A subway station steel pipe column detection method based on infrared sensor

Abstract

The application relates to the field of image processing, and provides a subway station steel pipe column detection method based on an infrared sensor, which comprises the following steps: collecting an RGB image and an infrared image of a subway station steel pipe column, and performing semantic segmentation on the RGB image and the infrared image, so that the RGB image and the infrared image only contain the steel pipe column; determining the surface abnormality degree of the steel pipe column based on the RGB image, wherein the surface abnormality degree represents visible defects on the surface of the steel pipe column; determining the conduction abnormality degree of the steel pipe column based on the infrared image, wherein the conduction abnormality degree represents defects inside the steel pipe column; determining the type of an abnormal area based on the surface abnormality degree and the conduction abnormality degree; and determining a defect area based on the type of the abnormal area. The method can accurately obtain defect type information of the steel pipe column, complete abnormality detection of the subway station steel pipe column, and facilitate subsequent operation and maintenance.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method for detecting steel pipe columns in subway stations based on infrared sensors. Background Technology

[0002] In the context of subway steel pipe column inspection, defects such as corrosion, bending deformation, fatigue fracture, and detachment may occur during the use of subway steel pipe columns. These defects increase the safety risks of the steel pipe columns, necessitating regular maintenance and inspection to ensure the safe operation of subway stations. Therefore, utilizing computer vision technology for regular defect inspection of subway steel pipe columns undoubtedly saves significant operating and maintenance costs. During the defect inspection process of subway steel pipe columns, machine vision devices are often used to obtain RGB and infrared images of the steel pipe columns for analysis of defect areas. However, in actual use scenarios, the heat conduction phenomenon caused by the movement of objects around the steel pipe columns is often overlooked, making it difficult to distinguish defect types and significantly affected by imaging conditions.

[0003] Due to the heat conduction phenomenon in the steel pipe column area of ​​the subway station, the upper and lower regions of the infrared image of the steel pipe column show a gradual temperature gradient change. The degree of local anomaly at the edge of the area is difficult to measure accurately, which means that the accuracy of defect detection in the existing technology is insufficient. Summary of the Invention

[0004] This invention provides a method for detecting steel pipe columns in subway stations based on infrared sensors. This method can accurately obtain information on the defect types of steel pipe columns, complete the detection of abnormal conditions of steel pipe columns in subway stations, and facilitate subsequent operation and maintenance.

[0005] In a first aspect, this application provides a method for detecting steel pipe columns in subway stations based on infrared sensors, including:

[0006] RGB and infrared images of steel pipe columns in subway stations are acquired, and semantic segmentation is performed on the RGB and infrared images so that the RGB and infrared images contain only the steel pipe columns.

[0007] The degree of surface anomaly of the steel pipe column is determined based on the RGB image, wherein the degree of surface anomaly characterizes visible defects on the surface of the steel pipe column; and the degree of conduction anomaly of the steel pipe column is determined based on the infrared image, wherein the degree of conduction anomaly characterizes internal defects of the steel pipe column.

[0008] The type of the abnormal region is determined based on the degree of surface anomaly and the degree of conduction anomaly.

[0009] Defect areas are determined based on the type of abnormal areas.

[0010] In an optional embodiment, determining the degree of surface anomalies of the steel pipe column based on the RGB image includes:

[0011] Convert the RGB image to an RGB grayscale image;

[0012] The RGB grayscale image is processed using an adaptive local threshold segmentation algorithm to obtain multiple binary images;

[0013] Multiple connected components are obtained based on the multiple binary images, and each connected component represents a surface anomaly region.

[0014] The degree of surface anomaly is calculated based on the area of ​​the surface anomaly region, the difference between the gray value inside the surface anomaly region and the gray value of the neighborhood of the surface anomaly region.

[0015] In an optional embodiment, the degree of surface anomaly is calculated based on the area of ​​the surface anomaly region, the difference between the gray values ​​within the surface anomaly region and the gray values ​​of the neighborhood of the surface anomaly region, including:

[0016]

[0017] Among them, SD i S represents the degree of surface anomaly in the i-th surface anomaly region. i This represents the area size of the i-th surface anomaly region. It represents the average gray value of the set of pixels in the outer neighborhood of the surface anomaly region. This represents the average gray value within the abnormal area on the surface.

