Power centralized train umbrella skirt state detection method and system

Abstract

The invention relates to the field of rail transit image processing, in particular to a power centralized train umbrella skirt state detection method and system, and the method comprises the steps: collecting a roof insulator image of a to-be-detected train, carrying out the insulator segmentation of the roof insulator image, and obtaining a first insulator foreground region; performing defect detection on the first insulator foreground area by adopting a deep learning positioning network, and judging the defect type of the train umbrella skirt according to a defect detection result; and giving an alarm prompt according to the defect type, so that a maintainer carries out rechecking and rectification according to the content of the alarm prompt. Umbrella skirt defects in a complex environment are detected by adopting the segmentation model and the positioning model, the segmentation model extracts an insulator area with shielding and a complex background from a to-be-detected image, and the positioning network positions the defects in the insulator area, so that the detection precision of umbrella skirt damage and chipping is effectively improved, and the detection efficiency is improved. And error detection caused by a complex background is avoided.

Description

Technical Field

[0001] The present invention relates to the field of rail transit image processing, and in particular to a method and system for detecting the state of a shed skirt of a power-centralized train. Background Art

[0002] Insulators are important high-voltage components in the roof pantograph structure. During operation, insulators are easily damaged, broken, and cracked on the shed surface due to long-term impact of strong winds and vibrations. If the shed damage is not discovered in time, the creepage distance will be shortened, resulting in abnormal flashover of the EMU. In severe cases, it will cause railway transportation to be interrupted, thus affecting the normal operation of the train. Therefore, it is of great significance to detect the state of the insulator shed.

[0003] At present, most of the shed damage detection uses the naked eye to observe the changes in the appearance of the insulator to make a preliminary assessment of the insulator state. The workload is large, the detection efficiency is low, and there is a lack of reliable and efficient automated detection methods. A few rely on image processing technology to intelligently identify the damage of the insulator shed. The drone-mounted camera combines the deep separable convolution and the lightweight network of the double-layer pyramid structure to detect the shed damage in real time. However, the roof insulator is located in a complex environment with occlusions, insulators are connected in series and in parallel, and the multi-channel feature extraction of a single network cannot effectively extract the insulator area and shed fault location from the complex environment. Therefore, there is a low detection accuracy. The camera equipment of the drone is only suitable for shooting static insulators in open areas and requires professional personnel to operate. If the roof insulator of a moving train is detected in real time, the use of drones will seriously affect driving safety.

[0004] In summary, the existing technology has the technical problems of low detection efficiency and low detection accuracy in the presence of occlusion and complex background. Summary of the invention

[0005] In view of this, the present invention provides a method and system for detecting the state of shed skirts of a power-centralized train, aiming to solve all or part of the above-mentioned technical problems.

[0006] In order to solve the above technical problems, the technical solution of the present invention is to provide a method for detecting the state of a shed skirt of a power-centralized train, comprising:

[0007] Collect the roof insulator image of the train to be inspected;

[0008] Performing insulator segmentation on the roof insulator image to obtain a first insulator foreground area;

[0009] Using a deep learning positioning network to perform defect detection on the foreground area of ​​the first insulator, and judging the defect type of the train shed according to the defect detection result;

[0010] An alarm is issued according to the defect type, so that maintenance personnel can review and rectify according to the content of the alarm.

[0011] As an implementation manner, performing insulator segmentation on the roof insulator image to obtain a first insulator foreground region includes:

[0012] Performing pixel expansion on the roof insulator image;

[0013] The expanded insulator image is fed into the UNet deep learning segmentation network;

[0014] Use the insulator Mask in the segmentation result as the first mask to obtain the first insulator foreground area in the original image.

[0015] As an implementation mode, the method of using a deep learning positioning network to perform defect detection on the foreground area of ​​the first insulator, and judging the defect type of the train shed according to the defect detection result, includes:

[0016] The trained YOLOv5 deep learning positioning network is used to locate defects in the foreground area of ​​the first insulator. If a damaged or missing ROI area is located, a confidence determination is performed.

