An anomaly detection method for virtual simulation of anesthesiology crisis skills

Abstract

The present invention relates to the field of outlier detection technology, and specifically to an anomaly detection method for virtual simulation of crisis skills in anesthesiology, by determining the local feature code of each pixel in the grayscale image of each frame of virtual simulation image, thereby determining the feature point extraction index of each frame of grayscale image, and based on the feature point extraction index, a plurality of feature point groups of each frame of grayscale image are obtained. By analyzing the position change characteristics of each feature point in each feature point group, the noise insensitivity corresponding to the feature point group is determined, and based on the noise insensitivity, the low-sensitivity pixels in each frame of grayscale image are screened out, and finally the abnormal pixel detection in each frame of grayscale image is completed. The present invention effectively improves the accuracy of abnormal data detection in virtual simulation images by identifying low-sensitivity pixels belonging to scene feature information in each frame of virtual simulation image.

Description

Technical Field

[0001] The present invention relates to the technical field of outlier detection, and in particular to an anomaly detection method for virtual simulation of anesthesiology crisis skills. Background Art

[0002] Virtual simulation creates a virtual world of tactile perception, visual perception and information interaction. Virtual simulation of anesthesiology crisis skills has gradually been applied to medical teaching and practice, and has become an important means to improve the anesthesia ability of medical practitioners. In the virtual simulation of anesthesiology crisis skills, it is necessary to detect abnormal noise data in the virtual simulation image that can be processed by the computer, and denoise the detected abnormal data to eliminate the influence of abnormal data on the virtual simulation effect, thereby improving the quality of the virtual simulation image, so that medical practitioners can better view and learn.

[0003] As a common anomaly detection algorithm, the Isolation Forest Detection Algorithm can be used for abnormal data detection in virtual simulation images. However, when using the Isolation Forest Detection Algorithm for abnormal data detection, since different types of data points in the virtual simulation image have the same sensitivity, that is, all data points are treated indiscriminately, some scene feature information is easily misjudged as abnormal data, thereby reducing the accuracy of abnormal data detection in the virtual simulation image, and further affecting the quality of the final virtual simulation image. For example, the moving direction and edge details of the human outline are the most important scene feature information in the virtual simulation image. Due to its own grayscale change characteristics, it is easy to be identified as abnormal data. Summary of the invention

[0004] The object of the present invention is to provide an anomaly detection method for virtual simulation of anesthesiology crisis skills, so as to solve the problem of low accuracy in detecting abnormal data in existing simulated images.

[0005] In order to solve the above technical problems, the present invention provides an abnormality detection method for virtual simulation of anesthesiology crisis skills, comprising the following steps:

[0006] Obtaining a grayscale image of each frame of virtual simulation image, and determining a local feature code of each pixel in each frame of grayscale image;

[0007] Determine the information amount index of each frame of grayscale image according to the number of jumps of the element value in the local feature code, and determine the feature point extraction index of each frame of grayscale image according to the information amount index;

[0008] Extracting feature points from each frame of grayscale image according to the feature point extraction index to determine feature points in each frame of grayscale image, and grouping the feature points according to position distribution of the feature points to determine at least two feature point groups for each frame of grayscale image;

[0009] Determine the noise insensitivity corresponding to the feature point group according to the position change characteristics of each feature point in the feature point group, and determine the low-sensitivity pixel points in each frame of the grayscale image according to the noise insensitivity;

[0010] According to the low-sensitivity pixels, abnormality detection is performed on the pixels in each frame of grayscale image to determine the abnormal pixels in each frame of grayscale image.

[0011] Furthermore, the information content index of each frame of grayscale image is determined, including:

[0012] According to the jump times of the element values ​​in the local feature code, a local feature code whose element value jump times in each frame of the grayscale image is greater than a set jump times threshold is determined as a target local feature code;

[0013] The average value of the number of jumps of the element value in the target local feature code in each frame of grayscale image is determined, and the information content index of each frame of grayscale image is determined according to the number of jumps of the element value in the target local feature code in each frame of grayscale image and the difference between each local feature code in each frame of grayscale image and the average value.

[0014] Furthermore, the corresponding calculation formula for determining the information amount index of each frame of grayscale image is:

[0015] Where Id(x) represents the information content index of the x-th grayscale image; q(j) represents the number of jumps of the element value in the j-th target local feature code in the x-th grayscale image; M′ represents the total number of target local feature codes in the x-th grayscale image; q max represents the maximum number of jumps of the element value in the local feature code; p(i) represents the number of jumps of the element value in the i-th local feature code in the x-th frame grayscale image, represents the average number of jumps of the element value in the target local feature code in the x-th frame grayscale image; N represents the total number of local feature codes in the x-th frame grayscale image; ε1 represents the first coefficient value; Norm{} represents the normalization function.