[0018] In one optional embodiment, determining the degree of conduction anomaly of the steel pipe column based on the infrared image includes: converting the infrared image into an infrared grayscale image;

[0019] The infrared grayscale image is processed using an adaptive local threshold segmentation algorithm to obtain multiple binary images;

[0020] Multiple connected components are obtained based on the multiple binary images, and each connected component represents a region of conduction anomaly.

[0021] The degree of conduction abnormality is calculated based on the area of ​​the conduction abnormal region, the difference between the gray value inside the conduction abnormal region and the gray value of the neighborhood of the conduction abnormal region.

[0022] In an optional embodiment, the degree of conduction abnormality is calculated based on the area of ​​the conduction abnormality region, the difference between the gray values ​​inside the conduction abnormality region and the gray values ​​of the neighborhood of the conduction abnormality region, including:

[0023]

[0024] Among them, IDi S represents the degree of conduction abnormality in the i-th conduction abnormality region. i This represents the area of ​​the i-th transmission anomaly region. It represents the average gray value of the set of pixels in the outer neighborhood of the region with the abnormal conduction. This represents the average gray level within the region of conduction anomaly.

[0025] In an optional embodiment, determining the type of the anomalous region based on the degree of surface anomaly and the degree of conduction anomaly includes:

[0026] A coordinate system is established based on the surface anomaly degree and the conduction anomaly degree, with the horizontal axis representing the surface anomaly degree of the connected domain and the vertical axis representing the conduction anomaly degree of the connected domain. Points on the coordinate system represent anomaly regions.

[0027] Calculate the Euclidean distance between two anomalous regions, and determine the similarity between the two anomalous regions based on the Euclidean distance;

[0028] Based on the similarity, points on the coordinate system are clustered to obtain multiple clusters, and the type of the abnormal region corresponding to the cluster is determined.

[0029] In an optional embodiment, determining the similarity between the two anomalous regions based on the Euclidean distance includes:

[0030] Calculate the anomaly confidence level for each anomaly region;

[0031] The distance between the two abnormal regions is weighted based on the anomaly confidence level to obtain the weighted Euclidean distance.

[0032] The similarity between the two anomalous regions is determined based on the weighted Euclidean distance.

[0033] In an optional embodiment, if the weighted Euclidean distance is greater than a preset distance, the two abnormal regions are determined to belong to the same cluster; if the weighted Euclidean distance is less than the preset distance, the two abnormal regions are determined to belong to different clusters.

[0034] In an optional embodiment, determining the type of the anomalous region based on the cluster includes:

[0035] When the surface anomaly degree SD of an abnormal region in a cluster is greater than the conduction anomaly degree ID, the cluster does not have surface or internal anomalies.

[0036] When the surface anomaly degree SD of the abnormal region in the cluster is approximately equal to the conduction anomaly degree ID, the cluster belongs to the surface anomaly defect.

[0037] When the surface anomaly degree SD of an abnormal region in a cluster is less than the conduction anomaly degree ID, the cluster belongs to an internal anomaly defect.

[0038] In an alternative embodiment:

[0039] Calculate the anomaly confidence level for each anomaly region; including:

[0040]

[0041] Where DB represents the anomaly confidence level of the abnormal region, and d represents the minimum distance from the center pixel of the current abnormal region to the upper and lower boundaries of the image. Let g represent the average gray value of the i-th region among the three different regions of the image, and g represent the average gray value corresponding to the current abnormal region. The three different regions are the upper connected region, the middle region, and the lower connected region.

[0042] The distance between two anomalous regions is weighted based on the aforementioned anomaly confidence level to obtain a weighted Euclidean distance, including:

[0043]

[0044] in, For weighted Euclidean distance, D represents the Euclidean distance between two outlier regions in the coordinate system, and ΔDB represents the absolute value of the difference in confidence levels between outliers in the outlier regions.

