Ultrasonic therapeutic apparatus and its regulating system based on artificial intelligence parameter regulation

Abstract

The present invention discloses an ultrasonic therapeutic instrument based on artificial intelligence parameter adjustment and its adjustment system, which relates to the field of ultrasonic beauty technology, including a therapeutic instrument body, the outer surface of which is connected to an ultrasonic head through a wire; the ultrasonic head includes: a shell, an ultrasonic module is fixed inside the shell; the ultrasonic module is used to emit high-frequency ultrasonic waves to stimulate subcutaneous tissue; a camera is fixed in the shell, used to shoot video data, and based on the video data, determine whether the ultrasonic module exceeds the treatment area, if it exceeds the treatment area, an early warning instruction is issued; a vibrator is fixed in the shell, and is used to vibrate to remind the user based on the early warning instruction. By real-time monitoring of the position of the ultrasonic module, if it is detected that the ultrasonic module exceeds the treatment area, the ultrasonic power is reduced or the work is stopped, so as to avoid damage to the skin part that does not need ultrasonic beauty.

Description

Technical Field

[0001] The present invention belongs to the technical field of ultrasonic beauty, and in particular relates to an ultrasonic therapeutic apparatus and an adjustment system thereof based on artificial intelligence parameter adjustment. Background Art

[0002] With the rapid development of beauty technology, ultrasound therapy technology has been widely used in the beauty field, especially in skin tightening, anti-aging and body sculpting. Ultrasonic therapy devices mainly use high-frequency sound waves to non-invasively stimulate the deep layers of the skin, promote collagen reorganization, and thus achieve the effect of tightening the skin and improving the appearance of the skin.

[0003] For example, a Chinese patent with publication number CN118059399A discloses an ultrasonic beauty instrument, comprising: a body, the body comprising a casing and a movement installed in the casing, the outer surface of the casing being arranged in a curved surface; a functional output head, the functional output head being installed in the casing and electrically connected to the movement, the functional output head comprising an ultrasonic unit and an electric pulse unit arranged around the ultrasonic unit, the ultrasonic unit having a first skin-touch surface, the electric pulse unit having a second skin-touch surface arranged around the first skin-touch surface, the first skin-touch surface and the second skin-touch surface both protruding from the outer surface of the casing; the electric pulse unit comprising a first electrode group and a second electrode group installed on the second skin-touch surface, the plurality of electrode sheets of the first electrode group being arranged along the circumference of the ultrasonic unit, and the plurality of electrode sheets of the second electrode group being arranged along the circumference of the first electrode group; the patent stimulates the deep layer of the skin by emitting ultrasound through the ultrasonic unit;

[0004] At present, ultrasound beauty treatment is mainly used for facial skin, especially cheek area. Before ultrasound beauty treatment, the patient's face needs to be checked, the ultrasound beauty area needs to be located, and marked on the patient's face. The doctor performs ultrasound beauty treatment based on the marks, which requires the doctor to have strong enough professional skills and concentrate on controlling the function output head with his hands. This method is because the current ultrasound therapy device cannot automatically adjust the ultrasound beauty parameters based on the patient's face. Even if the doctor has strong enough professional skills, mistakes will still occur when performing ultrasound beauty treatment on the patient's face after working for a long time, resulting in damage to the skin part that does not need ultrasound beauty treatment.

[0005] To this end, the present invention proposes an ultrasonic therapeutic apparatus and an adjustment system thereof based on artificial intelligence parameter adjustment. Summary of the invention

[0006] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes an ultrasonic therapeutic apparatus and its regulating system based on artificial intelligence parameter regulation, which improves the early warning intelligence and safety.

[0007] To achieve the above object, an ultrasonic therapeutic apparatus based on artificial intelligence parameter adjustment is provided according to an embodiment of the present invention, comprising a therapeutic apparatus body, an outer surface of which is connected to an ultrasonic head via a wire; the ultrasonic head comprises:

[0008] A shell, wherein an ultrasonic module is fixed inside the shell;

[0009] Ultrasonic module, used to emit high-frequency ultrasonic waves to stimulate subcutaneous tissue;

[0010] A camera is fixed in the housing and is used to capture video data and determine whether the ultrasound module is beyond the treatment area based on the video data. If it is beyond the treatment area, an early warning instruction is issued;

[0011] The vibrator is fixed in the shell and is used to vibrate to remind the user based on the early warning instruction.

