A method and system for detecting cracks in cable sealing fireproof mortar

Abstract

A method for detecting cracks in fireproof sealant in cable sealing is disclosed. This method utilizes an RGB-D depth camera to capture images of the fireproof sealant area within a power protection cabinet, generating a target detection model for the fireproof sealant area and a crack detection model for the fireproof sealant. These models are then used to detect the presence of fireproof sealant areas or suspected crack areas. If a suspected crack area is found, the information is input into a crack fusion detection model. Based on the obtained crack fusion detection model, the results of the crack detection structure are output. If no abnormalities are found, the process ends; if a crack area is found, an alarm is reported. This invention offers faster detection and identification speeds, solves the problem of high false detection rates and impracticality in traditional fireproof sealant crack detection methods, reduces reliance on manual inspection and analysis, and improves the accuracy and efficiency of fireproof sealant crack detection in power protection cabinets.

Description

Technical Field

[0001] This invention belongs to the field of power equipment protection, specifically relating to a method and system for detecting cracks in cable sealing fireproof mud. Background Technology

[0002] Power protection cabinets are widely used in the construction of the State Grid. Grounding cables, as an important component of these cabinets, play a crucial role in the normal operation of the equipment. Cables enter the cabinet through external openings, and the gaps in these openings need to be sealed with fire-retardant putty. This fire-retardant putty effectively prevents external sources of fire, small animals, dust, smoke, etc., from entering the cabinet, ensuring its safety.

[0003] Fireproof sealant may develop cracks after prolonged use, leading to detachment and safety hazards. Regular crack inspections are necessary. However, power protection cabinets have numerous grounding cables, making manual inspection time-consuming and labor-intensive. Furthermore, close-range inspections during equipment operation pose significant safety risks. Image recognition technology can effectively address these issues. Image recognition technology typically uses traditional image analysis methods, such as edge feature extraction and support vector machine classification, as well as object detection based on deep learning neural networks, such as R-CNN and YOLO series networks. These methods utilize two-dimensional image information for analysis, lacking depth information, resulting in a high false detection rate and limiting practical application. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for detecting cracks in fireproof sealant used for cable sealing. This system can be embedded into a large-scale resource management platform to intelligently analyze the fireproof sealant portion of grounding cables inside power protection cabinets, reducing reliance on manual inspection and analysis and improving the accuracy and efficiency of fireproof sealant crack detection in power protection cabinets.

[0005] The present invention adopts the following technical solution.

[0006] A method for detecting cracks in cable sealing fireproof sealant includes the following steps:

[0007] Step S1: The RGB-D depth camera takes a picture to obtain a photo of the fireproof mud area in the power protection cabinet, and obtains the target detection model of the fireproof mud area and the crack detection model of the fireproof mud.

[0008] Step S2, Detection stage: Input the on-site image taken in step S1, and use the fireproof mud area target detection model and fireproof mud crack detection model provided in step S1 to detect whether there is a fireproof mud area or a suspected crack area. If there is a suspected crack area, the information is input to the crack fusion detection model.

[0009] Step S3: Based on the crack fusion detection model obtained from the input information in step S2, output the crack detection results.

[0010] Step S4: If no abnormality is found, the process ends; if a cracked area is found, an alarm is reported.

[0011] Furthermore, step S1 specifically includes:

[0012] Step S11: The collected photos of the fireproof mud area are used to train a convolutional neural network. Two models need to be trained, including a fireproof mud area detection model and a crack detection model.

[0013] Step S12: Using dataset annotation software, the fireproof mud area and the fireproof mud crack area are labeled respectively, and then data augmentation techniques are used to expand and create two datasets.

[0014] Step S13: Using the YOLO series of deep learning networks, load the two datasets provided in step S12 into the network for training, and use the model of the COCO large dataset as a pre-trained model to obtain the fireproof mud area detection model and the fireproof mud crack detection model.

[0015] Furthermore, step S2 specifically includes:

[0016] The fireproof putty area target detection model provided in step S1 is used to detect whether there is a fireproof putty area. If not, the detection process ends. If there is a fireproof putty area, an image of the fireproof putty area is captured. Then, the fireproof putty crack detection model provided in step S1 is used to detect whether there is a suspected crack target area. If there is no suspected crack target area, the process ends. The information input to the crack fusion detection model includes a color image, a depth image, fireproof putty area information, and crack area information.