[0045] The beneficial effects of this application, unlike existing technologies, include a method for detecting steel pipe columns in subway stations based on infrared sensors. This method comprises: acquiring RGB and infrared images of the steel pipe columns in the subway station, and performing semantic segmentation on the RGB and infrared images so that they contain only the steel pipe columns; determining the degree of surface anomalies of the steel pipe columns based on the RGB images, whereby the surface anomaly degree characterizes visible defects on the surface of the steel pipe columns; determining the degree of conduction anomalies of the steel pipe columns based on the infrared images, whereby the conduction anomaly degree characterizes internal defects in the steel pipe columns; determining the type of anomaly region based on the surface and conduction anomalies; and determining the defect region based on the type of anomaly region. This method can accurately obtain information on the defect type of the steel pipe columns, complete the detection of anomalies in the steel pipe columns of subway stations, and facilitate subsequent operation and maintenance. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating an embodiment of the method for detecting steel pipe columns in subway stations based on infrared sensors according to the present invention.

[0047] Figure 2 for Figure 1 A flowchart illustrating an embodiment of calculating the degree of surface anomaly in step S12;

[0048] Figure 3 for Figure 1 A flowchart illustrating an embodiment of calculating the degree of conduction anomaly in step S12. Detailed Implementation

[0049] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0050] The present application will now be described in detail with reference to the accompanying drawings and embodiments.

[0051] Please see Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the infrared sensor-based method for detecting steel pipe columns in subway stations according to the present invention, specifically including:

[0052] Step S11: Acquire RGB and infrared images of the steel pipe columns in the subway station, and perform semantic segmentation on the RGB and infrared images so that the RGB and infrared images contain only the steel pipe columns.

[0053] In the daily use and maintenance of subway facilities, steel pipe columns bear the loads of large passenger and vehicle traffic in subway stations. Over time, they may develop defects such as corrosion, partial fracture, and detachment. Besides visible surface defects, there may also be internal defects that are not visible on the surface. These defects increase the safety risks of the steel pipe columns. Therefore, in this scenario, image acquisition is used for defect detection of steel pipe columns.

[0054] Conventional machine vision devices are used to acquire RGB and infrared images of steel pipe columns in subway stations. However, the acquired images often contain many interfering areas. Semantic segmentation technology is used to segment the RGB and infrared images so that they only contain the area of ​​the subway steel pipe columns.

[0055] Step S12: Determine the degree of surface anomaly of the steel pipe column based on the RGB image, which characterizes the visible defects on the surface of the steel pipe column; and determine the degree of conduction anomaly of the steel pipe column based on the infrared image, which characterizes the defects inside the steel pipe column.

[0056] Specifically, in combination Figure 2 and Figure 3 This step will be explained in detail, among which... Figure 2 for Figure 1 A flowchart illustrating an embodiment of calculating the degree of surface anomaly in step S12; Figure 3for Figure 1 A flowchart illustrating an embodiment of calculating the degree of conduction anomaly in step S12. First, see... Figure 2 Methods for calculating the degree of surface anomalies include:

[0057] Step S21: Convert the RGB image to an RGB grayscale image.

[0058] The segmented RGB image is converted to grayscale to obtain an RGB grayscale image.

[0059] Step S22: Process the RGB grayscale image using an adaptive local threshold segmentation algorithm to obtain multiple binary images.

[0060] Based on the acquired RGB grayscale image, since the grayscale features of the abnormal areas on the surface are obvious, the adaptive local threshold segmentation algorithm is used to obtain the abnormal areas. The adaptive local threshold segmentation algorithm is a conventional method, and the specific process will not be described in detail.

[0061] Step S23: Obtain multiple connected components based on multiple binary images, each connected component representing a surface anomaly region.

[0062] This results in multiple connected components obtained from the segmented binary image, with each connected component representing a surface anomaly region.

[0063] Step S24: Calculate the degree of surface anomaly based on the area of ​​the surface anomaly region, the gray value inside the surface anomaly region, and the difference in gray value between the neighborhood of the surface anomaly region.

