Otoscope abnormity nondestructive detection system based on big data analysis

Abstract

The invention belongs to the technical field of nondestructive testing, and discloses an otoscope abnormity nondestructive testing system based on big data analysis, which comprises a data storage module, a key position determination module, a detection execution module, a key image extraction module, a quality parameter extraction module and a detection evaluation module, the method comprises the following steps: firstly, storing a plurality of predefined standard acquisition paths and historical acquisition data thereof; by analyzing the change of image features in the video stream, identifying a key spatial position with a high requirement on otoscope imaging in the ear canal; selecting a standard path as a test path, acquiring real-time acquisition data of the to-be-tested otoscope, and extracting an image corresponding to the key position from the real-time acquisition data; and extracting quality parameters from the key images, comparing the quality parameters with a normal distribution model established based on historical data, and generating a detection report containing abnormal image position distribution and parameter types. Accurate evaluation of the quality of the otoscope is achieved, and the actual performance of the otoscope can be comprehensively reflected.

Description

Technical Field

[0001] The present invention relates to the technical field of non-destructive testing, and more specifically, to an earscope abnormal non-destructive testing system based on big data analysis. Background Art

[0002] With the development of the precision of medical devices, as an important tool for ear examinations, the imaging quality of earscopes directly affects the accuracy of clinical diagnoses. However, there are obvious deficiencies in the existing earscope quality detection technologies: traditional detection methods mainly use static tests, ignoring the dynamic imaging performance of earscopes during the movement in the ear canal; at the same time, there is a lack of recognition and targeted evaluation of key anatomical positions in the ear canal, such as the curved part of the ear canal and the tympanic membrane area, which pose higher requirements for the imaging ability of earscopes; in addition, the detection system is not comprehensive enough in extracting image quality parameters and fails to establish a scientific evaluation model based on historical data. These problems lead to the lack of objectivity and comparability in the earscope quality detection results and cannot comprehensively reflect the actual performance of earscopes. Summary of the Invention

[0003] In order to overcome the problem that the earscope quality detection results in the prior art lack objectivity and comparability, the present invention proposes an earscope abnormal non-destructive testing system based on big data analysis to solve the above problems.

[0004] The present invention provides the following technical solutions: An earscope abnormal non-destructive testing system based on big data analysis, comprising: A data storage module, configured to store multiple predefined standard acquisition paths and a set of acquisition data composed of historical acquisition data corresponding to each standard acquisition path; A key position determination module, configured to analyze the set of acquisition data to determine a set of key spatial positions for each standard acquisition path; A detection execution module, configured to select a test execution path from the standard acquisition paths as the test execution path and obtain real-time acquisition data of the earscope to be tested according to the test execution path; A key image extraction module, configured to obtain a key image sequence from the real-time acquisition data according to the set of key spatial positions; A quality parameter extraction module, configured to extract the image quality parameters of each image from the key image sequence; A detection evaluation module, configured to generate a final detection report according to the image quality parameters of each image.

[0005] Preferably, the standard acquisition path is represented by a sequence of path points in three-dimensional space. A reference three-dimensional coordinate system is established with the first path point as the origin, and each path point is a coordinate point in the reference three-dimensional coordinate system. The acquisition data includes video stream data and corresponding spatial position data, where the video stream data is a sequence of images continuously acquired by the otoscope, and the spatial position data is the position coordinates of the otoscope in the reference three-dimensional coordinate system when the image is acquired.

[0006] Preferably, the step of analyzing the acquisition data set and determining the key spatial position set for each standard acquisition path includes: Processing each piece of acquisition data in the acquisition data set of each standard acquisition path, including: Extracting image features from the video stream data in the acquisition data; Calculating the outlier value of each image in the video stream data according to the extracted image features; According to the outlier value of each image, marking the spatial position corresponding to the image with an outlier value exceeding the preset threshold as the candidate key position of this piece of acquisition data; Integrating the candidate key positions of all acquisition data under the same standard acquisition path to form the candidate key position set of this standard acquisition path; Counting the occurrence frequency of the candidate key positions to determine the key spatial position set.