[0016] Furthermore, the feature point extraction index of each frame of grayscale image is determined, including:

[0017] Initialize the feature point extraction index of the first frame grayscale image, and perform negative correlation mapping on the information index of each frame grayscale image to obtain the information index mapping value of each frame grayscale image;

[0018] Each frame of grayscale image except the first frame of grayscale image is taken as the object grayscale image, and the feature point extraction index of each frame of object grayscale image is determined according to the information amount index mapping value of each frame of object grayscale image and the feature point extraction index of the previous frame of grayscale image of the frame of object grayscale image.

[0019] Further, determining at least two feature point groups of each frame of grayscale image includes:

[0020] The step of generating a feature point group is performed at least twice in each frame of grayscale image, and each of the steps of generating a feature point group includes:

[0021] Among the feature points currently remaining in each frame of grayscale image, a first target feature point is randomly determined, and a feature point within a neighborhood range of the first target feature point that is closest to the first target feature point is determined as a second target feature point, and a feature point within a neighborhood range of the first target feature point that is second closest to the first target feature point is determined as a third target feature point, and so on, until a termination condition is met, the termination condition being: the total number of determined target feature points reaches a set number threshold, or there are no remaining feature points within a neighborhood range of the first target feature point;

[0022] All the determined target feature points are regarded as a feature point group.

[0023] Further, determining the noise insensitivity corresponding to the feature point group includes:

[0024] According to the position of each feature point in the feature point group, curve fitting is performed on each feature point in the feature point group to obtain a fitting curve;

[0025] Determine a fitting point of each feature point in the feature point group in the fitting curve, and determine a tangent slope value corresponding to the fitting point;

[0026] Determine, according to the positions of each feature point in the feature point group, a first nearest neighbor feature point and a second nearest neighbor feature point of each feature point in the feature point group;

[0027] Determine the Euclidean distance between each feature point in the feature point group and its first nearest neighbor feature point to obtain a first distance corresponding to each feature point in the feature point group, and determine the Euclidean distance between each feature point in the feature point group and its second nearest neighbor feature point to obtain a second distance corresponding to each feature point in the feature point group;

[0028] Determine a non-structural index corresponding to the feature point group according to a difference between a tangent slope value corresponding to each feature point in the feature point group and a fitting point of its first nearest neighbor feature point, and a difference between a first distance and a second distance corresponding to each feature point in the feature point group;

[0029] Negative correlation mapping is performed on the non-structural indicators corresponding to the feature point group, and the negative correlation mapping result is used as the noise insensitivity corresponding to the feature point group.

[0030] Furthermore, the non-structural index corresponding to the feature point group is determined, and the corresponding calculation formula is:

[0031] Wherein, Cos(u) represents the non-structural index corresponding to the u-th feature point group in each frame of grayscale image; k m k represents the tangent slope value corresponding to the fitting point of the mth feature point in the uth feature point group in each frame of grayscale image; ′ m represents the tangent slope value corresponding to the fitting point of the first nearest neighbor feature point of the mth feature point in the uth feature point group in each frame of the grayscale image; m,1 represents the first distance corresponding to the mth feature point in the uth feature point group in each frame of grayscale image; l m,2 It represents the second distance corresponding to the mth feature point in the uth feature point group in each frame of grayscale image; M represents the total number of feature points in the uth feature point group in each frame of grayscale image; || represents the absolute value function; ε2 represents the second coefficient value; arctan() inverse tangent function.

[0032] Furthermore, low-sensitivity pixels in each frame of grayscale image are determined, including:

[0033] Comparing the noise insensitivity corresponding to each of the feature point groups in each frame of the grayscale image with a set insensitivity threshold, and determining the feature points in the feature point group corresponding to the noise insensitivity greater than the set insensitivity threshold as initial low-sensitivity pixels in each frame of the grayscale image;

[0034] Determine the neighboring pixel points of the initial low-sensitivity pixel points in each frame of grayscale image, and determine the neighboring pixel points as the extended low-sensitivity pixel points in each frame of grayscale image;

[0035] The initial low-sensitivity pixels and the expanded low-sensitivity pixels in each frame of grayscale image are used as the final low-sensitivity pixels in each frame of grayscale image.

[0036] Furthermore, determining abnormal pixels in each frame of grayscale image includes:

[0037] The isolation forest is used to perform anomaly detection on the pixel points in each frame of the grayscale image, and during the anomaly detection process, the average path corresponding to the low-sensitivity pixel points is set to a fixed value, so as to obtain the abnormal pixel points in each frame of the grayscale image, and the fixed value is not greater than the set path threshold.