[0064] In one embodiment, the degree of surface anomaly is calculated based on the area of ​​the surface anomaly region, the difference between the gray values ​​within the surface anomaly region and the gray values ​​of the neighborhood of the surface anomaly region, including:

[0065]

[0066] Among them, SD i This represents the degree of surface anomalousness of the i-th surface anomalous region, which is related to the size of the surface anomalous region and the absolute value of the gray-level difference in its neighboring area. The larger the area of ​​the surface anomalous region and the larger the absolute value of the gray-level difference in its neighboring area, the greater the degree of surface anomalousness of the surface anomalous region. Si represents the size of the i-th surface anomalous region. This represents the average grayscale value of the set of pixels in the outer neighborhood of the surface anomaly region. The set of pixels in the outer neighborhood can be obtained by traversing an operator of a smaller size. This represents the average gray value within the abnormal area on the surface. It represents the absolute value of the grayscale difference in the neighborhood of the abnormal area on the surface; This represents the minimum value of the product of the area of ​​the surface anomaly region and the absolute value of the gray level difference of its neighbors; it is used as a normalization parameter. This represents the maximum value of the product of the area of ​​the abnormal region on the surface and the absolute value of the gray level difference of its neighbors; it is used as a normalization parameter.

[0067] Please see Figure 3 Methods for calculating the degree of conduction anomalies include:

[0068] Step S31: Convert the infrared image into an infrared grayscale image.

[0069] The segmented infrared image is converted to grayscale to obtain an infrared grayscale image.

[0070] Step S32: Process the infrared grayscale image using an adaptive local threshold segmentation algorithm to obtain multiple binary images.

[0071] Based on the acquired infrared grayscale image, since the grayscale features of the abnormal areas on the surface are obvious, the adaptive local threshold segmentation algorithm is used to obtain the abnormal areas. The adaptive local threshold segmentation algorithm is a conventional method, and the specific process will not be described in detail.

[0072] Step S33: Obtain multiple connected components based on multiple binary images, each connected component representing a transmission anomalous region.

[0073] This results in multiple connected components obtained from the segmented binary image, with each connected component representing a surface anomaly region.

[0074] Step S34: Calculate the degree of conduction anomaly based on the area of ​​the conduction anomaly region, the difference between the gray values ​​inside the conduction anomaly region and the gray values ​​of the neighborhood of the conduction anomaly region.

[0075] In one embodiment, the degree of conduction anomaly is calculated based on the area of ​​the conduction anomaly region, the difference between the grayscale value inside the conduction anomaly region and the grayscale value of the neighborhood of the conduction anomaly region, including:

[0076]

[0077] Among them, ID i This represents the degree of conduction aberration in the i-th conduction aberration region, which is related to the size of the conduction aberration region and the absolute value of the gray-level difference in its neighboring areas. The larger the area of ​​the conduction aberration region and the larger the absolute value of the gray-level difference in its neighboring areas, the greater the degree of conduction aberration in the conduction aberration region; Si represents the size of the i-th conduction aberration region. It represents the average gray value of the set of pixels in the neighborhood outside the transmission anomaly region. The set of pixels in the neighborhood can be obtained by traversing the smaller size operator. This represents the average gray level within the region of conduction anomaly. This represents the absolute value of the grayscale difference in the neighborhood of the transmission anomaly region; This represents the minimum value of the product of the area of ​​the region with abnormal conduction and the absolute value of the gray level difference of its neighbors, and is used as a normalization parameter; This represents the maximum value of the product of the area of ​​the transmission anomaly region and the absolute value of the gray level difference of its neighbors, and it is used as a normalization parameter.

[0078] Because defective areas in steel pipe columns often exhibit structural differences, leading to energy loss or accumulation, in addition to visible surface defects, hotspot anomalies in infrared images indicate abnormal areas. Therefore, RGB grayscale images are used to obtain the degree of surface dust abnormalities, and infrared grayscale images are used to obtain the degree of conduction anomalies. There is a certain correlation between the internal and external anomaly degrees, and the degree of correlation can be used to determine the defect type. However, this judgment process has a confidence level, which is related to the location of the defect area; the closer to the heat conduction area, the higher the confidence level.