[0017] Furthermore, step S3 specifically includes:

[0018] Step S31: Obtain the depth map from the RGB-D camera;

[0019] Step S32: Pair the RGB image and the depth map by pixel to obtain the pixel coordinate system [u,v,d] of the image. Based on the camera parameters, derive the spatial point cloud coordinate system [x,y,z], where u is the horizontal coordinate, v is the vertical coordinate, and d is the depth coordinate.

[0020] Step S33: Based on the area range detected by the fireproof mud area detection model, a local point cloud map of the fireproof mud area is extracted;

[0021] Step S34: Using the Aggregate Hierarchical Clustering (AHC) algorithm, obtain each planar region in the point cloud map of the fireproof mud region. Determine the plane with the largest area in the point cloud map of the fireproof mud region as the plane where the fireproof mud is located. Calculate its plane normal vector and define it as the XOY plane. Rotate and transform the point cloud map to the XOY plane. If the suspected crack area output by the fireproof mud crack detection model is not within the XOY plane range, it means that it is not in the fireproof mud plane region and can be judged as a false detection and excluded. If there are still suspected crack areas, further inspection is carried out.

[0022] Step S35: After mapping the point cloud map of the suspected crack area onto the XOY plane, new [x,y,z] point cloud coordinates are obtained, where z is the depth coordinate of the fireproof mud plane. If a continuous local depth change area appears when scanning the area, it is determined that the area contains a crack.

[0023] Furthermore, step S31 specifically includes:

[0024] The depth map is then subjected to median filter noise reduction processing, namely:

[0025] g(x, y)=med{f(xk, yl), (k, l∈W)}

[0026] Where g(x,y) is the filtered image, f(x,y) is the original depth image, and W is the median filter template.

[0027] Furthermore, step S32 specifically includes:

[0028] The derivation process of the spatial point cloud coordinate system [x,y,z] includes:

[0029] z = d / s

[0030] x = (uc) x )·z / f x

[0031] y = (vc y )·z / f y

[0032] Among them, f x f y The focal length of the camera on both the x and y axes, c x c y The aperture center of the camera is indicated by 's', and the scaling factor of the depth map is indicated by 's'.

[0033] Furthermore, step S35 specifically includes:

[0034] Step S351: Map the z-axis depth coordinates of the XOY plane map [x,y,z] of the suspected crack area to a range of 1 to 10, representing 10 depth levels, to simplify subsequent calculations;

[0035] Step S352: Use the Seed-Filling algorithm to calculate the connectivity threshold of the depth map;

[0036] Step S353: Ra is the proportion of the area of ​​connected regions with depth greater than a to the total area. Based on the value of Ra, determine whether the image contains cracks.

[0037] Furthermore, step S352 specifically includes:

[0038] The steps of the Seed-Filling algorithm are as follows:

[0039] Step S3521: Scan the image until the current pixel B(x,y)==n;

[0040] Step S35211: Take B(x,y) as the seed, i.e., the pixel position, and assign it a label. Then push all the foreground pixels adjacent to the seed onto the stack.

[0041] Step S35212: Pop the top pixel from the stack, assign it the same label, and then push all the foreground pixels adjacent to the top pixel onto the stack.

[0042] Step S35213: Repeat step S35212 until the stack is empty, and obtain a connected region in image B. The pixel values ​​in this region are labeled as labels.

[0043] Step S3522: Repeat step S3521 until the scanning of n from 1 to 10 is completed. After the scanning is completed, all connected regions in image B are obtained.

[0044] Furthermore, step S353 specifically includes:

[0045]

[0046] Where Ra is the proportion of the area of ​​the connected domain with depth greater than a to the area of ​​the suspected crack region, Sn is the area of ​​the connected domain with a calculated depth of n, and S is the area of ​​the suspected crack region.

[0047] A cable sealing fireproof mud crack detection system is used in the aforementioned cable sealing fireproof mud crack detection method.

[0048] The beneficial effects of this invention are that, compared with the prior art, it can be embedded into a large-scale resource management platform to perform intelligent analysis on the fireproof sealant of the grounding cable inside the power protection cabinet, reducing the reliance on manual inspection and analysis and improving the accuracy and efficiency of fireproof sealant crack detection in power protection cabinets; by using an RGB-D depth camera to collect depth information, and performing comprehensive modeling and analysis of image and depth information, the detection accuracy of cable sealing fireproof sealant cracks is greatly improved, and it has a faster detection and recognition speed, solving the problem of high false detection rate and impracticality of traditional fireproof sealant crack detection. Attached Figure Description

[0049] Figure 1 This invention provides a target detection neural network structure for a method of detecting cracks in cable sealing fireproof mud.