[0007] Preferably, the extracting image features from the video stream data in the acquisition data includes: For each frame of image in the video stream data, extracting the color features of the image, including the color histograms of the RGB three channels; extracting the texture features of the image, including the gray-level co-occurrence matrix; extracting the shape features of the image, including the histogram of oriented gradients; The calculating the outlier value of each image in the video stream data according to the extracted image features includes: For each frame of image in the video stream: Denote it as the target image, obtain the previous frame of the target image and denote it as the previous image, and the next frame of the image and denote it as the next image: Calculating the Bhattacharyya distance of the color histograms between the target image and the previous image , and the Bhattacharyya distance of the color histograms between the target image and the next image ; Calculating the chi-square distance of the gray-level co-occurrence matrices between the target image and the previous image , and the chi-square distance of the gray-level co-occurrence matrices between the target image and the next image ; Calculating the KL divergence of the histograms of oriented gradients between the target image and the previous image , and the KL divergence of the histograms of oriented gradients between the target image and the next image ; Calculate the outliers of the target image using the following formula: , wherein, represents the outlier, , and are respectively the preset color weight value, texture weight value and shape weight value.

[0008] Preferably, the steps of statistically calculating the occurrence frequency of the candidate key positions and determining the final set of key spatial positions include: Map the candidate key positions of all acquisition data under the same standard acquisition path to the reference three-dimensional coordinate system; Establish a spatial grid in the reference three-dimensional coordinate system, and divide the coordinate space into several grid cells with equal volumes; For each grid cell, calculate the frequency of the candidate key positions it contains, that is, the ratio of the number of acquisition data containing the candidate key positions in the grid cell to the total number of acquisition data; Set a frequency threshold, and mark the grid cells with frequencies exceeding the frequency threshold as key cells; In each key cell, select the candidate key position with the most occurrences as the representative point of the cell; Form a set of key spatial positions by combining the representative points of all key cells.

[0009] Preferably, the obtaining of the key image sequence from the real-time acquisition data according to the set of key spatial positions includes: Receive the real-time video stream data and the corresponding spatial position data acquired by the otoscope to be tested along the test execution path; Adopt the same coordinate system definition method as the reference three-dimensional coordinate system, use the first path point of the test execution path as the origin, establish a three-dimensional rectangular coordinate system, and map the real-time acquired spatial position data and the set of key spatial positions to the established three-dimensional coordinate system; For each key spatial position in the set of key spatial positions, calculate the distance between all real-time acquired spatial positions and the key spatial position, and select the video frame corresponding to the spatial position with the smallest distance as the key image of the key spatial position; Organize the key images corresponding to all key spatial positions into a key image sequence according to the order of the video stream data.

[0010] Preferably, the steps of extracting the image quality parameters of each image from the key image sequence include: Extract the clarity parameters of each image, and the clarity parameters include calculating the variance of the Laplacian operator, the high-frequency energy ratio and the average value of the gradient amplitude; Extract the brightness parameters of each image. The brightness parameters include calculating the average brightness, the standard deviation of brightness, and the entropy value of the brightness histogram of the image; Extract the contrast parameters of each image. The contrast parameters include calculating the RMS contrast, the Weber contrast, and the Michelson contrast; Extract the color uniformity parameters of each image. The color uniformity parameters include calculating the mean ratio of the RGB three channels and the color saturation; Combine the sharpness parameters, brightness parameters, contrast parameters, and color uniformity parameters of each image into the image quality parameters of the image.

[0011] Preferably, the step of generating a final detection report according to the image quality parameters of each image includes: Obtain an image quality parameter normal distribution model established based on historical data. This model includes the normal value range, mean, standard deviation, and weight of each image quality parameter; For each image in the key image sequence, perform the following steps: Calculate the standardized deviation value of each image quality parameter. The standardized deviation value is obtained by subtracting the normal distribution mean from the actual parameter value and then dividing by the normal distribution standard deviation; According to the weight of each parameter, calculate the sum of the squares of the weighted standardized deviation values as the abnormal measurement value of the image; Compare the abnormal measurement value of the image with a preset threshold. If it exceeds the threshold, mark the image as an abnormal image and record its corresponding spatial position and abnormal parameter type; Generate a final detection report containing information on all abnormal images. The final detection report includes the number of abnormal images, the spatial position distribution, and the abnormal parameter type.

