Magnetic positioning control system for driving capsule endoscope in vitro

Abstract

The invention discloses a magnetic positioning control system for driving a capsule endoscope in vitro, which comprises a guider for driving the capsule endoscope, a magnetic positioning processor connected with the guider, and an upper computer connected with the magnetic positioning processor, the first permanent magnet is magnetically connected with a second permanent magnet arranged in the capsule endoscope, and the guider drives the capsule endoscope to move and / or rotate. The capsule endoscope handheld active control guider based on the external magnetic field driving principle can flexibly control the movement path and speed of the capsule endoscope, the accuracy and efficiency in the examination process are improved, and therefore the accuracy and efficiency of endoscopic examination are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of capsule endoscopes, and in particular to a magnetic positioning control system for driving an in vitro capsule endoscope. Background Art

[0002] Traditionally, the diagnosis of digestive tract diseases usually uses a push-pull endoscope for examination. The doctor inserts a hose with a camera device at the end into the patient's body from the mouth or anus, and uses the external push-pull action and the image sent back by the camera device to find and identify the lesion. Although this examination method is efficient and low-cost, the examination process is painful, prone to cross-infection, and the small intestine cannot be examined. The rapid development of passive wireless capsule endoscopes in the past decade has overcome some of the shortcomings of push-pull endoscopes. However, the passive capsule endoscope examination process is not subject to physician intervention, so there are also some problems, such as the inability to actively select the content of the gastrointestinal examination, easy retention in the human body, and failure before reaching the predicted lesion area. At the same time, the physician cannot actively select the observation site and the duration of continuous observation, which is easy to miss key image information, resulting in misdetection and missed detection of gastrointestinal diseases. This also poses great difficulties for the realization of live detection, targeted quantitative drug release, and minimally invasive surgical treatment in the future capsule robot endoscope medical examination and treatment process. In order to improve the accuracy and efficiency of gastrointestinal disease examinations and provide a theoretical and practical basis for future medical diagnosis and treatment, it is particularly important to achieve active and precise control of capsule endoscope robots. Active control of their forward, backward, anchoring and stable rotation in the gastrointestinal tract has become an important research direction. Institutions in various countries have actively studied a variety of active control schemes for capsule endoscopes, including shape memory alloy drive, spiral drive, double spiral drive, electrical stimulation drive principle, motor drive and hydraulic drive. These schemes all hope to integrate the drive into the endoscope capsule through miniaturization, which is bound to greatly increase the complexity and production cost of disposable capsules, increase inspection costs, and reduce system reliability.

[0003] In existing research, most electromagnetic propulsion systems change the position of permanent magnets through mechanical transmission mechanisms. The magnets inside the capsule will change their own motion forms according to the different positions of the external permanent magnets, thereby achieving control of the capsule. This system has six coupling axes, which can realize multi-degree-of-freedom movements of the capsule in the gastrointestinal tract, such as moving horizontally or longitudinally and rotating around its vertical or longitudinal axis. The position control mode of the input pulse signal is integrated into the human-machine interface to guide the capsule to move accurately. The above method can realize multi-degree-of-freedom movement of the capsule in the gastrointestinal tract to a certain extent by using an external permanent magnet to drive the capsule, and improve the endurance of the capsule. However, this method requires the design of a complex permanent magnet drive system, combined with mechanical mechanisms and electric transmission devices, which greatly increases the complexity and cost of the control process, making it difficult to achieve real-time and accurate control of the capsule. Summary of the invention

[0004] The purpose of the present invention is to solve the above technical problems and provide a magnetic positioning control system for driving an in vitro capsule endoscope.

[0005] One of the purposes of the present invention is to improve the accuracy and efficiency of endoscopic examinations by using machine learning or deep learning to establish a regression model using the collected spatial magnetic field information to predict the position and shape of the permanent magnet and improve the positioning range. Using a handheld controller with magnetic sensing and wireless transmission functions, the capsule endoscope can be positioned and moved more accurately, thereby improving the accuracy and efficiency of the inspection process;

[0006] One of the purposes of the present invention is to enhance the doctor's control over the examination process, so that the doctor can more flexibly control the movement path and speed of the capsule endoscope to conduct a more in-depth and detailed examination of a specific area;

[0007] One of the purposes of the present invention is to improve patient experience and safety. Compared with the traditional peristaltic drive method, magnetic control can reduce the stagnation and instability of the capsule in the digestive tract, thereby improving the comfort and safety of patients;

[0008] One of the purposes of the present invention is to expand the scope of application of endoscopic examination. The magnetic control technology can make it easier for the capsule endoscope to pass through the complex structures in the digestive tract, thereby expanding the scope of application of endoscopic examination, including deeper tissues and organs.

