Power channel foreign matter intelligent detection system and method based on multi-scale Retinex algorithm

Abstract

The invention provides a power channel foreign matter intelligent detection system and method based on a multi-scale Retinex algorithm, and the system comprises an image processing module which is used for processing a real-time remote sensing image through the multi-scale Retinex algorithm, and obtaining an enhanced image; the intelligent detection module is used for detecting the enhanced image by using a foreign matter detection model and outputting a space coordinate of a target foreign matter and a corresponding hidden danger type; the hidden danger risk assessment module is used for generating a dynamic risk assessment value of the target foreign body based on the space coordinate of the target foreign body, the corresponding hidden danger type and the power safety hidden danger space-time distribution rule; and the early warning visualization module is used for performing visual early warning display on the risk foreign matter mark on the real-time remote sensing image when the dynamic risk assessment value is greater than or equal to a preset threshold value. The collected power channel image is preprocessed through a multi-scale Retinex algorithm, the influence of uneven illumination is eliminated, the image quality is improved, and accurate detection of the power channel foreign matter is achieved in combination with an intelligent recognition algorithm.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power equipment detection, and in particular to a power channel foreign matter intelligent detection system and method based on a multi-scale Retinex algorithm. BACKGROUND

[0002] In modern society, electricity has become the core energy supporting human production and life. In daily life, from home lighting, household appliance operation to urban traffic signal maintenance, electricity supply is indispensable; in the industrial field, electricity drives the efficient operation of the production line, and is the cornerstone of the stable operation of key industries such as manufacturing and energy; in the communication field, the power guarantee of the base station determines the continuity and stability of information transmission. Therefore, the reliability of electricity supply is directly related to social order stability, economic development vitality and people's living quality.

[0003] As the key carrier of power transmission, the safe operation of the power channel is of great importance. However, with the acceleration of urbanization and the increase of agricultural production activities, the power channel faces many potential threats: during temporary construction, large machinery such as cranes and tower cranes may accidentally touch the power line or construction materials may accidentally fall on the line to form foreign matter; in the greenhouse planting area, lightweight materials such as plastic film and sunshade net are easily blown up and wrapped around the power line in windy weather; the construction of illegal buildings not only occupies the space of the power channel, but also may cause the power line to be squeezed or collided by the construction materials and equipment, and at the same time, the accumulation of sundries around the illegal buildings also easily forms power foreign matter. Once these foreign matters enter the power channel, they may cause line short circuit, trip and other faults, which may cause local power outage, affect normal life and production, and even cause serious safety accidents such as fire, resulting in huge economic losses and social impact.

[0004] At present, the detection of power channel foreign matter mainly adopts manual inspection and traditional image recognition technology. Manual inspection is low in efficiency and high in labor intensity, and it is difficult to realize real-time and comprehensive monitoring of the power channel, and the inspection personnel have high safety risks in complex environments; the traditional image recognition technology is easily affected by image quality under complex lighting conditions, and at the same time, in the face of dynamic changes of temporary construction scenes, complex backgrounds of greenhouse planting areas and chaotic environments around illegal buildings, the accuracy of foreign matter detection is low, which is difficult to meet the actual application requirements. Therefore, there is an urgent need for a technology that can accurately and efficiently detect power channel foreign matter in complex environments. SUMMARY

[0005] The application provides a power channel foreign matter intelligent detection system and method based on a multi-scale Retinex algorithm, which solves the above problems. The collected power channel image is preprocessed through the multi-scale Retinex algorithm, the influence of uneven illumination is eliminated, the image quality is improved, and the foreign matter detection model is optimized for the foreign matter detection difficulties in complex scenes such as temporary construction, greenhouse planting and illegal buildings, so that accurate detection of the power channel foreign matter is realized.

[0006] The application provides a power channel foreign matter intelligent detection system based on a multi-scale Retinex algorithm, which comprises:

[0007] An image processing module is configured to process real-time remote sensing images through a multi-scale Retinex algorithm to obtain enhanced images.

[0008] An intelligent detection module is configured to detect the enhanced images by using a foreign matter detection model and output the spatial coordinates of the detected target foreign matter and the corresponding hazard types.

[0009] A hazard risk assessment module is configured to generate a dynamic risk assessment value of the target foreign matter to the power channel based on the spatial coordinates of the target foreign matter, the corresponding hazard types and the spatiotemporal distribution law of power safety hazards.

[0010] A warning visualization module is configured to mark the target foreign matter as a risk foreign matter when the dynamic risk assessment value is greater than or equal to a preset threshold, and visually warn and display the risk foreign matter mark on the real-time remote sensing images.

[0011] Preferably, in a power channel foreign matter intelligent detection system based on a multi-scale Retinex algorithm, the image processing module comprises:

[0012] An image acquisition unit is configured to acquire a plurality of real-time remote sensing images of a target detection area.

