Medical endoscope system video distortion intelligent self-checking method, device and apparatus

Abstract

The application provides a medical endoscope system video distortion intelligent self-checking method, device and apparatus, and relates to the technical field of fault detection.The application extracts features of all video frames of the collected endoscope video to obtain distortion features of the video frames, classifies and rates the distortion features, takes the average value of all the video frames as the distortion classification and distortion rating of the endoscope video, and finally matches the fault type causing the distortion based on the distortion classification and distortion rating of the endoscope video, thereby meeting the requirements of timeliness and accuracy of the endoscope video scene.

Description

Technical Field

[0001] This invention relates to the field of fault detection technology, specifically to an intelligent self-testing method, equipment, and device for video distortion in a medical endoscope system. Background Technology

[0002] Endoscopic systems are commonly used tools for surgical video acquisition. However, video distortion is a common problem during the acquisition process, including noise, defocusing blur, motion blur, uneven lighting, and smoke. While some research has addressed these issues—for example, patent applications 2022105336865 and 2022114107013—have initially implemented fault analysis functions for endoscopic system videos. However, existing technologies still have the following shortcomings:

[0003] 1. Most existing intelligent self-testing methods are based on real-time and historical data. However, the timeliness of these methods is insufficient to meet the real-time requirements of some scenarios. Especially in scenarios such as medical endoscope systems that require rapid self-testing and fault diagnosis, video-based self-testing analysis methods are more urgent and necessary.

[0004] 2. Current intelligent self-testing methods, when performing distortion analysis, often struggle to determine the specific cause of system malfunctions from the analysis results. This can lead to difficulties in accurately locating and diagnosing faults, especially in endoscopic systems which require high precision and causal reasoning.

[0005] 3. Current methods for deep learning-based distortion analysis based on video quality are mostly applied to natural scenes, with limited application to videos from medical endoscopic systems. This situation results in a lack of effective video distortion analysis methods for the intelligent self-checking of medical endoscope systems, making it impossible to accurately diagnose and analyze distortion in videos.

[0006] Therefore, there is an urgent need for a distortion analysis method for endoscopic system videos in the medical field. Summary of the Invention

[0007] (a) Technical problems to be solved

[0008] To address the shortcomings of existing technologies, this invention provides a method, device, and apparatus for intelligent self-testing of video distortion in medical endoscope systems, solving the problems of timeliness, accuracy, and professionalism in intelligent self-testing of video distortion in medical endoscope systems.

[0009] (II) Technical Solution

[0010] To achieve the above objectives, the present invention provides the following technical solution:

[0011] Firstly, a method for intelligent self-detection of video distortion in a medical endoscope system is provided, the method comprising:

[0012] Obtain endoscopic video;

[0013] Feature extraction is performed on all video frames of the endoscopic video to obtain the distortion features of each video frame;

[0014] The distortion features are classified and rated for distortion.

[0015] The average value of all video frames is used as the distortion classification and distortion rating of the endoscopic video;

[0016] Based on the distortion classification and distortion rating of endoscopic videos, the fault types that cause the distortion are matched.

[0017] Furthermore, the endoscopic video has a fixed frame rate, and the RGB pixel values ​​of its video frames are normalized RGB pixel values.

[0018] Furthermore, feature extraction is performed on all video frames of the endoscopic video to obtain the distortion features of each video frame, including:

[0019] The first feature F is obtained by using the pre-trained ResNet-50 as the feature extractor. res ;

[0020] Based on the first feature F res Perform a multi-scale feature fusion operation to obtain the second feature F output by the multi-scale feature fusion at the l-th layer. mul_l Where l represents the corresponding layer in ResNet-50;

[0021] Based on the second feature F mul_l Perform patch embedding operation to obtain the third feature P l ;

[0022] The third feature P l The input is fed into the trained Swing Transformer network to obtain the fourth feature F. swin ;

[0023] For the fourth feature F swin Perform global average pooling to obtain the distortion features Z of the video frames. swin .

