Portable gastroscope based on edge computing device

Abstract

The invention provides a portable gastroscope based on an edge detection device, which comprises an endoscope probe and an endoscope tube body, and further comprises an edge calculation device, the input end of the edge calculation device is connected with one end of the endoscope tube body; the display equipment is connected with the output end of the edge computing equipment and is used for receiving and displaying the image from the edge computing equipment; wherein an intelligent detection module is arranged in the edge computing device, and the intelligent detection module is used for acquiring an optical image of the stomach collected by the endoscope probe, performing stomach disease diagnosis, marking a focus and a name and sending the focus and the name to the display device. According to the method, the intelligent detection method is set in the edge computing device to diagnose the stomach diseases, miniaturization of intelligent diagnosis of the stomach diseases is achieved, non-professional medical personnel can diagnose the stomach diseases of patients, and the problem that the non-professional personnel cannot diagnose the stomach diseases is solved.

Description

Technical Field

[0001] This application relates to the field of intelligent diagnostic assistance technologies, and more particularly, to a portable gastroscope based on an edge detection device. Background Art

[0002] Existing gastroscope devices are usually bulky and require professional medical personnel for operation. Especially in the wild or in areas with scarce medical resources, it is impossible to carry traditional gastroscope devices for examinations, and there is a lack of professional medical personnel support. In the wild environment, field workers are prone to stomach diseases or aggravated stomach conditions due to food and other reasons, but there is a lack of necessary stomach examination equipment under field conditions. Therefore, there is an urgent need for a portable, easy-to-operate device that can quickly perform stomach examinations so that field workers can conduct stomach examinations without the need for professional equipment and personnel. Summary of the Invention

[0003] Aiming at the deficiencies in the prior art, the purpose of this application is to provide a portable gastroscope based on an edge detection device, a portable and easy-to-operate gastroscope device, which can quickly and effectively perform stomach examinations in the absence of professional medical equipment and personnel, and solve the problems of the existing gastroscope devices being bulky, complex to operate, and lacking professional medical personnel.

[0004] In one aspect of this application, a portable gastroscope based on an edge detection device is provided, including: an endoscope probe and an endoscope tube body;

[0005] An edge computing device, the input end of the edge computing device is connected to one end of the endoscope tube body;

[0006] A display device, connected to the output end of the edge computing device, for receiving and displaying the image from the edge computing device;

[0007] Wherein, an intelligent detection module is provided in the edge computing device, and the intelligent detection module is used to acquire the optical image of the stomach collected by the endoscope probe, diagnose stomach diseases, and mark the lesions and names and send them to the display device.

[0008] Further, the endoscope probe is connected to the other end of the endoscope tube body, and a gas port, a light source, and a camera are provided at the endoscope probe;

[0009] The gas port is used to introduce external air to inflate the stomach during use;

[0010] The light source is used to provide brightness when detecting the stomach;

[0011] The camera is used to acquire stomach images during detection.

[0012] Furthermore, a wrapping component is provided on the outer wall of the endoscope probe; openings corresponding to the gas port, the light source, and the camera are provided on the wrapping component;

[0013] The wrapping component wraps the gas port, the light source, and the camera.

[0014] Furthermore, it further includes an airbag, which is connected to one end of the endoscope tube body close to the edge computing device and communicates with the gas port through the endoscope tube body, and is used to inject gas into the stomach through the gas port to inflate the stomach for observing the gastric mucosa.

[0015] Furthermore, a video data line, a light source power supply line, and an air tube are provided inside the endoscope tube body;

[0016] The video data line is connected to the camera and is used to transmit the optical image detected by the camera to the edge computing device;

[0017] The light source power supply line is connected to the light source and is used to supply power to the light source;

[0018] The air tube is arranged inside the endoscope tube body, one end is communicated with the gas port, and the other end is communicated with the airbag, and is used to send the gas in the airbag into the stomach through the gas port.

[0019] Furthermore, an airway opening is provided at a position one-third of the endoscope tube body close to the edge computing device, and the airbag is connected to the air tube through the airway opening.

[0020] Furthermore, it further includes a power supply. A video interface and a power supply interface are provided on the edge computing device. The video data line is connected to the edge computing device through the video interface;

[0021] The power supply is connected to the edge computing device through the power supply interface and is used to supply power to the edge computing device.

[0022] Furthermore, the endoscope tube body and the wrapping component are made of polyurethane material.

