A surgical behavior capture system based on fast and slow neural networks

Abstract

The present invention discloses a surgical behavior capture system based on a fast and slow neural network, and relates to the technical field of monitoring and analysis. Through a screen segmentation module, a screen annotation module and a data judgment module, movable objects and fixed objects in an image can be accurately identified, and important parts of a video image can be accurately captured by annotating the moving movable objects and judging the reasons for the changes of the movable objects. At the same time, through an importance degree calculation module and a continuous change judgment module, the degree of influence of the change of the current screen image on the target video step can be judged in real time, and through the continuous change judgment module, whether the current video image is a continuous change can be accurately judged when the video step is performed, and by performing the continuously changing video step, omissions can be avoided when the video step is performed again, and at the same time, the video can be accurately captured through the different movable object change process steps, avoiding the need for manual viewing and screening after the video is captured.

Description

Technical Field

[0001] The present invention relates to the technical field of monitoring and analysis, and in particular to a surgical behavior capturing system based on fast and slow neural networks. Background Art

[0002] In recent years, with the rapid development of computers, networks, image processing and transmission technologies, video surveillance technology for surgical procedures has also made great progress. Current surgical procedure video surveillance equipment usually records video continuously. Continuous recording makes the video files larger, and larger video files store massive amounts of video data. The resulting technical problem is that when capturing target video based on images, only video data within a period of time before and after the current image can be captured. This will result in the need for manual viewing and screening of the captured video after video capture. Summary of the invention

[0003] The purpose of the present invention is to provide a surgical behavior capture system based on fast and slow neural networks to solve the above-mentioned deficiencies in the prior art.

[0004] In order to achieve the above-mentioned object, the present invention provides the following technical solutions: a surgical behavior capture system based on a fast and slow neural network, comprising a user-side input module, a sampling interval setting module, a picture acquisition module, a screen segmentation module, a screen comparison module, a screen annotation module, a data judgment module, an importance calculation module, a continuous change judgment module, a video capture module, and a video capture authentication module;

[0005] The user-side input module is used for the user to input the surgical behavior image to be queried;

[0006] The sampling interval setting module is used to set the sampling interval, and the sampling interval is the playing time period of the video;

[0007] The image acquisition module is used to acquire images within a sampling interval;

[0008] The image segmentation module is used to perform image segmentation on the image within the sampling interval and the standard image based on the image segmentation model, and to segment the image within the sampling interval and the standard image into movable objects and fixed objects;

[0009] The image comparison module is used to determine whether the movable object in the image frame within the sampling interval has changed relative to the movable object in the standard image;

[0010] The picture annotation module is used to annotate the changed image pictures in the sampling interval as changed images and annotate the changed positions in the image pictures in the sampling interval based on the picture annotation model, and annotate the unchanged image pictures in the sampling interval as unchanged images;

[0011] The data judgment module is used to judge the reason why the movable object in the image marked as a changed image changes based on the judgment model;

[0012] The importance calculation module is used to calculate the importance of the current change based on the importance calculation model, and sort according to the importance, preset a detection threshold, retain the annotation for the change image whose importance of the current change is higher than the preset detection threshold, and delete the annotation for the change image whose importance of the current change is lower than the preset detection threshold;

[0013] The continuous change judgment module is used to judge whether the change of the change image with the mark retained in the importance calculation module is a continuous change;

[0014] The video capture module is used to capture the video with complete annotated changes according to the changes of the corresponding annotations in the picture annotation module for the changing images judged as continuously changing by the data judgment module;

[0015] The video capture authentication module performs video capture authentication on each video segment captured in the video capture module based on a video comparison authentication model.