[0017] If the confidence result is greater than the threshold, it is determined that the shed is suspected to be damaged or broken, otherwise the shed is normal;

[0018] The image to be detected that is determined to be suspected of being damaged or missing is compared with a pre-stored standard image for similarity difference, and based on the comparison result, it is determined that the image to be detected that is suspected of being damaged or missing is true.

[0019] As an implementation manner, the image to be detected that is determined to be suspected of being damaged or missing is compared with a pre-stored standard image for similarity difference, and according to the comparison result, determining that the image to be detected that is suspected of being damaged or missing is true includes:

[0020] The XOR operation and structural similarity algorithm are used to determine the similarity difference between the image to be detected and the ROI area in the standard image, where the value range of the structural similarity algorithm is [0,1];

[0021] If the XOR area and size in the ROI region are greater than the threshold and the similarity difference is less than the threshold, it is determined that the image to be detected that is suspected to be damaged or chipped is true, otherwise, the shed is normal.

[0022] As an implementation method, the method further includes: using a trained YOLOv5 deep learning positioning network to locate defects in the foreground area of ​​the first insulator, and if no damaged or missing ROI area is located, performing penetration damage detection on the shed skirt;

[0023] According to the result of the penetration wound detection, it is determined whether the state of the shed is normal.

[0024] As an implementation manner, the penetrating wound detection of the shed skirt includes:

[0025] Performing insulator segmentation on the pre-stored standard image to obtain a second insulator foreground area;

[0026] Performing an XOR operation on the first insulator foreground area and the second insulator foreground area to obtain an XOR result;

[0027] Whether the shed has a penetrating wound is determined based on the XOR result.

[0028] As an implementation manner, the performing insulator segmentation on the pre-stored standard image to obtain a second insulator foreground area includes:

[0029] Performing pixel expansion on the pre-stored standard image;

[0030] Send the expanded standard image to the UNet deep learning segmentation network;

[0031] Use the insulator Mask in the segmentation result as the second mask to obtain the second insulator foreground area in the original image.

[0032] As an implementation manner, before judging whether a penetrating injury occurs in the shed according to the XOR result, the method further includes:

[0033] Morphological processing is performed on the XOR result to filter out small convex parts and fill small holes in the image.

[0034] As an implementation manner, judging whether a penetrating injury occurs in the shed according to the XOR result includes:

[0035] The ROI region after morphological processing is used as a mask to obtain the ROI region in the image to be detected and the standard image, and the structural similarity algorithm is used to determine the similarity difference and brightness difference in the ROI region between the image to be detected and the standard image, where the value range of the structural similarity algorithm is [0,1];

[0036] If the similarity difference within the ROI area is less than the threshold, it is determined that the shed has a penetrating injury, otherwise, the shed is normal.

[0037] Correspondingly, the present invention further provides a system for detecting the state of a shed of a power-centralized train, which is applied to a method for detecting the state of a shed of a power-centralized train as described in any one of the above, comprising:

[0038] An image acquisition module, used to acquire images of roof insulators of the train to be inspected;

[0039] A region segmentation module, used for performing insulator segmentation on the roof insulator image to obtain a first insulator foreground region;

[0040] A defect detection module, used to perform defect detection on the foreground area of ​​the first insulator using a deep learning positioning network, and determine the defect type of the train shed according to the defect detection result;

[0041] The alarm prompt module is used to issue an alarm prompt according to the defect type, so that maintenance personnel can review and rectify according to the content of the alarm prompt.

[0042] The main improvements of the present invention are:

[0043] 1. By using the segmentation model and the positioning model to detect shed defects in complex environments, the segmentation model extracts the insulator area with occlusion and complex background from the image to be detected, and the positioning network locates the defects in the insulator area, effectively improving the detection accuracy of shed damage and falling blocks, and avoiding false detection caused by complex backgrounds.