[0012] The present invention provides an otoscope abnormal non-destructive detection system based on big data analysis, having the following beneficial effects: By analyzing the changes in the image features of a large amount of historical acquisition data, accurately locate the areas where significant changes occur in the visual features in the ear canal. These areas usually pose higher requirements on the imaging ability of the otoscope, such as areas with large changes in light conditions, areas that require high contrast or high clarity. By focusing on these key positions for evaluation, the system realizes more targeted quality detection, greatly improving the detection efficiency and accuracy.

[0013] By adopting a multi-dimensional extraction method for four types of parameters, namely clarity, brightness, contrast, and color balance, the key factors affecting the imaging quality of otoscopes are comprehensively covered. Based on the normal distribution model established from historical data, scientific evaluation criteria are set for each parameter, improving the objectivity and comparability of the detection results. The generated inspection report includes the location distribution and parameter types of abnormal images, providing precise diagnosis for quality control, effectively solving the deficiencies of traditional detection methods, and providing technical support for improving the quality of otoscope products and the accuracy of clinical diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a schematic diagram of the modules of an otoscope abnormal non-destructive detection system based on big data analysis according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0016] Embodiment 1 Please refer to Figure 1 , in this embodiment, an otoscope abnormal non-destructive detection system based on big data analysis includes: A data storage module for storing multiple predefined standard acquisition paths and a set of acquisition data composed of historical acquisition data corresponding to each standard acquisition path; The standard acquisition path is represented by a sequence of path points in three-dimensional space. A reference three-dimensional coordinate system is established with the first path point as the origin, and each path point is a coordinate point in the reference three-dimensional coordinate system; the acquisition data includes video stream data and corresponding spatial position data, where the video stream data is a sequence of images continuously acquired by the otoscope, and the spatial position data is the position coordinates of the otoscope in the reference three-dimensional coordinate system when the image is acquired.

[0017] In this embodiment, first, the establishment method of the standard acquisition path is determined. Multiple standard acquisition paths are defined for different ear canal inspection purposes. Each standard acquisition path can be obtained by inviting professional physicians to perform multiple standard operations using an otoscope with a spatial positioning sensor to obtain the original path data. After smoothing and path simplification of these data, each standard path is finally represented by a sequence of several path points.

[0018] Specifically, the standard acquisition path is represented by a sequence of path points in three-dimensional space. A reference three-dimensional coordinate system is established with the first path point as the origin, and each path point is a coordinate point in the reference three-dimensional coordinate system. For example, a certain standard path may include a series of three-dimensional coordinate points from the entrance of an adult male ear canal to the eardrum.

[0019] For each standard acquisition path, a reference three-dimensional coordinate system can be established with the first path point as the origin. The X-axis points in the direction of the deep ear canal, the Y-axis points in the direction above the ear canal, and the Z-axis is perpendicular to the X and Y axes, forming a right-handed coordinate system. The coordinate system is established using a spatial positioning system to ensure the consistency of position measurement.

[0020] Next, for each standard path, a large amount of historical acquisition data is stored. Each standard path stores multiple pieces of historical acquisition data, which can come from the normal ear canal examination records during the clinical verification process or the examination records of a specially built simulated ear canal environment. Each acquisition data includes two parts of information: video stream data and corresponding spatial position data.

[0021] The video stream data is stored in a high-definition resolution, appropriate frame rate, and using a standard video coding format. The spatial position data is collected synchronously with the video frames, using the same sampling frequency as the video, and stored in a structured data format.

[0022] The key position determination module is used to analyze the acquisition data set and determine the key spatial position set for each standard acquisition path; The steps of analyzing the acquisition data set and determining the key spatial position set for each standard acquisition path include: Processing each piece of acquisition data in the acquisition data set of each standard acquisition path, including: Extracting image features from the video stream data in the acquisition data; Calculating the outlier value of each image in the video stream data according to the extracted image features; According to the outlier value of each image, marking the spatial position corresponding to the image with an outlier value exceeding the preset threshold as the candidate key position of this piece of acquisition data; Integrating the candidate key positions of all acquisition data under the same standard acquisition path to form the candidate key position set of this standard acquisition path; Statistical frequency of occurrence of candidate key positions to determine the key spatial position set.