[0079] Since defective areas of steel pipe columns will exhibit hotspot anomalies in infrared images and grayscale anomalies in RGB grayscale images, the degree of surface anomaly can be obtained by analyzing the neighborhood grayscale features of RGB grayscale images. The degree of surface anomaly measures the visible defective areas on the surface of the steel pipe column. The degree of conduction anomaly can be obtained by analyzing the neighborhood grayscale features of infrared grayscale images. The degree of conduction anomaly measures the defective areas inside the steel pipe column.

[0080] Step S13: Determine the type of anomalous region based on the degree of surface anomaly and the degree of conduction anomaly.

[0081] Specifically, a coordinate system is established based on the degree of surface anomaly and the degree of conduction anomaly. It should be noted that the anomaly region is a connected region. In the coordinate system, the horizontal axis represents the degree of surface anomaly of the anomaly region, and the vertical axis represents the degree of conduction anomaly of the anomaly region. Points on the coordinate system represent anomaly regions. The Euclidean distance between two anomaly regions is calculated, and the similarity between the two anomaly regions is determined based on the Euclidean distance. Based on the similarity, the points on the coordinate system are clustered to obtain multiple clusters, and the type of anomaly region corresponding to each cluster is determined.

[0082] In one embodiment, the anomaly confidence level of each anomalous region is calculated; the distance between two anomalous regions is weighted based on the anomaly confidence level to obtain a weighted Euclidean distance; and the similarity between the two anomalous regions is determined based on the weighted Euclidean distance.

[0083] Specifically, because the mean-shift clustering algorithm is sensitive to initial clustering points and noise, the outlier confidence level is used as the weight of the traditional Euclidean distance to measure the similarity of outlier points in the transformed coordinate system, thereby increasing the distinguishability of data points represented by different regions. Therefore, the weighted Euclidean distance is:

[0084]

[0085] in, For weighted Euclidean distance, D represents the Euclidean distance between data points representing two connected components in the coordinate system, and ΔDB represents the absolute value of the difference in outlier confidence scores between the connected components.

[0086] If the weighted Euclidean distance is greater than a preset distance, the two abnormal regions are determined to belong to the same cluster; if the weighted Euclidean distance is less than the preset distance, the two abnormal regions are determined to belong to different clusters.

[0087] Furthermore, this embodiment requires calculating the anomaly confidence level for each abnormal region. Based on the acquired infrared grayscale image, the grayscale value of each pixel in the image represents the magnitude of heat distribution at the current location. The changes in grayscale distribution in the infrared grayscale image are mainly due to heat conduction on the steel pipe column. The grayscale value is high in the area connecting the upper and lower parts of the steel pipe column and diffuses towards the middle of the steel pipe column, where the grayscale value is lower. On the other hand, defective areas of the steel pipe column will appear as bright areas in the infrared grayscale image.

[0088] Therefore, edge detection algorithms are used to obtain edge images. Due to the imaging characteristics of infrared images, based on the traversal results of the obtained edge pixels, the image can be divided into three regions by the two longest edge line segments: the upper connected region, the middle region, and the lower connected region. The edge line segments in each region are connected, and region filling is performed to obtain multiple connected components. The anomaly confidence level DB of the abnormal region represented by the connected component is related to the region location and the absolute value of the gray-level difference of the neighboring area. Therefore, the anomaly confidence level is calculated for each connected component as an abnormal region; including:

[0089]

[0090] Where DB represents the anomaly confidence level of the abnormal region, and d represents the minimum distance from the center pixel of the current connected component to the upper and lower boundaries of the image. Let g represent the average gray value of the i-th region among the three different regions of the image, and g represent the average gray value of the current connected component. The three different regions are the upper connected region, the middle region, and the lower connected region.

[0091] Further, based on the clusters, the type of anomalous region is determined, specifically including: when the surface anomalousness degree SD of the anomalous region in the cluster is greater than the conduction anomalousness degree ID, the cluster does not have surface or internal anomalousness; when the surface anomalousness degree SD of the anomalous region in the cluster is approximately equal to the conduction anomalousness degree ID, the cluster belongs to surface anomalous defects; when the surface anomalousness degree SD of the anomalous region in the cluster is less than the conduction anomalousness degree ID, the cluster belongs to internal anomalous defects.