[0023] Furthermore, the intelligent detection module includes:

[0024] Data acquisition sub-module: Acquire gastric image data, and the gastric image data includes: the lesion site and name corresponding to the gastric image;

[0025] Training sub-module: Input the gastric image data into the YOLOV11 model for training to generate a YOLOV11 recognition model, and convert the YOLOV11 recognition model into a training weight.pt format file model;

[0026] Format conversion sub-module: According to the training weight.pt format file model, it is converted into a format that can be deployed on the edge computing device according to the used edge computing device.

[0027] Further, the format conversion sub-module is configured to:

[0028] Obtain the training weight.pt format file model of the YOLOV11 recognition model;

[0029] Perform FP32 to FP16 floating-point precision conversion on the training weight.pt format file model, reduce the calculation precision of the model weights and activation values to 16-bit floating-point format, and generate an intermediate model;

[0030] Perform INT8 integer precision quantization on the intermediate model, map it to the 8-bit integer value space, and generate a precision-quantized model;

[0031] Perform hardware adaptation optimization on the precision-quantized model to generate an.rknn format file that can be deployed on the edge computing device.

[0032] Compared with the prior art, the present application has at least one of the following beneficial effects:

[0033] 1. By setting an intelligent detection method in the edge computing device to diagnose gastric diseases in the present application, it is possible to realize the diagnosis of gastric diseases by non-professional medical personnel for patients, and solve the problem that non-professionals cannot diagnose gastric diseases.

[0034] 2. By using a portable edge computing device to detect gastric diseases in the present application, it is possible to miniaturize the intelligent diagnosis of gastric diseases, and solve the difficulty that non-medical personnel cannot perform gastroscopy in the wild environment. Description of the Drawings

[0035] By reading the detailed description of the non-restrictive embodiments with reference to the following drawings, other features, purposes, and advantages of the present application will become more obvious:

[0036] Figure 1 It is a structural diagram of a portable gastroscope based on an edge detection device according to an embodiment of the present application.

[0037] In the figure: 1. Endoscope probe; 11. Gas port; 12. Camera; 13. Light source; 2. Endoscope tube body; 21. Airway opening; 22. Video data line; 23. Light source power supply line; 3. Edge computing device; 31. Video interface; 32. Power interface; 4. Display device; 5. Airbag; 6. Trachea; 7. Power supply; 8. Wrapping component. Detailed Embodiments

[0038] The present application will be described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present application, but do not limit the present application in any form. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can be made. These all fall within the protection scope of the present application.

[0039] In the description of the embodiments of the present application, it should be understood that the orientation or positional relationships indicated by the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "bottom", "inner", "outer", "clockwise", "counterclockwise", etc. are based on the orientation or positional relationships shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present application.

[0040] In addition, the terms "first" and "second" are only used for descriptive purposes and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features.

[0041] In the description of the embodiments of the present application, the meaning of "a plurality" is two or more, unless otherwise specifically defined. In the present application, unless otherwise clearly specified and limited, the terms such as "installed", "connected", "connected to", "fixed" and the like should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a mechanical connection or an electrical connection; it may be directly connected, or indirectly connected through an intermediate medium, and it may be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific circumstances.

[0042] In the embodiments of the present application, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally further includes steps or units not listed, or optionally further includes other steps or units inherent to these processes, methods, products or devices.

[0043] Refer to Figure 1 As shown, a portable gastroscope based on an edge detection device in an embodiment of the present application includes: an endoscope probe 1 and an endoscope tube body 2.

[0044] It further includes an edge computing device 3, the input end of the edge computing device 3 is connected to one end of the endoscope tube body 2; a display device 4, which is connected to the output end of the edge computing device 3 and is used to receive and display the images from the edge computing device 3. An intelligent detection method is provided in the edge computing device 3, which is used to obtain the optical images of the stomach collected by the endoscope probe 1, diagnose stomach diseases, and mark the lesions and their names and send them to the display device 4.

[0045] In this application, by integrating the endoscope probe 1, the endoscope tube body 2, the edge computing device 3 and the display device 4, a compact and efficient diagnostic device for stomach diseases is formed. Through the intelligent detection method built in the edge computing device 3, it can obtain and analyze the optical images of the stomach collected by the endoscope probe 1 in real time, quickly and accurately diagnose stomach diseases, identify the lesion sites, without uploading the image information, and can directly mark the lesions and their names on the display device 4, improving the efficiency and accuracy of diagnosis, making the diagnosis process more intuitive and convenient, providing strong diagnostic support for non-professional medical personnel, with low cost, and at the same time bringing an efficient examination experience for patients.