[0023] Extracting image features from the video stream data in the acquisition data includes: For each frame of image in the video stream data, extracting the color features of the image, including the color histograms of the RGB three channels; extracting the texture features of the image, including the gray-level co-occurrence matrix; extracting the shape features of the image, including the histogram of oriented gradients; Calculating the outliers of each image in the video stream data based on the extracted image features includes: For each frame of image in the video stream: Denote it as the target image, obtain the previous frame image of the target image and denote it as the previous image, and the next frame image and denote it as the next image: Calculate the Bhattacharyya distance of the color histogram between the target image and the previous image , and the Bhattacharyya distance of the color histogram between the target image and the next image ; Calculate the chi-square distance of the gray-level co-occurrence matrix between the target image and the previous image , and the chi-square distance of the gray-level co-occurrence matrix between the target image and the next image ; Calculate the KL divergence of the histogram of oriented gradients between the target image and the previous image , and the KL divergence of the histogram of oriented gradients between the target image and the next image ; Use the following formula to calculate the outlier of the target image: , In the formula, represents the outlier, , and are respectively the preset color weight value, texture weight value and shape weight value.

[0024] The steps of statistically calculating the occurrence frequency of candidate key positions and determining the final set of key spatial positions include: Map the candidate key positions of all acquisition data under the same standard acquisition path to the reference three-dimensional coordinate system; Establish a spatial grid in the reference three-dimensional coordinate system and divide the coordinate space into several grid cells with equal volumes; For each grid cell, calculate the frequency of its containing candidate key positions, that is, the ratio of the number of acquisition data containing candidate key positions in the grid cell to the total number of acquisition data; Set a frequency threshold and mark the grid cells with frequencies exceeding the frequency threshold as key cells; In each key cell, select the candidate key position with the most occurrences as the representative point of the cell; Form a set of key spatial positions with the representative points of all key cells.

[0025] In this embodiment, first, obtain the acquisition data set corresponding to each standard acquisition path from the data storage module. For each standard acquisition path, analyze all its historical acquisition data to identify the key positions for otoscope imaging quality assessment.

[0026] For each piece of collected data, first, image features are extracted from the video stream data. Specifically, for each frame of the video stream, three types of features are extracted: color features, texture features, and shape features. The color features are represented by the color histogram of the RGB three channels, with each channel divided into 32 intervals, forming a 96-dimensional feature vector; the texture features are represented by the gray-level co-occurrence matrix, calculating the co-occurrence matrices in four directions (0°, 45°, 90°, 135°) and extracting statistics such as energy, contrast, and correlation; the shape features are represented by the histogram of gradient directions, dividing the gradient direction into 8 intervals and counting the gradient magnitudes within each interval. Next, the outliers of each frame of the image are calculated according to the extracted features. For each frame of the image in the video stream (except the first and last frames), it is used as the target image, and its previous frame and the next frame are obtained. The feature distances between the target image and the two adjacent frames are calculated respectively: the Bhattacharyya distance is used to measure the similarity of color features, the chi-square distance is used to measure the similarity of texture features, and the KL divergence is used to measure the similarity of shape features. Then, the outlier of the target image is calculated using the weighted summation method, and the weight values are preset according to the characteristics of otoscope imaging, with the color weight being 0.4, the texture weight being 0.4, and the shape weight being 0.2. According to the calculated outliers, the spatial positions corresponding to the images with outliers exceeding the preset threshold are marked as the candidate key positions of this piece of collected data. The preset threshold can be determined through statistical analysis of the normal otoscope collected data and can be set to a level that makes approximately 10% of the images be marked as candidate key positions.

[0027] For all the collected data under the same standard acquisition path, their candidate key positions are integrated to form a candidate key position set. To handle the minor differences in spatial positions, a spatial grid is established in the reference three-dimensional coordinate system, and the coordinate space is divided into grid cells with equal volumes. The size of the grid cells is determined according to the positioning accuracy of the otoscope and is usually set to 1 cubic millimeter.

[0028] For each grid cell, calculate the frequency of the candidate key positions it contains, that is, the ratio of the number of collected data containing candidate key positions within this grid cell to the total number of collected data. Set the frequency threshold to 0.6, and mark the grid cells with frequencies exceeding the threshold as key cells.

[0029] In each key cell, select the candidate key position that appears the most times as the representative point of this cell. The representative points of all key cells form the final set of key spatial positions.

