An intelligent detection and identification method for remote monitoring of contact network insulator defects

Abstract

The present invention relates to the technical field of insulator defect detection, and in particular to an intelligent detection and identification method for remotely monitoring contact network insulator defects, comprising: obtaining a one-stage cascade model for intercepting and correcting insulator image information, and a two-stage cascade model for selecting and calibrating insulator defect areas, periodically determining whether the operation of the cascade model is qualified based on the false detection rate of the two-stage cascade model, and adjusting the operating parameters of the cascade model when it is determined that the operation of the cascade model is abnormal. During the use of the cascade model, the operation of the cascade model is monitored and the model is calibrated in time, thereby improving the detection accuracy of defective insulators and improving the detection efficiency of defective insulators.

Description

Technical Field

[0001] The present invention relates to the technical field of insulator defect detection, and in particular to an intelligent detection and identification method for remotely monitoring contact network insulator defects. Background Art

[0002] Catenary arm insulators play a crucial role in providing electrical insulation and power support for electrified railways. Their insulation performance is crucial to the safe and stable operation of railway power supply systems. Catenary insulators operate in complex environments, and their surfaces are inevitably subject to external corrosion, resulting in varying degrees of defects. Typical examples include bird peck damage, shed brittle fracture, foreign matter adhesion, and cracks. Extended defects can cause insulation failure and trigger short-circuit faults, leading to widespread train outages. Therefore, long-term monitoring of the arm insulator condition is essential to ensure the safe operation of the catenary system.

[0003] Chinese Patent Publication No. CN118839245A discloses an insulator defect detection method based on an improved YOLOv8. The method includes designing a new C2f module based on ScConv and introducing GhostConv. In addition, the Adown module is also introduced. It can be seen that the above technical solution has the following problems: it is impossible to monitor the operation of the model during use and it is impossible to calibrate the model in a timely manner, which affects the detection accuracy of defective insulators. Summary of the Invention

[0004] To this end, the present invention provides an intelligent detection and identification method for remote monitoring of contact network insulator defects, which is used to overcome the problems in the prior art that the operation of the model cannot be monitored during use of the model, the model cannot be calibrated in time, and the detection accuracy of defective insulators is affected.

[0005] To achieve the above objectives, the present invention provides an intelligent detection and identification method for remotely monitoring contact network insulator defects, comprising:

[0006] S1, using the insulator area image information obtained by the insulator detector as the experimental dataset, training a one-stage cascade model to output the intercepted and corrected insulator image information;

[0007] S2, using the insulator image information as a training set to train a two-stage cascade model for selecting and calibrating insulator defect areas;

[0008] S3, inputting the insulator image information obtained by each insulator detector into the first-stage cascade model and the second-stage cascade model in sequence to obtain the insulator region image information after frame selection and calibration;

[0009] S4, periodically determining whether the cascade model is qualified based on the false detection rate of the two-stage cascade model with a preset monitoring period as a period, including:

[0010] When determining that the cascade model is operating abnormally, adjusting the operating parameters of the cascade model based on the identified difference, including using the insulator area image information obtained by the insulator detector as an experimental data set to calibrate the cascade model, or adjusting the layout density of the insulator detector to a corresponding value;

[0011] Alternatively, when it is determined that the cascade model is qualified, the two-stage cascade model is continuously used to perform frame selection and calibration on the image information of each insulator region.

[0012] Furthermore, the process of periodically determining whether the operation of the cascade model is qualified based on the false detection rate of the two-stage cascade model includes:

[0013] Comparing the obtained false positive rate with the first preset false positive rate and the second preset false positive rate respectively;

[0014] If the false detection rate is less than or equal to the first preset false detection rate, the cascade model is determined to be qualified, and the two-stage cascade model is continuously used to perform frame selection and calibration on the image information of each insulator area;

[0015] If the false detection rate is less than or equal to the second preset false detection rate and greater than the first preset false detection rate, the false detection rate of the current cycle is marked as an abnormal false detection rate, and whether the operation of the cascade model is qualified is determined in combination with the obtained historical false detection rate;

[0016] If the false positive rate is greater than the second preset false positive rate, determining that the operation of the cascade model is abnormal, and adjusting the operation parameters of the cascade model based on the recognition difference;

[0017] The ratio of the number of insulator image information obtained by the two-stage cascade model within the preset monitoring time to the preset number of insulators is determined as the recognition difference amount. Furthermore, the process of determining whether the operation of the cascade model is qualified in combination with the obtained historical false detection rate includes:

[0018] Calculate the ratio of the number of abnormal false positive rates in the historical false positive rate to the total number of historical false positive rates to obtain the abnormal frequency;

[0019] Calculate the ratio of the abnormal false detection rate of the current cycle to the average value of each historical false detection rate to obtain the abnormal deviation;

[0020] If the abnormal frequency is less than or equal to the preset abnormal frequency and the abnormal deviation is greater than the preset deviation, the cascade model is judged to be qualified, and the two-stage cascade model is continuously used to perform frame selection and calibration on the image information of each insulator area, and the preset monitoring time is adjusted to the corresponding value based on the abnormal deviation;

[0021] If the abnormal frequency is greater than a preset abnormal frequency and / or the abnormal deviation is less than or equal to a preset deviation, the operation of the cascade model is determined to be abnormal, and the operation parameters of the cascade model are adjusted based on the identified difference.

