Cable trench inspection cable defect detection method based on multi-mode double-flow algorithm

Abstract

The invention discloses a cable trench inspection cable defect detection method based on a multi-modal double-flow algorithm, belongs to the technical field of intelligent robots and machine vision, and aims at solving the problems that traditional inspection is poor in safety, insufficient in real-time performance, low in automatic detection precision and weak in robustness. According to the method, visible light and infrared thermal imaging images are synchronously collected based on a multi-mode imaging system carried by an inspection robot, and a double-assembly-line detection framework is constructed: for insulation layer damage, features are extracted through wavelet transform multi-scale decomposition, and a damage area is accurately identified in combination with multi-scale threshold screening, inter-class variance optimal threshold selection and topology analysis; for temperature abnormity, after a pseudo-color infrared image is converted into a grey-scale image, early-stage tiny temperature rise is captured by adopting local self-adaptive threshold judgment, morphological processing and area verification. The detection process runs in real time in the vehicle-mounted embedded unit, each frame processing is less than or equal to 50ms, high precision, non-contact performance and high robustness are achieved, the method is adaptive to severe environments, and a reliable scheme is provided for intelligent inspection of a power system.

Description

Technical Field

[0001] This invention relates to the intersection of intelligent robot technology and machine vision technology, and in particular to a cable defect detection method for cable trench inspection based on a multimodal dual-stream algorithm. Background Technology

[0002] Cable trenches are an indispensable and crucial infrastructure in power systems, primarily functioning to lay and protect a large number of cables carrying power transmission and control signals. Widely used in power plants, substations, and distribution systems, cable trenches house cables that perform critical tasks such as relay protection, automatic control, communication, and measurement, thus earning them the figurative title of the "nerve center" of the power system. Compared to traditional overhead lines, cables laid within cable trenches offer advantages such as smaller footprint, higher power supply reliability, and less susceptibility to external environmental interference. However, these advantages also come with complex operating environments and potential safety risks. Cable trenches typically suffer from insufficient sunlight, poor air circulation, high humidity, and large temperature gradients. The cable insulation, mostly made of polymer materials, is highly susceptible to insulation damage or internal overheating due to natural aging, mechanical damage, chemical corrosion, and rodent infestation during long-term operation. When defects form and develop, cables in cable trenches are highly susceptible to short circuits or even fires due to insulation failure or abnormal temperature rise. The fire can spread rapidly along the cable trench, causing large-scale power outages and significant economic losses in a short period of time. In severe cases, it may also affect the safe and stable operation of the power system.

[0003] For the detection and maintenance of cable operating status in cable trenches, current traditional methods mainly rely on manual inspection and periodic preventive testing. While manual inspection is intuitive, the narrow, enclosed environment of cable trenches with poor air quality presents significant safety risks for inspectors, including exposure to high temperatures, dust, toxic gases, and potential fire hazards. Furthermore, manual inspection is inefficient, relying on visual observation and experience to ensure stability and accuracy, easily leading to missed or false detections. Periodic insulation testing, while assessing the overall health of the cable, often requires power outages, causing power supply interruptions and indirect economic losses. In addition, these methods typically operate on a monthly or yearly basis, failing to detect defects in real time and creating "window periods" where new problems may arise between tests. Therefore, existing inspection and testing methods are insufficient in terms of real-time performance, safety, and accuracy, failing to meet the high requirements of modern power systems for the safe operation of infrastructure.

[0004] With the rapid development of robotics and machine vision, automated inspection has gradually become an important direction for cable trench maintenance. However, when existing automated inspection methods are applied to the complex and harsh environment of cable trenches, many technical bottlenecks still exist.

[0005] For example, CN108832727A describes a cable tunnel inspection robot system and its control method based on wireless charging. This solution provides continuous power to the inspection robot through wireless charging technology and enables remote control via a cloud terminal module, effectively improving inspection efficiency and safety. However, the adaptability of this solution in complex cable trench environments still needs improvement. Especially under conditions of insufficient light, high humidity, and high dust concentration, the performance of the vision sensor may be significantly affected, leading to a decrease in image acquisition quality and consequently affecting the accuracy of defect detection.

[0006] For example, CN111168696A describes a cable tunnel inspection system using an RGV robot. This system employs an RGV robot for automated inspection, combining it with a wireless communication base station and a backend server for data transmission and analysis, thus improving the level of automation in the inspection. However, the RGV robot is highly dependent on its track, making it difficult to adapt to the complex and varied terrain within cable trenches. Furthermore, its image processing capabilities are limited, making it difficult to accurately identify defects such as insulation damage and temperature anomalies under low signal-to-noise ratio conditions.

[0007] For example, CN114326524B describes a distributed cable tunnel inspection system and detection and evaluation method. This scheme uses distributed dynamic wireless charging terminals and multiple distributed monitoring points, combined with an inspection robot for data collection and analysis, to achieve comprehensive monitoring of the cable tunnel's operating status. However, this scheme still faces challenges when processing low signal-to-noise ratio images, especially in complex backgrounds within cable trenches, where target segmentation and feature extraction are difficult, easily leading to false positives and false negatives.

[0008] For example, CN119200617A describes an automated inspection method and system for cable tunnels based on quadruped robots. This solution utilizes the high adaptability and flexibility of quadruped robots, combined with various low-power, high-precision intelligent sensing and monitoring devices, to achieve automated inspection of cable tunnels. Although quadruped robots have good maneuverability in complex terrain, further improvements are needed in the accuracy and robustness of the algorithm in processing low signal-to-noise ratio images and accurately identifying subtle defects such as insulation layer damage and temperature anomalies.

[0009] Given the limitations of existing technologies in complex cable trench environments and the urgent need for safe and stable operation of power system infrastructure, this application proposes an innovative cable defect detection method, which is particularly necessary and urgent. This method, through multimodal visual sensing and dual-pipeline algorithm design, combined with advanced technologies such as wavelet transform threshold segmentation, adaptive threshold binarization, and topology analysis, effectively addresses the shortcomings of existing technologies in terms of detection accuracy, robustness, and real-time performance. This method can accurately identify defects such as insulation layer damage and temperature anomalies in complex environments, and also possesses advantages such as non-contact detection, high robustness, and real-time data processing, providing strong support for the safe and stable operation of power system infrastructure. In the context of the continuous expansion of the power system and increasingly stringent operational requirements, the proposal of this application undoubtedly has significant practical implications and broad application prospects. Summary of the Invention

[0010] The technical problem this invention aims to solve is to provide a cable defect detection method for cable trench inspection based on a multimodal dual-stream algorithm, addressing key technical issues in the field of cable trench inspection in power systems, namely insufficient detection accuracy, poor real-time performance, and low safety. Traditional manual inspection methods are inefficient, and inspectors face high temperatures, dust, toxic gases, and potential fire risks in the narrow, enclosed environment of cable trenches with poor air quality, which can easily endanger their personal safety. Furthermore, relying on manual visual observation and experience makes it difficult to guarantee the stability and accuracy of detection, easily leading to missed or false detections. Existing automated detection methods are susceptible to noise interference in complex cable trench environments, making it difficult to accurately identify defects such as cable insulation damage and abnormal temperatures.

[0011] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: This invention provides a multimodal dual-stream algorithm-based cable defect detection method for cable trench inspection. Using a cable trench inspection robot as a platform, a multimodal imaging system is deployed on the robot. This system includes a visible light camera and a thermal imager, capable of simultaneously acquiring visible light and infrared thermal images of the cable within the trench. The two types of images are input into an insulation damage detection pipeline and a temperature anomaly detection pipeline, respectively, constructing a dual-pipeline parallel detection framework. Finally, the detection results from both pipelines are integrated to complete comprehensive monitoring of cable defects. The entire detection process runs in real-time on the inspection robot's onboard embedded computing unit, with the processing time for each image frame controlled within 50 milliseconds, meeting the online detection requirements under normal robot inspection speeds.

[0012] 1. Methods for detecting cable insulation damage To address the challenges of low lighting conditions, complex backgrounds, and the high similarity between insulation damage characteristics and noise in cable trenches, this invention employs a multi-scale image analysis method based on wavelet transform, combined with topological structure screening, to achieve accurate detection. The specific steps are as follows: Wavelet transform multi-scale decomposition: When the inspection robot moves along the predetermined path, the visible light camera continuously acquires images f(x, y) of the cable surface. The mother wavelet function ψ(a, b) is selected, and the scale factor a and translation factor b are set. Wavelet transform is performed on the image f(x, y) to obtain the low-frequency component Lj and the high-frequency component Hj at different scales j. Among them, the low-frequency component Lj retains the overall outline and brightness distribution of the cable, while the high-frequency component Hj highlights the abrupt features such as insulation layer edges, cracks, and peeling.