[0024] Furthermore, the distortion features are classified and rated for distortion, including:

[0025] The classification layer based on the distortion classification task obtains the distortion classification of the distortion features; wherein, the distortion classification includes: noise, defocus blur, motion blur, uneven lighting, and smoke;

[0026] The classification layer based on the distortion rating task obtains the distortion rating of the distortion features.

[0027] Furthermore, during network training, the loss function is a weighted sum of the loss functions for the distortion classification task and the distortion rating task, where the loss function for the distortion classification task is the cross-entropy loss function, and the loss function for the distortion rating task is the mean squared error.

[0028] Furthermore, based on the distortion classification and rating of endoscopic videos, the types of faults causing the distortion are matched, including:

[0029] The fault types corresponding to distortion categories and distortion levels are obtained based on hidden Markov models.

[0030] Secondly, a smart self-checking device for video distortion in a medical endoscope system is provided, the device comprising:

[0031] The video processing module is used to acquire endoscopic videos;

[0032] The video frame distortion feature extraction module is used to extract features from all video frames of the endoscope video to obtain the distortion features of each video frame.

[0033] The video frame classification and rating module is used to classify and rate the distortion features.

[0034] The video classification and rating module is used to classify and rate the distortion of all video frames as the average value of the endoscopic video.

[0035] The fault type matching module is used for distortion classification and rating based on endoscopic video, matching the fault type that causes the distortion.

[0036] Furthermore, feature extraction is performed on all video frames of the endoscopic video to obtain the distortion features of each video frame, including:

[0037] The first feature F is obtained by using the pre-trained ResNet-50 as the feature extractor. res ;

[0038] Based on the first feature F res Perform a multi-scale feature fusion operation to obtain the second feature F output by the multi-scale feature fusion at the l-th layer. mul_l Where l represents the corresponding layer in ResNet-50;

[0039] Based on the second feature F mul_lPerform patch embedding operation to obtain the third feature P l ;

[0040] The third feature P l The input is fed into the trained Swing Transformer network to obtain the fourth feature F. swin ;

[0041] For the fourth feature F swin Perform global average pooling to obtain the distortion features Z of the video frames. swin .

[0042] Thirdly, a device is provided, comprising:

[0043] One or more processors;

[0044] Memory; and

[0045] One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including methods for performing intelligent self-testing methods for video distortion in medical endoscope systems as described in any one of claims 1-6.

[0046] (III) Beneficial Effects

[0047] This invention provides a method, device, and apparatus for intelligent self-checking of video distortion in a medical endoscope system. Compared with the prior art, it has the following advantages:

[0048] 1) This invention utilizes the Swin-Transformer network with multi-scale feature fusion to extract features of distorted videos and establishes a classification association between the types of distortion in medical endoscopy videos and distorted videos. At the same time, it obtains a specific assessment of the distortion level, thereby associating the types and severity of distortion.

[0049] 2) Based on the analysis method of Hidden Markov Model, this invention establishes a quantitative relationship between surgical video distortion classification, quality grade score and system failure, as well as assesses the relationship between various parameters related to system failure, thereby better understanding the cause of system failure and quickly and accurately locating and eliminating the fault, improving the reliability and timeliness of intelligent self-testing of medical laparoscopic system. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is a flowchart of an embodiment of the present invention;

[0052] Figure 2 This is a structural block diagram of an embodiment of the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are described clearly and completely. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] This application provides a method, device, and apparatus for intelligent self-testing of video distortion in medical endoscope systems, which solves the problems of timeliness, accuracy, and professionalism in intelligent self-testing of video distortion in medical endoscope systems.

[0055] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0056] Example 1:

[0057] like Figure 1 As shown, this invention provides an intelligent self-checking method for video distortion in a medical endoscope system. This method is executed by a computer and includes:

[0058] Obtain endoscopic video;

[0059] Feature extraction is performed on all video frames of the endoscopic video to obtain the distortion features of each video frame;

[0060] The distortion features are classified and rated for distortion.