[0012] Preferably, the method for determining whether the ultrasound module exceeds the treatment area based on video data includes:

[0013] Get the i-th frame in the video data, where i is an integer greater than 1;

[0014] Input i sets of frames into the trained anomaly recognition model and output the predicted label;

[0015] Based on the predicted label, it is determined whether the ultrasound module exceeds the treatment area.

[0016] Preferably, the training method of the anomaly recognition model is:

[0017] Pre-collect multiple groups of historical video data, obtain multiple groups of historical frames in the historical video data, detect each group of frames to obtain feature data of each group of frames, and generate corresponding actual annotations for each group of frame feature data;

[0018] When the ultrasound module in the frame exceeds the treatment area, the actual label is 0, when the ultrasound module in the frame is close to the treatment area, the actual label is 1, and when the ultrasound module in the frame is within the treatment area, the label is 2; each set of frame feature data is used as the input of the first machine learning model, the first machine learning model uses the predicted label of each set of frame feature data as the output, the actual label as the prediction target, and the minimization of the sum of the prediction accuracy of all frame feature data as the training target; wherein the calculation formula for the prediction accuracy is: k =(a k -w k ) 2 , where k is the number of the frame feature data, z k is the prediction accuracy, a kis the predicted annotation value corresponding to the k-th group of frame feature data, w k is the actual annotation corresponding to the kth group of frame feature data; the first machine learning model is trained until the sum of the prediction accuracies reaches convergence, the training is stopped and the trained first machine learning model is used as the abnormality recognition model, the first machine learning model is a naive Bayes model or a support vector machine model.

[0019] Preferably, the method for determining whether the ultrasound module exceeds the treatment area based on video data includes:

[0020] Get the i-th frame in the video data, where i is an integer greater than 1;

[0021] Input i sets of frames into the trained anomaly recognition model and output the predicted label;

[0022] Based on the predicted label, it is determined whether the ultrasound module exceeds the treatment area.

[0023] Preferably, if the predicted label is 0, the ultrasound module is judged to be beyond the treatment area, and an early warning instruction is issued, and the early warning level is level one;

[0024] If the predicted label is 1, the ultrasound module is judged to be close to the treatment area, and an early warning instruction is issued, and the early warning level is level 2;

[0025] If the predicted label is 2, it is determined that the ultrasound module is located within the treatment area and no warning instruction is issued.

[0026] Preferably, the vibrator includes two vibration modes;

[0027] Wherein, if the warning level is level one, the vibrator continues to vibrate;

[0028] If the warning level is level 2, the vibrator vibrates intermittently.

[0029] Preferably, there are at least four cameras, and the acquisition viewing angles of the four cameras are complementary.

[0030] Preferably, a reflector is fixed inside the housing, and a fill light is fixed at the bottom of the reflector.

[0031] An ultrasonic therapeutic apparatus adjustment system based on artificial intelligence parameter adjustment, which is used to adjust the ultrasonic therapeutic apparatus based on artificial intelligence parameter adjustment, and the adjustment system includes:

[0032] Video acquisition module, used to acquire video data;

[0033] The video processing module is used to obtain the i-th frame in the video data, where i is an integer greater than 1;

[0034] The anomaly recognition module is used to input the i-group frame images into the trained anomaly recognition model, output the predicted label, and determine whether to generate an early warning instruction based on the predicted label;

[0035] The control module is used to control the operation of the vibrator based on the early warning instruction, and control the ultrasonic module to reduce the ultrasonic power or stop working.

[0036] Preferably, the adjustment system further comprises:

[0037] The image processing module is used to obtain the frame image corresponding to the warning instruction that has not been generated, and obtain the grayscale values ​​of n pixels in the frame image, establish an analysis set with the n pixel grayscale values, and calculate the average value of the analysis set;

[0038] The first collection module is used to collect skin glossiness;

[0039] The second acquisition module is used to collect skin pressure values;

[0040] The skin recognition module is used to input the average value, glossiness and pressure value as skin condition data into the trained skin recognition model to obtain the ultrasonic power, which is used as the execution parameter of the ultrasonic module in the area where the current frame is located.