[0013] Multi-scale candidate threshold extraction: Utilizing the extreme value characteristics of wavelet decomposition coefficients, multi-scale analysis is performed on the coefficient histograms at each scale. When a certain wavelet decomposition coefficient... Simultaneously satisfy When the gray value corresponding to the coefficient is determined to be a local minimum of the image histogram, that is, a segmentation candidate threshold; after repeated detection across scales, a candidate threshold set {T1, T2, ..., Tn} is formed.

[0014] Optimal threshold selection and binarization: Introduce the inter-class variance maximization criterion to calculate the inter-class variance corresponding to each candidate threshold T (the inter-class variance is determined by the proportion of pixels in the two classes). , and the average grayscale of two types of pixels , (Decision), selecting the candidate threshold that maximizes the inter-class variance as the optimal threshold. and based on Image binarization is performed to obtain a binary image B(x, y), achieving preliminary separation of the cable body from the complex background.

[0015] Topology analysis and area screening: Contour detection is performed on the binary image B(x, y), all closed contours are extracted, and a parent-child contour hierarchy is constructed. Small closed inner contours nested within the outer contour of the cable body are identified as suspected damaged contours. The area S of the suspected damaged contours is calculated, and only when S is within a preset range... At that time, areas identified as areas of effective insulation layer damage are filtered out to eliminate isolated noise points and large invalid areas, thus ensuring detection accuracy.

[0016] 2. Methods for detecting abnormal cable temperature To address the issue of traditional global thresholding algorithms failing to capture early, minute temperature rises, this invention proposes an adaptive thresholding method based on local statistics, combined with morphological processing, to achieve accurate early hotspot capture. The specific steps are as follows: Infrared image preprocessing: The pseudo-color infrared image acquired by the thermal imager is converted into a grayscale image g(x,y). The grayscale value of this grayscale image has a linear relationship with the actual temperature of the cable.

[0017] Local Neighborhood Adaptive Threshold Determination: A neighborhood window W(x,y) is constructed for each pixel in the grayscale image g(x,y). The local mean within the window is calculated. If the grayscale value g(x,y) of the center pixel is greater than the sum of the local mean and the constant C, it is determined to be a suspected temperature anomaly point. The constant C is taken as the grayscale range (5 to 12 grayscale levels) corresponding to a temperature rise of 2℃ to 3℃, taking into account both detection sensitivity and noise resistance, to ensure that small temperature rises in the cable can be reliably identified.

[0018] Morphological processing: First, perform an erosion operation on the binary image composed of suspected abnormal points to remove isolated noise white points; then perform a dilation operation to connect adjacent abnormal pixels to form continuous hotspot regions and optimize the integrity of the detection results.

[0019] Connectivity analysis and area determination: Perform connectivity analysis on the morphologically processed image to extract candidate anomaly regions and calculate their areas S. When S ≥ a preset temperature anomaly area threshold... When the temperature is abnormal, it is identified as an effective temperature abnormality area, ensuring that the detection results have actual physical meaning and enabling timely early warning of slight heating of cable joints (temperature difference less than 3℃).

[0020] The cable defect detection method for cable trench inspection based on the multimodal dual-stream algorithm provided in this invention has the following beneficial effects: 1. This invention relies on multimodal visual sensing and a dual pipeline algorithm framework to effectively solve the problems of poor security, insufficient real-time performance, low detection accuracy, and weak adaptability to complex environments in existing cable trench defect detection methods. It achieves high-precision and robust real-time detection of cable insulation layer damage and temperature anomalies, providing a reliable technical solution for intelligent inspection of cable trenches in power systems.

[0021] 2. This invention enables high-precision, real-time detection of cable insulation layer damage and temperature anomalies in cable trenches, significantly improving the safety and reliability of power systems.

[0022] 3. Experiments show that this invention can maintain stable identification under various damage conditions such as slight cracks, peeling, and even conductor exposure on the cable surface. At the same time, it can immediately trigger an alarm when the cable joint is slightly heated (temperature difference less than 3°C), which can buy sufficient emergency time for power operation and maintenance personnel and effectively reduce fire hazards.

[0023] 4. This invention effectively solves the problems of insufficient detection accuracy and poor robustness in the existing technology through dual-modal imaging and dual-pipeline algorithm design, providing strong support for the intelligent and automated transformation of power system inspection.

[0024] 5. This invention overcomes the limitations of traditional manual inspection, such as poor safety and low efficiency, as well as the technical bottleneck of existing automated inspection, which is prone to failure under low signal-to-noise ratio and low contrast images, and achieves non-contact high-precision inspection.

[0025] 6. The insulation layer damage detection method of the present invention can effectively distinguish false defects such as cable surface texture and oil stains, and the temperature anomaly detection method can accurately capture early small temperature rises, greatly reducing the rate of missed detection and false detection.

[0026] 7. The detection process of this invention has real-time performance and high robustness, and can be adapted to the complex and harsh environment of cable trenches. It provides data support for predictive maintenance and proactive operation and maintenance of power systems, promotes the transformation of cable trench inspection towards intelligence and automation, and ensures the safe and stable operation of power infrastructure.

[0027] 8. This invention ensures the personal safety of inspection personnel by using an inspection robot equipped with a multimodal imaging system to complete automated detection, replacing the traditional method of manually entering cable trenches for inspection. This avoids personnel being exposed to dangerous environments such as high temperatures, dust, toxic gases, and potential fires, fundamentally eliminating the personal safety hazards of manual inspection.

[0028] 9. This invention eliminates the "window period" of detection and realizes real-time monitoring. Compared with the traditional monthly / yearly preventive testing, the cable condition can be detected in real time during the robot inspection process, and new cable defects formed between two periodic tests can be detected in a timely manner, thus completely solving the hidden danger of the "window period" of traditional detection methods.

[0029] 10. This invention is adaptable to complex and harsh environments and has strong anti-interference capabilities. Targeting the environmental characteristics of cable trenches such as insufficient lighting, high humidity, high dust concentration, and cluttered background, the algorithm can effectively process low signal-to-noise ratio images, breaking through the technical bottlenecks of traditional automated detection being susceptible to noise interference and difficult target segmentation, and has extremely strong adaptability to complex environments.

[0030] 11. This invention improves the accuracy of insulation layer damage detection. Based on wavelet transform multi-scale decomposition technology, it can separate the macroscopic structure of the cable body and the detailed features of the damaged area at different scales, achieving high-precision separation of the cable body and the damaged edge, and solving the problem of detection failure of traditional global threshold segmentation under complex texture interference.

[0031] 12. This invention filters out false defects, reduces the false detection and missed detection rate of insulation damage detection, and introduces a topological structure analysis and area range screening mechanism to accurately identify the real damage area nested within the outline of the cable body. It effectively filters out false defects formed by dust, oil stains and natural textures on the cable surface, and greatly improves the accuracy of insulation layer damage detection.

[0032] 13. This invention achieves accurate capture of early temperature anomalies in cables. It adopts an adaptive threshold thermal imaging detection method based on local statistics, which eliminates the dependence of traditional global threshold algorithms on bimodal / multimodal histograms. It can stably identify early minute temperature rises of only 2℃~3℃ in cables, solves the problem of failure of traditional methods under single-peak grayscale distribution, and buys sufficient time for defect early warning.

[0033] 14. This invention optimizes the reliability of temperature anomaly detection results. Through morphological processing of corrosion followed by expansion, isolated false hot spots in infrared thermal imaging can be eliminated. At the same time, adjacent temperature anomaly pixels are connected to form continuous hot spot areas, providing more physically meaningful detection results for subsequent anomaly determination.

[0034] 15. This invention achieves full coverage detection of two types of core defects, constructs a dual pipeline detection framework of visible light and infrared thermal imaging, realizes complementary detection of physical defects of insulation layer damage and thermal defects of abnormal temperature, with no detection blind spots, and can fully grasp the operating health status of the cable.

[0035] 16. This invention meets the engineering requirements for online real-time detection. The detection process runs on an on-board embedded computing unit, and the processing time for each frame of image is controlled within 50 milliseconds, which can match the normal inspection speed of the inspection robot and meet the engineering requirements for online real-time detection during cable trench inspection.

[0036] 17. This invention avoids interference with the power supply. This invention is a non-contact detection method, which does not require power outage operation like traditional preventive testing. It can complete defect detection while the cable is running normally, effectively reducing indirect economic losses caused by power outages.

[0037] 18. This invention reduces the probability of major power accidents and can provide accurate early warnings of early defects such as slight overheating of cable joints and minor cracks in the insulation layer, making it easier for operation and maintenance personnel to intervene and deal with them in advance. It significantly reduces the risk of short circuits and fires caused by cable insulation failure and abnormal temperature rise, and ensures the stable operation of the power system.