[0044] 2. The segmentation model and traditional difference detection algorithm are used to detect the penetration damage of the shed. Since the occlusion of the high-voltage flexible wire is similar to the occlusion of the penetration damage, the segmentation model is used to extract the insulator area in the image to avoid the occlusion of the high-voltage flexible wire, thereby improving the accuracy of subsequent difference detection.

[0045] 3. When judging the damage, falling pieces and penetration of the shed, the SSIM structural similarity algorithm is used to judge the similarity with the standard image, so as to further improve the accuracy of the shed status detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0047] Figure 1 It is a schematic diagram of the steps of a method for detecting the state of a shed of a power-centralized train provided by an embodiment of the present invention;

[0048] Figure 2It is a schematic diagram of insulator segmentation and obtaining of the insulator Mask area in the original image provided by an embodiment of the present invention;

[0049] Figure 3 This is a schematic diagram of shed defect location detection provided by an embodiment of the present invention;

[0050] Figure 4 The present invention is a schematic structural diagram of a system for detecting the state of a shed skirt of a power-centralized train provided in accordance with an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to enable those skilled in the art to better understand the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0052] Reference Figure 1 , is a schematic diagram of the steps of a method for detecting the state of a shed of a power-centralized train provided by an embodiment of the present invention.

[0053] S11. Collecting images of roof insulators of the train to be inspected.

[0054] S12, performing insulator segmentation on the roof insulator image to obtain a first insulator foreground area.

[0055] To obtain the insulator image, due to the influence of factors such as registration, it is necessary to expand the pixels around the insulator. Then the expanded insulator image is sent to the UNet segmentation network, and the insulator Mask in the segmentation result is used as a mask to obtain the insulator area in the original image to avoid subsequent positioning errors due to occlusion and cause misdetection.

[0056] In this embodiment, if Figure 2 As shown, firstly, the roof insulator image of the train to be detected is obtained, the insulator Mask is obtained by segmentation as the first mask, and the first insulator foreground area in the original image is further obtained.

[0057] S13. Use a deep learning positioning network to perform defect detection on the foreground area of ​​the first insulator, and judge the defect type of the train shed based on the defect detection results.

[0058] Further, such as Figure 3As shown, the trained YOLOv5 deep learning positioning network is used to locate defects in the foreground area of ​​the first insulator. If the damaged or chipped ROI area is located, a confidence judgment is performed. If the confidence result is greater than the threshold, it is determined that the shed is suspected of being damaged or chipped, otherwise the shed is normal; the image to be detected that is determined to be suspected of being damaged or chipped is compared with the pre-stored standard image for similarity difference, and according to the comparison result, the image to be detected that is suspected of being damaged or chipped is determined to be true.

[0059] Specifically, an XOR operation and a structural similarity algorithm (SSIM) are used to determine the similarity difference between the image to be detected and the ROI region in the standard image, wherein the value range of the structural similarity algorithm is [0,1]; the larger the value, the more similar the two images are. If the XOR area and size in the ROI region are greater than the threshold and the similarity difference is less than the threshold, then the image to be detected that is suspected to be damaged or missing is determined to be true, otherwise, the shed is normal.

[0060] Furthermore, if the damaged or broken ROI area is not located, a penetration damage detection of the shed is performed, and whether the shed is in a normal state is determined based on the result of the penetration damage detection.

[0061] Furthermore, when performing penetration damage detection on the shed, the pre-stored standard image is first segmented by insulators to obtain a second insulator foreground area; the first insulator foreground area and the second insulator foreground area are XORed to obtain an XOR result; and whether the shed has a penetration damage is determined based on the XOR result. That is, the insulator mask of the image to be detected and the insulator mask of the standard image are XORed, which is the difference detection algorithm in the image processing method.

[0062] It should be noted that after obtaining the XOR result, it is necessary to perform morphological processing on the XOR result to filter out small protruding parts and fill small holes in the image.

[0063] Furthermore, the standard image is subjected to the same operation as the image to be detected to obtain the insulator region. Specifically, the pre-stored standard image is pixel-expanded; the expanded standard image is sent to the UNet deep learning segmentation network; the insulator Mask in the segmentation result is used as the second mask to obtain the second insulator foreground region in the original image. The purpose is also to prevent the complex background from interfering with the detection result.