[0009] The technical solution adopted by the present invention to solve the above-mentioned technical problems is: a magnetic positioning control system for driving an extracorporeal capsule endoscope, comprising a guide for driving the capsule endoscope, a magnetic positioning processor connected to the guide, and a host computer connected to the magnetic positioning processor. A first permanent magnet is provided in the guide, and the first permanent magnet is magnetically connected to a second permanent magnet provided in the capsule endoscope. The guide drives the capsule endoscope to move and / or rotate.

[0010] Preferably, a stepper motor is provided in the guide, a rotating shaft is provided on the stepper motor, and the first permanent magnet is provided on the rotating shaft. The rotation of the stepper motor drives the rotating shaft to rotate, and the rotation of the rotating shaft drives the first permanent magnet to rotate. The rotation of the first permanent magnet forms a steady-state magnetic field module, and the steady-state magnetic field module controls the second permanent magnet to drive the capsule endoscope to move and / or rotate.

[0011] Preferably, a processing module is provided in the guide, and the processing module is respectively connected to the stepping motor and the magnetic positioning processor.

[0012] Preferably, the guide is provided with control buttons and a display screen, and both the control buttons and the display screen are connected to the processing module.

[0013] Preferably, one end of the magnetic positioning processor is connected to a host computer for feeding back the 6-dimensional posture information of the capsule endoscope to the guide, and one end of the magnetic positioning processor is connected to a magnetic sensor for detecting the magnetic field strength. The host computer is provided with an MP-KAN architecture for receiving the magnetic field strength information sent by the magnetic sensor to the magnetic positioning processor and calculating the 6-dimensional posture information of the capsule endoscope based on the magnetic field strength information.

[0014] As a preferred embodiment, the MP-KAN architecture includes an input layer, a hidden layer, and an output layer; the magnetic sensor collects the original three-component magnetic field intensity in space and converts it into text information, and the host computer extracts the sensor data from the data text to construct the overall feature sequence of the sensor [Mag0 X ,Mag0 Y ,Mag0 Z ···Magn Z ] as the input layer, where the parameter Magn Z It represents the magnetic field strength of the nth sensor in the Z-axis direction of space; when the patient's position is not moving, the X position is always fixed, and the magnet position [Y, Z] is selected as the output layer.

[0015] Preferably, the MP-KAN architecture training method is also included, and the training method comprises the following steps:

[0016] Step 1: Use the host computer to quickly extract the target data in the text file, and then reconstruct the data into a 1×L structure, where L is the number of all sensor channels and the position of the magnet; by performing N experiments on each point in the plane, a data matrix of size 153×N×L can be obtained as the training data of the MP-KAN model;

[0017] Step 2: Set the activation function φ(x) = f(f) + B_spline(x), where f(x) represents the expression of f(x) as follows:

[0018] Where p represents the input dimension, q represents the output dimension, and x p represents the pth input parameter, φ q,p represents a related function with trainable parameters, Φ q Represents a one-dimensional function related to the output dimension;

[0019] B_spline(x) is the spline function that will be fitted to the curve through a piecewise continuous multi-segment curve, and each activation function will be initialized to B_spline(x)≈0;

[0020] The expression of B_spline(x) is as follows:

[0021] B_=∑;

[0022] Among them, b i As a trainable parameter, i is an intermediate variable;

[0023] According to the input activation function, the grid is updated in real time every twenty training steps. The grid interval is set to G = 3. At the node, for a network structure of l+1 layers, the activation value of the neuron will be the sum of all the incoming activation values. A general l-layer KAN network will be expressed as:

[0024] KAN(x)=(Φ L-1 …Φ L-2 …Φ1…Φ0)x;

[0025] Where Φ1 represents the function matrix corresponding to the lth KAN layer;

[0026] Step 3: Use the L1 norm of the activation function and add additional entropy regularization to achieve model sparsification; combined with the structure of the input layer and the output layer, the L1 norm of Φ is defined as the sum of the L1 norms of all activation functions, and its entropy is S(Φ), which is expressed as follows:

[0027]

[0028] The overall training goal L total is the predicted loss L pred Add L1 and entropy regularization to the KAN layer;

[0029]

[0030] Among them, L pred The root mean square error between the predicted position and the actual position is used to control the size of the overall regularization. By training the MP-KAN architecture, the MP-KAN architecture can be more accurate in calculating the 6-dimensional posture information of the capsule endoscope.