[0038] 19. This invention has strong engineering scalability, and the algorithm framework and hardware carrier are highly adaptable. It can be flexibly integrated into different models of cable trench inspection robots, which is convenient for large-scale promotion in power infrastructure in different scenarios such as power plants, substations, and power distribution systems.

[0039] 20. This invention promotes the transformation and upgrading of power operation and maintenance models. It can synchronize the detected defect location, type and other data to the power operation and maintenance management platform, providing accurate data support for predictive maintenance and proactive operation and maintenance of power systems, and promoting the transformation of cable trench operation and maintenance from "human experience-driven" to "intelligent data-driven".

[0040] 21. This invention improves the practicality and traceability of the detection results, quantifies and determines key parameters such as the area and temperature rise of the defect region, and the detection results have clear physical meaning. At the same time, it can retain images and data records, providing a traceable basis for subsequent defect cause analysis and operation and maintenance strategy optimization. Attached Figure Description

[0041] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart illustrating the overall technical process of Embodiment 2 of the present invention; Figure 2 This is a schematic diagram of the deployment method of the multimodal imaging system in Embodiment 2 of the present invention; Figure 3 This is a physical image of the thermal infrared sensor of Embodiment 2 of the present invention; Figure 4 This is a physical image of the visible light sensor according to Embodiment 2 of the present invention; Figure 5 This is a flowchart of the insulation layer damage detection algorithm of Embodiment 2 of the present invention; Figure 6 This is a flowchart of the cable temperature anomaly detection algorithm in Embodiment 2 of the present invention; Figure 7 This is a diagram of a laboratory simulated electrical trench environment according to Embodiment 3 of the present invention; Figure 8 This is a physical diagram of the vehicle-mounted embedded computing unit according to Embodiment 3 of the present invention; Figure 9 This is a diagram showing the results of cable anomaly detection in Embodiment 3 of the present invention. Detailed Implementation

[0042] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments: Example 1 This embodiment provides a cable defect detection method for cable trench inspection based on a multimodal dual-stream algorithm, as detailed below: In terms of overall design, this embodiment uses an inspection robot as a carrier. A multimodal imaging system is deployed within the cable trench to acquire visible light and infrared thermal images in real time, which are then used as inputs for a dual-pipeline inspection method. The entire inspection method revolves around the core issue of "how to accurately extract defect areas in noisy environments and complex backgrounds." Its innovation lies in proposing a highly targeted algorithm flow. For insulation layer damage, traditional global threshold segmentation performs poorly under complex texture interference. The wavelet transform thresholding method proposed in this embodiment effectively separates the cable body from the damage edge at different scales using multi-resolution characteristics, achieving high-precision defect extraction. For temperature anomalies, traditional thermal imaging processing methods based on global histograms completely fail under unimodal distributions. The adaptive thresholding method proposed in this embodiment can dynamically determine the threshold within a local neighborhood, thereby sensitively capturing early, minute hot spots.

[0043] Firstly, in the cable insulation layer damage detection method, the core idea of ​​this embodiment is multi-scale image analysis based on wavelet transform. When the inspection robot moves along a predetermined path within the cable trench, its onboard visible light camera continuously acquires images of the cable surface. Let the image acquired at a certain moment be f(x, y). Wavelet transform can decompose the image signal into components of different scales, thereby extracting macroscopic structure and detailed features respectively. The wavelet transform formula is as follows: (1).

[0044] In the formula, ψ(a, b) is the mother wavelet function, a is the scaling factor, which controls the scaling of the wavelet in the frequency domain, and b is the translation factor, which determines the position of the wavelet in the time domain (or image space). By performing multi-scale decomposition on the image, the low-frequency part Lj and the high-frequency part Hj at scale j can be obtained. The low-frequency part Lj mainly preserves the overall outline and brightness distribution of the cable, representing the smooth structure of the cable body; while the high-frequency part Hj contains abrupt information such as edges, cracks, and peeling, and the salience of damage is highlighted in the high-frequency part.

[0045] In real-world scenarios like cable trenches with complex backgrounds and extremely poor lighting conditions, traditional global thresholding methods often fail to effectively separate the cable body from the background, let alone extract minute damage. Therefore, this embodiment takes a unique approach, utilizing the extreme value characteristics of coefficients at each scale after wavelet decomposition, and searching for candidate threshold points through multi-scale analysis of the coefficient histogram. The mathematical criteria can be expressed as follows: (2); in, These are the wavelet decomposition coefficients. That is, in the wavelet decomposition coefficient sequence, when a certain coefficient... At the same time, it is smaller than its adjacent coefficients. , When the gray value corresponding to a point is considered a local minimum of the image histogram, it is a candidate threshold for segmentation. By repeatedly detecting across scales, a set of candidate thresholds {T1, T2, ..., Tn} can be obtained.

[0046] Next, this embodiment introduces the between-class variance maximization criterion to select the final threshold. The expression for the between-class variance is: (3); In the formula, , These represent the proportions of the two pixel classes after being divided by the threshold T. , Let be the average gray level of the two classes of pixels. Intuitively, the larger the inter-class variance, the stronger the difference between the two classes of pixels classified by the threshold, and the better the segmentation effect. Therefore, we choose to make The threshold that reaches the maximum value is taken as the optimal threshold. Thus, the final binarized B(x, y) is obtained: (4).

[0047] After the above series of steps, the cable body is effectively separated from the complex background, and the edges of potential damaged areas are initially highlighted. However, in real and complex industrial environments, cable surfaces are often covered with dust, oil, or have various textures, all of which can create false defects in the binarization results. To address this issue, this embodiment further introduces topological analysis as a refining and filtering method. Specifically, the algorithm performs contour detection on the binary image, extracting all closed contours. According to topological principles, the outer contour of the cable usually corresponds to its main body edge, while insulation damage (such as holes or peeling) will appear as small closed inner contours nested within this outer contour. By constructing and analyzing this hierarchical relationship between parent and child contours, the algorithm can very effectively distinguish real damaged areas from isolated noise points that do not belong to the cable body. Finally, for each identified inner contour, the algorithm calculates its area: (5).

[0048] Only areas falling within this range are identified as valid damage zones. This design allows the method to filter out both excessively small noise points and excessively large invalid areas, thereby significantly improving detection accuracy. Through this process, this embodiment can maintain stable identification performance under various damage conditions on the cable surface, including minor cracks, peeling, and even conductor exposure.

[0049] Secondly, in the cable temperature anomaly detection method, this embodiment addresses the problem of traditional algorithms failing in the early defect stage by proposing an adaptive thresholding method based on local statistics. The original image acquired by the thermal imager is in pseudo-color format and needs to be converted to a grayscale image g(x, y), where the grayscale value has a linear correspondence with the actual temperature. Unlike the global threshold, the adaptive method no longer relies on the histogram of the entire image, but instead constructs a neighborhood window W(x, y) for each pixel and calculates the local mean of that neighborhood: (6).

[0050] in, This represents the number of pixels in the neighboring window. This mean value represents the "normal temperature level" of the local environment. The center pixel is considered to meet the following condition: (7).

[0051] If the point is not identified as a temperature anomaly, it is considered the foreground; otherwise, it is considered the background. A constant C is used to balance sensitivity and noise immunity. A smaller value for C allows the algorithm to capture weaker hotspots but makes it more susceptible to noise interference; a larger value for C makes the algorithm more robust but reduces its sensitivity to early defects. Through experiments in different scenarios, this embodiment provides a reasonable parameter selection range to ensure reliable detection even when the cable temperature is only 2°C to 3°C higher than the ambient temperature.

[0052] After obtaining the initial binarization result, this embodiment introduces a morphological processing step. First, an erosion operation is used to remove isolated white dots, eliminating false hotspots caused by random noise. Then, a dilation operation is used to connect adjacent white pixels, forming continuous hotspot regions. The binary image after morphological processing is clearer and suitable for further analysis. Next, candidate anomaly regions are extracted through connected component analysis, and their areas S are calculated. The final judgment criterion is: (8).

[0053] in, A temperature anomaly area threshold is set; only areas exceeding this threshold are considered valid temperature anomaly zones, ensuring the detection results have practical physical significance. Experiments show that this method can accurately detect localized temperature rises caused by joint overheating and internal short circuits, and can also trigger an alarm immediately when the cable joint is only slightly overheating (i.e., the temperature difference is less than 3°C), providing power maintenance personnel with ample emergency response time and significantly reducing fire hazards.