[0038] Furthermore, feature points are extracted from each frame of grayscale image to determine the feature points in each frame of grayscale image, including:

[0039] The SIFT algorithm is used to extract feature points from each frame of grayscale image, and the feature point extraction index of each frame of grayscale image is used as the initial variance when the SIFT algorithm extracts feature points, thereby obtaining the feature points in each frame of grayscale image.

[0040] The present invention has the following beneficial effects: the present invention obtains the grayscale image of each frame of virtual simulation image, and identifies the texture features around each pixel in the grayscale image to determine the local feature code of each pixel. Based on the jump times of the element value in the local feature code, the grayscale chaos and complexity in the grayscale image are analyzed to determine the amount of information contained in the grayscale image, thereby determining the information amount index of each frame of grayscale image, and then adaptively extracting features from each frame of grayscale image according to the information amount index, thereby accurately extracting feature points that meet the characteristics of different image scenes. For the convenience of analysis, feature points with similar positions are divided into a group, thereby obtaining multiple feature point groups. By analyzing the position change characteristics of each feature point in each feature point group, the noise insensitivity corresponding to different feature point groups is determined, and the noise insensitivity reflects the possibility of the feature point group belonging to the scene feature information, thereby accurately identifying the low-sensitivity pixel points belonging to the scene feature information. Based on the low-sensitivity pixel points determined to belong to the scene feature information, the pixel points in each frame of grayscale image are detected for abnormality, so as to avoid misjudging the low-sensitivity pixel points as abnormal data, and finally accurate abnormal data can be obtained. The present invention accurately identifies low-sensitivity pixels belonging to scene feature information in each frame of virtual simulation image, thereby avoiding the scene feature information being misjudged as abnormal data, and effectively improving the accuracy of abnormal data detection in virtual simulation images. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1The present invention is a flowchart of an abnormality detection method for virtual simulation of anesthesiology crisis skills according to an embodiment of the present invention. DETAILED DESCRIPTION

[0043] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the technical solutions proposed by the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0044] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. In addition, all parameters or indicators in the formulas involved in this article are normalized values ​​that eliminate the influence of dimensions.

[0045] In order to solve the problem of inaccurate data anomaly detection in the existing anesthesiology crisis skills virtual simulation, this embodiment provides an anomaly detection method for anesthesiology crisis skills virtual simulation, and the corresponding flowchart of the method is as follows: Figure 1 As shown, the following steps are included:

[0046] Step S1: obtaining a grayscale image of each frame of virtual simulation image, and determining a local feature code of each pixel in each frame of grayscale image.

[0047] In the virtual simulation of crisis skills in anesthesiology, virtual simulation images of different frames are collected and grayed to obtain grayscale images of each frame of virtual simulation images. For each frame of grayscale image, the LBP (Local Binary Pattern) algorithm is used to extract the local texture features of each pixel point, and the LBP code of each pixel point is determined, and the LBP code is the local feature code. In this embodiment, when the LBP algorithm is used to extract the local texture features of each pixel point, a circular LBP operator with a radius of 3 is used.

[0048] Step S2: determining the information amount index of each frame of grayscale image according to the number of jumps of the element value in the local feature code, and determining the feature point extraction index of each frame of grayscale image according to the information amount index.

[0049] For each pixel in each frame of grayscale image, the information index of each frame of grayscale image can be determined by analyzing the number of jumps of the element value in the LBP code of the pixel. The number of jumps of the element value in the LBP code refers to the total number of jumps of the adjacent element values ​​in the LBP code, which characterizes the grayscale changes of the surrounding pixels of the corresponding pixel relative to the corresponding pixel. For example, the number of jumps of the element value in the LBP code 01010101 is 7, while the number of jumps of the element value in the LBP code 11110000 is 1.

[0050] Preferably, in this embodiment, determining the information amount index of each frame of grayscale image includes: determining the local feature code whose element value jump times in each frame of grayscale image is greater than a set jump times threshold value as the target local feature code according to the jump times of the element value in the local feature code; determining the average number of jump times of the element value in the target local feature code in each frame of grayscale image, and determining the information amount index of each frame of grayscale image according to the jump times of the element value in the target local feature code in each frame of grayscale image and the difference between each local feature code in each frame of grayscale image and the average value. Among them, the specific value of the jump times threshold value can be reasonably set according to the actual situation. In this embodiment, the value of the jump times threshold value is set to 4.

[0051] Specifically, the information content index of each frame of grayscale image is determined, and the corresponding calculation formula is:

[0052] Where Id(x) represents the information content index of the x-th grayscale image; q(j) represents the number of jumps of the element value in the j-th target local feature code in the x-th grayscale image; M′ represents the total number of target local feature codes in the x-th grayscale image; q max represents the maximum number of jumps of the element value in the local feature code; p(i) represents the number of jumps of the element value in the i-th local feature code in the x-th frame grayscale image, represents the average number of jumps of the element value in the target local feature code in the x-th frame grayscale image; N represents the total number of local feature codes in the x-th frame grayscale image; ε1 represents the first coefficient value, which is used to prevent the denominator from being zero, and its value is 0.01; Norm{} represents the normalization function.

