer (in this case, the initial convolutional layer). If a previous convolutional layer exists within the current convolutional layer, the input of the current convolutional layer is the output of the previous convolutional layer. The Softmax transform is a function that converts a real vector into a probability distribution.

[0089] In this embodiment, the classification model can be trained using a labeled training set. By continuously adjusting model parameters and optimizing the algorithm, the accuracy of grading and classifying lesions can be improved. Based on this, different massage strategies can be set to guide the massage claws in the massage component to perform different massage techniques and intensities. Specifically, this embodiment can be applied to the recognition of infrared thermal images of the human back. In this case, the ResNet (residual network) architecture of the convolutional neural network (CNN) is selected as the basic model. The process of image recognition and classification using the ResNet architecture is as follows:

[0090] The convolution layer is used to process the image of the identified lesion. The convolution operation extracts features through the above formula: , the ResNet architecture introduces the skip connection, so that the output of each layer in the convolutional neural network not only depends on the convolution operation result of this layer, but also directly adds the input of the previous layer. For details, refer to the above formula: , after the convolution operation and residual connection, the ReLU activation function is applied: After establishing a mapping relationship and undergoing multiple layers of convolution and residual connections, the feature map is flattened and classified through a fully connected layer. The output formula of the fully connected layer can refer to the above formula: Finally, the lesion category can be confirmed by referring to the above formula: The ResNet architecture has a very deep network layer and can effectively extract features from images. Its residual connection helps solve the gradient vanishing problem in deep networks and is suitable for processing complex image data such as infrared thermal imaging.

[0091] Furthermore, in order to improve the adaptability of the above classification model and improve the accuracy of lesion classification, the initial convolution layer in the above classification model can be set to Convolutional layer, stride 1, number of convolution kernels is 64;

[0092] The pooling window size of the maximum pooling layer in the classification model is .

[0093] It should be noted that in this embodiment, the conventional stride of the initial convolution layer is adjusted to 1 in this embodiment, which can avoid losing key information. The number of convolution kernels in the initial convolution layer is set to 64, which can balance the computational complexity and feature representation ability when extracting features, avoid making the computation too complicated, and effectively capture the basic feature elements in the image. In addition, this embodiment sets the conventional maximum pooling layer size in the above classification model to Adjust to When used with infrared thermal imaging images, since the infrared thermal imaging images themselves are small in size, a larger pooling window will excessively reduce the image size, affecting subsequent feature extraction. The window can reduce the data dimension while balancing the preservation of sufficient feature information in the image.

[0094] For the training process of the model in this embodiment, when collecting training set images, the collected images can be randomly rotated within the range of -15° to 15°. Within this angle range, the diversity of the samples can be increased, simulating images taken at different shooting angles, while not causing excessive distortion of the images and losing the original feature patterns, thereby expanding the sample size. In the horizontal and vertical directions, the images are randomly translated with a maximum translation distance of 10% of the image width and image height. This allows the model to learn the positional invariance of features within a certain range, adapting to scenarios in practical applications where the captured images may have certain positional deviations. In addition, the image scaling ratio can be adjusted, with random scaling between 0.8 and 1.2 times the original image size, so that the trained model can adapt to image features at different magnification and reduction ratios.

[0095] The model training process uses the Adam (Adaptive Moment Estimation) optimizer and a cosine annealing learning rate to adjust the training process. The initial learning rate is set to 0.001. This allows the model parameters to be updated with a relatively appropriate step size during the initial training phase, avoiding problems such as model convergence failure due to an excessively large learning rate or slow training due to a too small learning rate. According to the cosine annealing formula, by gradually reducing the learning rate within each training cycle, with the cycle set to 10 epochs (training rounds), the model can quickly approach the optimal solution early in training, facilitating fine-tuning in the later stages of training and avoiding fluctuations near the optimal solution. In addition, in this embodiment, the loss value and accuracy are output to the log file once every 10 batches, which makes it convenient to check the training progress of the model at any time. Visualization tools (such as TensorBoard, which is a visualization tool mainly used to display and analyze the training process and structure of machine learning models) can be used to draw loss curves and accuracy curves, which can intuitively observe the changing trends in the training process. When abnormal fluctuations in the loss curve are found or the accuracy stagnates for a long time, the training strategy can be adjusted in time.