[0022] Furthermore, the process of adjusting the operating parameters of the cascade model based on the identified difference includes:

[0023] If the recognition difference is less than or equal to the preset recognition difference, the time nodes at which the insulator detector obtains the image information of each insulator region with calibration errors are obtained, and the concentrated time interval of each time node is determined;

[0024] The image information of a preset number of insulator regions obtained by the insulator detector within a concentrated time interval is used as an experimental data set to calibrate the cascade model;

[0025] The preset correction amount is adjusted to a corresponding value based on the identified difference amount.

[0026] Furthermore, the process of adjusting the operating parameters of the cascade model based on the identified difference amount further includes:

[0027] If the recognition difference is greater than the preset recognition difference, the cascade model is adjusted based on the recognition accuracy evaluation value determined in combination with the image information of each insulator region;

[0028] The average value of the area of ​​the selected area of ​​the image information of each insulator region is calculated to obtain the recognition accuracy evaluation value;

[0029] If the recognition accuracy evaluation value is less than or equal to the preset recognition accuracy evaluation value, the layout density of the insulator detectors is adjusted to a corresponding value based on the average value of the distances between the insulator detectors corresponding to the image information of the insulator regions with calibration errors;

[0030] If the recognition accuracy evaluation value is greater than the preset recognition accuracy evaluation value, the insulator area image information obtained by a preset number of correction insulator detectors is selected as the experimental data set to calibrate the cascade model, and the preset correction number is adjusted to the corresponding value based on the recognition accuracy evaluation value.

[0031] Furthermore, the specific process of obtaining the false positive rate includes:

[0032] Input the preset verification number of insulator image information obtained by the insulator detector into the cascade model to obtain the image information of the insulator area after frame selection and calibration;

[0033] The number of image information of the insulator area with calibration errors is determined, and the ratio of the number of calibration errors to the preset number of checks is calculated to obtain the false detection rate.

[0034] Furthermore, the preset monitoring duration is adjusted to a corresponding value based on the abnormal deviation amount, wherein:

[0035] The reduction in the preset monitoring time is proportional to the amount of abnormal deviation.

[0036] Further, the preset correction amount is adjusted to a corresponding value based on the identified difference amount, wherein,

[0037] The amount of preset corrections increases inversely proportional to the amount of the identified discrepancy.

[0038] Furthermore, the layout density of the insulator detectors is adjusted to a corresponding value based on the average value of the distances between the insulator detectors corresponding to the image information of the insulator regions with calibration errors, wherein:

[0039] The increase in deployment density is inversely proportional to the average distance.

[0040] Furthermore, the preset correction quantity is adjusted to a corresponding value based on the recognition accuracy evaluation value, wherein,

[0041] The increase in the preset correction number is proportional to the recognition accuracy evaluation value.

[0042] Compared with the prior art, the beneficial effect of the present invention lies in obtaining a one-stage cascade model for intercepting and correcting insulator image information, and a two-stage cascade model for selecting and calibrating insulator defect areas, periodically determining whether the operation of the cascade model is qualified based on the false detection rate of the two-stage cascade model, and adjusting the operating parameters of the cascade model when it is determined that the operation of the cascade model is abnormal, monitoring the operation of the cascade model during the use of the cascade model, and calibrating the model in time, thereby improving the detection accuracy of defective insulators and improving the detection efficiency of defective insulators.

[0043] Furthermore, the cascade model's false positive rate is used to determine whether it is operating properly. The false positive rate represents the accuracy of the cascade model's detection. When the false positive rate is less than or equal to a first preset false positive rate, the cascade model's detection accuracy is high and it can accurately select and calibrate the image information of the insulator region. In this case, the cascade model is deemed to be operating properly. When the false positive rate is greater than a second preset false positive rate, the false positive rate is relatively high and the cascade model's false positives are severe. The operating parameters of the cascade model are further adjusted based on the amount of identified differences. Promptly identifying any operational anomalies in the model ensures the stability and reliability of the cascade model and further improves the accuracy of insulator defect detection.