[0031] Preferably, in step 2, the data collected by the magnetic positioning system sometimes has the problem of gradient disappearance and the problem that the negative part gradient is zero when the relu function is selected as the activation function. The silu function is selected as the basis function f(x), and the expression of f(x) is as follows:

[0032]

[0033] This not only ensures the nonlinearity and continuous differentiability of the activation function, but also makes the activation function defined in the range from negative infinity to positive infinity.

[0034] As a preference, in the three-layer structure of MP-KAN, the number of neurons in the hidden layer is set to 5, and the number of neurons in the input layer and the output layer are set to 30 and 2 respectively. The number of spline curves is set to 3, the number of grid intervals is set to 3, and LBFGS is used as the optimizer for 100 epochs training.

[0035] Preferably, the magnetic positioning processor is provided with a pin with ADC sampling function, and the pin is connected to the magnetic sensor.

[0036] The beneficial effects of the present invention are as follows: the handheld active control guide for capsule endoscope based on the external magnetic field driving principle can flexibly control the movement path and speed of the capsule endoscope, improve the accuracy and efficiency during the inspection process, and thus improve the accuracy and efficiency of endoscopic inspection. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a system flow chart of the present invention.

[0038] Figure 2 It is a schematic diagram of magnetic field distribution.

[0039] Figure 3 It is a schematic diagram of the variation of the magnetic flux density in the XY direction between the first permanent magnet and the second permanent magnet with the distance.

[0040] Figure 4 It is a schematic diagram of the variation of the magnetic flux density in the XZ direction between the first permanent magnet and the second permanent magnet with the distance.

[0041] Figure 5 This is a simulation diagram of the magnetic field strength principle after the first permanent magnet is rearranged and combined using the Halbach array.

[0042] Figure 6 Schematic diagram of the magnetization direction of the first permanent magnet.

[0043] Figure 7 It is a cross-sectional view when the first permanent magnet rotates.

[0044] Figure 8 It is a schematic diagram of the magnetic field flux density.

[0045] Fig. 9 It is a schematic diagram of the magnetic field flux density in the other direction.

[0046] Fig.10 It is a schematic diagram of the structure of the guide.

[0047] Fig.11 This is a schematic diagram of the first type of first permanent magnet driving the second permanent magnet to rotate.

[0048] Fig.12 This is a schematic diagram of the second type of first permanent magnet driving the second permanent magnet to rotate.

[0049] Fig.13 This is a schematic diagram of the third type in which the first permanent magnet drives the second permanent magnet to rotate.

[0050] Fig.14 This is a schematic diagram of the fourth type of first permanent magnet driving the second permanent magnet to rotate.

[0051] Fig.15 This is a schematic diagram of the MP-KAN architecture.

[0052] In the figure: 1. guide, 11. stepper motor, 12. rotating shaft, 13. first permanent magnet, 14. control button, 15. processing module, 16. display screen, 2. magnetic positioning processor, 3. host computer, 4. magnetic sensor, 5. second permanent magnet. DETAILED DESCRIPTION

[0053] The present invention is further described below in conjunction with the accompanying drawings and implementation modes.

[0054] like Figure 1-15 As shown, the present invention is a magnetic positioning control system for driving an in vitro capsule endoscope, including a guide 1 for driving the capsule endoscope, a magnetic positioning processor 2 connected to the guide 1, and a host computer 3 connected to the magnetic positioning processor 2. A first permanent magnet 13 is provided in the guide 1, and the first permanent magnet 13 is magnetically connected to a second permanent magnet 5 provided in the capsule endoscope. The guide 1 drives the capsule endoscope to move and / or rotate. The magnetic positioning processor 2 is an STM32 MCU. The first permanent magnet 13 is a neodymium iron boron permanent magnet. A cubic neodymium iron boron permanent magnet with a size of 50mm*50mm*30mm is selected as the magnetic source of the guide 1, with a surface magnetic flux density of 3900Gauss, axial magnetization, and a rotating magnetic field. As shown Figure 2 As shown, the distribution of the magnetic field generated by the permanent magnet in space is depicted, where the streamlines represent the magnetic lines of force and the background color represents the distribution of the magnetic field. The denser the magnetic lines of force, the greater the magnetic field intensity.