[0096] When evaluating the trained model on the validation set, the accuracy (the ratio of correctly predicted examples to the total number of examples in the validation set), recall (the ratio of correctly predicted positive examples to the total number of true positive examples), and F1-score (a comprehensive measure of accuracy and recall, where F1 = 2 * (accuracy * recall) / (accuracy + recall)) are calculated on the validation set. These evaluations are performed on the validation set after each epoch of training. For hyperparameter tuning, the initial number of iterations can be set to 50, a value determined based on experience and an estimate of the dataset size. During training, observe the changes in validation set metrics. If the model converges after around 30 epochs (model convergence means no more significant improvement in metrics), training can be stopped early. If the model has not converged after 50 epochs, the number of iterations can be increased appropriately, but care should be taken to avoid overfitting. The optimal number of iterations that achieves good performance on the validation set can be determined through repeated experiments. In order to achieve accurate identification and classification of lesion sites, this embodiment can adopt a classification model based on the ResNet50 architecture. The formula of the classification model is as follows:

[0097] ,in, The image of the lesion detected by infrared thermal imaging. The coordinates of the lesion can be set as arrive These areas usually show abnormal temperature or abnormal morphology. The system can classify the lesion location, including the type and severity of the lesion. Based on the severity of the lesion, the system can adjust parameters such as massage intensity and duration to improve the precision and personalization of the massage strategy.

[0098] Furthermore, in order to improve the accuracy of identifying the lesion site and avoid misjudgment, the following steps may be included before the lesion site is identified:

[0099] Step S21: Acquire target morphological features of the target object;

[0100] The above-mentioned recognition model trained based on the deep learning algorithm is used to analyze the target heat distribution map and identify the lesion site, which may include the following steps:

[0101] Step S22: Analyze the target heat distribution map and extract the characteristics of the temperature abnormality area;

[0102] Step S23: combining the temperature anomaly region characteristics with the target morphological characteristics to obtain a comprehensive feature vector;

[0103] Step S24: Input the comprehensive feature vector into the recognition model trained based on the deep learning algorithm to identify the planar position information of the lesion site.

[0104] It should be noted that the intelligent massage method described in this embodiment is generally applied to the human back. Therefore, the target morphological features described above may include morphological features such as back contour and muscle texture, which can be implemented through morphological image processing technology, helping to further identify the lesion site. In this embodiment, by fusing and analyzing the characteristics of the abnormal temperature area with the target morphological features to identify the lesion site, it is possible to avoid identifying the target normal heating area as the lesion site, thereby avoiding misjudgment of the recognition model and improving the accuracy of identifying the lesion site.

[0105] Furthermore, in order to ensure that the recognition model is trained successfully and improve the recognition and extraction effect of the lesion site, the process of training the recognition model based on the deep learning algorithm may include the following steps:

[0106] Step S31: collecting multiple target heat distribution maps, and dividing the collected target heat distribution maps into a training set and a validation set; the target heat distribution maps in the training set are marked with the lesion site;

[0107] Step S32: using the training set to train the recognition model, and using the trained recognition model to recognize the validation set to generate a recognition result;

[0108] Step S33: Calculate an index for comprehensively measuring the accuracy and recall rate based on the recognition results, and determine whether the recognition model training is completed based on the index for comprehensively measuring the accuracy and recall rate.

[0109] In this embodiment, the training set and the validation set can be obtained by collecting pictures, and a large number of back infrared thermal images that meet the requirements can be collected as target heat distribution maps. The above back infrared thermal images can be referred to Figure 2 and Figure 3 , Figure 2 This is an example of a back infrared thermal imaging diagram provided by an embodiment of the present invention. Figure 3 This is another example of back infrared thermal imaging provided by an embodiment of the present invention. Figure 2 and Figure 3 The lesion sites are marked with red boxes and can be used as training sets. For example, the collection objects can be clearly defined as workers between the ages of 25 and 65 to ensure that the images are representative and extensive. Then, the collected images can be preprocessed, including but not limited to preprocessing the collected images using median filtering and Gaussian filtering to remove noise interference in the image, improve image quality, and improve the clarity of blood vessels and tendons, so as to facilitate subsequent training of recognition models to identify lesion sites. In addition, the present application can also enhance the areas with abnormal temperature in the image through methods such as histogram equalization and contrast stretching.