[0064] Furthermore, the ROI area after morphological processing is used as a mask to obtain the ROI area in the image to be detected and the standard image, and the structural similarity algorithm is used to judge the similarity difference and brightness difference in the ROI area between the image to be detected and the standard image. The value range of the structural similarity algorithm is [0,1]. The larger the value, the more similar the two images are. If the similarity difference in the ROI area is less than the threshold, it is determined that the shed has a penetrating wound, otherwise, the shed is normal.

[0065] S14. Issue an alarm based on the defect type so that maintenance personnel can review and rectify the problem based on the alarm content.

[0066] According to the detection situation, the shed status can be divided into three types: normal shed, broken / lost shed and shed penetration damage. Among them, broken / lost shed and shed penetration damage are defect types. When the defects of broken / lost shed or shed penetration damage are detected, alarm prompts including but not limited to sound and light alarms can be used to enable maintenance personnel to quickly determine what the defect type is, so as to conduct a review at the first time, and perform manual maintenance and rectification on the sheds that are verified to be correct. Improve the efficiency and safety of shed defect handling.

[0067] An embodiment of the present invention discloses a method for detecting the state of the shed of a power-centralized train. The method includes collecting the roof insulator image of the train to be detected, performing insulator segmentation on the roof insulator image, and obtaining a first insulator foreground area; using a deep learning positioning network to perform defect detection on the first insulator foreground area, and judging the defect type of the train shed according to the defect detection result; and giving an alarm prompt according to the defect type, so that maintenance personnel can review and rectify according to the content of the alarm prompt. By using a segmentation model and a positioning model to detect shed defects in a complex environment, the segmentation model extracts the insulator area with occlusion and complex background from the image to be detected, and the positioning network locates the defects in the insulator area, effectively improving the detection accuracy of shed damage and falling pieces, and avoiding false detection caused by complex backgrounds.

[0068] Correspondingly, such as Figure 4 As shown, an embodiment of the present invention further provides a system for detecting the state of a shed skirt of a power-centralized train, comprising:

[0069] An image acquisition module, used to acquire images of roof insulators of the train to be inspected;

[0070] A region segmentation module, used for performing insulator segmentation on the roof insulator image to obtain a first insulator foreground region;

[0071] A defect detection module, used to perform defect detection on the foreground area of ​​the first insulator using a deep learning positioning network, and determine the defect type of the train shed according to the defect detection result;

[0072] The alarm prompt module is used to issue an alarm prompt according to the defect type, so that maintenance personnel can review and rectify according to the content of the alarm prompt.

[0073] The above is a method and system for detecting the state of a power-centralized train shed skirt provided in an embodiment of the present invention. The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the various embodiments can be referenced to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description. It should be pointed out that for ordinary technicians in this technical field, without departing from the principle of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

[0074] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented with electronic hardware, computer software or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention. The steps of the method or algorithm described in conjunction with the embodiments disclosed herein can be directly implemented with a software module executed by a hardware or processor, or a combination of the two. The software module can be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the technical field.

Claims

1. A method for detecting the state of a shed skirt of a power-centralized train, characterized in that: include: Collect the roof insulator image of the train to be inspected; Performing insulator segmentation on the roof insulator image to obtain a first insulator foreground area; Using a deep learning positioning network to perform defect detection on the foreground area of ​​the first insulator, and judging the defect type of the train shed according to the defect detection result; An alarm is issued according to the defect type, so that maintenance personnel can review and rectify according to the content of the alarm.

2. A method for detecting the state of a shed skirt of a power-centralized train according to claim 1, characterized in that: The step of performing insulator segmentation on the roof insulator image to obtain a first insulator foreground region includes: Performing pixel expansion on the roof insulator image; The expanded insulator image is fed into the UNet deep learning segmentation network; Use the insulator Mask in the segmentation result as the first mask to obtain the first insulator foreground area in the original image.