[0050] Figure 2 This invention provides a crack fusion detection model for a method of detecting cracks in cable sealing fireproof mud.

[0051] Figure 3 The overall flowchart of a method for detecting cracks in cable sealing fireproof mud provided by the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, other embodiments obtained by those skilled in the art without creative effort are all within the protection scope of this invention.

[0053] according to Figure 2 and 3 As shown, the present invention provides a method for detecting cracks in cable sealing fireproof sealant, comprising the following steps:

[0054] Step S1: Take photos of the fireproof mud area in the power protection cabinet by means of manual inspection or other inspection equipment such as inspection robots, fixed cameras, etc., using an RGB-D depth camera to obtain photos of the fireproof mud area and fireproof mud crack detection model.

[0055] Step S11: The collected fireproof mud area photos are used to train a convolutional neural network. Two models need to be trained: 1. Fireproof mud area detection model, used to detect and extract the area where the fireproof mud is located; 2. Crack detection model, used to detect cracks.

[0056] Step S12: Using dataset annotation software, the fireproof mud area and the fireproof mud crack area are labeled respectively. Then, data augmentation techniques such as mirroring, distortion, and cropping are used to expand the dataset and create two datasets. The samples of the fireproof mud area are used to train the fireproof mud area detection model and the fireproof mud crack detection model respectively.

[0057] Step S13, use a YOLO series deep learning network, the network structure is as follows: Figure 1 As shown, the two datasets provided in step S12 are loaded into the network for training. The model of the COCO large dataset is used as the pre-trained model to improve the generalization ability and accelerate the convergence speed of the model. Finally, the target detection model of fireproof mud area and the crack detection model of fireproof mud are obtained.

[0058] Step S2, Detection Phase: Input the on-site image captured by the RGB-D depth camera in Step S1. First, use the fireproof clay area target detection model provided in Step S1 to detect whether there is a fireproof clay area. If not, the detection process ends. If there is a fireproof clay area, capture an image of the fireproof clay area and use the fireproof clay crack detection model provided in Step S1 to detect it. If there is no suspected crack target area, the process ends. If there is a suspected crack target area, collect the color image, depth image, fireproof clay area information, and crack area information as input information for the subsequent crack fusion detection model.

[0059] Step S3: Based on the crack fusion detection model obtained from the input information in step S2, output the crack detection results. Step S3 specifically includes:

[0060] Step S31: Obtain the depth map from the RGB-D camera. Since noise such as holes may exist during the acquisition of the depth map using an RGB-D camera, noise reduction processing is required. A popular median filter can be used for noise reduction to improve the accuracy of the results. The median filter is defined as:

[0061] g(x, y)=med{f(xk, yl), (k, l∈W)}

[0062] Where g(x,y) is the filtered image, f(x,y) is the original depth image, and W is the median filter template.

[0063] Step S32: Pair the RGB image and depth map by pixel to obtain the image's pixel coordinate system [u,v,d] (u,v are the horizontal and vertical coordinates, and d is the depth coordinate). Combined with the camera parameters, the spatial point cloud coordinate system [x,y,z] is derived through matrix transformation. The derivation formula is as follows:

[0064] z = d / s

[0065] x = (uc)x )·z / f x

[0066] y = (vc y )·z / f y

[0067] Among them, f x f y The focal length of the camera on both the x and y axes, c x c y The aperture center of the camera is indicated by 's', and the scaling factor of the depth map is indicated by 's'.

[0068] Step S33: Based on the area range detected by the fireproof mud area detection model, a local point cloud map of the fireproof mud area is extracted.

[0069] Step S34: Using the Aggregate Hierarchical Clustering (AHC) algorithm, obtain the planar regions in the point cloud map of the fireproof mud area. The plane with the largest area in the point cloud map within the fireproof mud area can be identified as the plane containing the fireproof mud. Calculate its plane normal vector, designate it as the XOY plane, and rotate the point cloud map to the XOY plane. If the suspected crack area output by the fireproof mud crack detection model is not within the XOY plane, it indicates that it is not within the fireproof mud planar region and can be determined as a false detection and excluded. If there are still suspected crack areas, further verification is performed.