[0061] The average value of all video frames is used as the distortion classification and distortion rating of the endoscopic video;

[0062] Based on the distortion classification and distortion rating of endoscopic videos, the fault types that cause the distortion are matched.

[0063] The beneficial effects of this embodiment are:

[0064] This embodiment extracts features from all video frames of the acquired endoscopic video to obtain the distortion features of each video frame; then, it classifies and rates the distortion features; the average value of all video frames is used as the distortion classification and rating of the endoscopic video; finally, based on the distortion classification and rating of the endoscopic video, it matches the fault type causing the distortion. This meets the timeliness and accuracy requirements of endoscopic video scenarios.

[0065] The implementation process of the embodiments of the present invention will be described in detail below:

[0066] In this embodiment of the invention, a system video acquisition module is included. This module may consist of a medical endoscope, an image acquisition card, a signal processor, etc. It can be flexibly configured according to specific needs to obtain high-quality video signals for data processing and analysis.

[0067] In practical implementation, the first step is to use the system's video acquisition module to obtain the video data to be extracted. This involves connecting the video interface of the endoscope device (SDI, BNC, AV terminal, S-terminal, VGA, DVI, HDMI, etc.) and converting different types of video interfaces into a unified MP4 or other consistent video file format. Alternatively, the video can be recorded on an SD card or external hard drive using a video capture box, and then imported into other computing devices for processing and analysis. The video duration T (in seconds) is set according to the hardware conditions, method, and computing environment, and is usually a positive integer not less than 10. The number of images N extracted per second should be no less than 25, and the screen resolution should not be lower than 512×288.

[0068] Secondly, before feature extraction, the video should be preprocessed to handle endoscopic video inputs with different parameters. Specific steps include: dividing the video stream into segments of equal duration using ffmpeg or OpenCV as input video data (typically 10 seconds), and normalizing the video frame rate to a fixed fps. i ,fps i A value of 25 is typically used to ensure that subsequent model algorithms can handle videos with different frame rates. Extract images from each 1-second interval of the video clip. Where H is the height of the video frame, W is the width of the video frame, and the pixel values ​​of the RGB three channels are integers from 0 to 255. The image resolution is adjusted to 224×224, and the pixel values ​​are normalized as follows:

[0069]

[0070] in,

[0071] I0 represents the normalized pixel values ​​of the RGB three channels of the video frame;

[0072] x raw This represents the original RGB three-channel pixel values ​​of the video frame before normalization;

[0073] `mean` and `std` are the mean and variance of the pixel values, respectively, and need to be set comprehensively based on the characteristics of the video from the medical endoscope system. For example:

[0074] mean = [0.485, 0.456, 0.406]

[0075] std = [0.229, 0.224, 0.225]

[0076] Then, data augmentation methods are used, with random rotation applied to each image.

[0077] In this embodiment of the invention, a distortion feature extraction module is also included. The purpose of this module is to extract features from the acquired video signal of the medical endoscope system in order to analyze and evaluate the distortion in the video. This module can use various algorithms and techniques, such as deep learning methods based on Transformer networks, to extract feature information such as color, texture, and shape from the video signal.

[0078] In practice, distortion feature extraction includes the following steps:

[0079] First, using a ResNet-50 pre-trained on the Cholec80 dataset as a feature extractor, the first feature F extracted by ResNet is obtained. res , can be represented as:

[0080] F res =ResNet(I0)

[0081] Wherein, ResNet represents the feature extractor;

[0082] I0 represents the normalized pixel values ​​of the video frame as input to the feature extractor;

[0083] Then, a multi-scale feature fusion operation is performed, which can be expressed as:

[0084]

[0085] in,

[0086] F res_l This represents the output of ResNet at layer l;

[0087] Conv 1×1 This represents a 1×1 convolution operation;

[0088] UpSample indicates an upsampling operation;

[0089] F mul_l It is the second feature output from the l-th layer by multi-scale feature fusion;

[0090] Then, the second feature F output after fusing the multi-scale features is... mul_l The patch is divided into non-overlapping patches, resulting in a patch embedding, which can be represented as:

[0091] P l =Embed(F mul_l )

[0092] in,

[0093] Embed refers to the Patch Embedding operation; the Patch Embedding operation divides the input image into individual image patches according to the PatchSize.