[0053] In the above information quantity index calculation formula, The mean of the ratio of the number of jumps of the target local feature code with a larger number of jumps to the maximum number of jumps. The closer the mean is to 1, the more chaotic the distribution of pixel grayscale values ​​in the corresponding grayscale image is, the greater the amount of information contained, and the larger the value of the corresponding information index. It indicates the closeness between the number of jumps of the local feature code of the pixel points in the entire corresponding grayscale image and the average number of high jumps. The higher the closeness, the higher the number of jumps of the local feature code of the pixel points in the entire corresponding grayscale image. The more complex the overall distribution of the corresponding grayscale image is, the greater the amount of information it contains, and the larger the value of the corresponding information index.

[0054] According to the information amount index of each frame of grayscale image determined above, the feature point extraction index of each frame of grayscale image can be determined. Preferably, in this embodiment, the feature point extraction index of each frame of grayscale image is determined, including: initializing the feature point extraction index of the first frame of grayscale image, and performing negative correlation mapping on the information amount index of each frame of grayscale image to obtain the information amount index mapping value of each frame of grayscale image; taking each frame of grayscale image except the first frame of grayscale image as the object grayscale image, and determining the feature point extraction index of each frame of object grayscale image according to the information amount index mapping value of each frame of object grayscale image and the feature point extraction index of the previous frame of grayscale image of the frame of object grayscale image. Among them, the initialization value of the feature point extraction index of the first frame of grayscale image can be adaptively and reasonably set according to the conventional needs when the sift (Scale-invariant feature transform) algorithm is used to obtain feature points later, for example, the value of the initialization value can be set to 0.2.

[0055] Preferably, in this embodiment, the feature point extraction index of each frame of grayscale image is determined, and the corresponding calculation formula is:

[0056] σ(x)=[1-Id(x)]×σ(x-1); wherein σ(x) represents the feature point extraction index of the x-th frame grayscale image; Id(x) represents the information amount index of the x-th frame grayscale image; σ(x-1) represents the feature point extraction index of the x-1-th frame grayscale image.

[0057] In the above-mentioned feature point extraction index, when the amount of information contained in the grayscale image is greater, that is, the distribution of the pixel grayscale values ​​in the grayscale image is more complex and chaotic, then when the SIFT algorithm is used to extract the feature points of the grayscale image, a smaller initial variance should be used, and the value of the feature point extraction index should be smaller, so as to retain the details of the image with a large amount of information. Therefore, by negatively mapping the information index, the information index mapping value [1-Id(x)] is obtained, and finally the corresponding feature point extraction index is obtained. At the same time, the smaller the value of the feature point extraction index of the current frame of grayscale image, the higher the amount of information it contains. In order to retain this feature as much as possible, it is necessary to further reduce the feature point extraction index, and the value of the feature point extraction index corresponding to the current frame of grayscale image should be smaller.

[0058] Step S3: extract feature points from each grayscale image frame according to the feature point extraction index, determine the feature points in each grayscale image frame, and group the feature points according to the position distribution of the feature points to determine at least two feature point groups for each grayscale image frame.

[0059] According to the feature point extraction index of each frame of grayscale image, feature points are extracted from each frame of grayscale image to determine the feature points in each frame of grayscale image. Preferably, in this embodiment, determining the feature points in each frame of grayscale image includes: extracting feature points from each frame of grayscale image using the SIFT algorithm, and using the feature point extraction index of each frame of grayscale image as the initial variance when the SIFT algorithm extracts feature points, thereby obtaining the feature points in each frame of grayscale image. By determining the initial variance when the SIFT algorithm is used to extract feature points from the grayscale image based on the amount of information of different frames of grayscale image, the feature points finally extracted can have more scene features of different frames of grayscale image.

[0060] Since the key rendering part of the Anesthesiology Crisis Skill is the outline of the person, it is necessary to determine whether the area composed of feature points has the characteristics of the outline of the person. The outline characteristics here can be understood as the area composed of feature points being sufficiently regular. The more the area composed of feature points has this outline characteristic, the higher the structure is, and the more likely it is to belong to scene feature information. Correspondingly, the confidence of such feature points should be higher, and their sensitivity to noise should be lower. The corresponding anomaly score should be lowered in subsequent anomaly detection, so as to avoid misjudgment of scene feature information.