[0110] It should be noted that in this embodiment, the recognition model training is completed after the comprehensive measurement of the accuracy and recall rate reaches the standard value. In this embodiment, the comprehensive measurement of the accuracy and recall rate can be set as the ratio of the product of 2 times the accuracy and recall rate divided by the sum of the accuracy and recall rates; the accuracy rate is the ratio of the number of correctly predicted samples in the validation set to the total number of samples in the validation set, and the recall rate is the ratio of the number of correctly predicted positive samples in the validation set to the number of true positive samples.

[0111] Furthermore, in order to ensure that the identified lesion site is accurately extracted, the above-mentioned recognition model trained based on the deep learning algorithm is used to analyze the target heat distribution map and identify the planar position information of the lesion site, which may include the following steps:

[0112] Step S41: Analyzing and processing the target heat distribution map using the recognition model to mark the lesion site;

[0113] Step S42: using the YOLOv8 target detection algorithm to identify and extract the planar position information of the lesion site.

[0114] It should be noted that in this embodiment, after the lesion site is identified and marked using the recognition model, the YOLOv8 (YOLOv8 is a YOLO series target detection algorithm) target detection algorithm can be used to extract the planar position information of the lesion site, so that when the lesion category of the lesion site is determined using the trained classification model, the planar position information of the identified lesion site can be input into the classification model. Specifically, the image information of the lesion site can be input into the classification model.

[0115] S103: Obtaining depth image information corresponding to the lesion site based on the lesion recognition result.

[0116] In this embodiment, based on the planar position information of the lesion part in the above-mentioned lesion identification results, the depth image information corresponding to the lesion part is determined. The planar position information and the depth image information of the lesion part can be combined to determine the three-dimensional position positioning coordinates of the lesion part, which is convenient for subsequent control of the massage component to accurately position the massage part.

[0117] S104: Positioning the lesion site based on the lesion recognition result and the depth image information, and controlling the massage component to massage the lesion site according to the positioning result.

[0118] In this embodiment, based on the lesion type determined above, a corresponding massage strategy can be determined. After the massage component is controlled to move to the corresponding lesion location, the massage strategy can be executed to alleviate the lesion. This embodiment does not limit the specific method for determining the corresponding massage strategy based on the lesion type; any massage strategy that can alleviate the lesion will suffice. For example, the lesion type can be set to include vascular obstruction in the area, muscle sclerosis in the area, or abnormal temperature in the area.

[0119] (1) When blood vessel blockage is detected in a region, the classification model can determine the severity of the blood vessel blockage and select a kneading massage operation based on the severity of the blood vessel blockage.

[0120] (2) When muscle stiffness is detected in an area, the above classification model can select a massage operation for tapping therapy.

[0121] (3) When abnormal temperature is detected in the area, the above classification model can select the heating operation.

[0122] In this embodiment, a mapping table can be constructed corresponding to the above-determined lesion categories and corresponding execution strategies. When it is necessary to generate a corresponding execution strategy based on the lesion category, the mapping table can be directly called to determine the execution strategy corresponding to the lesion category.

[0123] Furthermore, in order to ensure that the massage component moves smoothly to the lesion site and improve the accuracy of controlling the movement of the massage component, the lesion site is located based on the planar position information of the lesion site and the depth image information, and the massage component is controlled to massage the lesion site according to the positioning result, which may include the following steps:

[0124] Step S51: determining the relative position between the lesion site and the massage component based on the planar position information and depth image information of the lesion site;

[0125] Step S52: determining a corresponding massage strategy based on the lesion type;

[0126] Step S53: Based on the relative position, the massage component is controlled to move to the lesion site to execute the massage strategy.