[0094] P l It is the output of the Patch Embedding operation at layer l;

[0095] Then, the output P of Patch Embedding l The input is fed into the trained Swing Transformer network for processing, which can be represented as:

[0096] F swin_l =MLP(Norm(SW-MSA(Norm(P) l ))))

[0097]

[0098] Where l is the level, P l It is the output of Patch Wmbedding at layer l;

[0099] Norm is the Layer Normalization operation;

[0100] MLP stands for Multilayer Perceptron;

[0101] F swin_l It is the output of the Swing Transformer network at layer l;

[0102] F swin It is the total output of the Swing Transformer network;

[0103] SW-MSA is a sliding window multi-head self-attention operation, and its specific operation is obtained by the following formula:

[0104]

[0105]

[0106] Where Q, K, and V are the query, key, and value, respectively, which are obtained by linear transformations of the input features, d k It is the dimension of the key;

[0107] Concat is a join operation, W O It is the weight of the linear transformation, head i It is the output of the i-th head;

[0108] Finally, the total output F of the Swing Transformer network is... swin Performing global average pooling, the final distortion feature can be represented as:

[0109] Z swin =GlobalAvgPool(F swin )

[0110] Where GlobalAvgPool represents the global average pooling operation;

[0111] Thus, the distortion features of a single video frame can be obtained; similarly, the distortion features of all video frames in the video can be obtained.

[0112] In this embodiment of the invention, a distortion classification and rating module is also included. The purpose of this module is to classify and rate distortion features in order to quantitatively analyze the distortion present in the video signal. This module uses a deep multi-task learning model method to classify and evaluate the distortion in the video signal.

[0113] In practical implementation, separate classification layers are designed for distortion classification tasks and distortion rating tasks:

[0114] The classification layer of a distortion classification task can be represented as:

[0115] p class =Softmax(MLP) class (Z swin ))

[0116] Where, p class This indicates the distortion classification result of the video frame;

[0117] Specifically, the distortion of any video frame can be classified as follows: noise, out-of-focus blur, motion blur, uneven lighting, and smoke.

[0118] The classification layer of a distortion rating task can be represented as:

[0119] p score =MLP score (Z swin )

[0120] Where, p score This indicates the distortion rating of the video frame.

[0121] Specifically, the distortion level of any video frame is scored from 1 to 4, with 4 representing the most severe distortion.

[0122] For the video as a whole, the average value of all video frames (N frames in total) needs to be taken as the distortion classification and distortion rating result, which can be expressed as:

[0123]

[0124]

[0125] Furthermore, during training, a joint loss function is defined, which is a weighted sum of the loss functions for the distortion classification task and the distortion rating task. The cross-entropy loss function is used as the loss function for the distortion classification task, and the mean squared error is used as the loss function for the distortion rating task. This can be expressed as:

[0126] L = w class *CrossEntropyLoss(p class ,y class )+w score *MeanSquaredError(p score ,y score )

[0127] Among them, w class and w score This refers to the weight of each task, which can be set according to the importance of the task;

[0128] y class and y score These are labels for distortion classification and distortion levels.

[0129] In this embodiment of the invention, a distortion fault matching module is also included. The purpose of this module is to analyze the distortion situation and match possible system fault types. This module can use various fault diagnosis algorithms and techniques, such as analysis methods based on hidden Markov models, to analyze and predict the distortion situation, thereby helping medical staff or biomedical engineers locate possible fault types.

[0130] In practical implementation, the obtained video distortion classification C and distortion rating result S can be used as observations; the fault type T (T={t1,t2,…,t…) can be used as the basis for further observation. m} is a set of fault types) as hidden states, and then a Hidden Markov Model (HMM) is established to calculate the relationship between T, C, and S.