[0054] Within the overall dual-pipeline inspection framework, this embodiment fully leverages the complementarity of visible light and infrared imaging. Visible light imaging focuses on identifying physical defects such as insulation damage, while infrared thermal imaging is sensitive to abnormal cable temperatures. The combination of the two enables comprehensive monitoring of potential cable risks. The inspection process runs in real-time on an onboard embedded computing unit, with the processing time for each frame controlled within 50 milliseconds, ensuring the robot's online inspection capabilities at normal inspection speeds.

[0055] Example 2 In another preferred embodiment, based on Embodiment 1 above, this embodiment provides a cable defect detection method for cable trench inspection using a multimodal dual-stream algorithm. Its core idea is to utilize multimodal visual sensing and a dual-pipeline algorithm framework to achieve high-precision real-time detection of cable insulation damage and temperature anomalies in complex cable trench environments. This method combines the advantages of wavelet transform threshold segmentation and adaptive threshold binarization, overcoming the limitations of traditional manual inspection methods (low efficiency, high risk) and existing automated detection methods (prone to failure under noise interference and low-contrast images). It provides a reliable intelligent inspection solution for power systems. Compared with traditional image processing methods, this embodiment not only has significant advantages in robustness, sensitivity, and real-time performance, but also demonstrates strong scalability and practical value in engineering applications.

[0056] The overall technical process of this embodiment is as follows: Figure 1 As shown, the entire process from image acquisition to defect confirmation is covered: the inspection robot, equipped with a multimodal imaging system, enters the cable trench and moves along a preset path; a visible light camera and an infrared thermal imager simultaneously acquire images of the cable surface and temperature distribution; the image data is transmitted to the vehicle-mounted embedded computing unit, and then enters the insulation layer damage detection pipeline and the temperature anomaly detection pipeline, respectively; after the two pipelines process the data in parallel, preliminary defect detection results are output; combined with topology analysis and area constraints, valid defects are screened, and finally a defect report is generated and transmitted to the ground station.

[0057] The multimodal imaging system consists of an M384 thermal infrared sensor and an IMX577 wide-angle industrial camera, mounted side-by-side at the end of an inspection robotic arm, facing the cable direction, and deployed as follows: Figure 2As shown. The cable trench structure is characterized by its enclosed, narrow, and multi-layered distribution, making it difficult for fixed monitoring methods to cover all critical locations. A robotic arm, as a highly flexible and programmable end effector, can carry a multi-modal imaging system deep into various target areas within the trench, enabling dynamic deployment and precise observation, and achieving accurate detection of multi-level and multi-angle cable targets. The M384 thermal infrared sensor weighs approximately 82g and has dimensions of 42mm×42mm×43mm; its infrared resolution is 324×288, and it can achieve a minimum temperature measurement distance of 10cm with an adjustable temperature range of -20°C to 350°C. Figure 3 As shown. The visible light sensor used is the IMX577 wide-angle industrial camera, weighing approximately 150g, with dimensions of 42mm × 42mm × 22mm, a resolution of 3840 × 2880 (up to 12 megapixels), autofocus, and multi-angle lens selection, such as... Figure 4 As shown.

[0058] In terms of overall design, this embodiment uses an inspection robot as a carrier. A multimodal imaging system is deployed within the cable trench to acquire visible light and infrared thermal images in real time, which are then used as inputs for a dual-pipeline inspection method. The entire inspection method revolves around the core issue of "how to accurately extract defect areas in noisy environments and complex backgrounds." Its innovation lies in proposing a highly targeted algorithm flow. For insulation layer damage, traditional global threshold segmentation performs poorly under complex texture interference. The wavelet transform thresholding method proposed in this embodiment effectively separates the cable body from the damage edge at different scales using multi-resolution characteristics, achieving high-precision defect extraction. For temperature anomalies, traditional thermal imaging processing methods based on global histograms completely fail under unimodal distributions. The adaptive thresholding method proposed in this embodiment can dynamically determine the threshold within a local neighborhood, thereby sensitively capturing early, minute hot spots.

[0059] Firstly, in the cable insulation layer damage detection method, the core idea of ​​this embodiment is based on multi-scale image analysis using wavelet transform. The insulation layer damage detection algorithm flow is as follows: Figure 5 As shown.

[0060] As the inspection robot moves along a predetermined path within the cable trench, its onboard visible light camera continuously captures images of the cable surface. Let the image captured at a certain moment be f(x, y). Wavelet transform can be used to decompose the image signal into components of different scales, thereby extracting macroscopic structure and detailed features. The wavelet transform formula is as follows: (1).

[0061] Where ψ(a, b) is the mother wavelet function, a is the scaling factor, controlling the scaling of the wavelet in the frequency domain, and b is the translation factor, determining the position of the wavelet in the time domain (or image space). By performing multi-scale decomposition on the image, the low-frequency component Lj and the high-frequency component Hj at scale j can be obtained. The low-frequency component Lj mainly preserves the overall outline and brightness distribution of the cable, representing the smooth structure of the cable body; while the high-frequency component Hj contains abrupt information such as edges, cracks, and peeling, and the salience of damage is highlighted in the high-frequency component.

[0062] In real-world scenarios like cable trenches with complex background elements and extremely poor lighting conditions, traditional global thresholding segmentation methods often fail to effectively separate the cable body from the background, let alone extract minute damage. Therefore, this embodiment utilizes the extreme value characteristics of coefficients at each scale after wavelet decomposition, and searches for candidate threshold points through multi-scale analysis of the coefficient histogram. The mathematical criteria can be expressed as follows: (2); in, These are the wavelet decomposition coefficients. That is, in the wavelet decomposition coefficient sequence, when a certain coefficient... At the same time, it is smaller than its adjacent coefficients. , When the gray value corresponding to a point is considered a local minimum of the image histogram, it is a candidate threshold for segmentation. By repeatedly detecting across scales, a set of candidate thresholds {T1, T2, ..., Tn} can be obtained.

[0063] Next, this embodiment introduces the between-class variance maximization criterion to select the final threshold. The expression for the between-class variance is: (3); In the formula, , These represent the proportions of the two pixel classes after being divided by the threshold T. , Let be the average gray level of the two classes of pixels. Intuitively, the larger the inter-class variance, the stronger the difference between the two classes of pixels classified by the threshold, and the better the segmentation effect. Therefore, we choose to make The threshold that reaches the maximum value is taken as the optimal threshold. Thus, the final binarized B(x, y) is obtained: (4).

[0064] After the above series of steps, the cable body is effectively separated from the complex background, and the edges of potential damaged areas are initially highlighted. However, in real and complex industrial environments, cable surfaces are often covered with dust, oil, or have various textures, all of which can create false defects in the binarization results. To address this issue, this embodiment further introduces topological analysis as a refining and filtering method. Specifically, the algorithm performs contour detection on the binary image, extracting all closed contours. According to topological principles, the outer contour of the cable usually corresponds to its main body edge, while insulation damage (such as holes or peeling) will appear as small closed inner contours nested within this outer contour. By constructing and analyzing this hierarchical relationship between parent and child contours, the algorithm can very effectively distinguish real damaged areas from isolated noise points that do not belong to the cable body. Finally, for each identified inner contour, the algorithm calculates its area: (5).

[0065] Only areas falling within this range are identified as valid damage zones. This design allows the method to filter out both excessively small noise points and excessively large invalid areas, thereby significantly improving detection accuracy. Through this process, this embodiment can maintain stable identification performance under various damage conditions on the cable surface, including minor cracks, peeling, and even conductor exposure.

[0066] Secondly, in the cable temperature anomaly detection method, this embodiment addresses the problem of traditional algorithms failing in the early defect stage by proposing an adaptive threshold method based on local statistics. The cable temperature anomaly detection algorithm flow is as follows: Figure 6 As shown.

[0067] The raw images acquired by thermal imagers are in pseudo-color format and need to be converted to grayscale images g(x, y) first. The grayscale values ​​have a linear relationship with the actual temperature. Unlike global thresholding, the adaptive method no longer relies on the histogram of the entire image. Instead, it constructs a neighborhood window W(x, y) for each pixel and calculates the local mean of that neighborhood. (6).

[0068] in, This represents the number of pixels in the neighboring window. This mean value represents the "normal temperature level" of the local environment. The center pixel is considered to meet the following condition: (7).

[0069] If the point is not identified as a temperature anomaly, it is considered the foreground; otherwise, it is considered the background. A constant C is used to balance sensitivity and noise immunity. A smaller value for C allows the algorithm to capture weaker hotspots but makes it more susceptible to noise interference; a larger value for C makes the algorithm more robust but reduces its sensitivity to early defects. Through experiments in different scenarios, this embodiment provides a reasonable parameter selection range to ensure reliable detection even when the cable temperature is only 2°C to 3°C higher than the ambient temperature.