[0070] Step S35: After mapping the point cloud map of the suspected crack area onto the XOY plane, new [x, y, z] point cloud coordinates are obtained, where z is the depth coordinate of the fireproof mud plane. Scanning this area, if continuous local abrupt changes in depth occur, it can be determined that the area contains a crack. The specific method is as follows:

[0071] Step S351: Map the z-axis depth coordinates of the XOY plane map [x,y,z] of the suspected crack area to a range of 1 to 10, representing 10 depth levels, to simplify subsequent calculations.

[0072] Step S352: Using the Seed-Filling algorithm, calculate the connectivity threshold of the depth map. The algorithm steps are as follows:

[0073] Step S3521: Scan the image until the current pixel B(x,y)==n;

[0074] Step S35211: Take B(x,y) as a seed (pixel position) and assign it a label, then push all the foreground pixels adjacent to the seed onto the stack;

[0075] Step S35212: Pop the top pixel from the stack, assign it the same label, and then push all the foreground pixels adjacent to the top pixel onto the stack.

[0076] Step S35213: Repeat step S35212 until the stack is empty;

[0077] At this point, a connected region in image B is found, and the pixel values ​​within this region are labeled as labels;

[0078] Step S3522, repeat step S3521, until the scan of n from 1 to 10 is completed;

[0079] After the scan is complete, all connected regions in image B can be obtained;

[0080] Step S353: Calculate the area Sn of the connected region with depth n, the area of ​​the suspected crack region S, and the area of ​​the connected region with depth greater than a as a percentage of the area of ​​the suspected crack region Ra.

[0081]

[0082] Experiments showed that 'a' can be set to 8, and 'Ra' is greater than the threshold of 0.3, indicating that the image contains cracks. This means that in the suspected crack area, regions with a depth level greater than 8 account for more than 0.3% of the total area. This region is a continuous local abrupt change in depth.

[0083] Step S4: If no abnormality is found, the process ends; if a cracked area is found, an alarm is reported.

[0084] A cable sealing fireproof mud crack detection system is used in the aforementioned cable sealing fireproof mud crack detection method.

[0085] The beneficial effects of this invention are that, compared with the prior art, it can be embedded into a large-scale resource management platform to perform intelligent analysis on the fireproof sealant of the grounding cable inside the power protection cabinet, reducing the reliance on manual inspection and analysis and improving the accuracy and efficiency of fireproof sealant crack detection in power protection cabinets; it performs comprehensive modeling and analysis of image information and depth information, which greatly improves the detection accuracy of cable sealing fireproof sealant cracks and has a faster detection and recognition speed, solving the problem of high false detection rate and impracticality of traditional fireproof sealant crack detection.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for detecting cracks in cable sealing fireproof sealant, characterized in that, Includes the following steps: Step S1: The RGB-D depth camera takes a picture to obtain a photo of the fireproof mud area in the power protection cabinet, and obtains the target detection model of the fireproof mud area and the crack detection model of the fireproof mud. Step S2, Detection stage: Input the on-site image taken in step S1, and use the fireproof mud area target detection model and fireproof mud crack detection model provided in step S1 to detect whether there are fireproof mud areas and suspected crack areas. If a suspected crack area is found, the information is input to the crack fusion detection model. Step S3: Based on the crack fusion detection model obtained from the input information in step S2, output the crack detection results. Step S3 specifically includes: Step S31: Obtain the depth map from the RGB-D camera; Step S32: Pair the RGB image and the depth map by pixel to obtain the pixel coordinate system [u,v,d] of the image. Based on the camera parameters, derive the spatial point cloud coordinate system [x,y,z], where u is the horizontal coordinate, v is the vertical coordinate, and d is the depth coordinate. Step S33: Based on the area range detected by the fireproof mud area detection model, a local point cloud map of the fireproof mud area is extracted; Step S34: Using the Aggregate Hierarchical Clustering (AHC) algorithm, obtain each planar region in the point cloud map of the fireproof mud region. Determine the plane with the largest area in the point cloud map of the fireproof mud region as the plane where the fireproof mud is located. Calculate its plane normal vector and define it as the XOY plane. Rotate and transform the point cloud map to the XOY plane. If the suspected crack area output by the fireproof mud crack detection model is not within the XOY plane range, it means that it is not in the fireproof mud plane region, and it is judged as a false detection and excluded. If there are still suspected crack areas, further inspection is performed. Step S35: After mapping the point cloud map of the suspected crack area onto the XOY plane, a new [x,y,z] point cloud coordinate is obtained, where z is the depth coordinate of the fireproof mud plane. If a continuous local depth change area appears when scanning the area, it is determined that the area contains a crack. Step S4: If no abnormality is found, the process ends; if a cracked area is found, an alarm is reported.