[0055] like Figure 3 and 4As shown, when the distance between the first permanent magnet 13 on the guide 1 and the second permanent magnet 5 built into the capsule endoscope is 30mm, the magnetic flux density at the midpoint can still reach 3000Gauss, which is sufficient to establish a stable magnetic connection to ensure stable control of the capsule movement. When the magnetic flux density is effectively transmitted between the two permanent magnets, the distance of 30mm requires higher technical operation requirements. The Halbach array is used to rearrange and combine the external first permanent magnet 13. The Halbach array is composed of trapezoidal first permanent magnets 13, each of which has a height of 30mm and a thickness of 20mm. Through this arrangement, the Halbach array can enhance the magnetic field strength in a unit direction. Figure 6 It is a projection diagram on the xy plane, and the arrow indicates the direction of the magnetization intensity M. Figure 5 , 9 , 10 show the principle simulation results of the Halbach array. The simulation results show that the rotor rearranged by the Halbach array can still provide a magnetic flux density of up to 0.3T when it is 50mm away from the second permanent magnet 5 built in the capsule, thereby effectively controlling the second permanent magnet 5 built in the capsule. By optimizing the design of the Halbach array, we can not only increase the effective distance between the rotor and the capsule, but also reduce the volume of the rotor to a certain extent and improve the efficiency of the stepper motor 11 of the guide 1. By optimizing the design of the first permanent magnet 13 using the Halbach array, the magnetic flux density transfer capacity is effectively improved. The simulation results verify the effectiveness of the design in enhancing the magnetic field strength and extending the control distance. Future research can further optimize the Halbach array structure to achieve higher magnetic field strength and longer control distance, thereby improving the overall performance of the handheld device.

[0056] A stepper motor 11 is provided in the guide 1, and a rotating shaft 12 is provided on the stepper motor 11. A first permanent magnet 13 is provided on the rotating shaft 12. The rotation of the stepper motor 11 drives the rotating shaft 12 to rotate, and the rotation of the rotating shaft 12 drives the first permanent magnet 13 to rotate. The rotation of the first permanent magnet 13 forms a steady-state magnetic field module, and the steady-state magnetic field module controls the second permanent magnet 5 to drive the capsule endoscope to move and / or rotate.

[0057] The guide 1 is provided with a processing module 15, which is respectively connected to the stepper motor 11 and the magnetic positioning processor 2. The processing module 15 is a single chip microcomputer, which adopts STM32F103C8T6.

[0058] The guide 1 is provided with a control button 14 and a display screen 16, and both the control button 14 and the display screen 16 are connected to the processing module 15. The single-chip microcomputer STM32F103C8T6 controls the start / stop and rotation speed of the stepper motor through the driver chip ULN2003 of the stepper motor 11. The rotation direction of the stepper motor 11 is controlled by switching the high and low levels of the direction control end of the driving circuit. The rotation speed of the stepper motor 11 varies within the range of 0.1 to 10.0 r / s, and the rotation speed value is displayed by the LED digital tube. The four control buttons respectively control the start / stop, acceleration, deceleration, and clockwise / counterclockwise rotation of the stepper motor 11.

[0059] One end of the magnetic positioning processor 2 is connected to a host computer 3 for feeding back the capsule endoscope 6-dimensional posture information to the guide 1, and one end of the magnetic positioning processor 2 is connected to a magnetic sensor 4 for detecting the magnetic field strength. The host computer 3 is provided with an MP-KAN architecture for receiving the magnetic field strength information sent by the magnetic sensor 4 to the magnetic positioning processor 2 and calculating the capsule endoscope 6-dimensional posture information according to the magnetic field strength information. Assume that 12 magnetic sensors 4 are evenly installed around the outside of the human body and connected to the ADC function pin of the STM32 MCU.

[0060] The magnetic positioning processor 2 is provided with a pin with an ADC sampling function, and the pin is connected to the magnetic sensor 4 .