[0131] The calculation method for each intermediate variable can be expressed as follows:

[0132] State transition probability:

[0133] a ij =P(T) t+1 =j|T t =i)

[0134] Among them, a ij Let represent the probability that the fault type changes from i to j at time t;

[0135] P represents the probability of failover;

[0136] T t Indicates the fault type at time t;

[0137] Observation probability:

[0138] b i (c,s)=P(C=c,S=s|T=i)

[0139] b i (c,s) represents the probability of observing a distortion classification of c and a distortion rating of s when the fault type is i;

[0140] Prior state probabilities:

[0141] π i =P(T1=i)

[0142] π i This represents the probability that the fault type is i at the initial moment;

[0143] Introduction of prior knowledge:

[0144]

[0145] in, λ is prior knowledge, and λ is a weighting parameter between 0 and 1, used to balance the influence of observed data and prior knowledge.

[0146] Next, we use a recursive formula to calculate the probability distribution of fault types:

[0147]

[0148] in,

[0149] δ t(i) indicates that at time t, state i is observed and O1, O2, ..., O t The probability of the most likely state sequence;

[0150] Observe the sequence O = (C1, S1), (C2, S2), ..., (C N ,S N ), with a length of N;

[0151] variable δ t (i) represents a state i at time t and observations of O1, O2, ..., O t The probability of the most likely state sequence.

[0152] Finally, the forward-backward algorithm is used to calculate the most likely state sequence:

[0153] Forward variables:

[0154]

[0155] Where, α t (i) indicates that at time t, state i is observed and O1, O2, ..., O t The probability of;

[0156] Backward variables:

[0157]

[0158] Where, β t-1 (i) At time t-1, state i and observation O t O t+1 ,…,O N The probability of;

[0159] Final state sequence probability:

[0160]

[0161] By using the final state sequence probability, the matching relationship between the distortion type C and the distortion level S and the fault type T can be obtained, thus obtaining the final fault type T.

[0162] Based on the intelligently identified laparoscopic fault types, the system can provide corresponding suggestions according to the existing fault detection manual, such as cleaning the lens or adjusting the focus. It collects evaluations of the effectiveness of these self-inspection suggestions to continuously optimize them. This module intelligently improves the accuracy and reliability of the laparoscopic system's self-inspection, providing users with efficient and considerate self-inspection services and ensuring the safe and stable operation of the medical endoscope system.

[0163] Example 2:

[0164] like Figure 2As shown, a smart self-checking device for video distortion in a medical endoscope system includes:

[0165] The video processing module is used to acquire endoscopic videos;

[0166] The video frame distortion feature extraction module is used to extract features from all video frames of the endoscope video to obtain the distortion features of each video frame.

[0167] The video frame classification and rating module is used to classify and rate the distortion features.

[0168] The video classification and rating module is used to classify and rate the distortion of all video frames as the average value of the endoscopic video.

[0169] The fault type matching module is used for distortion classification and rating based on endoscopic video, matching the fault type that causes the distortion.

[0170] Example 3:

[0171] An apparatus comprising:

[0172] One or more processors;

[0173] Memory; and

[0174] One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including steps for performing the following:

[0175] Obtain endoscopic video;

[0176] Feature extraction is performed on all video frames of the endoscopic video to obtain the distortion features of each video frame;

[0177] The distortion features are classified and rated for distortion.

[0178] The average value of all video frames is used as the distortion classification and distortion rating of the endoscopic video;

[0179] Based on the distortion classification and distortion rating of endoscopic videos, the fault types that cause the distortion are matched.

[0180] It is understood that the intelligent self-testing device and equipment for video distortion of medical endoscope systems provided in the embodiments of the present invention correspond to the above-mentioned intelligent self-testing method for video distortion of medical endoscope systems. The explanations, examples, and beneficial effects of the relevant content can be referred to the corresponding content in the intelligent self-testing method for video distortion of medical endoscope systems, and will not be repeated here.

[0181] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0182] 1) This invention utilizes the Swin-Transformer network with multi-scale feature fusion to extract features of distorted videos and establishes a classification association between the types of distortion in medical endoscopy videos and distorted videos. At the same time, it obtains a specific assessment of the distortion level, thereby associating the types and severity of distortion.