[0044] Furthermore, when the false detection rate is less than or equal to the second preset false detection rate and greater than the first preset false detection rate, the operation of the model is judged in combination with the historical false detection rate. The abnormal frequency characterizes the frequency of the abnormal false detection rate in the historical false detection rate. When the abnormal frequency is less than or equal to the preset abnormal frequency, the occurrence of the abnormal false detection rate is accidental. The abnormal deviation characterizes the degree of deviation of the abnormal false detection rate of the current cycle from the average value of the historical false detection rate. When the abnormal deviation is greater than the preset deviation, the current abnormal false detection rate deviates significantly from the historical average level; when the abnormal frequency is low and the cascade model suddenly deviates significantly from the historical average level, the abnormal false detection rate is high due to extreme environment or emergency situation, and the model can still operate normally. In this case, the preset monitoring time is adjusted for further observation. Comprehensively evaluate the stability of the model, avoid misjudgment due to accidental factors, improve the accuracy of the model evaluation, and further improve the detection accuracy of defective insulators.

[0045] Furthermore, the operating parameters of the cascade model are adjusted based on the recognition difference. The recognition difference reflects the recognition performance of the cascade model in a single stage. When the recognition difference is less than or equal to a preset recognition difference, the number of insulators identified by the single stage cascade model differs significantly from the actual number of insulators to be detected in the area. In this case, the model has problems during certain time periods and cannot accurately identify insulators due to external environmental influences. In this case, a concentrated time interval is determined and image information within this interval is obtained to calibrate the model. This is a targeted calibration. When the recognition difference is greater than the preset recognition difference, the recognition difference is large. In this case, the model is further adjusted based on the recognition accuracy evaluation value, thereby improving the adaptability and accuracy of the model. The recognition accuracy evaluation value represents the general situation of the area selected by the model. When the recognition accuracy evaluation value is less than or equal to the preset recognition accuracy evaluation value, the area selected by the model is generally small. Although the model can still accurately identify defects and anomalies in small areas, the image information obtained is incomplete due to insufficient insulator detector placement density. In this case, the insulator detector placement density is adjusted. When the recognition accuracy evaluation value exceeds the preset recognition accuracy evaluation value, the model selects a larger area. In this case, due to the abnormal model recognition accuracy, small area anomalies are not accurately identified, resulting in false detection. In this case, the insulator area image information is used as the experimental data set to calibrate the cascade model. This optimizes the number of insulator detectors deployed, improves the accuracy of insulator defect detection, and further enhances the efficiency of defective insulator detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a flowchart of the steps of an intelligent detection and identification method for remote monitoring of contact network insulator defects according to an embodiment of the present invention;

[0047] Figure 2A logic decision diagram for determining whether the operation of the cascade model is qualified according to the false detection rate of the cascade model according to an embodiment of the present invention;

[0048] Figure 3 A logic decision diagram for determining whether the operation of the cascade model is qualified in combination with the historical false positive rate according to an embodiment of the present invention;

[0049] Figure 4 This is a logic decision diagram for adjusting the operating parameters of the cascade model based on the identification difference amount according to an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are merely used to explain the present invention and are not intended to limit the present invention.

[0051] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0052] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside", and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0053] Furthermore, it should be noted that, in the description of the present invention, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0054] See also Figure 1 、 Figure 2 、 Figure 3 as well as Figure 4 As shown, they are respectively a flowchart of the steps of an intelligent detection and identification method for remote monitoring of contact network insulator defects according to an embodiment of the present invention, a logic decision diagram for determining whether the operation of the cascade model is qualified based on the false detection rate of the cascade model, a logic decision diagram for determining whether the operation of the cascade model is qualified based on the historical false detection rate, and a logic decision diagram for adjusting the operating parameters of the cascade model based on the recognition difference amount; an intelligent detection and identification method for remote monitoring of contact network insulator defects according to an embodiment of the present invention comprises:

[0055] S1, using the insulator area image information obtained by the insulator detector as the experimental dataset, training a one-stage cascade model to output the intercepted and corrected insulator image information;

[0056] S2, using the insulator image information as a training set to train a two-stage cascade model for selecting and calibrating insulator defect areas;

[0057] S3, inputting the insulator image information obtained by each insulator detector into the first-stage cascade model and the second-stage cascade model in sequence to obtain the insulator region image information after frame selection and calibration;

[0058] S4, periodically determining whether the cascade model is qualified based on the false detection rate of the two-stage cascade model with a preset monitoring period as a period, including:

[0059] When determining that the cascade model is operating abnormally, adjusting the operating parameters of the cascade model based on the identified difference, including using the insulator area image information obtained by the insulator detector as an experimental data set to calibrate the cascade model, or adjusting the layout density of the insulator detector to a corresponding value;

[0060] Alternatively, when it is determined that the cascade model is qualified, the two-stage cascade model is continuously used to perform frame selection and calibration on the image information of each insulator region.