[0030] Through the above steps, a set of key spatial positions is determined for each standard acquisition path. These positions usually correspond to regions where significant changes in visual features occur within the ear canal and are important reference points for evaluating the imaging quality of otoscopes. These regions generally pose higher requirements on the imaging capabilities of otoscopes, such as areas with large variations in light conditions, high contrast, or high clarity. In subsequent otoscope quality inspections, these key positions will serve as key evaluation areas. By analyzing the imaging performance of the otoscope at these positions, the imaging quality and working performance of the otoscope can be comprehensively evaluated to ensure the reliability of the device.

[0031] A detection execution module, configured to select a test execution path from the standard acquisition paths as the test execution path, and obtain real-time acquisition data of the otoscope under test according to the test execution path; In this embodiment, first, a standard acquisition path can be selected from the data storage module as the test execution path according to the model and use of the otoscope under test. The selection process can be manually specified through the user interface or automatically selected according to preset rules.

[0032] During the test, a standardized ear canal model is usually used. This model is made according to the anatomical structure of the human ear canal and includes key structures such as the curvature of the ear canal and the eardrum. The ear canal model is fixed on a bracket and its position remains unchanged.

[0033] The otoscope under test records its three-dimensional spatial position in real time while collecting video stream data. The position sensor uses an optical tracking system or an electromagnetic positioning system to ensure the accuracy of spatial position measurement.

[0034] Through the above steps, real-time acquisition data of the otoscope under test along the test execution path is obtained, including video stream data and spatial position data, providing basic data for subsequent key image extraction and quality parameter analysis.

[0035] A key image extraction module, configured to obtain a key image sequence from the real-time acquisition data according to the set of key spatial positions; Obtaining a key image sequence from the real-time acquisition data according to the set of key spatial positions includes: Receiving the real-time video stream data collected by the otoscope under test along the test execution path and the corresponding spatial position data; Using the same coordinate system definition method as the reference three-dimensional coordinate system, taking the first path point of the test execution path as the origin, establishing a three-dimensional rectangular coordinate system, and mapping the real-time acquired spatial position data and the set of key spatial positions into the established three-dimensional coordinate system; For each key spatial position in the set of key spatial positions, calculate the distance between all the real-time acquired spatial positions and this key spatial position, and select the video frame corresponding to the spatial position with the smallest distance as the key image of this key spatial position; Organize the key images corresponding to all key spatial positions into a key image sequence according to the order of the video stream data.

[0036] In this embodiment, first, receive the real-time video stream data and the corresponding spatial position data collected by the otoscope under test along the test execution path. These data are collected by the detection execution module and transmitted to the key image extraction module, which contains a complete record of the movement process of the otoscope under test in the standard ear canal model.

[0037] Next, use the same coordinate system definition method as the reference three-dimensional coordinate system, with the first path point of the test execution path as the origin, to establish a three-dimensional rectangular coordinate system. This method ensures the consistency of the test data and the historical data in the spatial reference system. Map the real-time collected spatial position data and the set of key spatial positions into the established three-dimensional coordinate system, so that the two sets of data can be compared within the same coordinate space.

[0038] For each key spatial position in the set of key spatial positions, calculate the Euclidean distance between all the real-time collected spatial positions and this key spatial position. By calculating the three-dimensional distance between spatial coordinate points, find the actual collected point closest to each key position.

[0039] After calculating all the distances, select the video frame corresponding to the spatial position with the smallest distance as the key image of this key spatial position. This ensures that the selected image is the one closest to the predefined key position in terms of spatial position, which is convenient for comparative analysis with historical data.

[0040] Finally, organize the key images corresponding to all key spatial positions into a key image sequence according to the time order of the video stream data. This chronological arrangement maintains the natural process of ear canal examination and helps to observe the trend of image quality changes in subsequent analysis.

[0041] Through the above steps, a key image sequence corresponding to the key spatial positions is extracted from the real-time collected data of the otoscope under test. These key images represent the imaging performance of the otoscope at specific challenging positions and will be used for subsequent quality parameter extraction and performance evaluation.