[0070] After obtaining the initial binarization result, this embodiment introduces a morphological processing step. First, an erosion operation is used to remove isolated white dots, eliminating false hotspots caused by random noise. Then, a dilation operation is used to connect adjacent white pixels, forming continuous hotspot regions. The binary image after morphological processing is clearer and suitable for further analysis. Next, candidate anomaly regions are extracted through connected component analysis, and their areas S are calculated. The final judgment criterion is: (8).

[0071] in, A temperature anomaly area threshold is set; only areas exceeding this threshold are considered valid temperature anomaly zones, ensuring the detection results have practical physical significance. Experiments show that this method can accurately detect localized temperature rises caused by joint overheating and internal short circuits, and can also trigger an alarm immediately when the cable joint is only slightly overheating (i.e., the temperature difference is less than 3°C), providing power maintenance personnel with ample emergency response time and significantly reducing fire hazards.

[0072] Within the overall dual-pipeline inspection framework, this embodiment fully leverages the complementarity of visible light and infrared imaging. Visible light imaging focuses on identifying physical defects such as insulation damage, while infrared thermal imaging is sensitive to abnormal cable temperatures. The combination of the two enables comprehensive monitoring of potential cable risks. The inspection process runs in real-time on an onboard embedded computing unit, with the processing time for each frame controlled within 50 milliseconds, ensuring the robot's online inspection capabilities at normal inspection speeds.

[0073] The core focus of this embodiment is the innovative design and engineering implementation of targeted algorithms for dual pipelines. Addressing the pain points of detecting two core defect types in the complex environment of cable trenches, it achieves a dual breakthrough of "accurate identification + efficient processing": For the subtle features and complex texture interference of insulation layer damage, an innovative algorithm combining wavelet transform multi-scale decomposition and inter-class variance maximization threshold screening is adopted to overcome the high false positive and false negative rates of traditional global segmentation algorithms; for the unimodal gray-scale distribution characteristics of the small temperature rise in the early stage of temperature anomalies, an adaptive threshold method based on local statistics is proposed to solve the problem of failure of traditional global threshold algorithms. Simultaneously, the algorithm ensures a balance between detection accuracy and real-time performance through clear parameter calibration (such as filtering window, decomposition scale, temperature mapping coefficient, etc.) and standardized step-by-step processing flow, possessing strong operability and engineering reusability. To clearly present the core technology and implementation details, the specific implementation steps of the core algorithms of the two pipelines are explained in detail below: The specific implementation steps of the wavelet transform threshold segmentation algorithm are as follows: For cable insulation layer damage detection, the algorithm first preprocesses the image acquired by visible light: after converting the color image to grayscale, Gaussian filtering (σ=1.5, 5×5 window) is used to suppress noise interference, and Contrast Limited Adaptive Histogram Equalization (CLAHE) is used to enhance the contrast between the damaged area and the background; then, the db4 mother wavelet is selected for 3-level multi-scale decomposition, extracting and fusing high-frequency components in the horizontal, vertical and diagonal directions, preserving the damage edge features; based on the grayscale histogram of the high-frequency image, local minima are selected as candidate thresholds. If the number of candidate thresholds is insufficient, the decomposition scale is increased to level 4; if the number is too large, adjacent thresholds are merged (difference ≤ 5), and the optimal threshold is determined from the candidate set by the inter-class variance maximization criterion, completing the high-frequency image binarization; finally, contour detection is performed on the binary image, filtering the inner contour nested inside the outer contour of the cable, and valid damaged areas are verified by area constraints (50~5000 pixels), and the detection results are finally output. Partial code examples are as follows: import cv2 import numpy as np import pywt def wavelet_threshold_segmentation(image_path): # Step 1: Image Preprocessing img = cv2.imread(image_path) gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) # Gaussian Filter gaussian = cv2.GaussianBlur(gray, (5, 5), 1.5) # CLAHE Contrast Enhancement clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8)) enhanced = clahe.apply(gaussian) # Step 2: Wavelet Transform Multiscale Decomposition wavelet = 'db4' level = 3 coeffs = pywt.wavedec2(enhanced, wavelet, level=level) L3, (H3, V3, D3) = coeffs[0], coeffs[1] # Merging high-frequency components high_freq = np.sqrt(H3**2 + V3**2 + D3**2) high_freq = cv2.normalize(high_freq, None, 0, 255, cv2.NORM_MINMAX).astype(np.uint8) # Step 3: Candidate Threshold Filtering hist = cv2.calcHist([high_freq], [0], None,

[256] , [0, 256]) candidates = [] for l in range(1, 255): if hist[l-1]>hist[l] and hist[l] <hist[l+1]: candidates.append(l) # Threshold simplification if len(candidates) <= 3: coeffs = pywt.wavedec2(enhanced, wavelet, level=4) L4, (H4, V4, D4) = coeffs[0], coeffs[1] high_freq = np.sqrt(H4**2 + V4**2 + D4**2) high_freq = cv2.normalize(high_freq, None, 0, 255, cv2.NORM_MINMAX).astype(np.uint8) hist = cv2.calcHist([high_freq], [0], None,

[256] , [0, 256]) candidates = [] for l in range(1, 255): if hist[l-1]>hist[l] and hist[l]<hist[l+1]: candidates.append(l) elif len(candidates)>10: sorted_candidates = sorted(candidates) merged = [sorted_candidates[0]] for t in sorted_candidates[1:]: if t - merged[-1]>5:​​​​​​​​​​​​​​​​​​​​​​​if w0 == 0 or w1 == 0: continue mu0 = np.mean(high_freq[background]) mu1 = np.mean(high_freq[foreground]) var = w0 * w1 * (mu0 - mu1)**2 if var>max_var: max_var = var optimal_T = T # Binarization _, B1 = cv2.threshold(high_freq, optimal_T, 255, cv2.THRESH_BINARY) # Step 5: Topology Analysis and Area Verification contours, hierarchies = cv2.findContours(B1, cv2.RETR_CCOMP,cv2.CHAIN_APPROX_SIMPLE) valid_defects = [] for i, contour in enumerate(contours): # Filter inner contours (hierarchies[0][i][3] != -1 indicates that there is a parent contour) if hierarchies[0][i][3] != -1: area = cv2.contourArea(contour) if 50 <= area <= 5000: valid_defects.append(contour) # Draw the valid damaged area result = img.copy() cv2.drawContours(result, valid_defects, -1, (0, 0, 255), 2) return result, len(valid_defects) # Function call example if __name__ == "__main__": defect_img, defect_count = wavelet_threshold_segmentation("cable_defect.jpg") cv2.imwrite("detection_result.jpg", defect_img) print(f"Number of detected damaged areas: {defect_count}") The specific steps of the adaptive threshold binarization algorithm are as follows: For cable temperature anomaly detection, the algorithm first reads the 16-bit grayscale image output by the infrared thermal imager (grayscale values ​​are linearly correlated with temperature), and removes random noise through a 3×3 median filter; the 16-bit image is mapped to an 8-bit grayscale image to simplify calculation, and a mapping relationship between grayscale values ​​and actual temperature is established (T=0.04×g16+b, b=-20); a 15×15 neighborhood window is used to traverse the image, and the local mean of each pixel is calculated. A temperature difference constant C=15 is set (corresponding to an actual temperature rise of 2~3℃). If the pixel grayscale value is higher than the local mean + C, it is marked as a suspected hot spot, generating a preliminary binary image; morphological processing is performed on the binary image: first, isolated noise is removed by erosion using a 3×3 rectangular structuring element, and then adjacent hot spots are connected by expansion using a 5×5 rectangular structuring element; finally, connected regions are detected, and regions with an area ≥100 pixels are selected as effective temperature anomaly regions. The actual temperature is calculated based on the temperature mapping relationship, and the result is output. Partial code examples are as follows: import cv2 import numpy as np def adaptive_threshold_thermal(image_path, k=0.04, b=-20): # Step 1: Image Preprocessing thermal_img = cv2.imread(image_path, cv2.IMREAD_ANYDEPTH) # Read 16-bit infrared image # Median filtering for noise reduction median_blur = cv2.medianBlur(thermal_img, 3) # Step 2: Grayscale Conversion and Temperature Mapping g_min = np.min(median_blur) g_max = np.max(median_blur) g_8 = ((median_blur - g_min) / (g_max - g_min) * 255).astype(np.uint8) # Step 3: Adaptive Threshold Calculation window_size = 15 C = 15# corresponds to an actual temperature difference of 2-3℃. rows, cols = g_8.shape B2 = np.zeros((rows, cols), dtype=np.uint8) # Boundary Fill pad_size = window_size / / 2 padded = cv2.copyMakeBorder(g_8, pad_size, pad_size, pad_size, pad_size, cv2.BORDER_REPLICATE) for i in range(rows): for j in range(cols): # Extract Neighborhood Window window = padded[i:i+window_size, j:j+window_size] local_mean = np.mean(window) if g_8[i, j]>local_mean + C: B2[i, j] = 255 # Step 4: Morphological Processing kernel_erode = np.ones((3, 3), np.uint8) kernel_dilate = np.ones((5, 5), np.uint8) B_erode = cv2.erode(B2, kernel_erode, iterations=1) B_dilate = cv2.dilate(B_erode, kernel_dilate, iterations=1) # Step 5: Connectivity Analysis and Area Verification contours, _ = cv2.findContours(B_dilate, cv2.RETR_EXTERNAL,cv2.CHAIN_APPROX_SIMPLE) valid_hotspots = [] hotspot_temps = [] T_temp = 100# Area threshold (pixels) For contour in contours: area = cv2.contourArea(contour) if area>= T_temp: valid_hotspots.append(contour) # Calculate the average temperature of the region mask = np.zeros_like(thermal_img) cv2.drawContours(mask, [contour], -1, 255, -1) mean_gray = np.mean(thermal_img[mask == 255]) mean_temp = k * mean_gray + b hotspot_temps.append(mean_temp) # Drawing Results result = cv2.cvtColor(g_8, cv2.COLOR_GRAY2BGR) cv2.drawContours(result, valid_hotspots, -1, (0, 255, 0), 2) # Temperature label for i, (contour, temp) in enumerate(zip(valid_hotspots, hotspot_temps)): x, y = contour[0][0] cv2.putText(result, f"Temp: {temp:.1f}℃", (x, y), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 0, 0), 1) return result, len(valid_hotspots), hotspot_temps # Function call example if __name__ == "__main__": hotspot_img, hotspot_count, temps = adaptive_threshold_thermal("thermal_image.tif") cv2.imwrite("hotspot_detection.jpg", hotspot_img) print(f"Number of detected temperature anomaly areas: {hotspot_count}") print(f"Temperature of abnormal area: {temps}") Example 3 In another preferred embodiment, based on Embodiment 2 above, this embodiment provides a cable defect detection method for cable trench inspection using a multimodal dual-stream algorithm, used to verify the implementation effect of the present invention. The simulated cable trench experimental environment has dimensions of 100cm × 40cm × 85cm (length × width × height). Figure 7 As shown, the internal setup includes the following scenarios: 1. Cable layout: Due to laboratory safety concerns, a cable tube is wrapped around a foam rod to simulate a high-voltage cable. Four layers are laid, arranged horizontally, and dust and oil stains are attached to the surface to simulate actual pollution.