[0013] An image denoising unit is configured to perform denoising processing on the plurality of real-time remote sensing images based on a preset filtering algorithm.

[0014] An image enhancement unit is configured to perform multi-scale decomposition on the denoised real-time remote sensing images to obtain a plurality of image layers of different scales.

[0015] Based on the Retinex algorithm, each image layer is processed to obtain a plurality of light uniform image layers, and the plurality of light uniform image layers are fused to obtain an enhanced image.

[0016] Preferably, in a power channel foreign matter intelligent detection system based on a multi-scale Retinex algorithm, the intelligent detection module comprises:

[0017] The model training unit is configured to train a preset deep learning module based on the multi-scene image training set, and generate the foreign matter detection model.

[0018] The foreign matter detection unit is configured to detect the enhanced image based on the foreign matter detection model, virtually mark the detected target foreign matter on the remote sensing image, and determine the foreign matter type of the marked target foreign matter.

[0019] Meanwhile, the spatial coordinates of the target foreign matter are output according to the virtual marking result.

[0020] The list generation unit is configured to generate a foreign matter positioning list in the target detection area based on the spatial coordinates of the target foreign matter and the corresponding foreign matter type, and send the list to the power operation and maintenance center for display.

[0021] Preferably, in the power channel foreign matter intelligent detection system based on the multi-scale Retinex algorithm, the model training unit comprises:

[0022] The data acquisition subunit is configured to acquire data based on big data acquisition technology for various abnormal scenes and normal scenes of the power channel.

[0023] The training set generation subunit is configured to establish corresponding scene image data sets after cleaning the acquired massive images.

[0024] According to the acquisition keywords and the acquisition image power channel positions, the images in each scene image data set are marked with foreign matters, and combined with artificial quantitative quality inspection, the correctly labeled multiple scene image data sets are obtained.

[0025] The images in the correctly labeled multiple scene image data sets are cross-combined to generate multiple multi-scene image sets, and the corresponding multiple multi-scene image training sets are generated based on the correctly labeled multiple multi-scene image data sets and the multiple multi-scene image sets.

[0026] The model training subunit is configured to continuously train the preset learning model according to each multi-scene image training set until the optimal solution is reached, and output the foreign matter detection model.

[0027] Preferably, in the power channel foreign matter intelligent detection system based on the multi-scale Retinex algorithm, the foreign matter detection unit further comprises:

[0028] The channel foreign matter detection subunit is configured to use the foreign matter detection model to focus on detecting the power channel on the enhanced image, and when detecting that there is a foreign matter on the power channel or the power channel is in contact with the foreign matter, quickly positioning the position of the foreign matter or the contact position based on the power pole code, obtaining the operation and maintenance coordinates, and sending the operation and maintenance coordinates to the early warning visualization module for visual early warning.

[0029] A contact foreign matter tracking subunit is configured to continuously focus on the location of the foreign matter or the contact location before the maintenance personnel arrive at the scene when the foreign matter is detected on the power channel or the power channel is in contact with the foreign matter.

[0030] Preferably, in the power channel foreign matter intelligent detection system based on the multi-scale Retinex algorithm, the hidden danger risk assessment module comprises:

[0031] A distribution law analysis unit is configured to collect foreign matter-channel influence data of foreign matters in multiple scenes, and perform feature extraction on the foreign matter-channel influence data to obtain spatio-temporal distribution features corresponding to foreign matters in different scenes;

[0032] Based on the spatio-temporal distribution features, the severity of the damage of foreign matters in different scenes to the power channel under different solar terms and climates is determined, and a time weight factor of the corresponding scene foreign matter is generated;

[0033] A position distribution determination unit is configured to determine the distribution features of power lines in the surrounding area of the target foreign matter and the index coefficient of the importance degree of the corresponding lines based on the spatial coordinates of the target foreign matter and the channel position coordinate cluster of the power channel in the target detection area;

[0034] A first evaluation unit is configured to determine the relative distance between the target foreign matter and each directional power channel based on the power line distribution features, and obtain distance coefficients of multiple directions of the target foreign matter according to the relative distance and the minimum suitable distance;

[0035] After calculating the quotient value between the maximum relative distance and the relative distance and performing normalization processing, the risk weight factors of multiple directions are obtained, and the distance risk coefficient of the target foreign matter is obtained according to the distance coefficients of multiple directions and the corresponding risk weight factors;

[0036] Based on the distance risk coefficient and the index coefficient, the basic risk index of the target foreign matter is calculated, and the basic risk index of the target foreign matter in the current time period is obtained in combination with the time weight factor;

[0037] An intelligent setting unit is configured to mark the target foreign matter as a segment key detection foreign matter in the current time when the basic risk index is greater than a preset value;

[0038] Otherwise, the target foreign matter is marked as an ordinary detection foreign matter in the current time period;

[0039] And based on the key detection foreign matter and the ordinary detection foreign matter marking, the detection frequency of each target foreign matter in the current time period is set respectively.