[0041] Preferably, the training method of the skin recognition model includes:

[0042] Pre-collecting g groups of historical skin condition data, and the corresponding ultrasonic power when collecting the historical skin condition data;

[0043] converting a set of historical skin condition data and the ultrasound power corresponding to the historical skin condition data into a set of feature vectors;

[0044] The feature vector is used as the input of the second machine learning model, and the second machine learning model uses the predicted ultrasound power for each set of historical skin condition data as the output, and minimizes the sum of the prediction accuracy of all ultrasound powers as the training goal; wherein the calculation formula for the prediction accuracy is: f c =(b c -m b ) 2 , where c is the number of the feature vector, f c is the prediction accuracy, b c is the predicted ultrasonic power value corresponding to the cth group of eigenvectors, m b is the actual ultrasonic power corresponding to the cth group of feature vectors; the second machine learning model is trained until the sum of the prediction accuracies reaches convergence, and the training is stopped, and the trained second machine learning model is used as the skin recognition model, and the second machine learning model is a naive Bayes model or a support vector machine model.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] The present invention monitors the position of the ultrasound module in real time. If it is detected that the ultrasound module is beyond the treatment area, the ultrasound power is reduced or the operation is stopped, thereby avoiding damage to the skin part that does not require ultrasound beauty treatment.

[0047] Secondly, the present invention collects the patient's facial data, and based on the patient's facial data, adjusts the ultrasonic power in real time as an execution parameter of the ultrasonic module, thereby providing personalized beauty parameters for different patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a schematic diagram of the overall structure of the ultrasonic therapeutic apparatus based on artificial intelligence parameter adjustment of the present invention;

[0049] Figure 2 It is a schematic diagram of the exploded structure of the ultrasonic head in the ultrasonic therapeutic device based on artificial intelligence parameter adjustment of the present invention;

[0050] Figure 3 It is a schematic diagram of the top view structure of the camera in the ultrasonic therapeutic apparatus based on artificial intelligence parameter adjustment of the present invention;

[0051] Figure 4 It is a schematic diagram of the simulated working of the camera in the ultrasonic therapeutic apparatus based on artificial intelligence parameter adjustment of the present invention;

[0052] Figure 5 The schematic diagram of the ultrasonic therapeutic apparatus adjustment system based on artificial intelligence parameter adjustment of the present invention is shown in FIG.

[0053] Description of the drawings: 1. Therapeutic device body; 2. Ultrasonic head; 21. Shell; 22. Ultrasonic module; 23. Camera; 24. Vibrator; 25. Fill light; 251. Reflector. DETAILED DESCRIPTION

[0054] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] Example 1

[0056] See also Figure 1 and Figure 2As shown, the camera 23 is fixed to the outer surface of the ultrasound module 22 and is used to capture video data. Based on the video data, it is determined whether the ultrasound module 22 exceeds the treatment area. If it exceeds the treatment area, an early warning instruction is issued, and the vibrator 24 works based on the early warning instruction. The vibrator 24 is fixed in the housing 21 and is used to emit vibration to remind the user. For example, Figure 4 As shown, the video data collected by the camera 23 includes a screen, point A is the edge of the ultrasound module 22, and point B is the marking line. If the collected video data is as follows Figure 4 As shown, it means that the ultrasound module 22 is about to exceed the treatment area.

[0057] For further information, see Figure 3 and Figure 4 As shown, at least four cameras 23 are provided, and the acquisition viewing angles of the four cameras 23 are complementary, the purpose is to Figure 4 In the figure, if point A moves toward point B again, the ultrasound module 22 will cover the marking line. At this time, if there is only one camera 23 collecting a video data perspective, misjudgment may occur, because the video data only includes the edge of the ultrasound module 22 and the skin, which is the same as the picture when the ultrasound module 22 is in the treatment area. However, if different perspectives are complementary, the ultrasound module 22 can be captured pressing on the marking line or exceeding the marking line.

[0058] Specifically, further, the training method of the anomaly recognition model is to obtain multiple groups of historical video data in an experimental environment, and obtain multiple groups of historical frame pictures in the historical video data, and the technical personnel in this field detect each group of frame pictures to obtain feature data of each group of frame pictures, and generate corresponding actual annotations for each group of frame picture feature data;

[0059] When the ultrasound module 22 exceeds the treatment area in the frame, the actual label is 0, when the ultrasound module 22 is close to the treatment area in the frame, the actual label is 1, and when the ultrasound module 22 is located in the treatment area in the frame, the label is 2; each set of frame feature data is used as the input of the first machine learning model, and the first machine learning model uses the predicted label of each set of frame feature data as the output, the actual label as the prediction target, and the minimization of the sum of the prediction accuracy of all frame feature data as the training target; wherein the calculation formula for the prediction accuracy is: k =(a k -w k ) 2 , where k is the number of the frame feature data, z k is the prediction accuracy, a k is the predicted annotation value corresponding to the k-th group of frame feature data, w kis the actual annotation corresponding to the kth group of frame feature data; the first machine learning model is trained until the sum of the prediction accuracies reaches convergence. It should be noted that the convergence standard is determined by technical personnel in this field according to actual conditions. The training is stopped and the first machine learning model obtained by training is used as an abnormality recognition model. The first machine learning model is a naive Bayes model or a support vector machine model.