2. The method for detecting cracks in cable sealing fireproof sealant according to claim 1, characterized in that, Step S1 specifically includes: Step S11: The collected photos of the fireproof mud area are used to train a convolutional neural network. Two models need to be trained, including a fireproof mud area detection model and a crack detection model. Step S12: Using dataset annotation software, the fireproof mud area and the fireproof mud crack area are labeled respectively, and then data augmentation techniques are used to expand and create two datasets. Step S13: Using the YOLO series of deep learning networks, load the two datasets provided in step S12 into the network for training, and use the model of the COCO large dataset as a pre-trained model to obtain the fireproof mud area detection model and the fireproof mud crack detection model.

3. The method for detecting cracks in cable sealing fireproof sealant according to claim 1, characterized in that, Step S2 specifically includes: The fireproof putty area target detection model provided in step S1 is used to detect whether there is a fireproof putty area. If not, the detection process ends. If there is a fireproof putty area, an image of the fireproof putty area is captured. Then, the fireproof putty crack detection model provided in step S1 is used to detect whether there is a suspected crack target area. If there is no suspected crack target area, the process ends. The information input to the crack fusion detection model includes color images, depth maps, fireproofing mud area information, and crack area information.

4. The method for detecting cracks in cable sealing fireproof sealant according to claim 1, characterized in that, Step S31 specifically includes: The depth map is then subjected to median filter noise reduction processing, namely: g(x,y)=med{f(xk,yl),(k,l∈W)} Where g(x,y) is the filtered image, f(x,y) is the original depth image, and W is the median filter template.

5. The method for detecting cracks in cable sealing fireproof sealant according to claim 1, characterized in that, Step S32 specifically includes: The derivation process of the spatial point cloud coordinate system [x,y,z] includes: z = d / s x=(u-c x )·z / f x y=(v-c y )·z / f y Among them, f x f y The focal length of the camera on both the x and y axes, c x c y The aperture center of the camera is indicated by 's', and the scaling factor of the depth map is indicated by 's'.

6. The method for detecting cracks in cable sealing fireproof sealant according to claim 1, characterized in that, Step S35 specifically includes: Step S351: Map the z-axis depth coordinates of the XOY plane map [x,y,z] of the suspected crack area to a range of 1 to 10, representing 10 depth levels, to simplify subsequent calculations; Step S352: Use the Seed-Filling algorithm to calculate the connectivity threshold of the depth map; Step S353: Ra is the proportion of the area of ​​connected regions with depth greater than a to the total area. Based on the value of Ra, determine whether the image contains cracks.

7. The method for detecting cracks in cable sealing fireproof sealant according to claim 6, characterized in that, Step S352 specifically includes: The steps of the Seed-Filling algorithm are as follows: Step S3521: Scan the image until the current pixel B(x,y)==n; Step S35211: Take B(x,y) as the seed, i.e., the pixel position, and assign it a label. Then push all the foreground pixels adjacent to the seed onto the stack. Step S35212: Pop the top pixel from the stack, assign it the same label, and then push all the foreground pixels adjacent to the top pixel onto the stack. Step S35213: Repeat step S35212 until the stack is empty, and obtain a connected region in image B. The pixel values ​​in this region are labeled as labels. Step S3522: Repeat step S3521 until the scanning of n from 1 to 10 is completed. After the scanning is completed, all connected regions in image B are obtained.

8. The method for detecting cracks in cable sealing fireproof sealant according to claim 6, characterized in that, Step S353 specifically includes: Where Ra is the proportion of the area of ​​the connected domain with depth greater than a to the area of ​​the suspected crack region, Sn is the area of ​​the connected domain with a calculated depth of n, and S is the area of ​​the suspected crack region.

9. A cable sealing fireproof sealant crack detection system, characterized in that: The system is used for the cable sealing fireproof mud crack detection method according to any one of claims 1-8.

Images