[0061] The MP-KAN architecture includes an input layer, a hidden layer, and an output layer; the magnetic sensor 4 collects the original spatial three-component magnetic field intensity and converts it into text information, and the upper computer 3 extracts the sensor data from the data text to construct the overall feature sequence of the sensor as the input layer.

[0062] Also included is a MP-KAN architecture training method, the training method comprising the following steps:

[0063] Step 1: Use the host computer 3 to quickly extract the target data in the text file, and then reconstruct the data into a 1×L structure, where L is the number of all sensor channels and the position of the magnet; by performing N experiments on each point in the plane, a data matrix of size (153×N)×L can be finally obtained as the training data of the MP-KAN model;

[0064] Step 2: Set the activation function φ(x) = f(x) + B_spline(x), where f(f) represents the expression of f(x) as follows:

[0065] Where p represents the input dimension, q represents the output dimension, and x p represents the pth input parameter, φ q,p represents a related function with trainable parameters, Φ qRepresents a one-dimensional function related to the output dimension;

[0066] B_spline(x) is the spline function that will be fitted to the curve through a piecewise continuous multi-segment curve, and each activation function will be initialized to B_spline(x)≈0;

[0067] The expression of B_spline(x) is as follows:

[0068] B_spline(x)=Σ i b i B i (x);

[0069] Among them, b i As a trainable parameter, i is an intermediate variable;

[0070] According to the input activation function, the grid is updated in real time every twenty training steps. The grid interval is set to G = 3. At the node, for a network structure of l+1 layers, the activation value of the neuron will be the sum of all the incoming activation values. A general l-layer KAN network will be expressed as:

[0071] KAN(x)=(Φ L-1 …Φ L-2 …Φ1…Φ0)x;

[0072] Where Φ1 represents the function matrix corresponding to the lth KAN layer;

[0073] Step 3: Use the L1 norm of the activation function and add additional entropy regularization to achieve model sparsification; combined with the structure of the input layer and the output layer, the L1 norm of Φ is defined as the sum of the L1 norms of all activation functions, and its entropy is S(Φ), which is expressed as follows:

[0074]

[0075] The overall training goal L total is the predicted loss L pred Add L1 and entropy regularization to the KAN layer;

[0076]

[0077] Among them, L pred is the root mean square error between the predicted position and the actual position, and λ is used to control the size of the overall regularization.

[0078] In step 2, the data collected by the magnetic positioning system sometimes has the problem of gradient disappearance and the problem that the negative part of the gradient is zero when the relu function is selected as the activation function. The silu function is selected as the basis function f(x), and the expression of f(x) is as follows:

[0079]

[0080] In the three-layer structure of MP-KAN, the number of neurons in the hidden layer is set to 5, and the number of neurons in the input layer and output layer is set to 30 and 2 respectively.

[0081] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0082] It should be noted that when an element is referred to as being "fixed on" or "provided on" another element, it can be directly located on the other element or indirectly located on the other element. When an element is referred to as being "connected to" another element, it can be directly connected to the other element or indirectly connected to the other element.

[0083] It should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside" and "outside" etc., indicating directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings and are only for the convenience of describing the present invention, and do not indicate that a device or element must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as a limitation on the present invention.

[0084] In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating relative importance or indicating the number of technical features. In the description of the present invention, "plurality" means two or more, unless otherwise clearly and specifically defined.

Claims

1. A magnetic positioning control system for driving an in vitro capsule endoscope, characterized in that: The invention comprises a guide (1) for driving a capsule endoscope, a magnetic positioning processor (2) connected to the guide (1), and a host computer (3) connected to the magnetic positioning processor (2). A first permanent magnet (13) is arranged in the guide (1), and the first permanent magnet (13) is magnetically connected to a second permanent magnet (5) arranged in the capsule endoscope. The guide (1) drives the capsule endoscope to move and / or rotate.

2. The magnetic positioning control system for driving an in vitro capsule endoscope according to claim 1, characterized in that: A stepper motor (11) is arranged inside the guide (1), a rotating shaft (12) is arranged on the stepper motor (11), a first permanent magnet (13) is arranged on the rotating shaft (12), the stepper motor (11) rotates to drive the rotating shaft (12), the rotating shaft (12) rotates to drive the first permanent magnet (13), the first permanent magnet (13) rotates to form a steady-state magnetic field module, and the steady-state magnetic field module controls the second permanent magnet (5) to drive the capsule endoscope to move and / or rotate.