[0074] 2. Defect setup: Artificially create three types of insulation layer defects (micro-crack: 5mm in length, 0.2mm in width; surface peeling: 20mm² in area; conductor exposure: 3mm in diameter); simulate abnormal temperature defects using a hot water cup.

[0075] To address the application requirements of cable trench inspection robots performing real-time cable defect detection and localization in complex environments, the system needs to complete multi-source information processing, including high-resolution image acquisition, thermal infrared data fusion, and gas sensor data fusion, within a limited space. The computing unit must possess the following characteristics: high computing power and real-time performance, supporting real-time inference of defect detection algorithms at over 30 FPS; low power consumption and high reliability, meeting the robot's battery power requirements and ensuring stable operation over extended periods; rich I / O interfaces, supporting peripherals such as visible light sensors, infrared imaging modules, temperature and humidity sensors, and gas concentration sensors; and software ecosystem compatibility: running Ubuntu 20.04 / 22.04 and the ROS2Humble framework, supporting AI (Artificial Intelligence) acceleration libraries such as TensorRT (TensorRuntime), CUDA (Compute Unified Device Architecture), and cuDNN (NVIDIA CUDA® Deep Neural Network library). Therefore, this system selects the NVIDIA Jetson AGX Orin as the vehicle-mounted embedded computing unit. Figure 8 As shown in Table 1, its main performance indicators are as follows.

[0076] Table 1 Main Performance Indicators

[0077] Cable defects and temperature anomalies were detected in a laboratory environment and compared with traditional methods (traditional Otsu method (global threshold segmentation), Canny algorithm based on edge detection, and manual inspection). The results of insulation layer damage detection are shown in Table 2, and the results of temperature anomaly detection are shown in Table 3.

[0078] Table 2 Comparison of Insulation Layer Damage Detection Performance

[0079] Note: The processing time for a single frame during manual inspection is calculated based on the actual inspection efficiency (approximately 10 seconds / meter).

[0080] Table 3 Comparison of Temperature Anomaly Detection Performance

[0081] The method in this embodiment achieves an average accuracy of 94.8% in insulation layer damage detection and 96.5% in temperature anomaly detection, significantly higher than traditional algorithms and manual inspection. The single-frame processing time is controlled within 50ms (38ms for insulation layer damage and 42ms for temperature anomalies), meeting the real-time requirements of robot inspection (inspection speed 1km / h). The false detection rate is ≤2.0%, and the missed detection rate is ≤3.2%, superior to traditional algorithms, and requires no manual intervention, reducing maintenance costs. Cable anomaly detection results are as follows... Figure 9 As shown.

[0082] In summary, the cable defect detection method proposed in this embodiment effectively solves the problems of insufficient detection accuracy and poor robustness in existing technologies through dual-modal imaging and dual-pipeline algorithm design. By organically combining wavelet transform threshold segmentation and adaptive threshold binarization, this embodiment not only achieves effective separation of complex backgrounds and low-contrast targets at the image segmentation level, but also ensures the reliability and practicality of the results at the defect determination level through topological analysis and area constraint mechanisms. This embodiment has broad application prospects in engineering practice, and can promote the transformation of power system inspection from manual to intelligent and automated methods, providing strong protection for the safe operation of power infrastructure.

[0083] In addition to being applicable to cable trench inspection, this embodiment can also be extended to defect detection in similar scenarios: Cable tunnel inspection: For tunnel environments with long distances, high humidity and multiple cable layouts, the multimodal imaging system and detection algorithm can be directly reused, only requiring adjustment of the pan-tilt angle to adapt to the tunnel height (3~5m). Overhead cable inspection: A multimodal imaging system is deployed on drones to perform aerial inspections of overhead cables, targeting defects such as insulation damage and joint overheating. By adjusting the camera focal length (12mm) and threshold parameters (area threshold expanded to 1000 pixels), aerial inspections can be achieved. Cable inspection inside switchgear: Adapted to the enclosed and low-light environment of switchgear, it uses a miniaturized multimodal imaging module (6mm lens focal length) to detect insulation damage and temperature abnormalities of cable terminals inside the switchgear. Optical cable inspection: To address defects such as sheath damage and exposed optical fibers in optical cables, the wavelet decomposition scale (j=2) and area threshold (S∈[30, 3000]) are adjusted to achieve high-precision detection of optical cable defects; Industrial pipeline inspection: Extended to the detection of surface cracks, corrosion damage and temperature anomalies in industrial pipelines (such as steam pipelines and gas pipelines), only requiring modification of temperature mapping parameters and defect area thresholds.

[0084] In a preferred embodiment, the multimodal imaging system includes a visible light camera and a thermal imager. As the inspection robot moves along a predetermined path, the visible light camera continuously acquires visible light images f(x, y) of the cable surface at a preset frame rate. This configuration ensures high-precision capture of surface defects, stains, and other appearance issues on the cable. The thermal imager simultaneously acquires temperature distribution data T(x, y) of the cable surface. Through dual-modal data fusion analysis, potential overheating hazards and internal insulation defects are accurately located.

[0085] In the preferred embodiment, the specific process of wavelet transform multi-scale decomposition in step 1 is as follows: A mother wavelet function ψ(a, b) is selected, and a scale factor a and a translation factor b are set. A wavelet transform is then performed on the visible light image f(x, y) to obtain low-frequency components Lj and high-frequency components Hj at different scales j. The low-frequency component Lj preserves the overall outline and brightness distribution of the cable, while the high-frequency component Hj highlights abrupt changes in the cable insulation layer, such as edges, cracks, and peeling. This setup ensures comprehensive and accurate feature extraction of the cable image, providing a reliable basis for subsequent image processing and analysis. By separating the high and low frequency components at different scales, the cable condition can be analyzed in more detail, and the location and characteristics of potential defects can be accurately identified.

[0086] In the preferred embodiment, the specific criterion for multi-scale candidate threshold extraction in step 1 is: in the wavelet decomposition coefficient sequence, if a certain coefficient... At the same time, it is smaller than its adjacent coefficients. , If the gray value corresponding to the coefficient is a local minimum of the image histogram, it is used as a candidate threshold for segmentation. Through repeated detection across scales, a candidate threshold set {T1, T2, ..., Tn} is obtained. This setting effectively avoids the misjudgment problem of local minima at a single scale and improves the robustness of the candidate thresholds through multi-scale fusion. Step 2 further optimizes the entropy value of the set {T1, T2, ..., Tn} to select the threshold that maximizes the inter-class variance as the final segmentation point.