[0016] Further, the following steps are included:

[0017] S1, a user inputs a surgical behavior image to be queried through an input terminal, and the surgical behavior image to be queried is used as a standard image;

[0018] S2, set the sampling interval;

[0019] S3, acquiring an image within the sampling interval, and performing screen segmentation on the image within the sampling interval and the standard image based on a screen segmentation model, and segmenting the image within the sampling interval and the standard image into a movable object and a fixed object;

[0020] S4, judging whether the movable object in the image frame within the sampling interval has changed relative to the movable object in the standard image based on the image comparison model, marking the changed image frame within the sampling interval as a changed image and marking the changed position within the image frame within the sampling interval based on the image annotation model, and marking the unchanged image frame within the sampling interval as an unchanged image;

[0021] S5, judging the reason why the movable object in the changed image in step S4 changes based on the judgment model, wherein the reason for the change is not limited to human factors, environmental factors, and light factors;

[0022] S6, calculating the importance of the current change based on the importance calculation model, and sorting according to the importance, presetting a detection threshold, retaining the annotation for the change image whose importance of the current change is higher than the preset detection threshold, and deleting the annotation for the change image whose importance of the current change is lower than the preset detection threshold;

[0023] S7, determining whether the change of the marked change image retained in step S6 is a continuous change;

[0024] S8, for the changing image judged as continuously changing in step S7, according to the change of the corresponding annotation in step S4, intercept the video with complete annotated changes, for the changing images with multiple annotated changes, intercept and classify and store according to different annotations, wherein the specific standard for video interception is a complete change process of the annotation of the changing image;

[0025] S9, performing video capture authentication on each video segment captured in step S8 based on the video comparison authentication model.

[0026] Furthermore, the step S3 performs screen segmentation on the image within the sampling interval and the standard image based on screen segmentation, specifically comprising the following steps:

[0027] A1, segment the image, mark one part of the image as fixed objects as N, and the other part as movable objects as M. Array F[n] is used to record the labels;

[0028] A2, scan the fixed object and movable object area pixel by pixel, obtain the angle value x1 of each pixel in the color vector graph, and set the value of F[x1] to 1;

[0029] A3, scan the M area containing the movable object point by point, obtain the angle value x2 corresponding to the non-pixel point, calculate F[x2], if the value of the point is not 1, retain it, otherwise remove it;

[0030] The movable object photo is set as P, and P(x,y) is defined as the brightness value of the point with internal coordinates (x,y) on the movable object, which is converted into polar coordinates using the formula:

[0031] P'(r, θ)=P(r sinθ+x0, r cosθ+y0)

[0032] Where P'(γ,θ) represents the brightness change value of the point with internal coordinates (γ,θ);

[0033] In order to reduce the unknown influence of local features such as cracks of different degrees on movable objects on segmentation, we first perform Gaussian filtering on the image, then extract the brightness channel and hue channel from the three color attributes for polar coordinate transformation. After completion, we use the edge detection operator to perform an edge detection and use a vertical edge detector to detect the image;

[0034] The detection result is converted back to the rectangular coordinate system using the following formula, and then a morphological operation is performed to obtain a closed edge:

[0035]

[0036] By identifying different closed edges, the specific shape of movable objects can be detected, which facilitates image segmentation.

[0037] Furthermore, the generation of the importance model in step S6 specifically includes the following steps:

[0038] B1, randomized initialization variation data set;

[0039] B2, calculate the data fitness, based on the weight value and initial threshold in the back propagation neural network determined by the individual, the specific calculation formula is as follows:

[0040]

[0041] Where a is the number of nodes of the change data, yb is the predicted data of the b-th node, ob is the actual data of the b-th node, and c is the coefficient;

[0042] B3, selection operation, using the roulette method to perform the selection operation, and then select individuals with good fitness from the variation data to form a new variation data group. The specific formula for the probability of individual selection is as follows:

[0043]

[0044] Where c is the coefficient, N is a random number, Pb represents the data fitness of the b-th node, and Pj represents the data fitness of the j-th node;

[0045] B4, cross operation, randomly select two change data for cross operation, take the xth change data ax and the yth change data ay at the jth position as an example, the specific formula is as follows:

[0046]

[0047] Where z is a random number between [0, 1];

[0048] B5, mutation operation, randomly selects a variation data from the variation data group, and performs mutation operation according to a preset probability to obtain new variation data;

[0049] B6, using the obtained optimal value to update the weight threshold, training to obtain the current change importance model.