[0061] Specifically, the insulator area image information obtained by the insulator detector is input into the first-stage cascade model to obtain the insulator image information after cropping and correction; the insulator image information is input into the two-stage cascade model, the two-stage cascade model identifies the defects in the insulator image information, and calibrates and selects the defective area, and maps the coordinates of the defective area in the insulator image information to the insulator area image information, and obtains the insulator area image information after the defective area is calibrated and selected, and obtains the detection result of the cascade model.

[0062] Specifically, in S1, the insulator area image information used as the experimental data set is the image information actually obtained by the insulator detector in the railway department. Labelme software is used to label all the insulator area image information obtained by the insulator detector, and labels of defect type and defect location are added to the insulators in each image information. The labeled image information is randomly divided into a training set and a validation set in a ratio of 7:3 to train a one-stage cascade model for identifying, locating, cropping, and correcting target insulators in a global image.

[0063] Specifically, there is no limitation on the specific structure of the insulator detector, which may be a contact network suspension state detection and monitoring device. This is a prior art and will not be described in detail.

[0064] Specifically, the one-stage cascade model uses the YOLOv8 target detection algorithm to identify insulators and locate their positions in the insulator area image information, intercepts the insulators in the insulator area image information, and uses the Canny algorithm and Hough transform algorithm to correct the intercepted insulators. This is an existing technology and will not be described in detail.

[0065] Specifically, in S2, the insulator image information output by the first-stage cascade model is used as a training set to train the YOLO v8 instance segmentation algorithm to obtain a two-stage cascade model; the two-stage cascade model identifies, selects, and calibrates five types of surface defects of the insulator, and maps the defect coordinates to the insulator region image information using a homography matrix as the final output of the insulator surface defect detection result. The five types of surface defects include bird pecking damage, shed brittle fracture, cracks, foreign matter attachment, and bird droppings.

[0066] Specifically, a one-stage cascade model for intercepting and correcting insulator image information and a two-stage cascade model for selecting and calibrating insulator defect areas are obtained. Whether the operation of the cascade model is qualified is periodically determined based on the false detection rate of the two-stage cascade model. When the operation of the cascade model is determined to be abnormal, the operating parameters of the cascade model are adjusted. During the use of the cascade model, the operation of the cascade model is monitored and the model is calibrated in time, thereby improving the detection accuracy of defective insulators and improving the detection efficiency of defective insulators.

[0067] Specifically, the process of periodically determining whether the operation of the cascade model is qualified based on the false detection rate of the two-stage cascade model includes:

[0068] Comparing the obtained false positive rate with the first preset false positive rate and the second preset false positive rate respectively;

[0069] If the false detection rate is less than or equal to the first preset false detection rate, the cascade model is determined to be qualified, and the two-stage cascade model is continuously used to perform frame selection and calibration on the image information of each insulator area;

[0070] If the false detection rate is less than or equal to the second preset false detection rate and greater than the first preset false detection rate, the false detection rate of the current cycle is marked as an abnormal false detection rate, and whether the operation of the cascade model is qualified is determined in combination with the obtained historical false detection rate;

[0071] If the false positive rate is greater than the second preset false positive rate, determining that the operation of the cascade model is abnormal, and adjusting the operation parameters of the cascade model based on the recognition difference;

[0072] The ratio of the amount of insulator image information obtained by the two-stage cascade model within the calculated preset monitoring time to the preset number of insulators is determined as the recognition difference amount.

[0073] Specifically, the preset number of insulators is the number of insulators actually set in the preset monitoring area.

[0074] Specifically, it can be understood that in the process of determining whether the operation of the cascade model is qualified based on the false detection rate of the two-stage cascade model, the false detection rate represents the false detection situation of the output result of the two-stage cascade model.

[0075] It should be noted that the data in this embodiment are obtained by comprehensive analysis and evaluation of the historical test data and corresponding historical test results of the three months before the current test by the method described in the present invention. It can be understood by those skilled in the art that the method described in the present invention can be used to determine the single parameter by selecting the value with the highest proportion as the preset standard parameter based on the data distribution, using weighted summation to use the obtained value as the preset standard parameter, or other selection methods, as long as the system described in the present invention can clearly define the different specific situations in the single determination process through the obtained values.

[0076] Specifically, the first preset false positive rate is selected within the interval [0.04, 0.07], and the second preset false positive rate is selected within the interval [0.1, 0.13].