[0046] Specifically, when using this application for stomach examination, first insert the endoscope probe 1 into the patient's stomach through the endoscope tube body 2. The light source 13 in the endoscope probe 1 provides illumination, and the camera 12 captures the optical images of the stomach. The optical images are transmitted through the endoscope tube body 2 to the input end of the edge computing device 3. The intelligent detection method in the edge computing device 3 analyzes and processes the received images, identifies and diagnoses stomach diseases. The intelligent detection method will automatically mark the position of the lesion and attach the corresponding name. Finally, the processed images and diagnostic information are sent to the display device 4, realizing low-cost diagnosis of stomach diseases for non-professional medical personnel, and the detection can be achieved without the need for a network, improving the efficiency and accuracy of stomach disease diagnosis.

[0047] In some specific embodiments, the endoscope probe 1 is connected to the other end of the endoscope tube body 2. The endoscope probe 1 is provided with a gas port 11, a light source 13 and a camera 12. The gas port 11 is used to introduce external air to inflate the stomach during use. The light source 13 is used to provide brightness when detecting the stomach. The camera 12 is used to obtain stomach images during detection.

[0048] Among them, the light source 13 is an LED lamp.

[0049] Specifically, the outer wall of the endoscope probe 1 is provided with a wrapping member 8; the wrapping member 8 is provided with openings corresponding to the gas port 11, the light source 13 and the camera 12; the wrapping member 8 wraps the gas port 11, the light source 13 and the camera 12.

[0050] In this application, a wrapping component 8 is separately provided on the outer wall of the endoscope probe 1, and the wrapping component 8 wraps the gas port 11, the light source 13, and the camera 12, which facilitates separate replacement after use and avoids infecting the stomach.

[0051] It should be noted that the endoscope tube body 2 and the wrapping component 8 are made of polyurethane material.

[0052] In some possible embodiments, an airbag 5 is further included, which is connected to one end of the endoscope tube body 2 close to the edge computing device 3 and is communicated with the gas port 11 through the endoscope tube body 2, and is used to inject gas into the stomach through the gas port 11 to inflate the stomach and observe the gastric mucosa.

[0053] In the above embodiment, by adding an airbag 5 to the gastroscope and connecting it to the gas port 11 through the endoscope tube body 2, an appropriate amount of gas can be injected into the stomach during the examination, causing the gastric wall to expand, more clearly exposing the details of the gastric mucosa, facilitating the camera 12 to examine the gastric information, obtaining the state of the gastric mucosa more comprehensively, improving the recognition accuracy of the intelligent detection method for lesions, and ensuring the reliability of the diagnosis result.

[0054] Specifically, during use, first insert the portable gastroscope into the patient's stomach through the endoscope probe 1. Then, press the airbag 5 connected to the endoscope tube body 2. The airbag 5 is connected to the gas port 11 through the channel inside the endoscope tube body 2. When the airbag 5 is inflated, the gas is injected into the stomach through the gas port 11. As the gas is injected, the gastric wall gradually expands, and the details of the gastric mucosa become more clearly visible; the camera 12 captures and records the images of the stomach and transmits them to the edge computing device 3 for analysis. Through the intelligent detection method, the gastric mucosa in the images is examined in detail, and any possible lesions are identified and marked; finally, the processed images and diagnostic information are sent to the display device 4 to realize the diagnosis of gastric diseases for patients by non-professional medical personnel, solving the problem that non-professionals cannot diagnose gastric diseases in case of emergency.

[0055] In some specific embodiments, a video data line 22, a light source power supply line 23, and an air tube 6 are provided inside the endoscope tube body 2.

[0056] The video data line 22 is connected to the camera 12 and is used to transmit the optical images detected by the camera 12 to the edge computing device 3; the light source power supply line 23 is connected to the light source 13 and is used to supply power to the light source 13; the air tube 6 is arranged inside the endoscope tube body 2, one end of which is communicated with the gas port 11, and the other end is communicated with the airbag 5, and is used to send the gas in the airbag 5 into the stomach through the gas port 11.

[0057] By arranging the video data line 22, the light source power supply line 23 and the air tube 6 inside the endoscope tube body 2, it is possible to avoid infecting the stomach when the endoscope probe 1 is inserted into the stomach. At the same time, it supports the endoscope probe 1. Among them, the endoscope tube body 2 is made of polyurethane material and wraps the video data line 22, the light source power supply line 23 and the air tube 6.