[0050] Furthermore, the step S7 of determining whether the important change is a continuous change specifically includes the following steps:

[0051] C1, first, detect each sub-annotation data spectrum in the image annotation result based on the local model;

[0052] C2, if the posterior score of a sub-annotated spectrogram exceeds the threshold, it is considered a local change;

[0053] C3, the system records the start time and end time of continuous local changes. The start time is the left edge of the window corresponding to the first local change, and the end time is the right edge of the window corresponding to the last local change.

[0054] C4, when continuous local changes occur, it indicates that there are continuous changes in the image annotations within the time period from the start time to the end time, that is, the important changes currently detected are continuous changes;

[0055] C5, if no continuous local changes occur, the currently detected important changes are considered as discontinuous changes;

[0056] C6, re-inputting the non-continuous change image annotations in step C5 into step C1 for subsequent operations until there is no continuous change in the important change annotations.

[0057] Furthermore, the video capture authentication in step S9 specifically includes the following steps:

[0058] D1, in response to each video segment captured in step S8, extracting image data that meets the specified image condition from each video segment; any image data that meets the specified image condition carries at least identification information and timestamp information of the beacon device that sends the image data;

[0059] D2, if extracted, obtaining the specified data in the extracted image data; wherein the specified data at least includes the identification information and timestamp information of the beacon device that sent the extracted image data;

[0060] D3, obtaining target identification information of each video segment intercepted in step S8, and target timestamp information of the video;

[0061] D4, performing a matching analysis on the target timestamp information and the timestamp information in the specified data to obtain a first analysis result, and performing a matching analysis on the target identification information and the identification information in the specified data to obtain a second analysis result;

[0062] D5, based on the first analysis result and the second analysis result, determining the authentication result of the video; wherein the authentication result is used to indicate whether each video segment captured in the step S8 is fake.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] The present invention provides a surgical behavior capture system based on a fast and slow neural network. The movable objects and fixed objects in the image can be accurately identified through a screen segmentation module, a screen annotation module and a data judgment module. The important parts of the video image can be accurately captured by annotating the moving movable objects and judging the reasons for the changes in the movable objects. At the same time, the importance calculation module and the continuous change judgment module can be used to judge in real time the degree of influence of the change of the current screen image on the target video step. The continuous change judgment module can accurately judge whether the current video image is continuously changing when performing the video step. By performing the continuously changing video step, omissions can be avoided when performing the video step again. At the same time, the video can be accurately captured through the different movable object change process steps, avoiding the need for manual viewing and screening after video capture. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0066] Figure 1 A system structure block diagram provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0067] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0068] Example embodiments will be described more fully below with reference to the accompanying drawings, but the example embodiments may be embodied in different forms and should not be construed as limited to the embodiments set forth herein. On the contrary, the purpose of providing these embodiments is to make the present disclosure thorough and complete and to enable those skilled in the art to fully understand the scope of the present disclosure.

[0069] In the absence of conflict, the various embodiments of the present disclosure and the various features therein may be combined with each other.

[0070] As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0071] The terms used herein are only used to describe specific embodiments and are not intended to limit the present disclosure. As used herein, the singular forms "a", "an" and "the" are also intended to include the plural forms, unless the context clearly indicates otherwise. It will also be understood that when the terms "comprising" and / or "made of" are used in this specification, the presence of the features, wholes, steps, operations, elements and / or components is specified, but the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or groups thereof is not excluded.

[0072] The embodiments described herein may be described with reference to plan views and / or cross-sectional views by means of idealized schematic diagrams of the present disclosure. Therefore, the example illustrations may be modified according to manufacturing techniques and / or tolerances. Therefore, the embodiments are not limited to the embodiments shown in the accompanying drawings, but include modifications of the configurations formed based on the manufacturing process. Therefore, the regions illustrated in the accompanying drawings have schematic properties, and the shapes of the regions shown in the figures illustrate the specific shapes of the regions of the elements, but are not intended to be limiting.

[0073] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those of ordinary skill in the art. It will also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and the present disclosure, and will not be interpreted as having an idealized or overly formal meaning unless explicitly defined as such herein.

[0074] See also Figure 1 A surgical behavior capture system based on fast and slow neural networks includes a user-side input module, a sampling interval setting module, a picture acquisition module, a screen segmentation module, a screen comparison module, a screen annotation module, a data judgment module, an importance calculation module, a continuous change judgment module, a video capture module, and a video capture authentication module.