[0077] Specifically, the specific process of obtaining the false positive rate includes:

[0078] Inputting a preset number of verification insulator image information obtained by the insulator detector into the first-stage cascade model and the second-stage cascade model in sequence to obtain the insulator area image information after frame selection and calibration;

[0079] The number of image information of the insulator area with calibration errors is determined, and the ratio of the number of calibration errors to the preset number of checks is calculated to obtain the false detection rate.

[0080] Specifically, Labelme software can be used to annotate a preset number of calibration insulator area image information obtained by the insulator detector, and the insulator area image information annotated by the Labelme software is overlapped and compared with the insulator area image information after the frame selection calibration output by the cascade model, and the ratio of the area of ​​the intersection of the two framed areas to the area of ​​the union of the two framed areas is used as the comparison overlap; the insulator area image information with a comparison overlap lower than the preset overlap is determined as the insulator area image information with frame selection calibration errors.

[0081] Specifically, the cascade model's false detection rate is used to determine whether it is operating properly. The false detection rate represents the accuracy of the cascade model's detection. When the false detection rate is less than or equal to the first preset false detection rate, the cascade model's detection accuracy is high and it can more accurately select and calibrate the image information of the insulator area. In this case, the cascade model is considered to be operating properly. When the false detection rate is greater than the second preset false detection rate, the false detection rate is relatively high and the cascade model's false detection situation is more serious. The operating parameters of the cascade model are further adjusted based on the amount of recognition difference. Timely detection of model operation anomalies ensures the stability and reliability of the cascade model and further improves the accuracy of insulator defect detection.

[0082] Specifically, the process of determining whether the cascade model is qualified based on the historical false positive rate includes:

[0083] Calculate the ratio of the number of abnormal false positive rates in the historical false positive rate to the total number of historical false positive rates to obtain the abnormal frequency;

[0084] Calculate the ratio of the abnormal false detection rate of the current cycle to the average value of each historical false detection rate to obtain the abnormal deviation;

[0085] If the abnormal frequency is less than or equal to the preset abnormal frequency and the abnormal deviation is greater than the preset deviation, the cascade model is judged to be qualified, and the two-stage cascade model is continuously used to perform frame selection and calibration on the image information of each insulator area, and the preset monitoring time is adjusted to the corresponding value based on the abnormal deviation;

[0086] If the abnormal frequency is greater than a preset abnormal frequency and / or the abnormal deviation is less than or equal to a preset deviation, the operation of the cascade model is determined to be abnormal, and the operation parameters of the cascade model are adjusted based on the identified difference.

[0087] Specifically, the preset abnormal frequency is selected within the interval [0.15, 0.21], and the preset deviation amount P0 is selected within the interval [1.18, 1.23].

[0088] Specifically, there is no limit on the number of historical false detection rates selected for determining abnormal frequencies and abnormal deviations. It can be understood that in order to obtain the occurrence pattern of abnormal false detection rates and to determine whether accidental abnormal false detection rates occur due to extreme environments or emergencies, the number of false detection rates selected in historical data should be no less than 30. This will not be repeated here.

[0089] Specifically, when the false detection rate is less than or equal to the second preset false detection rate and greater than the first preset false detection rate, the operation of the model is judged in combination with the historical false detection rate. The abnormal frequency represents the frequency of the abnormal false detection rate in the historical false detection rate. When the abnormal frequency is less than or equal to the preset abnormal frequency, the occurrence of the abnormal false detection rate is accidental. The abnormal deviation represents the degree of deviation of the abnormal false detection rate of the current cycle from the average value of the historical false detection rate. When the abnormal deviation is greater than the preset deviation, the current abnormal false detection rate deviates significantly from the historical average level; when the abnormal frequency is low and the cascade model suddenly deviates significantly from the historical average level, the abnormal false detection rate is high due to extreme environment or emergency situation. The model can still operate normally. In this case, the preset monitoring time is adjusted for further observation. Comprehensively evaluate the stability of the model, avoid misjudgment due to accidental factors, improve the accuracy of the model evaluation, and further improve the detection accuracy of defective insulators.

[0090] Specifically, the process of adjusting the operating parameters of the cascade model based on the identified difference includes:

[0091] If the recognition difference is less than or equal to the preset recognition difference, the time nodes at which the insulator detector obtains the image information of each insulator region with calibration errors are obtained, and the concentrated time interval of each time node is determined;

[0092] The image information of a preset number of insulator regions obtained by the insulator detector within a concentrated time interval is used as an experimental data set to calibrate the cascade model;

[0093] The preset correction amount is adjusted to a corresponding value based on the identified difference amount.