[0061] Based on the above analysis, in order to facilitate the subsequent determination of whether the area composed of feature points in each frame of grayscale image has the characteristics of human contour, for the feature points extracted in each frame of grayscale image, these feature points are grouped according to the position distribution of these feature points to determine each feature point group. Preferably, in this embodiment, determining each feature point group includes: performing at least two feature point group generation steps in each frame of grayscale image, each of the feature point group generation steps includes: randomly determining the first target feature point among the feature points currently remaining in each frame of grayscale image, and determining the feature point closest to the first target feature point within the neighborhood range of the first target feature point as the second target feature point, determining the feature point within the neighborhood range of the first target feature point that is the second closest to the first target feature point as the third target feature point, and so on, until the termination condition is met, the termination condition is: the total number of determined target feature points reaches the set number threshold, or there are no remaining feature points within the neighborhood range of the first target feature point; all the determined target feature points are regarded as a feature point group.

[0062] For ease of understanding, specifically, in each frame of grayscale image, a feature point is randomly selected from all feature points, and the feature point is used as the first target feature point. Since there are other feature points in its neighborhood, the first target feature point is used as the center and the set length is used as the radius to determine the neighborhood range of the circle. The set length can be reasonably set according to the actual situation and is not limited here. The feature point with the smallest Euclidean distance from the first target feature point is marked and calculated within the neighborhood, and the feature point with the smallest Euclidean distance is used as the second target feature point. Then, the feature point with the second smallest Euclidean distance from the first target feature point is marked and calculated within the neighborhood, and the feature point with the second smallest Euclidean distance is used as the third target feature point, and so on, until the termination condition is met, that is, the total number of target feature points determined reaches the set number threshold, or there are no remaining feature points within the neighborhood of the first target feature point. At this time, all the target feature points determined constitute a feature point group, thereby completing a feature point group generation step. The set number threshold can be reasonably set according to the actual situation. For example, the value of the set number threshold can be set to 100. After obtaining the first feature point group, for all the remaining feature points, randomly select a feature point from the remaining feature points and use it as the first target feature point. Then, according to the above feature point group generation steps, determine the next feature point group until all feature points are divided into different feature point groups. Since the number of feature points in a frame of grayscale image is limited, the feature points in each frame of grayscale image can be divided into 2 to 3 feature point groups.

[0063] Step S4: determining the noise insensitivity corresponding to the feature point group according to the position change characteristics of each feature point in the feature point group, and determining the low-sensitivity pixel points in each frame of the grayscale image according to the noise insensitivity.

[0064] In each frame of grayscale image, a Cartesian coordinate system is constructed with the lower left corner of the image as the origin, the horizontal right direction of the position where the image is placed when it is most suitable for viewing as the positive direction of the x-axis, and the vertical upward direction as the positive direction of the y-axis, so that the coordinate values ​​of each feature point in the grayscale image of this frame can be determined. The position coordinates of each feature point in each feature point group are analyzed to determine the noise insensitivity corresponding to each feature point group. Preferably, in this embodiment, determining the noise insensitivity corresponding to each feature point group includes: performing curve fitting on each feature point in the feature point group according to the position of each feature point in the feature point group to obtain a fitting curve; determining the fitting points of each feature point in the feature point group in the fitting curve, and determining the tangent slope values ​​corresponding to the fitting points; determining the first nearest neighbor feature point and the second nearest neighbor feature point of each feature point in the feature point group according to the position of each feature point in the feature point group; determining the Euclidean distance between each feature point in the feature point group and its first nearest neighbor feature point to obtain the feature point. The first distance corresponding to each feature point in the group is determined, and the Euclidean distance between each feature point in the feature point group and its second nearest neighbor feature point is determined to obtain the second distance corresponding to each feature point in the feature point group; according to the difference between the tangent slope values ​​corresponding to the fitting points of each feature point in the feature point group and its first nearest neighbor feature point, and the difference between the first distance and the second distance corresponding to each feature point in the feature point group, the non-structural index corresponding to the feature point group is determined; negative correlation mapping is performed on the non-structural index corresponding to the feature point group, and the negative correlation mapping result is used as the noise insensitivity corresponding to the feature point group.

[0065] For ease of understanding, specifically, in each frame of grayscale image, for each feature point group, a cubic spline curve is used to perform curve fitting on each feature point in the feature point group, thereby obtaining a fitting curve. The fitting points of each feature point in the fitting curve are determined, each feature point has the same horizontal coordinate as its corresponding fitting point, and the tangent slope value corresponding to each fitting point in the fitting curve is determined. At the same time, the first nearest neighbor feature point and the second nearest neighbor feature point of each feature point in the feature point group are determined, the first nearest neighbor feature point refers to the other feature point in the feature point group with the smallest Euclidean distance to the corresponding feature point, and the second nearest neighbor feature point refers to the other feature point in the feature point group with the second smallest Euclidean distance to the corresponding feature point, and the Euclidean distance between each feature point in the feature point group and its first nearest neighbor feature point is used as the first distance, and the Euclidean distance between each feature point in the feature point group and its second nearest neighbor feature point is used as the second distance.