[0075] The user-side input module is used for the user to input the surgical behavior image to be queried;

[0076] The sampling interval setting module is used to set the sampling interval, and the sampling interval is the playing time period of the video;

[0077] The image acquisition module is used to acquire images within a sampling interval;

[0078] The image segmentation module is used to perform image segmentation on the image within the sampling interval and the standard image based on the image segmentation model, and to segment the image within the sampling interval and the standard image into movable objects and fixed objects.

[0079] The picture comparison module is used to determine whether the movable object in the image picture within the sampling interval has changed relative to the movable object in the standard image; the picture comparison module can use a deep learning model (such as a convolutional neural network CNN) to extract features from the image and calculate the similarity of the features for comparison.

[0080] The picture annotation module is used to annotate the changed image pictures in the sampling interval as changed images and annotate the changed positions in the image pictures in the sampling interval based on the picture annotation model, and annotate the unchanged image pictures in the sampling interval as unchanged images;

[0081] The data judgment module is used to judge the reason why the movable object in the image marked as a changed image changes based on the judgment model.

[0082] The importance calculation module is used to calculate the importance of the current change based on the importance calculation model, and sort according to the importance, preset a detection threshold, retain the annotation for the change image whose importance of the current change is higher than the preset detection threshold, and delete the annotation for the change image whose importance of the current change is lower than the preset detection threshold;

[0083] The continuous change judgment module is used to judge whether the change of the change image with the mark retained in the importance calculation module is a continuous change;

[0084] The video capture module is used to capture the video with complete annotated changes according to the changes of the corresponding annotations in the picture annotation module for the changing images judged as continuously changing by the data judgment module;

[0085] The video capture authentication module performs video capture authentication on each video segment captured in the video capture module based on a video comparison authentication model.

[0086] The picture segmentation module, the picture labeling module and the data judgment module can accurately identify the movable objects and fixed objects in the image, and the important parts of the step video image can be accurately stepped by labeling the moving movable objects and judging the reasons for the changes of the movable objects. At the same time, the importance calculation module and the continuous change judgment module can judge in real time the degree of influence of the change of the current picture image on the target video step, and the continuous change judgment module can accurately judge whether the current video image is continuously changing when performing the video step. By stepping the continuously changing video, omissions can be avoided when performing the video step again. At the same time, the video step can be accurately performed through the steps of different movable object change processes, avoiding the need for manual viewing and screening after the video step.

[0087] The following steps are involved:

[0088] S1, a user inputs a surgical behavior image to be queried through an input terminal, and the surgical behavior image to be queried is used as a standard image;

[0089] S2, set the sampling interval;

[0090] S3, acquiring an image within the sampling interval, and performing screen segmentation on the image within the sampling interval and the standard image based on a screen segmentation model, and segmenting the image within the sampling interval and the standard image into a movable object and a fixed object;

[0091] S4, judging whether the movable object in the image frame within the sampling interval has changed relative to the movable object in the standard image based on the image comparison model, marking the changed image frame within the sampling interval as a changed image and marking the changed position within the image frame within the sampling interval based on the image annotation model, and marking the unchanged image frame within the sampling interval as an unchanged image;

[0092] S5, judging the reason why the movable object in the changed image in step S4 changes based on the judgment model, wherein the reason for the change is not limited to human factors, environmental factors, and light factors;

[0093] S6, calculating the importance of the current change based on the importance calculation model, and sorting according to the importance, presetting a detection threshold, retaining the annotation for the change image whose importance of the current change is higher than the preset detection threshold, and deleting the annotation for the change image whose importance of the current change is lower than the preset detection threshold;

[0094] S7, determining whether the change of the marked change image retained in step S6 is a continuous change;

[0095] S8, for the changing image judged as continuously changing in step S7, according to the change of the corresponding annotation in step S4, intercept the video with complete annotated changes, for the changing images with multiple annotated changes, intercept and classify and store according to different annotations, wherein the specific standard for video interception is a complete change process of the annotation of the changing image;

[0096] S9, performing video capture authentication on each video segment captured in step S8 based on the video comparison authentication model.

