Magnetic control method and system for capsule endoscopy

Abstract

The invention relates to the technical field of capsule endoscopy magnetic control, in particular to a capsule endoscopy magnetic control method and system.The method comprises the following steps that a normally-on light source loaded at the front end of a capsule endoscopy is turned on, and all tissue positions of a tissue model are shot by controlling the capsule endoscopy to move in the tissue model; obtaining a group of tissue images for positioning focus tissues; determining position information of the focus tissue through a target detection model established based on a neural network in combination with a knowledge distillation technology; and according to the position information of the focus tissue, controlling the capsule endoscopy to move until the shooting view faces the focus tissue, and synchronously starting a supplementary light source to carry out supplementary light shooting on the shooting view to obtain an enhanced image of the focus tissue. According to the method, each tissue position of the tissue model is shot through active magnetic control traversal, the position information of the focus tissue is determined through the target detection model established based on the neural network in combination with the knowledge distillation technology, and accurate positioning of the magnetic control capsule on the lesion position is realized.

Description

Technical Field

[0001] The present invention relates to the field of magnetic control technology of capsule endoscope, and in particular to a magnetic control method and system of capsule endoscope. Background Art

[0002] Wireless capsule endoscopy has the advantages of being non-invasive, painless, and free of cross-infection, overcoming the shortcomings of traditional endoscopes such as gastroscopy, and has been used as a first-line examination method for small intestinal diseases in many parts of the world. The capsule endoscope component usually includes the capsule itself, with a built-in micro camera.

[0003] Currently, most capsule endoscopes are passive and move along with the peristalsis of the gastrointestinal tract. They are unstable and uncertain. The lack of active control makes it difficult for the capsule to accurately locate the lesion and obtain high-resolution images of the lesion. They are not suitable for examinations of stomachs with larger and more complex cavities. Summary of the invention

[0004] The object of the present invention is to provide a magnetic control method and system for capsule endoscopy to solve the technical problem in the prior art that the lack of active control makes it difficult for the capsule to accurately locate the lesion and difficult to obtain a high-resolution image of the lesion.

[0005] In order to solve the above technical problems, the present invention specifically provides the following technical solutions:

[0006] A magnetic control method for capsule endoscopy, comprising the following steps:

[0007] Turn on the constant light source loaded at the front end of the capsule endoscope, and control the capsule endoscope to move in the tissue model so that the shooting field of view is directed toward various tissue positions in the tissue model, thereby obtaining a set of tissue images for locating the lesion tissue;

[0008] In a set of tissue images, the location information of the lesion tissue is determined through the target detection model based on neural network combined with knowledge distillation technology;

[0009] According to the location information of the lesion tissue, the capsule endoscope is controlled to move until the shooting field of view faces the lesion tissue, and the supplementary light source loaded at the front end of the capsule endoscope is simultaneously turned on to fill in the shooting field of view to obtain an enhanced image of the lesion tissue.

[0010] As a preferred solution of the present invention, the tissue image has position information, and the position information includes the position coordinates of the capsule endoscope and the rotation angle of the capsule endoscope lens.

[0011] As a preferred solution of the present invention, the method for constructing the target detection model includes:

[0012] Selecting a plurality of images from a group of tissue images as sample images;

[0013] Acquire the position information of the sample image, and use conventional endoscopic examination to locate the position information of the sample image on the tissue model to take tissue photos, so as to obtain a high-resolution image of the sample image;

[0014] Marking the lesion tissue detection result in the high-resolution image of the sample image, using the high-resolution image of the sample image as an input item of the YOLO V3 network, and using the lesion tissue detection result in the high-resolution image of the sample image as an output item of the YOLO V3 network;

[0015] The input items of the YOLO V3 network and the output items of the YOLO V3 network are trained through the YOLO V3 network to obtain a high-resolution detection model that detects lesion tissue in the high-resolution image of the sample image;

[0016] The sample image is used as the input of the classifier, and the output of the high-resolution detection model is used as the output of the classifier;