[0058] The airbag 5 is connected to the gas port 11 through the air tube 6 and is separately arranged inside the endoscope tube body 2, avoiding contact between the air tube 6 and the video data line 22 and the light source power supply line 23, which may cause infection to the stomach and improving the safety of the portable gastroscope.

[0059] Specifically, an airway opening 21 is provided at one-third of the position of the endoscope tube body 2 close to the edge computing device 3, and the airbag 5 is connected to the air tube 6 through the airway opening 21.

[0060] Specifically, a video data line 22, an LED power supply line, and an airway are arranged inside the endoscope tube body 2, and the outside is wrapped by polyurethane material. It is mainly used to support the endoscope probe 1. During operation, it can be inserted into the stomach along the human esophagus. One end is connected to the endoscope probe 1, and the other end is connected to the power supply and video interface 31 of the edge detection device. An airway opening 21 is left at the 1 / 3 position close to the edge detection device and is connected to the airbag 5. Air can be inhaled from the outside, and the gas can be injected into the stomach along the airway by squeezing to inflate the stomach to observe the gastric mucosa. During use, power is supplied to the LED lamp through the LED power supply line, the optical image of the stomach collected by the camera 12 is received through the video data line 22, and a judgment is made on the gastric condition through the set intelligent detection method, and the lesion and its name are marked in the optical image (picture). The image information marked with the lesion information is sent to the display device 4 through the video data line 22.

[0061] In some possible embodiments, a power supply 7 is further included. A video interface 31 and a power supply interface 32 are provided on the edge computing device 3. The video data line 22 is connected to the edge computing device 3 through the video interface 31; the power supply 7 is connected to the edge computing device 3 through the power supply interface 32 to supply power to the edge computing device 3.

[0062] In some possible embodiments, the intelligent detection module includes:

[0063] Data acquisition sub-module: Acquire gastric image data, and the gastric image data includes: the lesion site and name corresponding to the gastric image;

[0064] Training sub-module: Input the gastric image data into the YOLOV11 model for training to generate a YOLOV11 recognition model, and convert the YOLOV11 recognition model into a training weight.pt format file model;

[0065] Format conversion sub-module: According to the training weight.pt format file model, it is converted into a format that can be deployed on the edge computing device according to the edge computing device used.

[0066] Specifically, the gastric images are obtained through the data acquisition sub-module, and the lesion types (such as ulcers, polyps, tumors, etc.) are classified and labeled; then, the training sub-module inputs the labeled data set into YOLOv11 for training, and adopts adaptive data augmentation (such as random lighting, rotation) and hard negative mining strategy to generate a.pt weight file with the ability to detect multi-class lesions, and adjusts the model hyperparameters based on the mAP index of the validation set; according to the hardware architecture of the target edge device (edge computing device based on the rk3588s chip), then, the format conversion sub-module uses the corresponding SDK (such as RKNN Toolkit or TensorRT) to convert the.pt model into an optimized inference format, and deploys the converted model to the edge computing device to achieve real-time inference function.

[0067] In the above embodiments of the present application, by deploying a lesion recognition model based on YOLOv11 on the edge computing device, the efficiency and decentralization of gastric disease detection are realized. Combining the multi-label annotation data of the lesion location and name greatly improves the detection accuracy and generalization ability of complex gastric lesions; by converting the model into a format adapted to the edge device (such as.rknn based on NPU or.tflite of the embedded inference framework), the local computing power can be used to directly perform inference on the gastroscope, reducing the dependence on the cloud server, reducing network latency and ensuring data privacy and security.

[0068] In some specific embodiments, the format conversion sub-module is configured to: obtain the training weight.pt format file model of the YOLOV11 recognition model; perform floating-point precision conversion from FP32 to FP16 on the training weight.pt format file model, reduce the calculation precision of the model weights and activation values to 16-bit floating-point format, and generate an intermediate model; perform INT8 integer precision quantization on the intermediate model, map it to the 8-bit integer value space, and generate a precision-quantized model; perform hardware adaptation optimization on the precision-quantized model to generate an.rknn format file that can be deployed on the edge computing device.