[0060] More specifically, the method for judging whether the ultrasound module 22 exceeds the treatment area based on video data includes obtaining i groups of frames of video data, where i is an integer greater than 1; inputting the i groups of frames into a trained abnormality recognition model and outputting a predicted label; if the predicted label is 0, it is judged that the ultrasound module 22 exceeds the treatment area, and an early warning instruction is issued, and the early warning level is level one, that is, the ultrasound module 22 needs to stop working immediately, reduce the ultrasonic power or return to the treatment area; if the predicted label is 1, it is judged that the ultrasound module 22 is close to the treatment area, and an early warning instruction is issued, and the early warning level is level two, that is, the operator is reminded to pay attention; if the predicted label is 2, it is judged that the ultrasound module 22 is located in the treatment area, and no early warning instruction is issued, that is, it is in the instruction area and no early warning is required.

[0061] Further, the vibrator 24 includes two vibration modes;

[0062] If the warning level is level one, the vibrator 24 continues to vibrate, that is, continues to remind the operator until the ultrasound module 22 returns to the treatment area;

[0063] If the warning level is level two, the vibrator 24 will vibrate intermittently to intermittently remind the operator so that the operator can slow down the treatment speed at the edge of the treatment area to avoid the warning level being raised to level one.

[0064] It is worth mentioning that see Figure 2 As shown, a reflector 251 is fixed inside the housing 21, and a fill light 25 is fixed at the bottom of the reflector 251. In this embodiment, the reflector 251 is diffusely reflective. The purpose is that since the camera 23 is inside the housing 21, when working, there is no external light source provided to the camera 23, and the fill light 25 illuminates the inside of the housing 21 through diffuse reflection of the reflector 251, thereby avoiding overexposure of the camera 23 when shooting, and ensuring the stability of video data acquisition.

[0065] Example 2

[0066] Based on Example 1, refer to Figure 1 , Figure 2 and Figure 5 As shown, the ultrasonic therapeutic apparatus adjustment system based on artificial intelligence parameter adjustment is used to adjust the above-mentioned ultrasonic therapeutic apparatus based on artificial intelligence parameter adjustment, and the adjustment system includes:

[0067] A video acquisition module, used to acquire video data. In this embodiment, the video acquisition module is implemented based on the camera 23;

[0068] The video processing module is used to obtain the i-th frame in the video data, where i is an integer greater than 1;

[0069] The abnormality recognition module is used to input i groups of frames into the trained abnormality recognition model, output a prediction label, and determine whether to generate a warning instruction based on the prediction label. The training method of the abnormality recognition model is to obtain multiple groups of historical video data in an experimental environment, and obtain multiple groups of historical frames in the historical video data. The technicians in this field detect each group of frames to obtain feature data of each group of frames, and generate corresponding actual annotations for each group of frame feature data;

[0070] When the ultrasound module 22 exceeds the treatment area in the frame, the actual label is 0, when the ultrasound module 22 is close to the treatment area in the frame, the actual label is 1, and when the ultrasound module 22 is located in the treatment area in the frame, the label is 2; each set of frame feature data is used as the input of the first machine learning model, and the first machine learning model uses the predicted label of each set of frame feature data as the output, the actual label as the prediction target, and the minimization of the sum of the prediction accuracy of all frame feature data as the training target; wherein the calculation formula for the prediction accuracy is: k =(a k -w k ) 2 , where k is the number of the frame feature data, z k is the prediction accuracy, a k is the predicted annotation value corresponding to the k-th group of frame feature data, w k is the actual annotation corresponding to the k-th group of frame feature data; the first machine learning model is trained until the sum of the prediction accuracies reaches convergence. It should be noted that the convergence standard is determined by those skilled in the art according to the actual situation, the training is stopped and the first machine learning model obtained by training is used as the abnormality recognition model, and the first machine learning model is a naive Bayes model or a support vector machine model;

[0071] The control module is used to control the operation of the vibrator 24 based on the early warning instruction, and control the ultrasound module 22 to reduce the ultrasound power or stop working, that is, to remind the operator, or actively control the ultrasound module 22; the purpose of this embodiment is to automatically determine whether the ultrasound module 22 is in abnormal operation, that is, whether it exceeds the treatment area, through artificial intelligence. If it is in abnormal operation, the ultrasound power of the ultrasound module 22 can be actively controlled, and the operator can be reminded.