[0017] The high-resolution detection model is used to guide the training of the classifier, thereby obtaining a low-resolution detection model that detects lesion tissue in the sample image;

[0018] Among them, the loss function of the low-resolution detection model is:

[0019] L l =H(g l ,g r )+H(g l ,g h )-H(G l ,G h );

[0020] Where, L l is the loss function of the low-resolution detection model, g l is the lesion tissue detection result output by the low-resolution detection model, g r is the true value of the lesion tissue detection result, g h is the lesion tissue detection result output by the high-resolution detection model, G l The image G is the input of the low-resolution detection model. h The image input to the high-resolution detection model, H(g l ,g r ) is g l and g r The cross entropy between l ,g h ) is g l and g h The cross entropy between l ,G h ) is Gl and G h The cross entropy between .

[0021] As a preferred embodiment of the present invention, the method for acquiring the location information of the lesion tissue comprises:

[0022] Input each tissue image in a group of tissue images into the low-resolution detection model one by one to obtain the lesion tissue detection result in each tissue image;

[0023] The lesion tissue detection result is the position information of the tissue image where the lesion tissue exists, as the position information of the lesion tissue;

[0024] The position information of the tissue image in which the lesion tissue detection result indicates that the lesion tissue does not exist is used as the position information of the non-lesion tissue.

[0025] As a preferred embodiment of the present invention, the method for controlling the capsule endoscope to move to a shooting field of view toward the lesion tissue includes:

[0026] First, the capsule endoscope is controlled to move to the position coordinates of the capsule endoscope included in the position information of the lesion tissue;

[0027] Then, the capsule endoscope lens is controlled to rotate at the rotation angle of the capsule endoscope lens included in the position information of the lesion tissue so that the imaging field of the capsule endoscope faces the lesion tissue.

[0028] As a preferred solution of the present invention, the supplementary light source and the permanent light source both rotate synchronously with the capsule endoscope lens, so that the light source range coincides with the shooting field of view.

[0029] As a preferred solution of the present invention, during the shooting of a group of tissue images for locating lesion tissue, the permanent light source is in the on state and the supplementary light source is in the off state; during the shooting of an enhanced image of the lesion tissue, both the permanent light source and the supplementary light source are in the on state.

[0030] As a preferred embodiment of the present invention, the present invention provides a magnetic control system for a capsule endoscope, which is applied to the magnetic control method for a capsule endoscope, and the system comprises:

[0031] Capsule endoscope, used to obtain tissue images by motion photography in a tissue model, and to obtain enhanced images of lesion tissues;

[0032] A magnetic control device for controlling the movement of the capsule endoscope in the tissue model;

[0033] The positioning unit is used to obtain a set of tissue images for locating the lesion tissue according to the movement of the capsule endoscope in the tissue model, and determine the location information of the lesion tissue through a target detection model established based on a neural network combined with knowledge distillation technology.

[0034] As a preferred solution of the present invention, the capsule endoscope comprises a light source assembly, a lens assembly, a positioning assembly, a magnetic assembly and a transmission assembly, wherein the light source assembly comprises a constant light source, a supplementary light source and a light source controller, wherein:

[0035] The permanent light source is used to provide light source for capturing tissue images;

[0036] The supplementary light source is used to provide light source for taking enhanced images of lesion tissues;

[0037] The lens assembly is used to capture tissue images and enhanced images of lesion tissues;

[0038] The positioning component is used to locate the position coordinates of the capsule endoscope and the rotation angle of the capsule endoscope lens;

[0039] The magnetic component is used to move under the control of the magnetic field generated by the magnetron device;

[0040] The transmission component is used to transmit the tissue image and the enhanced image of the lesion tissue obtained by the lens component to the positioning unit, and transmit the position information of the lesion tissue calculated by the positioning unit to the transmission magnetic control device, the light source controller and the lens component.