[0127] It should be noted that, in this embodiment, the depth image information is combined to determine the relative position between the location of the lesion and the current position of the massage component. Based on this relative position, the massage component can be precisely moved to the lesion to execute the corresponding massage strategy, thereby improving the efficiency of massage at the lesion. This embodiment does not limit the massage component to a specific location at the lesion when executing the corresponding massage strategy; as long as the massage component can alleviate the lesion at the lesion when executing the massage strategy, it is sufficient. In this embodiment, an optimal position within the lesion can be selected as the endpoint for the massage component to move to the corresponding lesion. In this embodiment, the process of collecting depth image information can be accomplished using a depth camera. As a visual sensor, the depth camera can capture three-dimensional information in a scene. By emitting light and measuring the time difference or phase difference of reflected light, the depth camera calculates the distance from the object to the camera, thereby generating a depth image. This process involves converting the coordinates of the lesion in the depth camera's coordinate system to the intelligent robotic arm's coordinate system, enabling the intelligent robotic arm to accurately reach the corresponding lesion.

[0128] In this embodiment, the aforementioned recognition model guides the positioning of the intelligent robotic arm. During the positioning planning process, the YOLOv8 object detection algorithm is first used to identify and extract the location of the lesion within the target heat map. A depth camera is then used in conjunction with the 6D pose estimation algorithm Pix2Pose (a deep learning-based pose estimation algorithm) to determine the precise location of the lesion in three-dimensional space. The optimal massage point is then determined using the G2L-Net positioning point detection algorithm (a deep learning algorithm for real-time 6D pose estimation). This typically involves a global-to-local network structure and processing of 3D point clouds. The optimal massage point can be set at the center of the lesion. Finally, the motion planning stage utilizes dynamic motion primitives (DMP) (a mathematical framework for robotic motion control that can generate and imitate complex motion trajectories) to generate a smooth trajectory from the current position of the massage component to the optimal massage point on the lesion. Alternatively, imitation learning can be used to learn a grasping strategy from an operator's demonstration to ensure that the massage component accurately and efficiently executes the massage strategy. In addition, the motion planning of the intelligent robotic arm can also take into account the workspace limitations and joint motion range of the intelligent robotic arm to avoid collisions with the intelligent robotic arm and ensure that the intelligent robotic arm can safely and efficiently execute the massage strategy. When the intelligent robotic arm executes the motion planning, the various joints of the intelligent robotic arm rotate or move according to the instructions of the main control component, and the massage claws at the end operate according to the massage strategy. In this embodiment, the massage technique and massage intensity in the massage strategy are adjusted according to the severity of the lesion analyzed by the classification model. During the entire process, the sensors of the intelligent robotic arm will continuously feedback the actual motion status of the joints, and the main control component will make real-time adjustments based on this feedback information to ensure that the intelligent robotic arm can accurately follow the preset motion trajectory. After completing the massage strategy, the intelligent robotic arm can perform adjustment operations as needed and finally return to its initial position, waiting for the next task instruction.

[0129] Furthermore, in order to ensure smooth execution of the massage strategy and improve the effect of the massage operation, the above-mentioned relative position-based control of the massage component to move to the lesion site to execute the massage strategy may include:

[0130] Based on the relative position, the massage component is controlled to move to the center of the lesion area to execute the massage strategy.

[0131] It should be noted that in this embodiment, the center of the lesion is used as the optimal position for executing the massage strategy in the lesion. In a specific application scenario, after determining the type and severity of the lesion, the system begins to guide the intelligent robotic arm to perform precise positioning and massage operations. After the depth camera is aligned, the coordinates of the rectangular area (lesion site) detected by the infrared thermal imaging camera are combined with the three-dimensional spatial coordinates obtained by the depth camera to expand the two-dimensional image information into three-dimensional coordinates. arrive , and then calculate the center point of the lesion, which can be substituted into the formula:

[0132] ,in, The z coordinate represents the center point of the lesion. The intelligent robotic arm moves based on this coordinate, ensuring that the massage claws at the end are precisely positioned at the center of the lesion. The intelligent robotic arm then descends vertically, gradually lowering the z coordinate until it detects contact with the body (i.e., z = 0). At this point, the intelligent robotic arm implements the corresponding massage strategy based on the lesion type and severity output by the classification model. For example, the system can be configured to apply a gentle "kneading" action to areas with mild blood vessel blockage, while applying increased pressure and extending the massage time for more severe blockages. For areas with abnormal temperatures, the system prioritizes "heating" to help improve blood circulation. Furthermore, the intelligent robotic arm can continuously adjust massage intensity and technique based on real-time feedback to ensure effective operation and user comfort.