[0087] In the preferred embodiment, the specific process for selecting the optimal threshold for maximizing the inter-class variance in step 1 is as follows: Calculate the inter-class variance after dividing the image pixels using the candidate threshold T, wherein the inter-class variance expression satisfies the condition based on the proportion of pixels in the two classes. , and the average gray level of two types of pixels , The variance calculation logic selects the candidate threshold that maximizes the inter-class variance as the optimal threshold. And based on the optimal threshold Image binarization yields a binary image B(x, y); the above settings effectively improve image segmentation accuracy, especially suitable for scenes with significant gray-level differences between the target and background. Calculating the inter-class variance by iterating through all candidate thresholds ensures... By achieving global optimality and avoiding interference from local extrema, the final binary image B(x,y) can clearly distinguish between the foreground and background regions.

[0088] In the preferred embodiment, the specific process of the topology analysis in step 1 is as follows: perform contour detection on the binary image B(x, y) to extract all closed contours; construct the parent-child hierarchy of contours, and determine small closed inner contours nested inside the outer contour of the cable body as suspected damaged contours; calculate the area S of the suspected damaged contours, and only when the area S is within a preset range... When the area is confirmed as a valid area of ​​insulation layer damage, the above settings can effectively distinguish between actual damage and image noise interference, improving detection accuracy. Step 2 then uses image registration technology to locate the corresponding position in the original color image for the confirmed damaged area, further analyzing the morphological characteristics of the damaged edge to help determine the severity of the damage.

[0089] In a preferred embodiment, the preset area range The value of needs to meet the requirements of filtering isolated noise points and eliminating invalid large areas, among which A threshold of not less than 3 times the area of ​​a single noisy pixel. The threshold is set to not exceed 1 / 20 of the cable body outline area. This setting ensures that during image processing, minute noise interference can be accurately filtered out to avoid it being misjudged as a valid feature, while also preventing the incorrect removal of relevant areas of the cable body. This ensures that subsequent analysis is based on accurate and effective image data, thereby improving the overall detection accuracy and reliability.

[0090] In the preferred embodiment, step 2, the specific process of infrared thermal imaging image preprocessing, involves converting the pseudo-color infrared image acquired by the thermal imager into a grayscale image g(x, y). The grayscale values ​​of the grayscale image g(x, y) have a linear correlation with the actual cable temperature. This setting ensures that the image temperature information accurately reflects the actual temperature distribution of the cable. Next, the grayscale image g(x, y) is filtered and denoised to eliminate noise interference, improve image quality, and provide a reliable data foundation for subsequent accurate extraction of cable temperature features and fault diagnosis.

[0091] In the preferred embodiment, the specific process of the local neighborhood adaptive threshold determination in step 2 is as follows: a neighborhood window W(x,y) is constructed for each pixel in the grayscale image g(x,y), and the local mean within the neighborhood window is calculated; if the grayscale value g(x,y) of the center pixel is greater than the sum of the local mean and the constant C, it is determined to be a suspected temperature anomaly point, otherwise it is determined to be background; the value of the constant C is matched with the sensitivity of cable temperature rise monitoring to ensure that a local temperature rise of 2℃~3℃ can be detected; the above settings can effectively distinguish between real temperature anomalies and background noise interference. The detection accuracy can be further optimized by dynamically adjusting the neighborhood window size W(x,y). When the number of pixels in the window increases, the stability of the local mean calculation is improved, but the compensation coefficient of the constant C needs to be adjusted synchronously to maintain the sensitivity range of 2℃~3℃.

[0092] In the preferred embodiment, the value range of the constant C is the grayscale value range corresponding to a temperature rise of 2℃ to 3℃ in the grayscale image, specifically 5 to 12 grayscale levels. This setting ensures that subtle temperature rise changes can be effectively distinguished during image processing, while avoiding noise interference caused by too many grayscale levels. In practical applications, the parameters can be flexibly adjusted within this range to optimize the detection effect according to the equipment accuracy and environmental conditions.

[0093] In the preferred embodiment, the specific process of morphological processing in step 2 is as follows: First, an erosion operation is performed on the binary image composed of suspected temperature anomalies to remove isolated noise white points; then, a dilation operation is performed to connect adjacent temperature anomaly pixels, forming continuous hotspot regions. This setting effectively improves the accuracy of hotspot region detection and avoids misjudgments caused by noise interference or pixel dispersion. Simultaneously, by adjusting the kernel size and iteration count of the erosion and dilation operations, the processing effect can be further optimized to adapt to the temperature anomaly monitoring needs in different scenarios. Continuous hotspot regions provide a more reliable data foundation for subsequent temperature anomaly level assessment, helping to accurately locate the location of potential high-temperature hazards.

[0094] In the preferred embodiment, step 2, the specific process of connected component analysis and area threshold determination, is as follows: Connected component analysis is performed on the morphologically processed image to extract candidate anomaly regions and calculate their areas S; when the area S ≥ the temperature anomaly area threshold... At that time, it was identified as an area of ​​effective temperature anomaly. This is a preset minimum hotspot area threshold with practical physical meaning; the above settings effectively eliminate false detection areas caused by image noise or minor interference, ensuring the accuracy of the detection results. Furthermore, it can be flexibly adjusted according to the actual needs of different application scenarios. Values ​​are adjusted to accommodate varying sensitivity requirements in areas of temperature anomalies.

[0095] In the preferred embodiment, the detection process of the dual-pipeline algorithm framework runs in real time on the vehicle-mounted embedded computing unit of the inspection robot. The overall processing time for each frame of image is controlled within 50 milliseconds, meeting the online detection requirements of the inspection robot at its normal inspection speed. This configuration ensures both detection accuracy and improved inspection efficiency, enabling the robot to quickly identify equipment anomalies. Furthermore, the algorithm framework also features adaptive adjustment capabilities, allowing for parameter optimization based on different scenarios, further enhancing the stability and reliability of the detection.

[0096] In the preferred embodiment, step 3 involves the following comprehensive monitoring method: when both an insulation layer damage area and a temperature anomaly area are detected simultaneously, or when the area of ​​a single defect exceeds a preset alarm threshold, the audible and visual alarm module of the inspection robot is triggered, and the defect location and type information are uploaded to the power operation and maintenance management platform. This setup ensures that defects are detected and addressed promptly, preventing the escalation of faults. Simultaneously, maintenance personnel can quickly obtain accurate information through the platform, plan emergency repairs in advance, and rush to the site with appropriate tools and spare parts, greatly improving the efficiency and safety of power operation and maintenance.

[0097] In summary, the cable defect detection method proposed in this embodiment, through its dual-modal imaging and dual-pipeline algorithm design, effectively overcomes the problems of poor detection accuracy and weak robustness in existing technologies. By cleverly integrating wavelet transform threshold segmentation and adaptive threshold binarization, this embodiment not only achieves effective differentiation of complex backgrounds and low-contrast targets in the image segmentation stage, but also ensures the accuracy and practicality of the results in the defect determination stage through topological analysis and area constraint mechanisms. This embodiment shows great promise in engineering applications, helping to transform power system inspection from a manual mode to an intelligent and automated mode, thus building a solid defense for the stable operation of power infrastructure.

[0098] In terms of technological innovation, this method utilizes a multimodal visual sensing and dual-pipeline algorithm architecture to integrate visible light and infrared thermal imaging, achieving complementary detection of physical defects and thermal defects related to cable insulation layer damage. This eliminates detection blind spots, unlike traditional single-modal or single-defect-type detection methods. Multi-scale decomposition technology based on wavelet transform is used for insulation layer damage detection, enabling the separation of the macroscopic structure of the cable body from the detailed features of the damaged area at different scales, achieving high-precision separation and overcoming the problem of traditional global threshold segmentation failing under complex texture interference. Simultaneously, topological structure analysis and area interval filtering mechanisms are introduced to filter out false defects in insulation layer damage detection, accurately identifying the real damaged areas and effectively eliminating false defects formed by dust, oil, and natural textures on the cable surface, thus improving detection accuracy. Furthermore, an adaptive threshold thermal imaging detection method based on local statistics is employed to accurately capture early temperature anomalies in the cable, eliminating the dependence of traditional global threshold algorithms on bimodal / multimodal histograms and stably identifying early, minute temperature rises of only 2℃~3℃ in the cable. Furthermore, by employing a morphological process of corrosion followed by expansion, the reliability of temperature anomaly detection results is optimized, isolated false hot spots in infrared thermal imaging are eliminated, and adjacent temperature anomaly pixels are connected to form continuous hot spot regions.