[0094] Specifically, there is no limitation on the specific method for determining the concentrated time interval. For example, the time nodes can be sorted chronologically, and a sliding window can be used. A sliding window of one hour can be used on the time axis to count the number of time nodes within the window. The time period corresponding to windows with a number exceeding a preset threshold can be determined as the concentrated time interval. Alternatively, the time nodes can be sorted chronologically, and the number of time nodes within one hour of each time node can be counted. The time period with a larger number of time nodes can be determined as the concentrated time interval.

[0095] Specifically, the preset recognition difference amount B0 is selected within the interval [0.94, 0.97].

[0096] Specifically, the process of adjusting the operating parameters of the cascade model based on the identified difference amount further includes:

[0097] If the recognition difference is greater than the preset recognition difference, the cascade model is adjusted based on the recognition accuracy evaluation value determined in combination with the image information of each insulator region;

[0098] The average value of the area of ​​the selected area of ​​the image information of each insulator region is calculated to obtain the recognition accuracy evaluation value;

[0099] If the recognition accuracy evaluation value is less than or equal to the preset recognition accuracy evaluation value, the layout density of the insulator detectors is adjusted to a corresponding value based on the average value of the distances between the insulator detectors corresponding to the image information of the insulator regions with calibration errors;

[0100] If the recognition accuracy evaluation value is greater than the preset recognition accuracy evaluation value, the insulator area image information obtained by a preset number of correction insulator detectors is selected as the experimental data set to calibrate the cascade model, and the preset correction number is adjusted to the corresponding value based on the recognition accuracy evaluation value.

[0101] Specifically, the preset recognition accuracy evaluation value J0 is selected within the interval [420, 480], and the unit of the preset recognition accuracy evaluation value is pixel.

[0102] Specifically, when the recognition accuracy evaluation value is greater than the preset recognition accuracy evaluation value, the insulator detector acquisition time of the edge region image information selected for calibrating the cascade model is not limited and can be acquired at a random time node, which will not be repeated here.

[0103] Specifically, the layout density of the insulator detectors is the ratio of the number of layout insulator detectors to the area of ​​the area to be detected.

[0104] Specifically, the operating parameters of the cascade model are adjusted based on the recognition difference. The recognition difference reflects the recognition performance of the cascade model in a single stage. When the recognition difference is less than or equal to the preset recognition difference, the number of insulators identified by the single stage cascade model differs significantly from the actual number of insulators to be detected in the area. In this case, the model has problems during certain time periods and cannot accurately identify insulators due to external environmental influences. In this case, a concentrated time interval is determined and image information within this interval is obtained to calibrate the model. When the recognition difference is greater than the preset recognition difference, the recognition difference is large. In this case, the model is further adjusted based on the recognition accuracy evaluation value, improving the adaptability and accuracy of the model. The recognition accuracy evaluation value represents the general situation of the area selected by the model. When the recognition accuracy evaluation value is less than or equal to the preset recognition accuracy evaluation value, the area selected by the model is generally small. Although the model can still accurately identify defects and anomalies in small areas, the image information obtained is incomplete due to insufficient insulator detector placement density. In this case, the insulator detector placement density is adjusted. When the recognition accuracy evaluation value exceeds the preset recognition accuracy evaluation value, the model selects a larger area. In this case, due to the abnormal model recognition accuracy, small area anomalies are not accurately identified, resulting in false detection. In this case, the insulator area image information is used as the experimental data set to calibrate the cascade model. This optimizes the number of insulator detectors deployed, improves the accuracy of insulator defect detection, and further enhances the efficiency of defective insulator detection.

[0105] Specifically, the specific process of obtaining the false positive rate includes:

[0106] Inputting a preset number of verification insulator image information obtained by the insulator detector into the two-stage cascade model to obtain the image information of the insulator area after frame selection and calibration;

[0107] The number of image information of the insulator area with calibration errors is determined, and the ratio of the number of calibration errors to the preset number of checks is calculated to obtain the false detection rate.

[0108] Specifically, the preset monitoring duration is adjusted to a corresponding value based on the abnormal deviation amount, where:

[0109] The reduction in the preset monitoring time is proportional to the amount of abnormal deviation.

[0110] In this embodiment, optionally,

[0111] Comparing the abnormal deviation with a first preset deviation comparison threshold and a second preset deviation comparison threshold;

[0112] If the abnormal deviation is less than or equal to the first preset deviation comparison threshold, the preset monitoring time is adjusted to 0.92 times the initial preset monitoring time;

[0113] If the abnormal deviation is less than or equal to the second preset deviation comparison threshold and greater than the first preset deviation comparison threshold, the preset monitoring time is adjusted to 0.82 times the initial preset monitoring time;

[0114] If the abnormal deviation is greater than the second preset deviation comparison threshold, the preset monitoring time is adjusted to 0.72 times the initial preset monitoring time;

[0115] The first preset deviation comparison threshold is 1.2P0, and the second preset deviation comparison threshold is 1.35P0.