[0069] In the above embodiments of the present application, through multi-stage quantization and hardware adaptation optimization, the deployment efficiency and inference performance of the model on edge computing devices have been significantly improved. First, converting the model from FP32 to FP16 can reduce the memory footprint by 50% and improve the computing speed, while maintaining high detection accuracy (suitable for real-time scenarios with low sensitivity to accuracy); after further performing INT8 quantization, the model size is reduced to 1 / 4 of the original FP32 model, and the computing power consumption is greatly reduced (suitable for battery-powered mobile medical devices). By optimizing the numerical mapping through the calibration dataset, the accuracy loss caused by low precision can be alleviated; finally, through hardware adaptation optimization (such as operator fusion, memory layout conversion), the model is fully matched to the computing architecture of the edge device (such as the NPU instruction set), realizing end-to-end inference acceleration. In typical scenarios, the inference speed is increased by 3-5 times, and it supports operation with a power consumption as low as 10W.

[0070] Specifically, taking the detection model set on the RK3588 edge hardware as an example: Before running the.rknn inference on the device, it is necessary to convert the deep learning model (such as the YOLO model) into the.rknn format.

[0071] 1. The model conversion process, the main process is as follows:

[0072] (1) Select a suitable original model

[0073] Common object detection models:

[0074] YOLOv5 / v8 (lightweight and efficient, suitable for embedded);

[0075] MobileNet-SSD (low-power adaptation);

[0076] Faster R-CNN (high detection accuracy but large computational volume);

[0077] (2) Model conversion (rknn-Toolkit2)

[0078] Use rknn-toolkit2 for model conversion, including the process of PyTorch / TensorFlow → rknn:

[0079] Step 1, load the original model (in formats such as.pt,.onnx,.pb, etc.);

[0080] Step 2, perform FP32 to FP16 quantization (reduce the computing precision and improve the operation efficiency);

[0081] Step 3, perform INT8 quantization (optional) (a calibration dataset needs to be provided to improve the NPU acceleration effect);

[0082] Step 4: Convert and export the.rknn file to obtain yolov5.rknn, which can be deployed to the RK3588 for running.

[0083] 2. The specific steps for model deployment (embedded device) include:

[0084] (1) Prepare the device environment

[0085] Install the RKNN-Toolkit2 running environment:

[0086] To run the.rknn model on the RK3588, you need to install rknn-toolkit2: pip install rknn-toolkit2

[0087] Configure the NPU runtime library:

[0088] Ensure that librknn_runtime.so exists in the / usr / lib directory. If it is missing, you need to manually copy the library file provided by Rockchip.

[0089] (2) Code deployment

[0090] Run the Python code on the RK3588, load the.rknn and perform inference.

[0091] 3. Target detection program running process

[0092] After successful conversion and deployment, the running process of the target detection code is as follows:

[0093] (1) Initialization

[0094] Load the RKNN model and initialize the NPU running environment;

[0095] (2) Read and preprocess the image

[0096] (a) Read the image

[0097] (b) Perform letter_box transformation

[0098] The YOLO model requires a fixed input image size, such as 640×640: The original image ratio can be maintained while filling the blank areas.

[0099] (3) Execute inference

[0100] Send the image to the RKNN for target detection. The outputs are the original outputs of the model and need post-processing.

[0101] 4. Post-processing:

[0102] (1) Parse the model output

[0103] The output of the object detection model usually includes: boxes (coordinates of the object bounding boxes), scores (object confidence levels), and classes (object categories).

[0104] Parsing method: It is achieved by parsing the network output and calculating the final confidence level.

[0105] (2) Non-Maximum Suppression (NMS)

[0106] Filter out redundant bounding boxes with too high overlap:

[0107] compute_iou() calculates the IoU (Intersection over Union) to determine whether the bounding boxes overlap too much.

[0108] 5. Result visualization

[0109] Draw the detection bounding boxes.

[0110] The model is run on edge hardware through the above process. The complete process is as follows: 1. Model conversion: PyTorch → ONNX → RKNN and quantization optimization. 2. Model deployment: Install the environment on RK3588 and load the.rknn. 3. Program running: Read the image, perform letter_box, load the model, perform inference, parse the output, perform NMS processing, calculate the object size, draw the detection bounding boxes, and upload the results.

[0111] The following further illustrates the present application in combination with specific application examples / comparative examples to better understand the above technical solutions of the present application. It should be understood that the following are only partial examples and are not used to limit the present application.