[0099] In terms of technological contributions, this method overcomes the limitations of traditional manual inspections, which suffer from low safety and efficiency, as well as the susceptibility of existing automated inspections to failure under low signal-to-noise ratio and low-contrast images. It achieves non-contact, high-precision inspection, ensuring the safety of inspection personnel, eliminating inspection "downtime," and adapting to complex and harsh environments. The constructed dual-pipeline inspection framework not only achieves full coverage detection of two types of core defects but also runs on an in-vehicle embedded computing unit, controlling the image processing time per frame to within 50 milliseconds, meeting the engineering requirements of online real-time inspection while avoiding interference with power supply. Quantitative determination of key parameters in defect areas gives the inspection results clear physical meaning and allows for the retention of image and data records, improving the practicality and traceability of the inspection results and providing a basis for subsequent defect cause analysis and maintenance strategy optimization. Synchronizing detection data to the power operation and maintenance management platform provides precise data support for predictive maintenance and proactive operation and maintenance of the power system, promoting the transformation of cable trench operation and maintenance from "human experience-driven" to "intelligent data-driven." It possesses strong engineering scalability, facilitating large-scale deployment in power infrastructure across various scenarios. By comprehensively utilizing multiple innovative technologies and methods, it revolutionizes multiple levels from detection methods and result processing to operation and maintenance model transformation, significantly reducing the probability of major power accidents, comprehensively improving the level and effectiveness of cable defect detection in cable trenches, and providing all-round protection for the safe and stable operation of the power system.

Claims

1. A cable defect detection method for cable trench inspection using a multimodal dual-stream algorithm, characterized in that, The method uses a cable trench inspection robot as a carrier and acquires visible light and infrared thermal images of cables in the cable trench in real time through a multimodal imaging system. Based on a dual pipeline algorithm framework, it realizes the detection of cable insulation layer damage and cable temperature anomaly, respectively. Specifically, it includes the following steps: Step 1: Perform the insulation layer damage detection process on the visible light image, and sequentially complete wavelet transform multi-scale decomposition, multi-scale candidate threshold extraction, inter-class variance maximization optimal threshold selection, binarization processing, topology analysis and area screening to obtain the insulation layer damage detection results; Step 2: Perform the temperature anomaly detection process on the infrared thermal imaging image, sequentially completing pseudo-color to grayscale preprocessing, local neighborhood adaptive threshold determination, morphological processing, connected region analysis and area threshold determination, to obtain the temperature anomaly detection result; Step 3: Integrate the detection results from Step 1 and Step 2 to achieve comprehensive monitoring of cable defects.

2. The cable defect detection method for cable trench inspection using the multimodal dual-stream algorithm according to claim 1, characterized in that: The multimodal imaging system includes a visible light camera and a thermal imager. When the visible light camera moves along a predetermined path with the inspection robot, it continuously acquires visible light images f(x, y) of the cable surface at a preset frame rate.

3. The cable defect detection method for cable trench inspection using the multimodal dual-stream algorithm according to claim 1, characterized in that, In step 1, the specific process of wavelet transform multi-scale decomposition is as follows: select the mother wavelet function ψ(a, b), set the scale factor a and the translation factor b, perform wavelet transform on the visible light image f(x, y) to obtain the low-frequency component Lj and the high-frequency component Hj at different scales j. The low-frequency component Lj preserves the overall outline and brightness distribution of the cable, while the high-frequency component Hj highlights the abrupt changes in the cable insulation layer, such as edges, cracks, and peeling.

4. The cable defect detection method for cable trench inspection using the multimodal dual-stream algorithm according to claim 3, characterized in that, The formula for the wavelet transform is: （1）。 5. The cable defect detection method for cable trench inspection using the multimodal dual-stream algorithm according to claim 3, characterized in that, In step 1, the specific criterion for multi-scale candidate threshold extraction is: in the wavelet decomposition coefficient sequence, if a certain coefficient... At the same time, it is smaller than its adjacent coefficients. , That is, satisfying: （2）； The gray value corresponding to the coefficient is a local minimum of the image histogram, which is used as a candidate threshold for segmentation. After repeated detection across scales, a candidate threshold set {T1, T2, ..., Tn} is obtained.

6. The cable defect detection method for cable trench inspection using the multimodal dual-stream algorithm according to claim 4, characterized in that, In step 1, the specific process for selecting the optimal threshold to maximize the inter-class variance is as follows: calculate the inter-class variance after dividing the image pixels with the candidate threshold T, and the inter-class variance expression satisfies the condition based on the proportion of pixels in the two classes. , and the average gray level of two types of pixels , The variance calculation logic selects the candidate threshold that maximizes the inter-class variance as the optimal threshold. And based on the optimal threshold Image binarization yields a binary image B(x, y).

7. The cable defect detection method for cable trench inspection using the multimodal dual-stream algorithm according to claim 6, characterized in that, The inter-class variance The expressions for the binary image B(x, y) are as follows: （3）； （4）。 8. The cable defect detection method for cable trench inspection using the multimodal dual-stream algorithm according to claim 6, characterized in that, In step 1, the specific process of the topology analysis is as follows: perform contour detection on the binary image B(x, y) to extract all closed contours; construct the parent-child hierarchy of contours, and determine small closed inner contours nested inside the outer contour of the cable body as suspected damaged contours; calculate the area S of the suspected damaged contours, and only when the area S is within a preset range... At that time, it was confirmed as an area of ​​effective insulation layer damage.

9. The cable defect detection method for cable trench inspection based on the multimodal dual-stream algorithm according to claim 8, characterized in that, The preset area range The value of needs to meet the requirements of filtering isolated noise points and eliminating invalid large areas, among which A threshold of not less than 3 times the area of ​​a single noisy pixel. The threshold is no more than 1 / 20 of the cable body outline area.

10. The cable defect detection method for cable trench inspection using the multimodal dual-stream algorithm according to claim 1, characterized in that, In step 2, the specific process of infrared thermal imaging image preprocessing is as follows: the pseudo-color infrared image acquired by the thermal imager is converted into a grayscale image g(x, y), and the grayscale value of the grayscale image g(x, y) has a linear correspondence with the actual temperature of the cable.

11. The cable defect detection method for cable trench inspection using the multimodal dual-stream algorithm according to claim 10, characterized in that, In step 2, the specific process of the local neighborhood adaptive threshold determination is as follows: construct a neighborhood window W(x,y) for each pixel in the grayscale image g(x,y), and calculate the local mean within the neighborhood window; if the grayscale value g(x,y) of the center pixel is greater than the sum of the local mean and the constant C, it is determined to be a suspected temperature anomaly point, otherwise it is determined to be background; wherein the value of the constant C is matched with the sensitivity of cable temperature rise monitoring to ensure that a local temperature rise of 2℃~3℃ can be detected.

12. The cable defect detection method for cable trench inspection based on the multimodal dual-stream algorithm according to claim 11, characterized in that: The value range of the constant C is the gray value range of the corresponding gray image with a temperature rise of 2℃~3℃, specifically 5~12 gray levels.

13. The cable defect detection method for cable trench inspection using the multimodal dual-stream algorithm according to claim 1, characterized in that, In step 2, the specific process of the morphological processing is as follows: first, perform an erosion operation on the binary image composed of suspected temperature anomalies to remove isolated noise white points; then, perform a dilation operation to connect adjacent temperature anomaly pixels to form a continuous hotspot region.

14. The cable defect detection method for cable trench inspection using the multimodal dual-stream algorithm according to claim 1, characterized in that, In step 2, the specific process of connected component analysis and area threshold determination is as follows: Perform connected component analysis on the morphologically processed image, extract candidate abnormal regions and calculate their area S; When the area S ≥ the temperature anomaly area threshold At that time, it was identified as an area of ​​effective temperature anomaly. This is a preset minimum hotspot area threshold with actual physical significance.

15. The cable defect detection method for cable trench inspection based on the multimodal dual-stream algorithm according to claim 1, characterized in that: The detection process of the dual pipeline algorithm framework runs in real time on the vehicle-mounted embedded computing unit of the inspection robot. The overall processing time for each frame of image is controlled within 50 milliseconds, which meets the online detection requirements of the inspection robot at normal inspection speed.

16. The cable defect detection method for cable trench inspection using the multimodal dual-stream algorithm according to claim 1, characterized in that, In step 3, the specific method of comprehensive monitoring is as follows: when a damaged area of ​​the insulation layer and an abnormal temperature area are detected at the same time, or when the area of ​​a single defect area exceeds the preset alarm threshold, the sound and light alarm module of the inspection robot is triggered, and the defect location and type information are uploaded to the power operation and maintenance management platform.

Images