[0066] On this basis, the non-structural index corresponding to the feature point group is determined according to the difference between the tangent slope value corresponding to each feature point in the feature point group and the fitting point of its first nearest neighbor feature point, and the difference between the first distance and the second distance corresponding to each feature point in the feature point group. The corresponding calculation formula is:

[0067] Wherein, Cos(u) represents the non-structural index corresponding to the u-th feature point group in each frame of grayscale image; k m k represents the tangent slope value corresponding to the fitting point of the mth feature point in the uth feature point group in each frame of grayscale image; ′ m represents the tangent slope value corresponding to the fitting point of the first nearest neighbor feature point of the mth feature point in the uth feature point group in each frame of the grayscale image; m,1 represents the first distance corresponding to the mth feature point in the uth feature point group in each frame of grayscale image; l m,2 It represents the second distance corresponding to the mth feature point in the uth feature point group in each frame of grayscale image; M represents the total number of feature points in the uth feature point group in each frame of grayscale image; || represents the absolute value function; ε2 represents the second coefficient value, which is used to prevent the denominator from being zero, and its value is 0.01; arctan() inverse tangent function.

[0068] In the calculation formula of the above non-structural index, when the tangent slope value between each feature point in the feature point group and the fitting point of its nearest neighbor feature point is closer, the ratio of the two is closer to 1, and the angle formed by the difference between the two is smaller, indicating that the structure of the area formed by the feature point group is stronger, the more it conforms to the characteristics of the character outline, and the more likely it is to belong to the scene feature information, and the corresponding non-structural index value is smaller. Similarly, when the Euclidean distance between each feature point in the feature point group and its nearest neighbor feature point is closer to the Euclidean distance between each feature point and its second nearest neighbor feature point, it also indicates that the structure of the area formed by the feature point group is stronger, and the corresponding non-structural index value is smaller.

[0069] After determining the non-structural index corresponding to each feature point group in the above manner, negative correlation mapping is performed on the non-structural index, and the negative correlation mapping result is used as the noise insensitivity corresponding to the feature point group. Preferably, in this embodiment, the noise insensitivity corresponding to each feature point group is determined, and the corresponding calculation formula is:

[0070] Cof(u)=Norm{Cos(u))};Sud(u)=exp(-Cof(u)); wherein, Cof(u) represents the unreliability corresponding to the u-th feature point group in each frame of grayscale image; Cos(u) represents the non-structural index corresponding to the u-th feature point group in each frame of grayscale image; Norm{} represents a normalization function, which is used to normalize Cos(u) to the range of 0 to 1; Sud(u) represents the noise insensitivity corresponding to the u-th feature point group in each frame of grayscale image; exp() represents an exponential function with a natural constant as the base.

[0071] In the above noise insensitivity calculation formula, when the structure of the area formed by the feature point group is stronger and more consistent with the characteristics of the character contour, the confidence of the feature point group is higher and the value of the unreliability corresponding to the feature point group is smaller. At this time, the sensitivity of the area formed by the feature point group to noise should be lower, and the corresponding value of the noise insensitivity corresponding to the feature point group is higher.

[0072] According to the noise insensitivity corresponding to each feature point group in each frame of grayscale image, the low-sensitivity pixel points in each frame of grayscale image can be determined. Preferably, in this embodiment, the low-sensitivity pixel points in each frame of grayscale image are determined, including: comparing the noise insensitivity corresponding to each feature point group in each frame of grayscale image with the set insensitivity threshold, and determining the feature points in the feature point group corresponding to the noise insensitivity greater than the set insensitivity threshold as the initial low-sensitivity pixel points in each frame of grayscale image; determining the neighborhood pixel points of the initial low-sensitivity pixel points in each frame of grayscale image, and determining the neighborhood pixel points as the extended low-sensitivity pixel points in each frame of grayscale image; using the initial low-sensitivity pixel and the extended low-sensitivity pixel in each frame of grayscale image as the final low-sensitivity pixel points in each frame of grayscale image. Among them, the value of the set insensitivity threshold can be reasonably set according to the actual situation. In this embodiment, the value of the set insensitivity threshold is set to 100. The neighborhood pixel points of the low-sensitivity pixel point refer to the pixel points near the low-sensitivity pixel point. In the present embodiment, the neighborhood pixel points refer to the 24 neighborhood pixel points of the low-sensitivity pixel point.

[0073] Step S5: performing abnormality detection on the pixels in each frame of grayscale image according to the low-sensitivity pixels, and determining the abnormal pixels in each frame of grayscale image.