[0112] Application Example 1: Field Gastric Diagnosis for a Field Exploration Team

[0113] When a certain exploration team was conducting an investigation in the mountains, a team member suddenly had stomach discomfort, showing symptoms such as stomachache and dizziness. Since the distance to the nearest hospital was relatively far, it was impossible to quickly admit the patient to the hospital. The exploration team used the present invention to insert the endoscope into the team member's stomach, turn on the LED light, and inflate the stomach with the airbag. The AI recognition function on the edge computing device was started, and based on the gastric image, the lesion and its name were determined and displayed on the display device. After AI recognition, it was determined that the lesion location was the duodenal bulb, with ulcerative bleeding, and the accuracy rate was 90%. The exploration team communicated with the doctor about the team member's condition by phone, and the doctor gave advice on fasting and, if conditions permitted, administered relevant hemostatic drugs to the patient and sent the patient to the hospital as soon as possible.

[0114] The specific embodiments of the present application have been described above. It should be understood that the present application is not limited to the above specific embodiments, and those skilled in the art can make various modifications or alterations within the scope of the claims, which does not affect the essence of the present application. The above preferred features can be used in any combination without conflict.

Claims

1. A portable gastroscope based on an edge detection device, comprising: An endoscope probe and an endoscope tube body, characterized in that it further comprises: An edge computing device, the input end of the edge computing device is connected to one end of the endoscope tube body; A display device, connected to the output end of the edge computing device, for receiving and displaying images from the edge computing device; Wherein, an intelligent detection module is provided in the edge computing device, and the intelligent detection module is used to obtain the optical image of the stomach collected by the endoscope probe, diagnose stomach diseases, and mark the lesions and names and send them to the display device.

2. The portable gastroscope based on an edge detection device according to claim 1, wherein The endoscope probe is connected to the other end of the endoscope tube body, and a gas port, a light source and a camera are provided at the endoscope probe; The gas port is used to introduce external air to inflate the stomach during use; The light source is used to provide brightness when detecting the stomach; The camera is used to obtain stomach images during detection.

3. The portable gastroscope based on an edge detection device according to claim 2, wherein A wrapping component is provided on the outer wall of the endoscope probe; openings corresponding to the gas port, the light source and the camera are provided on the wrapping component; The wrapping component wraps the gas port, the light source and the camera.

4. A portable gastroscope based on an edge detection device according to claim 2, characterized in that, It further comprises an airbag, which is connected to one end of the endoscope tube body close to the edge computing device and communicates with the gas port through the endoscope tube body, and is used to inject gas into the stomach through the gas port to inflate the stomach to observe the gastric mucosa.

5. A portable gastroscope based on an edge detection device according to claim 4, characterized in that, A video data line, a light source power supply line and an air tube are provided inside the endoscope tube body; The video data line is connected to the camera and is used to transmit the optical image detected by the camera to the edge computing device; The light source power supply line is connected to the light source and is used to supply power to the light source; The air tube is arranged inside the endoscope tube body, one end is communicated with the gas port, and the other end is communicated with the airbag, and is used to send the gas in the airbag into the stomach through the gas port.

6. The portable gastroscope based on an edge detection device according to claim 5, wherein, An airway opening is provided at a position one-third of the endoscope tube body close to the edge computing device, and the airbag is connected to the air tube through the airway opening.

7. A portable gastroscope based on an edge detection device according to claim 5, characterized in that, It further comprises a power supply, a video interface and a power supply interface are provided on the edge computing device, and the video data line is connected to the edge computing device through the video interface; The power supply is connected to the edge computing device through the power supply interface and is used to supply power to the edge computing device.

8. A portable gastroscope based on an edge detection device according to claim 3, characterized in that, The endoscope tube body and the wrapping component are made of polyurethane material.

9. A portable gastroscope based on an edge detection device according to claim 1, characterized in that, The intelligent detection module includes: A data acquisition sub-module: acquiring stomach image data, and the stomach image data includes: the lesion site and name corresponding to the stomach image; A training sub-module: inputting the stomach image data into the YOLOV11 model for training, generating a YOLOV11 recognition model, and converting the YOLOV11 recognition model into a training weight.pt format file model; A format conversion sub-module: converting the training weight.pt format file model into a format deployable on the edge computing device according to the edge computing device used.

10. A portable gastroscope based on an edge detection device according to claim 9, characterized in that, The format conversion sub-module is configured to: Obtain the training weight.pt format file model of the YOLOV11 recognition model; Perform FP32 to FP16 floating-point precision conversion on the training weight.pt format file model, reduce the calculation precision of the model weights and activation values to 16-bit floating-point format, and generate an intermediate model; Perform INT8 integer precision quantization on the intermediate model, map it to the 8-bit integer value space, and generate a model after precision quantization; Perform hardware adaptation optimization on the model after precision quantization to generate an.rknn format file that can be deployed on edge computing devices.

Images