[0133] Finally, in this embodiment, an interactive user interface can be developed within the intelligent massage system to facilitate viewing recognition results, adjusting massage parameters, and providing feedback. A monitoring system can also be established to regularly monitor system performance and conduct safety checks to ensure stable operation. Necessary technical support and training services can also be provided to users. User feedback can also be collected to address issues encountered during use and generate personalized massage plans for users.

[0134] The intelligent massage method provided by the embodiment of the present invention includes obtaining a target heat distribution map of a target object; identifying and analyzing the target heat distribution map to obtain a lesion recognition result of the target object; obtaining depth image information corresponding to the lesion site based on the lesion recognition result; locating the lesion site based on the lesion recognition result and the depth image information, and controlling a massage component to massage the lesion site according to the positioning result. The present invention obtains a lesion recognition result by identifying and analyzing the target heat distribution map, obtains depth image information corresponding to the lesion site using the lesion recognition result, and then uses the lesion recognition result and the depth image information to locate the lesion site, and controls the massage component to massage the lesion site. This can improve the accuracy of identification and positioning, reduce manual operations, and improve massage efficiency.

[0135] In addition, the embodiment of the present invention obtains the planar position information of the lesion part of the target object by utilizing the recognition model obtained by training based on the deep learning algorithm, and determines the lesion category and the depth image information corresponding to the lesion part by utilizing the classification model obtained by training based on the convolutional neural network architecture of the residual network architecture. By combining the above-mentioned planar position information of the lesion part and the above-mentioned depth image information, the three-dimensional position positioning coordinates of the lesion part can be determined, and the lesion category of the lesion part determined by the classification model can obtain the corresponding massage strategy, thereby ensuring that the image-based AI recognition model can realize the recognition of the lesion of the target object and obtain the lesion part information; by limiting the features output by the initial convolution layer and the non-initial convolution layer, the model is allowed to reuse features between different levels, thereby improving the utilization efficiency of features; by comparing the temperature abnormality area features with the target morphological features Feature fusion analysis can identify the lesion site, which can avoid identifying the target normal fever area as the lesion site, thereby avoiding misjudgment of the recognition model and improving the accuracy of identifying the lesion site; the recognition model is trained using the training set and the validation set, and the comprehensive measurement of accuracy and recall rate indicators is used to determine whether the recognition model training is completed, which can ensure the smooth training of the recognition model and improve the recognition and extraction effect of the lesion site; the lesion site is extracted using the YOLOv8 target detection algorithm to ensure the accuracy of the extracted and identified lesion site, and at the same time facilitate the subsequent determination of the lesion category based on the identified lesion site; by determining the relative position between the position of the lesion site and the current position of the current massage component, the massage component can be accurately moved to the lesion site based on the relative position to execute the corresponding massage strategy, thereby improving the efficiency of massaging the lesion site.

[0136] An intelligent massage system provided by an embodiment of the present invention is introduced below. The intelligent massage system described below and the intelligent massage method described above can be referenced to each other.

[0137] Please refer to Figure 4 , Figure 4 A schematic structural diagram of an intelligent massage system provided by an embodiment of the present invention may include:

[0138] Target heat distribution acquisition component 10, massage component 20, depth image information acquisition component, memory 30 and processor 40;

[0139] The target heat distribution acquisition component 10 and the depth image information acquisition component are both in communication with the processor 40, and are used to send the acquired target heat distribution map and depth image information of the target object to the processor 40 respectively; the processor 40 is in communication with the massage component 20, and is used to control the massage component 20 to massage the lesion area;

[0140] The memory 30 is used to store computer programs, and the processor 40 is used to execute the computer programs to implement the steps of the above-mentioned intelligent massage method.

[0141] The memory 30 , the processor 40 , and the communication interface 50 all communicate with each other via a communication bus 51 .

[0142] In an embodiment of the present invention, the memory 30 is used to store one or more programs. The program may include program code, and the program code includes computer operating instructions. In an embodiment of the present application, the memory 30 may store programs for implementing the following functions:

[0143] Obtaining a target heat distribution map of a target object;

[0144] Based on the image AI recognition model, the target heat distribution map is identified and analyzed to obtain the lesion recognition results of the target object;

[0145] Obtaining depth image information corresponding to the lesion site based on the lesion recognition result;

[0146] The lesion part is located based on the lesion recognition result and the depth image information, and the massage component is controlled to massage the lesion part according to the positioning result.