[0074] For low-sensitivity pixels in each frame of grayscale image, since they have stronger structure in the image, they are more likely to belong to scene feature information. Therefore, when performing anomaly detection on pixels in the grayscale image, their sensitivity to noise should be reduced to make their anomaly scores lower, so as to avoid misjudging such pixels as abnormal pixels, thereby ultimately improving the accuracy of abnormal pixel detection.

[0075] Based on the above analysis, according to the low-sensitivity pixels in each frame of grayscale image, anomaly detection is performed on the pixels in each frame of grayscale image to determine the abnormal pixels in each frame of grayscale image. Preferably, in this embodiment, determining the abnormal pixels in each frame of grayscale image includes: using an isolation forest to perform anomaly detection on the pixels in each frame of grayscale image, and setting the average path corresponding to the low-sensitivity pixels to a fixed value during the anomaly detection process, thereby obtaining the abnormal pixels in each frame of grayscale image, and the fixed value is not greater than the set path threshold. Among them, the set path threshold can be reasonably set according to actual conditions, and the fixed value is a value not greater than the set path threshold to ensure that the abnormal score of the low-sensitivity pixels in the image can be reduced. In this embodiment, the set path threshold and the fixed value are both set to 1.

[0076] In the process of using the isolation forest to detect abnormalities in the pixels of each frame of grayscale image, by setting the average path in the abnormal score corresponding to the low-sensitivity pixel to a small fixed value, and other pixels participate in the calculation normally, the abnormal pixels in each frame of grayscale image can be accurately obtained. In order to remove the influence of abnormal points on the virtual simulation effect, these abnormal pixels in each frame of grayscale image are subjected to mean filtering and denoising to obtain each frame of grayscale image after denoising, and finally obtain each frame of virtual simulation image after denoising based on each frame of grayscale image after denoising. Render each frame of virtual simulation image after denoising to improve the authenticity of the virtual simulation scene, so as to facilitate the subsequent medical review and learning.

[0077] It should be noted that the above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. An anomaly detection method for virtual simulation of anesthesiology crisis skills, characterized in that: The following steps are involved: Obtaining a grayscale image of each frame of virtual simulation image, and determining a local feature code of each pixel in each frame of grayscale image; Determine the information amount index of each frame of grayscale image according to the number of jumps of the element value in the local feature code, and determine the feature point extraction index of each frame of grayscale image according to the information amount index; Extracting feature points from each frame of grayscale image according to the feature point extraction index to determine feature points in each frame of grayscale image, and grouping the feature points according to position distribution of the feature points to determine at least two feature point groups for each frame of grayscale image; Determine the noise insensitivity corresponding to the feature point group according to the position change characteristics of each feature point in the feature point group, and determine the low-sensitivity pixel points in each frame of the grayscale image according to the noise insensitivity; According to the low-sensitivity pixels, abnormality detection is performed on the pixels in each frame of grayscale image to determine the abnormal pixels in each frame of grayscale image; Determining the noise insensitivity corresponding to the feature point group includes: According to the position of each feature point in the feature point group, curve fitting is performed on each feature point in the feature point group to obtain a fitting curve; Determine a fitting point of each feature point in the feature point group in the fitting curve, and determine a tangent slope value corresponding to the fitting point; Determine, according to the positions of each feature point in the feature point group, a first nearest neighbor feature point and a second nearest neighbor feature point of each feature point in the feature point group; Determine the Euclidean distance between each feature point in the feature point group and its first nearest neighbor feature point to obtain a first distance corresponding to each feature point in the feature point group, and determine the Euclidean distance between each feature point in the feature point group and its second nearest neighbor feature point to obtain a second distance corresponding to each feature point in the feature point group; Determine a non-structural index corresponding to the feature point group according to a difference between a tangent slope value corresponding to each feature point in the feature point group and a fitting point of its first nearest neighbor feature point, and a difference between a first distance and a second distance corresponding to each feature point in the feature point group; Negative correlation mapping is performed on the non-structural indicators corresponding to the feature point group, and the negative correlation mapping result is used as the noise insensitivity corresponding to the feature point group.

2. The anomaly detection method for virtual simulation of anesthesiology crisis skills according to claim 1, characterized in that: Determine the information content index of each frame of grayscale image, including: According to the jump times of the element values ​​in the local feature code, a local feature code whose element value jump times in each frame of the grayscale image is greater than a set jump times threshold is determined as a target local feature code; The average value of the number of jumps of the element value in the target local feature code in each frame of grayscale image is determined, and the information content index of each frame of grayscale image is determined according to the number of jumps of the element value in the target local feature code in each frame of grayscale image and the difference between each local feature code in each frame of grayscale image and the average value.