[0097] In step S3, the image within the sampling interval and the standard image are segmented based on the image segmentation, which specifically includes the following steps:

[0098] A1, segment the image, mark one part of the image as fixed objects as N, and the other part as movable objects as M. Array F[n] is used to record the labels;

[0099] A2, scan the fixed object and movable object area pixel by pixel, obtain the angle value x1 of each pixel in the color vector graph, and set the value of F[x1] to 1;

[0100] A3, scan the M area containing the movable object point by point, obtain the angle value x2 corresponding to the non-pixel point, calculate F[x2], if the value of the point is not 1, retain it, otherwise remove it;

[0101] The movable object photo is set as P, and P(x,y) is defined as the brightness value of the point with internal coordinates (x,y) on the movable object, which is converted into polar coordinates using the formula:

[0102] P'(r, θ)=P(r sinθ+x0, r cosθ+y0)

[0103] Where P'(γ,θ) represents the brightness change value of the point with internal coordinates (γ,θ);

[0104] In order to reduce the unknown influence of local features such as cracks of different degrees on movable objects on segmentation, we first perform Gaussian filtering on the image, then extract the brightness channel and hue channel from the three color attributes for polar coordinate transformation. After completion, we use the edge detection operator to perform an edge detection and use a vertical edge detector to detect the image;

[0105] The detection result is converted back to the rectangular coordinate system using the following formula, and then a morphological operation is performed to obtain a closed edge:

[0106]

[0107] By identifying different closed edges, the specific shape of movable objects can be detected, which facilitates image segmentation.

[0108] The generation of the importance model in step S6 specifically includes the following steps:

[0109] B1, random initialization variation data set;

[0110] B2, calculate the data fitness, based on the weight value and initial threshold in the back propagation neural network determined by the individual, the specific calculation formula is as follows:

[0111]

[0112] Where a is the number of nodes of the change data, yb is the predicted data of the b-th node, ob is the actual data of the b-th node, and c is the coefficient;

[0113] B3, selection operation, using the roulette method to perform the selection operation, and then select individuals with good fitness from the variation data to form a new variation data group. The specific formula for the probability of individual selection is as follows:

[0114]

[0115] Where c is the coefficient, N is a random number, Pb represents the data fitness of the b-th node, and Pj represents the data fitness of the j-th node;

[0116] B4, cross operation, randomly select two change data for cross operation, take the xth change data ax and the yth change data ay at the jth position as an example, the specific formula is as follows:

[0117]

[0118] Where z is a random number between [0, 1];

[0119] B5, mutation operation, randomly selects a variation data from the variation data group, and performs mutation operation according to a preset probability to obtain new variation data;

[0120] B6, using the obtained optimal value to update the weight threshold, training to obtain the current change importance model.

[0121] In step S7, determining whether the important change is a continuous change specifically includes the following steps:

[0122] C1, first, detect each sub-annotation data spectrum in the image annotation result based on the local model;

[0123] C2, if the posterior score of a sub-annotated spectrogram exceeds the threshold, it is considered a local change;

[0124] C3, the system records the start time and end time of continuous local changes. The start time is the left edge of the window corresponding to the first local change, and the end time is the right edge of the window corresponding to the last local change.

[0125] C4, when continuous local changes occur, it indicates that there are continuous changes in the image annotations within the time period from the start time to the end time, that is, the important changes currently detected are continuous changes;

[0126] C5, if no continuous local changes occur, the currently detected important changes are considered as discontinuous changes;

[0127] C6, re-inputting the non-continuous change image annotations in step C5 into step C1 for subsequent operations until there is no continuous change in the important change annotations.