[0147] In one possible implementation, the memory 30 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and applications required for at least one function, etc.; the data storage area may store data created during use.

[0148] In addition, the memory 30 may include a read-only memory and a random access memory, and provides instructions and data to the processor 40. A portion of the memory 30 may also include NVRAM. The memory 30 stores an operating system and operating instructions, executable modules, or data structures, or a subset or an extended set thereof. The operating instructions may include various operating instructions for implementing various operations. The operating system may include various system programs for implementing various basic tasks and processing hardware-based tasks.

[0149] The processor 40 may be a central processing unit (CPU), an application-specific integrated circuit, a digital signal processor, a field programmable gate array, or other programmable logic device. The processor 40 may be a microprocessor or any conventional processor. The processor 40 may call programs stored in the memory 30 .

[0150] The communication interface 50 may be an interface of a communication module, and is used to connect to other devices or systems.

[0151] Furthermore, in order to avoid contact with the target when obtaining the target heat distribution map and to ensure the accuracy of controlling the massage component to move to the corresponding lesion site, reference can be made to Figure 5 , Figure 5 This is a schematic diagram of the structure of another intelligent massage system provided by an embodiment of the present invention. The above-mentioned depth image information acquisition component is a depth camera 60;

[0152] The target heat distribution acquisition component 10 is an infrared thermal imaging camera 11, and the processor 40 includes an intelligent recognition component 41 and a main control component 42;

[0153] The intelligent recognition component 41 is respectively connected to the depth camera 60 and the infrared thermal imaging camera 11 for identifying the planar position information of the lesion site according to the acquired target heat distribution map, and determining the corresponding lesion category and the corresponding massage strategy;

[0154] The main control component 42 is respectively connected to the intelligent recognition component 41 and the massage component 20 for communicating with each other, and is used to receive the corresponding massage strategy determined by the intelligent recognition component 41 and control the massage component 20 to move to the corresponding position to execute the massage strategy.

[0155] It should be noted that in this embodiment, non-contact back detection is achieved through the infrared thermal imaging camera 11, which avoids interference and discomfort to the user. In addition, in order to ensure the convenience of detection, the infrared thermal imaging camera 11 and the depth camera 60 can be set on the same plane, which can ensure that the depth camera 60 and the infrared thermal imaging camera 11 share a set of coordinate systems, thereby improving the simplicity of calculation. Specifically, in this embodiment, the intelligent massage system as a whole can be set to consist of an infrared thermal imaging camera 11, an intelligent recognition component 41, an intelligent robotic arm 70, and a main control component 42. It can also include a depth camera 60, a massage component 20, a memory 30, a communication interface 50, and a communication bus 51 for completing mutual communication. It needs to be further explained that:

[0156] 1. Infrared thermal imaging camera 11: It uses a high-resolution infrared thermal imaging lens and is generally used to capture infrared thermal images of the human back, that is, the target heat distribution map. Since the back of the human body is blocked or has other lesions, it is often accompanied by abnormal temperatures. Therefore, the temperature difference changes reflect the blood flow and metabolism of the back, so as to preliminarily determine and locate the areas with blockages or other lesions.

[0157] 2. Intelligent recognition component 41: A deep learning algorithm is used to construct an intelligent recognition system. The system will receive the infrared thermal image captured by the infrared thermal imaging camera 11, automatically analyze and identify the infrared thermal image through the trained recognition model, and accurately mark the areas where blockages or other lesions exist. Then, the YOLOv8 target detection algorithm is used to extract the lesion site on the infrared thermal image. The classification model trained by the convolutional neural network based on the residual network architecture is used to classify the problem area to obtain the lesion category, and based on this, a personalized massage guidance plan, i.e., a massage strategy, is generated and transmitted to the main control component 42. Then, the depth camera 60 is used to generate depth image information. According to the coordinate position of the lesion site extracted by the YOLOv8 target detection algorithm, the position distance information between the intelligent robotic arm 70 and the lesion site is determined. Finally, the main control component 42 sends an instruction including the massage strategy and the movement path to the intelligent robotic arm 70, so that the massage component 20 moves to the target point.