3. A method for detecting the state of a shed skirt of a power-centralized train according to claim 1, characterized in that: The method of using the deep learning positioning network to perform defect detection on the foreground area of ​​the first insulator and judging the defect type of the train shed according to the defect detection result includes: The trained YOLOv5 deep learning positioning network is used to locate defects in the foreground area of ​​the first insulator. If a damaged or missing ROI area is located, a confidence determination is performed. If the confidence result is greater than the threshold, it is determined that the shed is suspected to be damaged or broken, otherwise the shed is normal; The image to be detected that is determined to be suspected of being damaged or missing is compared with a pre-stored standard image for similarity difference, and based on the comparison result, it is determined that the image to be detected that is suspected of being damaged or missing is true.

4. A method for detecting the state of a shed skirt of a power-centralized train according to claim 3, characterized in that: The comparing the image to be detected that is determined to be suspected of being damaged or missing with a pre-stored standard image for similarity difference, and determining that the image to be detected that is suspected of being damaged or missing is true according to the comparison result, includes: The XOR operation and structural similarity algorithm are used to determine the similarity difference between the image to be detected and the ROI area in the standard image, where the value range of the structural similarity algorithm is [0,1]; If the XOR area and size in the ROI region are greater than the threshold and the similarity difference is less than the threshold, it is determined that the image to be detected that is suspected to be damaged or chipped is true, otherwise, the shed is normal.

5. A method for detecting the state of a shed skirt of a power-centralized train according to claim 3, characterized in that: Also includes: The trained YOLOv5 deep learning positioning network is used to locate defects in the foreground area of ​​the first insulator. If the damaged or fallen ROI area is not located, the penetration damage detection of the shed is performed; According to the result of the penetration wound detection, it is determined whether the state of the shed is normal.

6. A method for detecting the state of shed skirts of a power-centralized train according to claim 5, characterized in that: The penetrating injury detection of the shed comprises: Performing insulator segmentation on the pre-stored standard image to obtain a second insulator foreground area; Performing an XOR operation on the first insulator foreground area and the second insulator foreground area to obtain an XOR result; Whether the shed has a penetrating wound is determined based on the XOR result.

7. A method for detecting the state of a shed skirt of a power-centralized train according to claim 6, characterized in that: The step of performing insulator segmentation on the pre-stored standard image to obtain a second insulator foreground area includes: Performing pixel expansion on the pre-stored standard image; Send the expanded standard image to the UNet deep learning segmentation network; Use the insulator Mask in the segmentation result as the second mask to obtain the second insulator foreground area in the original image.

8. A method for detecting the state of a shed skirt of a power-centralized train according to claim 6, characterized in that: Before judging whether the shed has a penetrating injury according to the XOR result, the method further includes: Morphological processing is performed on the XOR result to filter out small convex parts and fill small holes in the image.

9. A method for detecting the state of shed skirts of a power-centralized train according to claim 8, characterized in that: The step of judging whether a penetrating injury occurs to the shed according to the XOR result includes: The ROI region after morphological processing is used as a mask to obtain the ROI region in the image to be detected and the standard image, and the structural similarity algorithm is used to determine the similarity difference and brightness difference in the ROI region between the image to be detected and the standard image, where the value range of the structural similarity algorithm is [0,1]; If the similarity difference within the ROI area is less than the threshold, it is determined that the shed has a penetrating injury, otherwise, the shed is normal.

10. A system for detecting the state of a shed of a power-concentrated train, applied to a method for detecting the state of a shed of a power-concentrated train as claimed in any one of claims 1 to 9, characterized in that: include: An image acquisition module, used to acquire images of roof insulators of the train to be inspected; A region segmentation module, used for performing insulator segmentation on the roof insulator image to obtain a first insulator foreground region; A defect detection module, used to perform defect detection on the foreground area of ​​the first insulator using a deep learning positioning network, and determine the defect type of the train shed according to the defect detection result; The alarm prompt module is used to issue an alarm prompt according to the defect type, so that maintenance personnel can review and rectify according to the content of the alarm prompt.

Images