[0041] As a preferred solution of the present invention, the lens assembly and the light source controller both include rotating structures to control the rotation of the lens and the light source respectively.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] The present invention actively magnetically controls each tissue position of the photographed tissue model to obtain a set of tissue images for locating the lesion tissue, and then determines the position information of the lesion tissue through a target detection model based on a neural network combined with knowledge distillation technology, thereby achieving accurate positioning of the lesion position by the magnetically controlled capsule. At the same time, a supplementary light source is turned on to perform fill-in light photography on the lesion tissue, thereby obtaining a high-resolution image of the lesion position and achieving the purpose of enhancing the lesion tissue image. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the implementation methods or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0045] Figure 1 A flow chart of a magnetic control method for capsule endoscopy provided by an embodiment of the present invention;

[0046] Figure 2 A block diagram of a magnetic control system for a capsule endoscope provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions 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 creative work are within the scope of protection of the present invention.

[0048] like Figure 1 As shown, the present invention provides a magnetic control method for capsule endoscope, comprising the following steps:

[0049] Turn on the constant light source (low brightness to minimize the impact on the endurance of the capsule endoscope) loaded at the front end of the capsule endoscope, and control the capsule endoscope to move in the tissue model so that the shooting field of view is directed toward various tissue positions in the tissue model, so as to obtain a set of tissue images for locating the lesion tissue;

[0050] In a set of tissue images, the location information of the lesion tissue is determined through the target detection model based on neural network combined with knowledge distillation technology;

[0051] According to the location information of the lesion tissue, the capsule endoscope is controlled to move until the shooting field of view faces the lesion tissue, and the supplementary light source loaded at the front end of the capsule endoscope is simultaneously turned on to fill in the shooting field of view to obtain an enhanced image of the lesion tissue.

[0052] In order to make the capsule endoscope suitable for the examination of tissue models with larger and more complex cavities (such as stomach models), accurately locate the lesion in the tissue model and obtain the lesion tissue image to achieve the goal of lesion examination, the present invention adopts magnetically controlled actively driven capsule endoscope movement.

[0053] The magnetically controlled active capsule endoscope of the present invention drives the capsule endoscope to traverse and photograph various tissue positions of the tissue model, obtaining a set of tissue images for locating lesion tissues, that is, collecting images of various tissue parts on the tissue model in all directions and at all angles, including images of lesion tissues.

[0054] The present invention uses a target detection model established based on a neural network combined with knowledge distillation technology to perform target recognition processing on images of various tissue parts on a tissue model collected from all directions and angles, and screens out images of lesion tissues, thereby realizing image screening of lesion tissues.

[0055] The present invention assigns position information when taking an image, extracts the position information of the image containing the lesion tissue and converts it into the position information of the lesion tissue, thereby determining the position information of the lesion tissue and realizing accurate positioning of the lesion position by the magnetically controlled capsule.

[0056] In the present invention, the position information is determined by the capsule endoscope during shooting. Therefore, the position information can be directly used to control the movement of the capsule endoscope in the future to reach the location of the corresponding lesion tissue.

[0057] Knowledge distillation is a model compression technique mainly used in deep learning. It involves training a smaller model (student model) to imitate a larger, already trained model (teacher model). This approach not only utilizes the information of the true label, but also utilizes the prediction results (soft labels) of the teacher model to convey richer information. The goal of knowledge distillation is to make the student model close to or reach the level of the teacher model in performance, while reducing the complexity and number of parameters of the model, thereby improving the generalization ability and computational efficiency of the model.

[0058] The present invention utilizes knowledge distillation technology to obtain high-resolution tissue images in the tissue model through conventional endoscopic examinations (gastroscopy, colonoscopy), and constructs a neural network model for identifying lesion tissue in high-resolution tissue images. Through knowledge distillation, a neural network model is constructed to identify lesion tissue in low-resolution tissue images obtained by capsule endoscopy, thereby achieving the effect of obtaining lesion tissue recognition in high-resolution tissue images in low-resolution tissue images, and retaining the low resolution of the image, which can significantly improve the efficiency of target detection, thereby ensuring the recognition efficiency of lesion tissue, which meets the timeliness requirements of capsule endoscopy due to limited endurance.