[0116] Specifically, the preset correction amount is adjusted to a corresponding value based on the identified difference amount, wherein,

[0117] The amount of preset corrections increases inversely proportional to the amount of the identified discrepancy.

[0118] In this embodiment, optionally,

[0119] Comparing the recognition difference amount with a first preset recognition comparison threshold and a second preset recognition comparison threshold;

[0120] If the recognition difference is less than or equal to the first preset recognition comparison threshold, the preset correction amount is adjusted to 1.23 times the initial preset correction amount;

[0121] If the recognition difference is less than or equal to the second preset recognition comparison threshold and greater than the first preset recognition comparison threshold, the preset correction amount is adjusted to 1.17 times the initial preset correction amount;

[0122] If the recognition difference is greater than the second preset recognition comparison threshold, the preset correction amount is adjusted to 1.11 times the initial preset correction amount;

[0123] The first preset recognition and comparison threshold is 0.72B0, and the second preset recognition and comparison threshold is 0.85B0.

[0124] Specifically, the layout density of the insulator detectors is adjusted to a corresponding value based on the average value of the distances between the insulator detectors corresponding to the image information of the insulator regions with calibration errors, wherein:

[0125] The increase in deployment density is inversely proportional to the average distance.

[0126] In this embodiment, optionally,

[0127] The average value of the distances between the insulator detectors corresponding to the image information of the insulator regions with calibration errors is recorded as the average distance value;

[0128] Comparing the distance average with a first preset distance average and a second preset distance average;

[0129] If the average distance value is less than or equal to the first preset distance value, adjusting the layout density of the insulator detectors to 1.23 times the initial layout density;

[0130] If the distance average value is less than or equal to the second preset distance average value and greater than the first preset distance average value, adjusting the layout density of the insulator detectors to 1.18 times the initial layout density;

[0131] If the average distance value is greater than the second preset average distance value, adjusting the layout density of the insulator detectors to 1.12 times the initial layout density;

[0132] The first preset distance average is 10m, and the second preset distance average is 20m.

[0133] Specifically, the preset correction amount is adjusted to a corresponding value based on the recognition accuracy evaluation value, wherein,

[0134] The increase in the preset correction number is proportional to the recognition accuracy evaluation value.

[0135] In this embodiment, optionally,

[0136] Comparing the recognition accuracy evaluation value with a first preset evaluation value comparison threshold and a second preset evaluation value comparison threshold;

[0137] If the recognition accuracy evaluation value is less than or equal to the first preset evaluation value comparison threshold, the preset correction quantity is adjusted to 1.2 times the initial preset correction quantity;

[0138] If the recognition accuracy evaluation value is less than or equal to the second preset evaluation value comparison threshold and greater than the first preset evaluation value comparison threshold, the preset correction amount is adjusted to 1.3 times the initial preset correction amount;

[0139] If the recognition accuracy evaluation value is greater than the second preset evaluation value comparison threshold, the preset correction quantity is adjusted to 1.35 times the initial preset correction quantity;

[0140] The first preset evaluation value comparison threshold is 1.2J0, and the second preset evaluation value comparison threshold is 1.4J0.

[0141] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

[0142] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that the present invention is susceptible to various modifications and variations. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. An intelligent detection and identification method for remote monitoring of contact network insulator defects, characterized in that: include: S1, using the insulator area image information obtained by the insulator detector as the experimental dataset, training a one-stage cascade model to output the intercepted and corrected insulator image information; S2, using the insulator image information as a training set to train a two-stage cascade model for selecting and calibrating insulator defect areas; S3, inputting the insulator region image information obtained by each insulator detector into the first-stage cascade model and the second-stage cascade model in sequence to obtain the insulator region image information after frame selection calibration; S4, periodically determining whether the cascade model is qualified based on the false detection rate of the two-stage cascade model with a preset monitoring period as a period, including: Inputting a preset number of verification image information of the insulator area obtained by the insulator detector into the cascade model to obtain the image information of the insulator area after frame selection and calibration; Determine the number of image information of the insulator area with calibration errors, calculate the ratio of the number of calibration errors to the preset number of checks, and obtain the false detection rate; When determining that the cascade model is operating abnormally, adjusting the operating parameters of the cascade model based on the identified difference, including using the insulator area image information obtained by the insulator detector as an experimental data set to calibrate the cascade model, or adjusting the layout density of the insulator detector to a corresponding value; Alternatively, when the cascade model is determined to be qualified, the two-stage cascade model is continuously used to perform frame selection and calibration on the image information of each insulator region; The ratio of the amount of insulator image information obtained by the two-stage cascade model within the calculated preset monitoring time to the preset number of insulators is determined as the recognition difference amount.