[0128] The video capture authentication in step S9 specifically includes the following steps:

[0129] D1, in response to each video segment captured in step S8, extracting image data that meets the specified image condition from each video segment; any image data that meets the specified image condition carries at least identification information and timestamp information of the beacon device that sends the image data;

[0130] D2, if extracted, obtaining the specified data in the extracted image data; wherein the specified data at least includes the identification information and timestamp information of the beacon device that sent the extracted image data;

[0131] D3, obtaining target identification information of each video segment intercepted in step S8, and target timestamp information of the video;

[0132] D4, performing a matching analysis on the target timestamp information and the timestamp information in the specified data to obtain a first analysis result, and performing a matching analysis on the target identification information and the identification information in the specified data to obtain a second analysis result;

[0133] D5, based on the first analysis result and the second analysis result, determining the authentication result of the video; wherein the authentication result is used to indicate whether each video segment captured in the step S8 is fake.

[0134] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A surgical behavior capture system based on fast and slow neural networks, characterized by: It includes a user-side input module, a sampling interval setting module, a picture acquisition module, a picture segmentation module, a picture comparison module, a picture annotation module, a data judgment module, an importance calculation module, a continuous change judgment module, a video capture module, and a video capture authentication module; wherein: The user-side input module is used for the user to input the surgical behavior image to be queried; The sampling interval setting module is used to set the sampling interval, and the sampling interval is the playing time period of the video; The image acquisition module is used to acquire images within a sampling interval; The image segmentation module is used to perform image segmentation on the image within the sampling interval and the standard image based on the image segmentation model, and to segment the objects in the image within the sampling interval and the standard image into movable objects and fixed objects; The image comparison module is used to determine whether the movable object in the image within the sampling interval has changed relative to the movable object in the standard image; The picture annotation module is used to annotate the changed image pictures in the sampling interval as changed images and annotate the changed positions in the image pictures in the sampling interval based on the picture annotation model, and annotate the unchanged image pictures in the sampling interval as unchanged images; The data judgment module is used to judge the reason why the movable object in the image marked as a changed image changes based on the judgment model; The importance calculation module is used to calculate the importance of the current change based on the importance calculation model, and sort according to the importance, preset a detection threshold, retain the annotation for the change image whose importance of the current change is higher than the preset detection threshold, and delete the annotation for the change image whose importance of the current change is lower than the preset detection threshold; The continuous change judgment module is used to judge whether the change of the change image with the mark retained in the importance calculation module is a continuous change; The video capture module is used to capture the video with complete annotated changes according to the changes of the corresponding annotations in the picture annotation module for the changed images judged as continuously changing by the continuous change judgment module; The video capture authentication module performs video capture authentication on each video segment captured in the video capture module based on the video comparison authentication model; The generation of the importance calculation module specifically includes the following steps: B1, randomized initialization variation data set; B2, calculate the data fitness, based on the weight value and initial threshold in the back propagation neural network determined by the individual, the specific calculation formula is as follows: Where a is the number of nodes of the variable data, y b is the prediction data of the bth node, o b is the actual data of the bth node, and c is the coefficient; B3, selection operation, using the roulette method to perform the selection operation, and then select individuals with good fitness from the variation data to form a new variation data group. The specific formula for the probability of individual selection is as follows: Where c is the coefficient, N is a random number, Pb represents the data fitness of the b-th node, and Pj represents the data fitness of the j-th node; B4, cross operation, randomly select two variable data for cross operation, and use the xth variable data a at the jth position x and the yth change data a y For example, the specific formula is as follows: Where z is a random number between [0, 1]; B5, mutation operation, randomly selects a variation data from the variation data group, and performs mutation operation according to a preset probability to obtain new variation data; B6, using the obtained optimal value to update the weight threshold, training to obtain the current change importance model.

2. A surgical behavior capture system based on fast and slow neural networks according to claim 1, characterized in that: Based on the screen segmentation, the image within the sampling interval and the standard image are segmented, specifically including the following steps: A1, segment the image, mark the area with fixed objects as N, and the area with movable objects as M. Array F[n] is used to record the marks; A2, scan the fixed object and movable object area pixel by pixel, obtain the angle value x1 of each pixel in the color vector graph, and set the value of F[x1] to 1; A3, scan the M area containing the movable object point by point, obtain the angle value x2 corresponding to the non-pixel point, calculate F[x2], if the value of the point is not 1, retain it, otherwise remove it; The movable object photo is set as P, and P(x,y) is defined as the brightness value of the point with internal coordinates (x,y) on the movable object, which is converted into polar coordinates using the formula: P'(r, θ)=P(r sinθ+x0, r cosθ+y0) Where P'(γ,θ) represents the brightness change value of the point with internal coordinates (γ,θ); In order to reduce the impact of different degrees of cracks on movable objects on segmentation, we first perform Gaussian filtering on the image, then extract the brightness channel and hue channel from the three color attributes for polar coordinate conversion. After completion, we use the edge detection operator to perform an edge detection and use a vertical edge detector to detect the image; The detection result is converted back to the rectangular coordinate system using the following formula, and then a morphological operation is performed to obtain a closed edge: By identifying different closed edges, the specific shape of movable objects can be detected, which facilitates image segmentation.