[0158] 3. Main Control Component 42: The main control component 42 is primarily responsible for data exchange, data processing, and command issuance between various components. The infrared thermal imaging image captured by the infrared thermal imaging camera 11 can be initially analyzed and screened before being sent to the intelligent recognition component 41 for in-depth analysis and recognition. At this point, a communication connection can be established between the infrared thermal imaging camera 11 and the main control component 42. After determining the lesion area to be massaged, the position and distance information fed back by the depth camera 60 is used to locate the lesion area. Finally, instructions are sent to the intelligent robotic arm 70, allowing it to complete the massage according to the massage strategy.

[0159] 4. Intelligent robotic arm: It is connected to the infrared thermal imaging camera 11 through the FAKRA (a radio frequency connector system) interface. It is mainly responsible for receiving instructions from the main control component 42 and completing massage operations according to the massage strategy. A variety of massage techniques can be set, such as kneading, massaging, tapping, etc., which can relieve the diseased area.

[0160] During the entire massage process, the intelligent recognition component 41, the main control component 42, the infrared thermal imaging camera 11, the depth camera 60 and the intelligent robotic arm 70 cooperate with each other to form a complete set of lesion detection, classification and massage processes, which can be referred to Figure 6 , Figure 6 The following is a schematic diagram of the structural connection of an intelligent massage system provided by an embodiment of the present invention. After capturing an image, the infrared thermal imaging camera 11 first pre-processes the image and then transmits it to the intelligent recognition component 41. The intelligent recognition component 41 is obtained after model training. After determining the massage strategy and the coordinate points of the corresponding lesion site, the intelligent recognition component 41 can transmit the massage strategy and the data of the corresponding lesion site coordinate points to the main control component 42, so that the main control component 42 can issue instructions to control the intelligent robotic arm 70 to move to the position of the corresponding lesion site coordinate point and enable the massage component 20 to execute the corresponding massage strategy. The infrared thermal imaging camera 11 provides preliminary lesion site image acquisition, the classification model is responsible for classifying and rating the lesion site, the depth camera 60 provides accurate three-dimensional coordinate information, and the intelligent robotic arm 70 ultimately executes a personalized massage plan. This collaborative working method not only improves the accuracy of lesion identification and positioning, but also improves the targetedness of the massage, providing users with an intelligent and precise massage experience.

[0161] It should be further explained that Figure 4 The structure shown does not constitute a limitation on the intelligent massage system in the embodiment of the present application. In actual application, the intelligent massage system may include Figure 4 More or fewer components than shown, or combinations of certain components.

[0162] The computer-readable storage medium provided by an embodiment of the present invention is introduced below. The computer-readable storage medium described below and the intelligent massage method described above can be referenced to each other.

[0163] The present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the above-mentioned intelligent massage method are implemented.

[0164] The computer-readable storage medium may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc., which can store program codes.

[0165] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from the other embodiments. Reference can be made to the descriptions of the identical or similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and the relevant parts can be referred to the descriptions of the methods.

[0166] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0167] Finally, it should be noted that, in this document, relationships such as first and second, etc., are used solely to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0168] The above is a detailed introduction to the intelligent massage method, device, system and computer-readable storage medium provided by the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core ideas. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the contents of this specification should not be understood as limiting the present invention.

Claims

1. An intelligent massage method, characterized in that: include: Obtaining a target heat distribution map of a target object; Performing identification and analysis on the target heat distribution map to obtain a lesion identification result of the target object; Acquiring depth image information corresponding to the lesion site based on the lesion recognition result; The lesion site is located based on the lesion recognition result and the depth image information, and a massage component is controlled to massage the lesion site according to the positioning result.