[0059] After accurately locating the lesion tissue, the present invention adopts a supplementary light source to superimpose the supplementary brightness on the conventional brightness to enhance the brightness of the lesion tissue for photographing, thereby compensating for the defect of the limited light source of the capsule endoscope and being able to obtain better images at a higher brightness, thereby achieving the purpose of image enhancement of the lesion tissue.

[0060] The goal of the present invention is to obtain details of the lesion tissue, so the light source is supplemented only at the lesion tissue to achieve high-resolution lesion tissue images, avoid wasting capsule energy in unnecessary situations, and effectively ensure the endurance of the capsule endoscope.

[0061] The tissue image has position information, and the position information includes the position coordinates of the capsule endoscope and the rotation angle of the capsule endoscope lens.

[0062] The present invention uses a target detection model based on a neural network combined with knowledge distillation technology to perform target recognition processing on images of various tissue parts on the all-round and all-angle collected tissue model, and selects images of lesion tissues from them, thereby realizing image screening of lesion tissues, as follows:

[0063] The methods for building the target detection model include:

[0064] Selecting a plurality of images from a group of tissue images as sample images;

[0065] Obtaining the position information of the sample image (a ground-resolution image obtained under low brightness), and using conventional endoscopic examination to locate the position information of the sample image on the tissue model to take tissue photos, thereby obtaining a high-resolution image of the sample image;

[0066] Marking the lesion tissue detection result in the high-resolution image of the sample image (the lesion tissue detection result includes the presence of lesion tissue and the absence of lesion tissue), using the high-resolution image of the sample image as the input item of the YOLO V3 network, and using the lesion tissue detection result in the high-resolution image of the sample image as the output item of the YOLO V3 network;

[0067] The input items of the YOLO V3 network and the output items of the YOLO V3 network are trained through the YOLO V3 network to obtain a high-resolution detection model that detects lesion tissue in the high-resolution image of the sample image;

[0068] The YOLO V3 algorithm is an object detection algorithm that achieves object detection by dividing the input image into a grid and predicting bounding boxes and category probabilities directly in the grid. Here are some key information about YOLO V3: Network structure: YOLO V3 uses Darknet-53 as its backbone network, which is a deep convolutional neural network with 53 convolutional layers for extracting features of images. Multi-scale detection: In order to better detect objects of different sizes, YOLO V3 makes predictions at three different scales, which helps improve the detection performance of small objects. Logical classifier: In YOLO V3, a logistic classifier (logistic) is used instead of the softmax function for object classification. This change helps improve the performance of the model when dealing with multiple independent objects. Anchor Boxes: YOLO V3 uses anchor boxes to predict the location and size of bounding boxes, which can speed up the training process of the model and improve the accuracy of detection.

[0069] These characteristics enable YOLO V3 to achieve a good balance between speed and accuracy, especially when processing real-time video streams and camera inputs. Therefore, the present invention uses the YOLO V3 network to detect lesions.

[0070] The sample image is used as the input of a classifier (such as SVM algorithm, Bayesian classifier, random forest algorithm, etc.), and the output of the high-resolution detection model is used as the output of the classifier;

[0071] The high-resolution detection model is used to guide the training of the classifier, thereby obtaining a low-resolution detection model that detects lesion tissue in the sample image;

[0072] Among them, the loss function of the low-resolution detection model is:

[0073] L l =H(g l ,g r )+H(g l ,g h )-H(G l ,G h );

[0074] Where, L l is the loss function of the low-resolution detection model, g l is the lesion tissue detection result output by the low-resolution detection model, g r is the true value of the lesion tissue detection result, g h is the lesion tissue detection result output by the high-resolution detection model, G l The image G is the input of the low-resolution detection model. hThe image input to the high-resolution detection model, H(g l ,g r ) is g l and g r The cross entropy between l ,g h ) is g l and g h The cross entropy between l ,G h ) is G l and G h The cross entropy between .