2. The intelligent detection and identification method for remote monitoring of contact network insulator defects according to claim 1 is characterized in that: The process of periodically determining whether the operation of the cascade model is qualified based on the false detection rate of the two-stage cascade model includes: Comparing the obtained false positive rate with the first preset false positive rate and the second preset false positive rate respectively; If the false detection rate is less than or equal to the first preset false detection rate, the cascade model is determined to be qualified, and the two-stage cascade model is continuously used to perform frame selection and calibration on the image information of each insulator area; If the false detection rate is less than or equal to the second preset false detection rate and greater than the first preset false detection rate, the false detection rate of the current cycle is marked as an abnormal false detection rate, and whether the operation of the cascade model is qualified is determined in combination with the obtained historical false detection rate; If the false detection rate is greater than the second preset false detection rate, it is determined that the operation of the cascade model is abnormal, and the operation parameters of the cascade model are adjusted based on the recognition difference.

3. The intelligent detection and identification method for remote monitoring of contact network insulator defects according to claim 2, characterized in that: The process of determining whether the cascade model is qualified based on the historical false positive rate includes: Calculate the ratio of the number of abnormal false positive rates in the historical false positive rate to the total number of historical false positive rates to obtain the abnormal frequency; Calculate the ratio of the abnormal false detection rate of the current cycle to the average value of each historical false detection rate to obtain the abnormal deviation; If the abnormal frequency is less than or equal to the preset abnormal frequency and the abnormal deviation is greater than the preset deviation, the cascade model is judged to be qualified, and the two-stage cascade model is continuously used to perform frame selection and calibration on the image information of each insulator area, and the preset monitoring time is adjusted to the corresponding value based on the abnormal deviation; If the abnormal frequency is greater than a preset abnormal frequency and / or the abnormal deviation is less than or equal to a preset deviation, the operation of the cascade model is determined to be abnormal, and the operation parameters of the cascade model are adjusted based on the identified difference.

4. The intelligent detection and identification method for remote monitoring of contact network insulator defects according to claim 3, characterized in that: The process of adjusting the operating parameters of the cascade model based on the identified difference amount includes: If the recognition difference is less than or equal to the preset recognition difference, the time nodes at which the insulator detector obtains the image information of each insulator region with calibration errors are obtained, and the concentrated time interval of each time node is determined; The image information of a preset number of insulator regions obtained by the insulator detector within a concentrated time interval is used as an experimental data set to calibrate the cascade model; The preset correction amount is adjusted to a corresponding value based on the identified difference amount.

5. The intelligent detection and identification method for remote monitoring of contact network insulator defects according to claim 4, characterized in that: The process of adjusting the operating parameters of the cascade model based on the identified difference amount also includes: If the recognition difference is greater than the preset recognition difference, the cascade model is adjusted based on the recognition accuracy evaluation value determined in combination with the image information of each insulator region; The average area of ​​the selected area of ​​each insulator region image information is calculated to obtain the recognition accuracy evaluation value; If the recognition accuracy evaluation value is less than or equal to the preset recognition accuracy evaluation value, the layout density of the insulator detectors is adjusted to a corresponding value based on the average value of the distances between the insulator detectors corresponding to the image information of the insulator regions with calibration errors; If the recognition accuracy evaluation value is greater than the preset recognition accuracy evaluation value, the insulator area image information obtained by a preset number of correction insulator detectors is selected as the experimental data set to calibrate the cascade model, and the preset correction number is adjusted to the corresponding value based on the recognition accuracy evaluation value.

6. The intelligent detection and identification method for remote monitoring of contact network insulator defects according to claim 5, characterized in that: The preset monitoring time is adjusted to a corresponding value based on the abnormal deviation amount, wherein the reduction of the preset monitoring time is proportional to the abnormal deviation amount.

7. The intelligent detection and identification method for remote monitoring of contact network insulator defects according to claim 6, characterized in that: The preset correction amount is adjusted to a corresponding value based on the identified difference amount, wherein the preset correction amount is increased in inverse proportion to the identified difference amount.

8. The intelligent detection and identification method for remote monitoring of contact network insulator defects according to claim 7, characterized in that: Based on the average value of the distances between the insulator detectors corresponding to the image information of the insulator areas with calibration errors, the layout density of the insulator detectors is adjusted to a corresponding value, wherein the increase in the layout density is inversely proportional to the average value of the distances.

9. The intelligent detection and identification method for remote monitoring of contact network insulator defects according to claim 8, characterized in that: The preset correction amount is adjusted to a corresponding value based on the recognition accuracy evaluation value, wherein an increase in the preset correction amount is proportional to the recognition accuracy evaluation value.

Images