[0042] A quality parameter extraction module for extracting the image quality parameters of each image from the key image sequence; The steps of extracting the image quality parameters of each image from the key image sequence include: Extract the sharpness parameter of each image, and the sharpness parameter includes calculating the variance of the Laplacian operator, the high-frequency energy ratio, and the average value of the gradient magnitude; Extract the brightness parameter of each image, and the brightness parameter includes calculating the average brightness of the image, the brightness standard deviation, and the entropy value of the brightness histogram; Extract the contrast parameters of each image. The contrast parameters include calculating the RMS contrast, Weber contrast, and Michelson contrast; Extract the color balance parameters of each image. The color balance parameters include calculating the mean ratio of the RGB three channels and the color saturation; Combine the sharpness parameter, brightness parameter, contrast parameter, and color balance parameter of each image into the image quality parameter of that image.

[0043] In this embodiment, first, each image in the key image sequence is processed to extract four types of image quality parameters: sharpness parameter, brightness parameter, contrast parameter, and color balance parameter.

[0044] For the extraction of the sharpness parameter, three different measurement methods are implemented: one is to calculate the variance of the Laplacian operator, detect edges by applying a Laplacian filter to the image, and the larger the variance value, the clearer the image; the second is to calculate the high-frequency energy ratio, decompose the image into frequency components through Fourier transform, and calculate the proportion of the high-frequency part energy in the total energy; the third is to calculate the average value of the gradient magnitude, calculate the gradient magnitude of each point in the image through the Sobel operator and take the average, and the larger the value, the richer the image details.

[0045] For the extraction of the brightness parameter, first convert the color image to a grayscale image, and then calculate three indicators: one is to calculate the average brightness of the image, that is, the arithmetic mean of all pixel grayscale values; the second is to calculate the brightness standard deviation, which reflects the degree of dispersion of the image brightness distribution; the third is to calculate the entropy value of the brightness histogram, which reflects the complexity and information content of the brightness distribution.

[0046] For the extraction of the contrast parameter, calculate three different contrast indicators: one is to calculate the RMS contrast, that is, the ratio of the standard deviation to the average value of the image pixel values; the second is to calculate the Weber contrast, which is suitable for evaluating the contrast in local regions of the image; the third is to calculate the Michelson contrast, which is suitable for evaluating the contrast difference between large-area regions in the image.

[0047] For the extraction of the color balance parameter, first analyze the RGB three channels of the image and calculate two indicators: one is to calculate the mean ratio of the RGB three channels to evaluate the color balance; the second is to calculate the color saturation, that is, convert the image from the RGB space to the HSV space and calculate the average value of the saturation channel.

[0048] Finally, combine the above four types of parameters into a complete set of image quality parameters. Through the above steps, comprehensive quality parameters are extracted for each image in the key image sequence, and these parameters reflect the imaging quality of the otoscope at key positions from different angles, providing a quantitative basis for subsequent detection and evaluation.

[0049] A detection and evaluation module for generating a final detection report based on the image quality parameters of each image.

[0050] The steps of generating a final detection report based on the image quality parameters of each image include: Obtain a normal distribution model of image quality parameters established based on historical data, which includes the normal value range, mean, standard deviation, and their weights of each image quality parameter; For each image in the key image sequence, perform the following steps: Calculate the standardized deviation value of each image quality parameter, which is obtained by subtracting the actual value of the parameter from the normal distribution mean and then dividing by the normal distribution standard deviation; According to the weight of each parameter, calculate the sum of the squares of the weighted standardized deviation values as the anomaly measure value of this image; Compare the anomaly measure value of this image with a preset threshold. If it exceeds the threshold, mark this image as an abnormal image and record its corresponding spatial position and abnormal parameter type; Generate a final detection report containing information on all abnormal images. The final detection report includes the number of abnormal images, spatial position distribution, and abnormal parameter types.

[0051] In this embodiment, the detection and evaluation module generates a final detection report on the otoscope performance based on the image quality parameters obtained by the quality parameter extraction module. The specific implementation process is as follows: First, obtain a normal distribution model of image quality parameters established based on historical data. This model is constructed by analyzing the imaging quality parameters of a large number of qualified otoscopes at each key position and includes the normal value range, mean, standard deviation, and their weights of each image quality parameter. The weights in the model reflect the importance of different parameters for otoscope performance evaluation. For example, in a low-light environment, the weight of the brightness parameter may be higher; in a detailed observation scenario, the weight of the clarity parameter may be higher.