[0075] In the present invention, H(g l ,g r )+H(g l ,g h )-H(G l ,G h ) is the loss function, so that when the low-resolution detection model is trained, H(G l ,G h ) is larger, the greater the difference in image resolution, which is also called the lower the image resolution, the lower the H(g l ,g r ) is smaller, the image output value is closer to the true value, H(g l ,g h ), the smaller the image output value is, the closer it is to the recognition result of the high-resolution image. That is, the low-resolution detection model can obtain an output that is close to the true value and the recognition result of the high-resolution image on the lower resolution image, thus ensuring the recognition performance of the low-resolution detection model.

[0076] The present invention utilizes knowledge distillation technology to obtain high-resolution tissue images in the tissue model through conventional endoscopic examinations (gastroscopy, colonoscopy), and constructs a neural network model for identifying lesion tissue in high-resolution tissue images. Through knowledge distillation, a neural network model is constructed to identify lesion tissue in low-resolution tissue images obtained by capsule endoscopy, thereby achieving the effect of obtaining lesion tissue recognition in high-resolution tissue images in low-resolution tissue images, and retaining the low resolution of the image, which can significantly improve the efficiency of target detection, thereby ensuring the recognition efficiency of lesion tissue, which meets the timeliness requirements of capsule endoscopy due to limited endurance.

[0077] The present invention gives position information when the image is captured, extracts the position information of the image containing the lesion tissue as the position information of the lesion tissue, thereby realizing the determination of the position information of the lesion tissue and realizing the accurate positioning of the lesion position by the magnetically controlled capsule, as follows:

[0078] Methods for obtaining location information of lesion tissue include:

[0079] Input each tissue image in a group of tissue images into the low-resolution detection model one by one to obtain the lesion tissue detection result in each tissue image;

[0080] The lesion tissue detection result is the position information of the tissue image where the lesion tissue exists, as the position information of the lesion tissue;

[0081] The position information of the tissue image in which the lesion tissue detection result indicates that the lesion tissue does not exist is used as the position information of the non-lesion tissue.

[0082] Methods for controlling the movement of the capsule endoscope so that the imaging field of view faces the lesion tissue include:

[0083] First, the capsule endoscope is controlled to move to the position coordinates of the capsule endoscope included in the position information of the lesion tissue;

[0084] Then, the capsule endoscope lens is controlled to rotate at the rotation angle of the capsule endoscope lens included in the position information of the lesion tissue so that the imaging field of the capsule endoscope faces the lesion tissue.

[0085] The supplementary light source and the permanent light source rotate synchronously with the capsule endoscope lens, so that the light source range coincides with the shooting field of view.

[0086] During the shooting of a set of tissue images for locating lesion tissue, the permanent light source is on and the supplementary light source is off. During the shooting of the enhanced image of the lesion tissue, both the permanent light source and the supplementary light source are on.

[0087] After accurately locating the lesion tissue, the present invention adopts a supplementary light source to superimpose the supplementary brightness on the conventional brightness to enhance the brightness of the lesion tissue for photographing, thereby compensating for the defect of the limited light source of the capsule endoscope and being able to obtain better images at a higher brightness, thereby achieving the purpose of image enhancement of the lesion tissue.

[0088] Therefore, high-resolution images of lesion tissue can be obtained through capsule endoscopy, thereby improving the image acquisition performance of capsule endoscopy.

[0089] The goal of the present invention is to obtain details of the lesion tissue, so the light source is supplemented only at the lesion tissue to achieve high-resolution lesion tissue images, avoid wasting capsule energy in unnecessary situations, and effectively ensure the endurance of the capsule endoscope.