[0072] Furthermore, the regulating system also includes:

[0073] The image processing module is used to obtain the frame image corresponding to the warning instruction that has not been generated, and obtain the grayscale values ​​of n pixels in the frame image, where n is a value greater than 1. The n pixel grayscale values ​​establish an analysis set, and calculate the average value of the analysis set. This embodiment can reflect the patient's skin color and whether there are spots or blackheads on the patient's skin by analyzing the grayscale values ​​in the frame image, because spots or blackheads on the skin will affect the average value of the n pixel grayscale values;

[0074] The first acquisition module is used to collect skin glossiness. In this embodiment, the first acquisition module uses a photometer to analyze the skin glossiness to reflect the skin type, such as oily or dry. Different beauty parameters are used for different skin types.

[0075] The second acquisition module is used to collect skin pressure values. In this embodiment, the second acquisition module is a pressure sensor or a strain sensor, which is not specifically limited here. By collecting the skin pressure value, the elasticity of the skin can be reflected, and the elasticity can be used as an important reference indicator for ultrasonic beauty treatment;

[0076] The skin recognition module is used to input the average value, glossiness and pressure value as skin condition data into the trained skin recognition model to obtain the ultrasonic power, and the ultrasonic power is used as the execution parameter of the ultrasonic module 22 in the area where the current frame is located; illustratively, taking the first acquisition module and the second acquisition module installed on the outer surface of the ultrasonic module 22 as an example, the skin condition data currently collected is the next beauty area, so those skilled in the art can set the parameters to adjust the execution time, as well as the standardized beauty path, that is, the moving path and moving speed of the ultrasonic head 2, so that the currently predicted ultrasonic power can be accurately applied to the collected skin part; it can also be based on hardware such as a photoelectric encoder or an accelerometer. Taking a photoelectric encoder as an example, a photoelectric encoder or an accelerometer is installed on the ultrasonic head 2. The photoelectric encoder is a sensor that can measure position changes by sensing rotation or linear motion, and using this position as a reference position, the skin area adjustment parameters can be accurately detected, which will not be elaborated on here.

[0077] Specifically, the training method of the skin recognition model includes:

[0078] G groups of historical skin condition data are collected under the experimental environment. The historical skin condition data include historical average values, historical glossiness and historical pressure values, and the corresponding ultrasonic power when the skin condition data is collected. It is worth mentioning that the corresponding ultrasonic power is, and the technicians in this field treat the same historical skin condition data with multiple groups of different ultrasonic powers under the experimental environment, and observe the postoperative effects, and select a group of ultrasonic powers with the best postoperative effects as the ultrasonic power corresponding to the historical skin condition data;

[0079] converting a set of skin condition data and the ultrasound power corresponding to the skin condition data into a set of feature vectors;

[0080] The feature vector is used as the input of the second machine learning model. The second machine learning model uses the predicted ultrasound power for each set of skin condition data as output, and minimizes the sum of the prediction accuracy of all ultrasound powers as the training goal; wherein the calculation formula for the prediction accuracy is: f c =(b c -m b ) 2 , where c is the number of the feature vector, f c is the prediction accuracy, b c is the predicted ultrasonic power value corresponding to the cth group of eigenvectors, m b is the actual ultrasonic power corresponding to the cth group of feature vectors; the second machine learning model is trained until the sum of the prediction accuracies reaches convergence and the training is stopped. It should be noted that the convergence criterion is determined by technical personnel in this field according to actual conditions. The training is stopped and the trained second machine learning model is used as the skin recognition model. The second machine learning model is a naive Bayes model or a support vector machine model.