[0092] The above methods can be used to determine the type of abnormal region.

[0093] Step S14: Determine the defect area based on the type of abnormal area.

[0094] Since defective areas of steel pipe columns will exhibit hotspot anomalies in infrared images and grayscale anomalies in RGB grayscale images, the degree of surface anomaly can be obtained by analyzing the neighborhood grayscale features of RGB grayscale images. The degree of surface anomaly measures the visible defective areas on the surface of the steel pipe column. The degree of conduction anomaly can be obtained by analyzing the neighborhood grayscale features of infrared grayscale images. The degree of conduction anomaly measures the defective areas inside the steel pipe column.

[0095] Specifically, since each steel pipe column has a variety of defect types and generally contains only one or more defect areas, accurate defect type detection is carried out based on the obtained set of multiple defect types and the prior information of defect types.

[0096] Based on the obtained sets of multiple defect types, the morphological information of the surface abnormal region and the conduction abnormal region corresponding to the geometric center point is obtained, including its length, width and the area of ​​the overlapping area of ​​the inner and outer abnormal regions. This information is then matched with the prior information of known defect types to determine whether the defect region belongs to a variety of precise defect types such as cracks and detachment, thus completing the defect type detection.

[0097] The method of this application obtains the surface abnormal area and conduction abnormal area of ​​each defect area based on the defect detection results in the image data analysis process. In order to facilitate users to view the abnormal condition of the detected steel pipe column more intuitively, the defect type information and the outline of the surface abnormal area and the outline of the conduction abnormal area are visualized on the corresponding display, which facilitates the subsequent operation and maintenance of the subway station steel pipe column.

[0098] The advantages of this application over existing technologies are as follows: it can obtain the correlation between the degree of surface anomaly and the degree of conduction anomaly based on the neighborhood grayscale variation characteristics of a local area of ​​a steel pipe column, and distinguish the defect type information of the steel pipe column, which is more accurate than traditional infrared detection. Furthermore, it can weight the Euclidean distance in the clustering algorithm according to the anomaly confidence level of the anomaly region, making it more accurate and effective in distinguishing the types of regional characterization data points compared to traditional clustering algorithms.

[0099] The above are merely embodiments of this application and do not limit the scope of this patent application. Any equivalent structural or procedural changes made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of this application.

Claims

1. A method for detecting steel pipe columns in subway stations based on infrared sensors, characterized in that, include: RGB and infrared images of steel pipe columns in subway stations are acquired, and semantic segmentation is performed on the RGB and infrared images so that the RGB and infrared images contain only the steel pipe columns. The degree of surface anomaly of the steel pipe column is determined based on the RGB image, wherein the degree of surface anomaly characterizes visible defects on the surface of the steel pipe column; and the degree of conduction anomaly of the steel pipe column is determined based on the infrared image, wherein the degree of conduction anomaly characterizes internal defects of the steel pipe column. The type of the abnormal region is determined based on the degree of surface anomaly and the degree of conduction anomaly. Defect areas are determined based on the type of abnormal areas; Determining the type of the anomalous region based on the degree of surface anomaly and the degree of conduction anomaly includes: A coordinate system is established based on the surface anomaly degree and the conduction anomaly degree, with the horizontal axis representing the surface anomaly degree of the abnormal area and the vertical axis representing the conduction anomaly degree of the abnormal area. Points on the coordinate system represent abnormal areas. Calculate the Euclidean distance between two anomalous regions, and determine the similarity between the two anomalous regions based on the Euclidean distance; Based on the similarity, points on the coordinate system are clustered to obtain multiple clusters, and the type of the abnormal region corresponding to the cluster is determined. When the surface anomalousness of anomalous regions in a cluster Degree of conduction abnormality At that time, the clusters do not have surface or internal anomalies; When the surface anomalousness of anomalous regions in a cluster Degree of conduction abnormality At that time, the cluster belongs to surface anomaly defects; When the surface anomalousness of anomalous regions in a cluster Degree of conduction abnormality At that time, the cluster belongs to an internal abnormal defect.