[0090] like Figure 2 As shown, the present invention provides a magnetic control system for a capsule endoscope, which is applied to a magnetic control method for a capsule endoscope. The system includes:

[0091] Capsule endoscope, used to obtain tissue images by motion photography in a tissue model, and to obtain enhanced images of lesion tissues;

[0092] A magnetic control device for controlling the movement of the capsule endoscope in the tissue model;

[0093] The positioning unit is used to obtain a set of tissue images for locating the lesion tissue according to the movement of the capsule endoscope in the tissue model, and determine the location information of the lesion tissue through a target detection model established based on a neural network combined with knowledge distillation technology.

[0094] The capsule endoscope includes a light source assembly, a lens assembly, a positioning assembly, a magnetic assembly, and a transmission assembly. The light source assembly includes a constant light source, a supplementary light source, and a light source controller.

[0095] The constant light source is used to provide light source for taking tissue images;

[0096] The supplementary light source is used to provide light source for taking enhanced images of lesion tissues;

[0097] The lens assembly is used for taking tissue images and enhanced images of lesion tissues;

[0098] The positioning component is used to locate the position coordinates of the capsule endoscope and the rotation angle of the capsule endoscope lens;

[0099] The magnetic component is used to move under the control of the magnetic field generated by the magnetic control device;

[0100] The transmission component is used to transmit the tissue image obtained by the lens component and the enhanced image of the lesion tissue to the positioning unit, and transmit the position information of the lesion tissue calculated by the positioning unit to the transmission magnetic control device, the light source controller and the lens component.

[0101] The lens assembly and the light source controller both include rotating structures to control the rotation of the lens and the rotation of the light source respectively.

[0102] The present invention actively magnetically controls each tissue position of the photographed tissue model to obtain a set of tissue images for locating the lesion tissue, and then determines the position information of the lesion tissue through a target detection model based on a neural network combined with knowledge distillation technology, thereby achieving accurate positioning of the lesion position by the magnetically controlled capsule. At the same time, a supplementary light source is turned on to perform fill-in light photography on the lesion tissue, thereby obtaining a high-resolution image of the lesion position and achieving the purpose of enhancing the lesion tissue image.

[0103] The above embodiments are only exemplary embodiments of the present application and are not intended to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present application within the essence and protection scope of the present application, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present application.

Claims

1. A magnetic control method for capsule endoscopy, characterized in that: The following steps are involved: Turn on the constant light source loaded at the front end of the capsule endoscope, and control the capsule endoscope to move in the tissue model so that the shooting field of view is directed toward various tissue positions in the tissue model, thereby obtaining a set of tissue images for locating the lesion tissue; In a set of tissue images, the location information of the lesion tissue is determined through the target detection model based on neural network combined with knowledge distillation technology; According to the location information of the lesion tissue, the capsule endoscope is controlled to move until the shooting field of view faces the lesion tissue, and the supplementary light source loaded at the front end of the capsule endoscope is simultaneously turned on to fill in the shooting field of view to obtain an enhanced image of the lesion tissue.

2. The magnetic control method of capsule endoscopy according to claim 1, characterized in that: The tissue image has position information, and the position information includes the position coordinates of the capsule endoscope and the rotation angle of the capsule endoscope lens.

3. The magnetic control method of capsule endoscopy according to claim 2, characterized in that: The method for constructing the target detection model includes: Selecting a plurality of images from a group of tissue images as sample images; Acquire the position information of the sample image, and use conventional endoscopic examination to locate the position information of the sample image on the tissue model to take tissue photos, so as to obtain a high-resolution image of the sample image; Marking the lesion tissue detection result in the high-resolution image of the sample image, using the high-resolution image of the sample image as the input item of the YOLO V3 network, and using the lesion tissue detection result in the high-resolution image of the sample image as the output item of the YOLOV3 network; The input items of the YOLO V3 network and the output items of the YOLO V3 network are trained through the YOLO V3 network to obtain a high-resolution detection model that detects lesion tissue in the high-resolution image of the sample image; The sample image is used as the input of the classifier, and the output of the high-resolution detection model is used as the output of the classifier; The high-resolution detection model is used to guide the training of the classifier, thereby obtaining a low-resolution detection model that detects lesion tissue in the sample image; Among them, the loss function of the low-resolution detection model is: L l =H(g l ,g r )+H(g l ,g h )-H(G l ,G h ); Where, L l is the loss function of the low-resolution detection model, g l is the lesion tissue detection result output by the low-resolution detection model, g r is the true value of the lesion tissue detection result, g h is the lesion tissue detection result output by the high-resolution detection model, G l The image G is the input of the low-resolution detection model. h The image input to the high-resolution detection model, H(g l ,g r ) is g l and g r The cross entropy between l ,g h ) is g l and g h The cross entropy between l ,G h ) is G l and G h The cross entropy between .