3. The magnetic positioning control system for driving an in vitro capsule endoscope according to claim 2, characterized in that: A processing module (15) is arranged inside the guide (1), and the processing module (15) is respectively connected to the stepping motor (11) and the magnetic positioning processor (2).

4. The magnetic positioning control system for driving an in vitro capsule endoscope according to claim 3, characterized in that: The guide (1) is provided with a control button (14) and a display screen (16), and both the control button (14) and the display screen (16) are connected to a processing module (15).

5. The magnetic positioning control system for driving an in vitro capsule endoscope according to claim 1, characterized in that: One end of the magnetic positioning processor (2) is connected to a host computer (3) for feeding back 6-dimensional posture information of the capsule endoscope to the guide (1); one end of the magnetic positioning processor (2) is connected to a magnetic sensor (4) for detecting the magnetic field strength; the host computer (3) is provided with an MP-KAN architecture for receiving the magnetic field strength information sent by the magnetic sensor (4) to the magnetic positioning processor (2) and calculating the 6-dimensional posture information of the capsule endoscope based on the magnetic field strength information.

6. The magnetic positioning control system for driving an in vitro capsule endoscope according to claim 5, characterized in that: The magnetic positioning processor (2) is provided with a pin with an ADC sampling function, and the pin is connected to the magnetic sensor (4).

7. The magnetic positioning control system for driving an in vitro capsule endoscope according to claim 5, characterized in that: The MP-KAN architecture includes an input layer, a hidden layer, and an output layer; the magnetic sensor (4) collects the original spatial three-component magnetic field intensity and converts it into text information, and the upper computer (3) extracts sensor data from the data text to construct the overall feature sequence of the sensor as the input layer.

8. The magnetic positioning control system for driving an in vitro capsule endoscope according to claim 7, characterized in that: Also included is a MP-KAN architecture training method, the training method comprising the following steps: Step 1: Use the host computer (3) to quickly extract the target data in the text file, and then reconstruct the data into a 1×L structure, where L is the number of all sensor channels and the position of the magnet; by performing N experiments on each point in the plane, a data matrix of size (153×N)×L can be finally obtained as the training data of the MP-KAN model; Step 2: Set the activation function φ(x) = f(x) + B_spline(x), where f(x) represents the expression of f(x) as follows: Where p represents the input dimension, q represents the output dimension, and x p represents the pth input parameter, φ q,p represents a related function with trainable parameters, Φ q A function that represents a one-dimensional function related to the output dimension; B_spline(x) is the spline function that will be fitted to the curve through a piecewise continuous multi-segment curve, and each activation function will be initialized to B_spline(x)≈0; The expression of B_spline(x) is as follows: B_spline(x)=∑b i B i (x); Among them, b i As a trainable parameter, i is an intermediate variable; According to the input activation function, the grid is updated in real time every twenty training steps. The grid interval is set to G = 3. At the node, for a network structure of l+1 layers, the activation value of the neuron will be the sum of all the incoming activation values. A general l-layer KAN network will be expressed as: KAN(x)=(Φ L-1 …F L-2 …Φ1…Φ0)x; Where Φ1 represents the function matrix corresponding to the lth KAN layer; Step 3: Use the L1 norm of the activation function and add additional entropy regularization to achieve model sparsification; combined with the structure of the input layer and the output layer, the L1 norm of Φ is defined as the sum of the L1 norms of all activation functions, and its entropy is S(Φ), which is expressed as follows: The overall training goal L total is the predicted loss L pred Add L1 and entropy regularization to the KAN layer; Among them, L pred is the root mean square error between the predicted position and the actual position, and λ is used to control the size of the overall regularization.

9. The magnetic positioning control system for driving an in vitro capsule endoscope according to claim 8, characterized in that: In step 2, the data collected by the magnetic positioning system sometimes has the problem of gradient disappearance and the problem that the negative part of the gradient is zero when the relu function is selected as the activation function. The silu function is selected as the basis function f(f), and the expression of f(x) is as follows:

10. The magnetic positioning control system for driving an in vitro capsule endoscope according to claim 7, characterized in that: In the three-layer structure of MP-KAN, the number of neurons in the hidden layer is set to 5, and the number of neurons in the input layer and output layer is set to 30 and 2 respectively.

Images