3. The anomaly detection method for virtual simulation of anesthesiology crisis skills according to claim 2, characterized in that: The corresponding calculation formula for determining the information content index of each frame of grayscale image is: Where Id(x) represents the information content index of the x-th grayscale image; q(j) represents the number of jumps of the element value in the j-th target local feature code in the x-th grayscale image; M′ represents the total number of target local feature codes in the x-th grayscale image; q max represents the maximum number of jumps of the element value in the local feature code; p(i) represents the number of jumps of the element value in the i-th local feature code in the x-th frame grayscale image, represents the average number of jumps of the element value in the target local feature code in the x-th frame grayscale image; N represents the total number of local feature codes in the x-th frame grayscale image; ε1 represents the first coefficient value; Norm{} represents the normalization function.

4. The anomaly detection method for virtual simulation of anesthesiology crisis skills according to claim 1, characterized in that: Determine the feature point extraction index of each frame of grayscale image, including: Initialize the feature point extraction index of the first frame grayscale image, and perform negative correlation mapping on the information index of each frame grayscale image to obtain the information index mapping value of each frame grayscale image; Each frame of grayscale image except the first frame of grayscale image is taken as the object grayscale image, and the feature point extraction index of each frame of object grayscale image is determined according to the information amount index mapping value of each frame of object grayscale image and the feature point extraction index of the previous frame of grayscale image of the frame of object grayscale image.

5. The anomaly detection method for virtual simulation of anesthesiology crisis skills according to claim 1, characterized in that: Determine at least two feature point groups of each frame of grayscale image, including: The step of generating a feature point group is performed at least twice in each frame of grayscale image, and each of the steps of generating a feature point group includes: Among the feature points currently remaining in each frame of grayscale image, a first target feature point is randomly determined, and a feature point within a neighborhood range of the first target feature point that is closest to the first target feature point is determined as a second target feature point, and a feature point within a neighborhood range of the first target feature point that is second closest to the first target feature point is determined as a third target feature point, and so on, until a termination condition is met, the termination condition being: the total number of determined target feature points reaches a set number threshold, or there are no remaining feature points within a neighborhood range of the first target feature point; All the determined target feature points are regarded as a feature point group.

6. The anomaly detection method for virtual simulation of anesthesiology crisis skills according to claim 1, characterized in that: Determine the non-structural index corresponding to the feature point group, and the corresponding calculation formula is: Wherein, Cos(u) represents the non-structural index corresponding to the u-th feature point group in each frame of grayscale image; k m k represents the tangent slope value corresponding to the fitting point of the mth feature point in the uth feature point group in each frame of grayscale image; ′ m represents the tangent slope value corresponding to the fitting point of the first nearest neighbor feature point of the mth feature point in the uth feature point group in each frame of the grayscale image; m,1 represents the first distance corresponding to the mth feature point in the uth feature point group in each frame of grayscale image; l m,2 It represents the second distance corresponding to the mth feature point in the uth feature point group in each frame of grayscale image; M represents the total number of feature points in the uth feature point group in each frame of grayscale image; || represents the absolute value function; ε2 represents the second coefficient value; arctan() inverse tangent function.

7. The anomaly detection method for virtual simulation of anesthesiology crisis skills according to claim 1, characterized in that: Determine the low-sensitivity pixels in each frame of grayscale image, including: Comparing the noise insensitivity corresponding to each of the feature point groups in each frame of the grayscale image with a set insensitivity threshold, and determining the feature points in the feature point group corresponding to the noise insensitivity greater than the set insensitivity threshold as initial low-sensitivity pixels in each frame of the grayscale image; Determine the neighboring pixel points of the initial low-sensitivity pixel points in each frame of grayscale image, and determine the neighboring pixel points as the extended low-sensitivity pixel points in each frame of grayscale image; The initial low-sensitivity pixels and the expanded low-sensitivity pixels in each frame of grayscale image are used as the final low-sensitivity pixels in each frame of grayscale image.

8. The anomaly detection method for virtual simulation of anesthesiology crisis skills according to claim 1, characterized in that: Determine the abnormal pixels in each frame of grayscale image, including: The isolation forest is used to perform anomaly detection on the pixel points in each frame of the grayscale image, and during the anomaly detection process, the average path corresponding to the low-sensitivity pixel points is set to a fixed value, so as to obtain the abnormal pixel points in each frame of the grayscale image, and the fixed value is not greater than the set path threshold.

9. The anomaly detection method for virtual simulation of anesthesiology crisis skills according to claim 1, characterized in that: Extract feature points from each grayscale image frame to determine the feature points in each grayscale image frame, including: The SIFT algorithm is used to extract feature points from each frame of grayscale image, and the feature point extraction index of each frame of grayscale image is used as the initial variance when the SIFT algorithm extracts feature points, thereby obtaining the feature points in each frame of grayscale image.

Images