4. The magnetic control method of capsule endoscopy according to claim 3, characterized in that: The method for acquiring the position information of the lesion tissue comprises: Input each tissue image in a group of tissue images into the low-resolution detection model one by one to obtain the lesion tissue detection result in each tissue image; The lesion tissue detection result is the position information of the tissue image where the lesion tissue exists, as the position information of the lesion tissue; The position information of the tissue image in which the lesion tissue detection result indicates that the lesion tissue does not exist is used as the position information of the non-lesion tissue.

5. The magnetic control method of capsule endoscopy according to claim 4, characterized in that: Methods for controlling the movement of the capsule endoscope so that the imaging field of view faces the lesion tissue include: First, the capsule endoscope is controlled to move to the position coordinates of the capsule endoscope included in the position information of the lesion tissue; Then, the capsule endoscope lens is controlled to rotate at the rotation angle of the capsule endoscope lens included in the position information of the lesion tissue so that the imaging field of the capsule endoscope faces the lesion tissue.

6. The magnetic control method of capsule endoscopy according to claim 5, characterized in that: The supplementary light source and the permanent light source both rotate synchronously with the capsule endoscope lens, so that the light source range coincides with the shooting field of view.

7. A magnetic control method for capsule endoscopy according to claim 6, characterized in that: During the shooting of a set of tissue images for locating lesion tissue, the permanent light source is on and the supplementary light source is off. During the shooting of the enhanced image of the lesion tissue, both the permanent light source and the supplementary light source are on.

8. A magnetic control system for capsule endoscopy, characterized in that: A magnetic control method for a capsule endoscope according to any one of claims 1 to 7, the system comprising: Capsule endoscope, used to obtain tissue images by motion photography in a tissue model, and to obtain enhanced images of lesion tissues; A magnetic control device for controlling the movement of the capsule endoscope in the tissue model; The positioning unit is used to obtain a set of tissue images for locating the lesion tissue according to the movement of the capsule endoscope in the tissue model, and determine the location information of the lesion tissue through a target detection model established based on a neural network combined with knowledge distillation technology.

9. The magnetic control system of capsule endoscopy according to claim 8, characterized in that: The capsule endoscope includes a light source assembly, a lens assembly, a positioning assembly, a magnetic assembly and a transmission assembly. The light source assembly includes a constant light source, a supplementary light source and a light source controller. The permanent light source is used to provide light source for capturing tissue images; The supplementary light source is used to provide light source for taking enhanced images of lesion tissues; The lens assembly is used to capture tissue images and enhanced images of lesion tissues; The positioning component is used to locate the position coordinates of the capsule endoscope and the rotation angle of the capsule endoscope lens; The magnetic component is used to move under the control of the magnetic field generated by the magnetron device; The transmission component is used to transmit the tissue image and the enhanced image of the lesion tissue obtained by the lens component to the positioning unit, and transmit the position information of the lesion tissue calculated by the positioning unit to the transmission magnetic control device, the light source controller and the lens component.

10. The magnetic control system of capsule endoscopy according to claim 9, characterized in that: The lens assembly and the light source controller both include a rotating structure to control the rotation of the lens and the rotation of the light source respectively.

Images