[0052] For each image in the key image sequence, perform a standardized evaluation process. First, calculate the standardized deviation value of each image quality parameter. The standardization process is achieved by subtracting the actual value of the parameter from the mean in the normal distribution model and then dividing by the corresponding standard deviation, enabling comparison of parameters with different dimensions.

[0053] Next, according to the preset weights of each parameter, calculate the sum of the squares of the weighted standardized deviation values as the anomaly measure value of this image. This calculation method is similar to the Mahalanobis distance and can comprehensively consider the deviation degree and importance of each parameter.

[0054] Compare the calculated anomaly metric with a preset threshold. The preset threshold is typically set at a level where 5% of normal samples are mislabeled as anomalies, balancing detection sensitivity and specificity. If the anomaly metric exceeds the threshold, mark the image as an abnormal image and record its corresponding spatial location and the type of anomaly parameter. The type of anomaly parameter indicates which type of parameter (sharpness, brightness, contrast, or color balance) contributed the most to the abnormal value, helping to locate specific problems with the otoscope.

[0055] Finally, generate a final detection report containing information on all abnormal images. The report content includes: the number and proportion of abnormal images, reflecting the overall performance level of the otoscope; the spatial location distribution, showing where the otoscope has performance problems; and the analysis of the type of anomaly parameter, indicating the main types of problems with the otoscope (such as insufficient sharpness, insufficient brightness, etc.).

[0056] Through the above steps, the detection and evaluation module generates a comprehensive and objective otoscope performance detection report, providing a reliable basis for otoscope quality control and product improvement.

[0057] In several embodiments provided by the present invention, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only one way, and there can be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections between each other can be through some interfaces. The indirect couplings or communication connections of devices or units can be in electrical, mechanical, or other forms.

[0058] As described above, this is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention.

[0059] Finally: The above description is only the preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should all be included within the protection scope of the present invention.

Claims

1. An ear mirror abnormality non-destructive detection system based on big data analysis, characterized in that, Including: A data storage module for storing multiple predefined standard acquisition paths and an acquisition data set composed of historical acquisition data corresponding to each standard acquisition path; A key position determination module for analyzing the acquisition data set to determine a set of key spatial positions for each standard acquisition path; A detection execution module for selecting a test execution path from the standard acquisition paths as the test execution path and obtaining real-time acquisition data of the otoscope to be tested according to the test execution path; A key image extraction module for obtaining a key image sequence from the real-time acquisition data according to the set of key spatial positions; A quality parameter extraction module for extracting the image quality parameter of each image from the key image sequence; A detection evaluation module for generating a final detection report according to the image quality parameter of each image.

2. The aural mirror abnormal non-destructive detection system based on big data analysis according to claim 1, wherein, The standard acquisition path is represented by a sequence of path points in three-dimensional space. A reference three-dimensional coordinate system is established with the first path point as the origin, and each path point is a coordinate point in the reference three-dimensional coordinate system; the acquisition data includes video stream data and corresponding spatial position data, where the video stream data is a sequence of images continuously acquired by the otoscope, and the spatial position data is the position coordinates of the otoscope in the reference three-dimensional coordinate system when the image is acquired.

3. The ear mirror abnormal non-destructive detection system based on big data analysis according to claim 2, characterized in that, The steps of analyzing the acquisition data set to determine a set of key spatial positions for each standard acquisition path include: Processing each piece of acquisition data in the acquisition data set of each standard acquisition path, including: Extracting image features from the video stream data in the acquisition data; Calculating the outlier value of each image in the video stream data according to the extracted image features; According to the outlier value of each image, marking the spatial position corresponding to the image with an outlier value exceeding the preset threshold as the candidate key position of the piece of acquisition data; Integrating the candidate key positions of all acquisition data under the same standard acquisition path to form a set of candidate key positions of the standard acquisition path; Counting the occurrence frequency of the candidate key positions to determine the set of key spatial positions.