[0183] 2) Based on the analysis method of Hidden Markov Model, this invention establishes a quantitative relationship between surgical video distortion classification, quality grade score and system failure, as well as assesses the relationship between various parameters related to system failure, thereby better understanding the cause of system failure and quickly and accurately locating and eliminating the fault, improving the reliability and timeliness of intelligent self-testing of medical laparoscopic system.

[0184] It should be noted that, through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the prior art, can be embodied in the form of software products. These computer software products can be stored in computer-readable storage media, such as ROM / RAM, magnetic disks, optical disks, etc., and include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or certain portions of the embodiments. In this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. In the absence of further restrictions, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0185] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent self-detection of video distortion in a medical endoscope system, characterized in that, The method includes: Obtain endoscopic video; Feature extraction is performed on all video frames of the endoscopic video to obtain the distortion features of each video frame; the feature extraction includes: based on pre-trained... -50 is used as a feature extractor to obtain the first feature. Based on the first feature Perform multi-scale feature fusion operation to obtain the multi-scale feature fusion at the th scale. The second feature of the layer output ;in, express The corresponding level in -50; based on the second feature conduct Operation, to obtain the third feature ; the third feature Input to the trained The network yields a fourth feature. ; Regarding the fourth feature Perform global average pooling to obtain the distortion features of the video frames. ; The distortion features are classified and rated for distortion. The average value of all video frames is used as the distortion classification and distortion rating of the endoscopic video; The fault types corresponding to the types and levels of distortion are obtained based on the Hidden Markov Model, so as to match the fault types that cause the distortion.

2. The intelligent self-checking method for video distortion in a medical endoscope system as described in claim 1, characterized in that, The endoscopic video has a fixed frame rate, and the RGB pixel values ​​of its video frames are normalized RGB pixel values.

3. The intelligent self-checking method for video distortion in a medical endoscope system as described in claim 1, characterized in that, The distortion features are classified and rated for distortion, including: The classification layer based on the distortion classification task obtains the distortion classification of the distortion features; wherein, the distortion classification includes: noise, defocus blur, motion blur, uneven lighting, and smoke; The classification layer based on the distortion rating task obtains the distortion rating of the distortion features.

4. The intelligent self-checking method for video distortion in a medical endoscope system as described in claim 3, characterized in that, During network training, the loss function is a weighted sum of the loss functions for the distortion classification task and the distortion rating task. The loss function for the distortion classification task is the cross-entropy loss function, and the loss function for the distortion rating task is the mean squared error.

5. A smart self-checking device for video distortion in a medical endoscope system, characterized in that, The device includes: The video processing module is used to acquire endoscopic videos; The video frame distortion feature extraction module is used to extract features from all video frames of the endoscope video to obtain the distortion features of each video frame. The video frame classification and rating module is used to classify and rate the distortion features. The video classification and rating module is used to classify and rate the distortion of all video frames as the average value of the endoscopic video. The fault type matching module is used for distortion classification and rating based on endoscopic video, matching the fault type that causes the distortion.

6. The intelligent self-checking device for video distortion in a medical endoscope system as described in claim 5, characterized in that, Feature extraction is performed on all video frames of the endoscopic video to obtain the distortion features of each video frame, including: Based on pre-training -50 is used as a feature extractor to obtain the first feature. ; Based on the first feature Perform multi-scale feature fusion operation to obtain the multi-scale feature fusion at the th scale. The second feature of the layer output ;in, express The corresponding level in -50; Based on the second feature conduct Operation, to obtain the third feature ; The third feature Input to the trained The network yields a fourth feature. ; For the fourth feature Perform global average pooling to obtain the distortion features of the video frames. .

7. A smart self-checking device for video distortion in a medical endoscope system, characterized in that, include: One or more processors; Memory; as well as One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including methods for performing intelligent self-testing methods for video distortion in medical endoscope systems as described in any one of claims 1-6.

Images