2. The intelligent massage method according to claim 1, characterized in that: The recognition analysis of the target heat distribution map is implemented based on an image AI recognition model, which includes a recognition model and a classification model. The recognition analysis of the target heat distribution map to obtain the lesion recognition result of the target object includes: Analyzing the target heat distribution map using a recognition model trained based on a deep learning algorithm to identify the planar position information of the lesion site; Using a classification model obtained by training a convolutional neural network based on a residual network architecture, the lesion site is classified to determine the lesion category; Accordingly, obtaining depth image information corresponding to the lesion site based on the lesion recognition result includes: Based on the planar position information of the lesion part, acquiring the depth image information corresponding to the lesion part; Accordingly, locating the lesion site based on the lesion recognition result and the depth image information, and controlling the massage component to massage the lesion site according to the positioning result, includes: The lesion part is positioned based on the planar position information of the lesion part and the depth image information, and a massage component is controlled to massage the lesion part according to the positioning result.

3. The intelligent massage method according to claim 2, characterized in that: Positioning the lesion part based on the planar position information of the lesion part and the depth image information, and controlling a massage component to massage the lesion part according to the positioning result, including: determining a relative position between the lesion site and the massage component based on the planar position information of the lesion site and the depth image information; determining a corresponding massage strategy based on the lesion category; The massage component is controlled to move to the lesion site based on the relative position to execute the massage strategy.

4. The intelligent massage method according to claim 2, characterized in that: The classification model obtained by training a convolutional neural network based on a residual network architecture is used to classify the lesion site and determine the lesion category, including: Inputting the identified lesion image into the classification model, and outputting features of the lesion image through the convolution operation of each convolutional layer; wherein, if the current convolutional layer has a previous convolutional layer, the features output by the current convolutional layer are the sum of the output item of the current convolutional layer through the convolution operation and the input item of the previous convolutional layer; The features of the lesion area image are input into a fully connected layer, and the lesion category is determined according to the lesion category probability output by the fully connected layer.

5. The intelligent massage method according to claim 2, characterized in that: Before identifying the planar position information of the lesion site, the method further includes: Acquiring target morphological features of the target object; The method of analyzing the target heat distribution map using a recognition model trained based on a deep learning algorithm to identify the planar position information of the lesion site includes: Analyzing the target heat distribution map to extract features of abnormal temperature areas; Combining the temperature anomaly region characteristics with the target morphological characteristics to obtain a comprehensive feature vector; The comprehensive feature vector is input into the recognition model trained based on the deep learning algorithm to identify the planar position information of the lesion site.

6. The intelligent massage method according to claim 2, characterized in that: The process of training the recognition model based on the deep learning algorithm includes: Collecting multiple target heat distribution maps, and dividing the collected target heat distribution maps into a training set and a validation set; the target heat distribution maps in the training set are marked with lesion sites; Using the training set to train the recognition model, and using the trained recognition model to recognize the validation set to generate a recognition result; An index for comprehensively measuring the accuracy and recall rate is calculated based on the recognition result, and whether the recognition model is trained is determined based on the index for comprehensively measuring the accuracy and recall rate.

7. The intelligent massage method according to claim 2, characterized in that: The method of analyzing the target heat distribution map using a recognition model trained based on a deep learning algorithm to identify the planar position information of the lesion site includes: Analyzing and processing the target heat distribution map using the recognition model to mark the lesion site; The YOLOv8 target detection algorithm is used to identify and extract the planar position information of the lesion site.

8. An intelligent massage system, characterized in that: include: Target heat distribution acquisition component, massage component, depth image information acquisition component, memory and processor; The target heat distribution acquisition component and the depth image information acquisition component are both in communication with the processor, and are used to send the acquired target heat distribution map and depth image information of the target object to the processor respectively; The processor is in communication with the massage component to control the massage component to massage the lesion area; The memory is used to store a computer program, and the processor is used to execute the computer program to implement the steps of the intelligent massage method according to any one of claims 1 to 7.

9. The intelligent massage system according to claim 8, characterized in that: The depth image information acquisition component is a depth camera; The target heat distribution acquisition component is an infrared thermal imaging camera, and the processor includes an intelligent recognition component and a main control component; The intelligent recognition component is respectively connected to the depth camera and the infrared thermal imaging camera for identifying the planar position information of the lesion site according to the acquired target heat distribution map, and determining the corresponding lesion category and the corresponding massage strategy; The main control component is respectively connected to the intelligent recognition component and the massage component for communication, and is used to receive the corresponding massage strategy determined by the intelligent recognition component, and control the massage component to move to a corresponding position to execute the massage strategy.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the intelligent massage method according to any one of claims 1 to 7.

Images