[0081] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An ultrasonic therapeutic device based on artificial intelligence parameter adjustment, comprising a therapeutic device body (1), characterized in that: The outer surface of the therapeutic device body (1) is connected to an ultrasonic head (2) via a wire; the ultrasonic head (2) comprises: A housing (21), wherein an ultrasonic module (22) is fixed inside the housing (21); An ultrasound module (22) for emitting high-frequency ultrasound to stimulate subcutaneous tissue; A camera (23) is fixed in the housing (21) and is used to capture video data and determine whether the ultrasound module (22) exceeds the treatment area based on the video data. If the ultrasound module (22) exceeds the treatment area, an early warning instruction is issued; A vibrator (24), fixed in the housing (21), for vibrating to remind a user based on the early warning instruction; The method for determining whether the ultrasound module (22) exceeds the treatment area based on video data comprises: Get the i-th frame in the video data, where i is an integer greater than 1; Input i sets of frames into the trained anomaly recognition model and output the predicted label; Based on the predicted label, determining whether the ultrasound module (22) is beyond the treatment area; The training method of the anomaly recognition model is: Pre-collect multiple groups of historical video data, obtain multiple groups of historical frames in the historical video data, detect each group of frames to obtain feature data of each group of frames, and generate corresponding actual annotations for each group of frame feature data; When the ultrasound module (22) in the frame exceeds the treatment area, the actual label is 0, when the ultrasound module (22) in the frame is close to the treatment area, the actual label is 1, and when the ultrasound module (22) in the frame is located in the treatment area, the label is 2; each set of frame feature data is used as input to a first machine learning model, the first machine learning model uses the predicted label of each set of frame feature data as output, the actual label as a prediction target, and minimizing the sum of the prediction accuracies of all frame feature data as a training target; the first machine learning model is trained until the sum of the prediction accuracies converges, the training is stopped, and the first machine learning model obtained by training is used as an abnormality recognition model, the first machine learning model is a naive Bayes model or a support vector machine model; If the predicted label is 0, it is determined that the ultrasound module (22) is beyond the treatment area, and a warning instruction is issued, and the warning level is level one; If the predicted label is 1, it is determined that the ultrasound module (22) is close to the treatment area, and an early warning instruction is issued, and the early warning level is level 2; If the predicted label is 2, it is determined that the ultrasound module (22) is located in the treatment area and no warning instruction is issued; The vibrator (24) includes two vibration modes; Wherein, if the warning level is level one, the vibrator (24) continues to vibrate; If the warning level is level 2, the vibrator (24) vibrates intermittently.

2. The ultrasonic therapeutic apparatus based on artificial intelligence parameter adjustment according to claim 1, characterized in that: At least four cameras (23) are provided, and the acquisition viewing angles of the four cameras (23) are complementary.

3. The ultrasonic therapeutic apparatus based on artificial intelligence parameter adjustment according to claim 1, characterized in that: A reflector (251) is fixed inside the housing (21), and a fill light (25) is fixed at the bottom of the reflector (251).

4. An ultrasonic therapeutic apparatus adjustment system based on artificial intelligence parameter adjustment, which is used to adjust the ultrasonic therapeutic apparatus based on artificial intelligence parameter adjustment according to any one of claims 1 to 3, the adjustment system comprising: Video acquisition module, used to acquire video data; The video processing module is used to obtain the i-th frame in the video data, where i is an integer greater than 1; The anomaly recognition module is used to input the i-group frame images into the trained anomaly recognition model, output the predicted label, and determine whether to generate an early warning instruction based on the predicted label; A control module is used to control the vibrator (24) to work based on the early warning instruction, and to control the ultrasonic module (22) to reduce the ultrasonic power or stop working.

5. The ultrasonic therapeutic apparatus adjustment system based on artificial intelligence parameter adjustment according to claim 4 is characterized in that: The regulating system also includes: The image processing module is used to obtain the frame image corresponding to the warning instruction that has not been generated, and obtain the grayscale values ​​of n pixels in the frame image, establish an analysis set based on the n pixel grayscale values, and calculate the average value of the analysis set; The first collection module is used to collect skin glossiness; The second acquisition module is used to collect skin pressure values; The skin recognition module is used to input the average value, glossiness and pressure value as skin condition data into the trained skin recognition model to obtain the ultrasonic power, and the ultrasonic power is used as the execution parameter of the ultrasonic module (22) in the area where the current frame image is located.

6. The ultrasonic therapeutic apparatus adjustment system based on artificial intelligence parameter adjustment according to claim 4 is characterized in that: The training method of the skin recognition model includes: Pre-collecting g groups of historical skin condition data, and the corresponding ultrasonic power when collecting the historical skin condition data; converting a set of historical skin condition data and the ultrasound power corresponding to the historical skin condition data into a set of feature vectors; The feature vector is used as the input of the second machine learning model, and the second machine learning model takes the predicted ultrasonic power for each set of historical skin condition data as the output, and takes minimizing the sum of the prediction accuracies of all ultrasonic powers as the training goal; the second machine learning model is trained until the sum of the prediction accuracies reaches convergence, and the training is stopped and the trained second machine learning model is used as the skin recognition model. The second machine learning model is a naive Bayes model or a support vector machine model.

Images