2. The method for detecting steel pipe columns in subway stations based on infrared sensors according to claim 1, characterized in that, Determining the degree of surface anomalies of the steel pipe column based on the RGB image includes: Convert the RGB image to an RGB grayscale image; The RGB grayscale image is processed using an adaptive local threshold segmentation algorithm to obtain multiple binary images; Multiple connected components are obtained based on the multiple binary images, and each connected component represents a surface anomaly region. The degree of surface anomaly is calculated based on the area of ​​the surface anomaly region, the difference between the gray value inside the surface anomaly region and the gray value of the neighborhood of the surface anomaly region.

3. The method for detecting steel pipe columns in subway stations based on infrared sensors according to claim 2, characterized in that, The degree of surface anomaly is calculated based on the area of ​​the surface anomaly region, the difference between the gray values ​​within the surface anomaly region and the gray values ​​of its neighbors, including: ； in, It indicates the first The degree of surface anomaly in each surface anomaly region It indicates the first The size of each surface anomalous region It represents the average gray value of the set of pixels in the outer neighborhood of the surface anomaly region. This represents the average gray value within the abnormal area on the surface.

4. The method for detecting steel pipe columns in subway stations based on infrared sensors according to claim 1, characterized in that, Determining the degree of conduction anomaly of the steel pipe column based on the infrared image includes: converting the infrared image into an infrared grayscale image; The infrared grayscale image is processed using an adaptive local threshold segmentation algorithm to obtain multiple binary images; Multiple connected components are obtained based on the multiple binary images, and each connected component represents a region of conduction anomaly. The degree of conduction abnormality is calculated based on the area of ​​the conduction abnormal region, the difference between the gray value inside the conduction abnormal region and the gray value of the neighborhood of the conduction abnormal region.

5. The method for detecting steel pipe columns in subway stations based on infrared sensors according to claim 4, characterized in that, The degree of conduction abnormality is calculated based on the area of ​​the conduction abnormality region, the difference between the gray values ​​within the conduction abnormality region and the gray values ​​of its neighboring regions, including: ； in, It indicates the first The degree of conduction abnormality in each conduction abnormality region It indicates the first The size of each abnormal conduction region It represents the average gray value of the set of pixels in the outer neighborhood of the region with the abnormal conduction. This represents the average gray level within the region of conduction anomaly.

6. The method for detecting steel pipe columns in subway stations based on infrared sensors according to claim 1, characterized in that, Determining the similarity between the two anomalous regions based on the Euclidean distance includes: Calculate the anomaly confidence level for each anomaly region; The distance between the two abnormal regions is weighted based on the anomaly confidence level to obtain the weighted Euclidean distance. The similarity between the two anomalous regions is determined based on the weighted Euclidean distance.

7. The method for detecting steel pipe columns in subway stations based on infrared sensors according to claim 6, characterized in that, If the weighted Euclidean distance is less than a preset distance, the two abnormal regions are determined to belong to the same cluster; if the weighted Euclidean distance is greater than the preset distance, the two abnormal regions are determined to belong to different clusters.

8. The method for detecting steel pipe columns in subway stations based on infrared sensors according to claim 6, characterized in that, Calculate the anomaly confidence level for each anomaly region; include: ； in, This represents the anomaly confidence level of the abnormal region. This represents the minimum distance from the center pixel of the current abnormal region to the upper and lower boundaries of the image. The third region represents three different regions of the image. The average gray level of each region This represents the average gray value corresponding to the current abnormal region. The three different regions are the upper connected region, the middle region, and the lower connected region. The distance between two anomalous regions is weighted based on the aforementioned anomaly confidence level to obtain a weighted Euclidean distance, including: ； in, For weighted Euclidean distance, This represents the Euclidean distance between two outlier regions in the coordinate system. It represents the absolute value of the difference in confidence levels between outliers in an outlier region.

Images