3. The surgical behavior capture system based on fast and slow neural networks according to claim 1, characterized in that: The continuous change judgment module judges whether the change is a continuous change, specifically including the following steps: C1, first, detect each sub-annotation data spectrum in the image annotation result based on the local model; C2, if the posterior score of a sub-annotated spectrogram exceeds the threshold, it is considered a local change; C3, the system records the start time and end time of continuous local changes. The start time is the left edge of the window corresponding to the first local change, and the end time is the right edge of the window corresponding to the last local change. C4, when continuous local changes occur, it indicates that there are continuous changes in the image annotations within the time period from the start time to the end time, that is, the important changes currently detected are continuous changes; C5, if no continuous local changes occur, the currently detected important changes are considered as discontinuous changes; C6, re-inputting the non-continuous change image annotations in step C5 into step C1 for subsequent operations until there is no continuous change in the important change annotations.

4. The surgical behavior capture system based on fast and slow neural networks according to claim 3, characterized in that: Video capture authentication includes the following steps: D1, in response to each captured video segment, extracting image data that meets the specified image condition from each video segment; any image data that meets the specified image condition carries at least identification information and timestamp information of the beacon device that sends the image data; D2, if extracted, obtaining the specified data in the extracted image data; wherein the specified data at least includes the identification information and timestamp information of the beacon device that sent the extracted image data; D3, obtaining target identification information of each captured video segment and target timestamp information of the video; D4, performing a matching analysis on the target timestamp information and the timestamp information in the specified data to obtain a first analysis result, and performing a matching analysis on the target identification information and the identification information in the specified data to obtain a second analysis result; D5. Determine an authentication result of the video based on the first analysis result and the second analysis result; wherein the authentication result is used to indicate whether each captured video segment is fake.

5. A surgical behavior capture method based on a fast and slow neural network according to claim 1, characterized in that: The following steps are involved: S1, a user inputs a surgical behavior image to be queried through an input terminal, and the surgical behavior image to be queried is used as a standard image; S2, set the sampling interval; S3, acquiring an image within the sampling interval, and performing screen segmentation on the image within the sampling interval and the standard image based on a screen segmentation model, and segmenting objects in the image within the sampling interval and the standard image into movable objects and fixed objects; S4, judging whether the movable object in the image within the sampling interval has changed relative to the movable object in the standard image based on the picture comparison model; if there is a change, marking the changed image within the sampling interval as a changed image based on the picture annotation model, and annotating the changed position in the image within the sampling interval; if there is no change, marking the image within the sampling interval as an unchanged image; S5, judging the reason why the movable object in the changed image in step S4 changes based on the judgment model, wherein the reason for the change is not limited to human factors, environmental factors, and light factors; S6, calculating the importance of the current change based on the importance calculation model, and sorting according to the importance, and presetting a detection threshold, retaining the annotation for the change image whose importance of the current change is higher than the preset detection threshold, and deleting the annotation for the change image whose importance of the current change is lower than the preset detection threshold; S7, determining whether the change of the marked change image retained in step S6 is a continuous change; S8, for the changing image determined as continuously changing in step S7, according to the change of the corresponding annotation in step S4, intercept the video with the complete annotation change, and for the changing images with multiple annotation changes, intercept and classify and store them according to different annotations, wherein the specific standard for video interception is a complete change process of the annotation of the changing image; S9, performing video capture authentication on each video segment captured in step S8 based on the video comparison authentication model.

Images