4. The ear mirror abnormal non-destructive detection system based on big data analysis according to claim 3, characterized in that, The extracting image features from the video stream data in the acquisition data includes: For each frame of image in the video stream data, extracting the color features of the image, including the color histograms of the RGB three channels; extracting the texture features of the image, including the gray-level co-occurrence matrix; extracting the shape features of the image, including the histogram of gradient directions; The calculating the outlier value of each image in the video stream data according to the extracted image features includes: For each frame of image in the video stream: Denote it as the target image, obtain the previous frame of the target image and denote it as the previous image, and the next frame of the image and denote it as the next image: Calculate the Bhattacharyya distance of color histograms between the target image and the previous image , and the Bhattacharyya distance of color histograms between the target image and the subsequent image ; Calculate the chi-square distance of the gray-level co-occurrence matrix between the target image and the previous image , and the chi-square distance of the gray-level co-occurrence matrix between the target image and the subsequent image ; Calculate the KL divergence of the histograms of oriented gradients between the target image and the previous image , and the KL divergence of the histograms of oriented gradients between the target image and the subsequent image ; Calculating the outlier value of the target image using the following formula: ， In the formula, represents an outlier, , and are respectively the preset color weight value, texture weight value, and shape weight value.

5. The aural mirror abnormal nondestructive detection system based on big data analysis according to claim 4, wherein, The steps of counting the occurrence frequency of the candidate key positions to determine the final set of key spatial positions include: Mapping the candidate key positions of all acquisition data under the same standard acquisition path into the reference three-dimensional coordinate system; Establishing a spatial grid in the reference three-dimensional coordinate system and dividing the coordinate space into several grid cells with equal volumes; For each grid cell, calculate the frequency of candidate key positions it contains, i.e., the ratio of the number of acquisition data containing candidate key positions within the grid cell to the total number of acquisition data; Set a frequency threshold and mark the grid cells with a frequency exceeding the frequency threshold as key cells; In each key cell, select the candidate key position with the highest occurrence frequency as the representative point of the cell; Form a set of key spatial positions by combining the representative points of all key cells.

6. The ear mirror abnormal non-destructive detection system based on big data analysis according to claim 2, wherein, The step of obtaining the key image sequence from the real-time acquisition data according to the set of key spatial positions includes: Receive the real-time video stream data and corresponding spatial position data acquired by the otoscope to be tested along the test execution path; Adopt the same coordinate system definition method as the reference three-dimensional coordinate system, use the first path point of the test execution path as the origin to establish a three-dimensional rectangular coordinate system, and map the real-time acquired spatial position data and the set of key spatial positions into the established three-dimensional coordinate system; For each key spatial position in the set of key spatial positions, calculate the distance between all the real-time acquired spatial positions and this key spatial position, and select the video frame corresponding to the spatial position with the minimum distance as the key image of this key spatial position; Organize the key images corresponding to all key spatial positions into a key image sequence according to the order of the video stream data.

7. The ear mirror abnormality non-destructive detection system based on big data analysis according to claim 6, characterized in that, The step of extracting the image quality parameters of each image from the key image sequence includes: Extract the clarity parameter of each image, and the clarity parameter includes calculating the variance of the Laplacian operator, the high-frequency energy ratio, and the average value of the gradient magnitude; Extract the brightness parameter of each image, and the brightness parameter includes calculating the average brightness of the image, the brightness standard deviation, and the entropy value of the brightness histogram; Extract the contrast parameter of each image, and the contrast parameter includes calculating the RMS contrast, the Weber contrast, and the Michelson contrast; Extract the color balance parameter of each image, and the color balance parameter includes calculating the mean ratio of the RGB three channels and the color saturation; Combine the clarity parameter, brightness parameter, contrast parameter, and color balance parameter of each image into the image quality parameter of this image.

8. An ear mirror abnormality non-destructive detection system based on big data analysis according to claim 7, characterized in that, The step of generating the final detection report according to the image quality parameter of each image includes: Obtain the normal distribution model of the image quality parameters established based on historical data, and this model includes the normal value range, mean, standard deviation, and weight of each image quality parameter; For each image in the key image sequence, perform the following steps: Calculate the standardized deviation value of each image quality parameter, and the standardized deviation value is obtained by subtracting the normal distribution mean from the actual parameter value and then dividing by the normal distribution standard deviation; According to the weight of each parameter, calculate the sum of the squares of the weighted standardized deviation values as the abnormal metric value of this image; Compare the abnormal metric value of this image with a preset threshold. If it exceeds the threshold, mark this image as an abnormal image and record its corresponding spatial position and abnormal parameter type; Generate a final detection report containing information on all abnormal images, and the final detection report includes the number of abnormal images, the spatial position distribution, and the abnormal parameter type.

Images