[0040] Preferably, in the power channel foreign matter intelligent detection system based on the multi-scale Retinex algorithm, the hidden danger risk assessment module further comprises:

[0041] a second evaluation unit configured to determine, based on the hazard type corresponding to each target foreign matter, a power hazard breeding path of each target foreign matter in combination with the preset foreign matter database;

[0042] acquire a current presentation state of the target foreign matter;

[0043] when the similarity between the current presentation state and any link on the power hazard breeding path thereof is greater than or equal to a similarity threshold, generate a dynamic risk assessment value of the target foreign matter to the power channel in combination with the stage order of the current presentation state on the corresponding power hazard breeding path and the corresponding basic risk index.

[0044] Preferably, in a power channel foreign matter intelligent detection system based on a multi-scale Retinex algorithm, the distribution rule analysis unit comprises:

[0045] a weight factor generation subunit configured to collect annual historical power channel interference data of various hazard types in the target detection area, analyze the annual historical power channel interference data, and determine the comprehensive power hazard score corresponding to different scene foreign matters in different regions, the hazard score of each period, the annual incident frequency, and the incident frequency of each period, respectively;

[0046] based on the hazard score of different scene foreign matters in the current period and the comprehensive power hazard score, a first hazard severity corresponding to various scene foreign matters in the current period is calculated;

[0047] based on the incident frequency and the annual incident frequency of different scene foreign matters in the current period, a second hazard severity corresponding to various scene foreign matters in the current period is calculated;

[0048] and according to the annual historical power channel interference data, the regional incident proportion of different scene foreign matters in each sub-region in the target detection area is determined;

[0049] based on the regional incident proportion, the first hazard severity, the second hazard severity, and the weighting weight corresponding to the first hazard severity and the second hazard severity, the time weight factor of the corresponding scene foreign matter in different sub-regions is obtained.

[0050] Preferably, in a power channel foreign matter intelligent detection system based on a multi-scale Retinex algorithm, the early warning visualization module comprises:

[0051] a warning unit configured to generate a warning signal according to the hazard type and spatial coordinates or operation and maintenance coordinates of the risk foreign matter and send the warning signal to the power operation and maintenance center and the corresponding regional power special person in charge;

[0052] A visualization unit is configured to display the risk position or the operation and maintenance position highlighted based on the early warning signal on a monitoring terminal or a personal terminal;

[0053] A tracking operation and maintenance unit is configured to detect the position to be maintained, update the operation and maintenance state of the visualized operation and maintenance position after confirming the arrival of the operation and maintenance personnel, and record the arrival time of the operation and maintenance personnel;

[0054] A remote communication unit is configured to communicate online with the operation and maintenance personnel who arrive at the position to be maintained.

[0055] The application provides an intelligent power channel foreign matter detection method based on a multi-scale Retinex algorithm, which comprises the following steps:

[0056] The real-time remote sensing image is processed by the multi-scale Retinex algorithm to obtain an enhanced image;

[0057] The enhanced image is detected by using a foreign matter detection model, and the spatial coordinates of the detected target foreign matter and the corresponding hazard type are outputted;

[0058] Based on the spatial coordinates of the target foreign matter, the corresponding hazard type and the space-time distribution rule of the power safety hazard, a dynamic risk assessment value of the target foreign matter to the power channel is generated;

[0059] When the dynamic risk assessment value is greater than or equal to a preset threshold value, the target foreign matter is marked as a risk foreign matter, and a visual warning display of the risk foreign matter mark is performed on the real-time remote sensing image.

[0060] Compared with the prior art, the application has at least the following beneficial effects:

[0061] The application effectively eliminates the influence of complex lighting conditions, atmospheric interference and other factors on image quality, significantly enhances the contrast of foreign matter and background in the image, and improves the clarity of image details. Compared with the traditional method, the algorithm can more accurately preserve the subtle features in the power channel scene, so that the key information such as the outline and texture of the foreign matter can be clearly presented, providing high-quality data support for the accurate identification of the subsequent intelligent detection module, and greatly improving the accuracy of foreign matter detection. Then, the intelligent detection module uses a pre-trained foreign matter detection model to detect the enhanced image, which can quickly and accurately identify various foreign matters in the power channel and output their spatial coordinates and corresponding risk types. Not only can it efficiently process massive real-time remote sensing image data, but also can accurately classify different types of foreign matter (such as temporary construction equipment, greenhouse film, illegal buildings, etc.), providing an efficient and reliable technical means for power channel hazard investigation. Subsequently, the hazard risk assessment module generates a dynamic risk assessment value based on the spatial coordinates of the target foreign matter, the hazard type and the spatio-temporal distribution rule of power safety hazards. This assessment process fully considers the risk characteristics of the power channel in different time and space dimensions, as well as the severity of the damage that various foreign matters may cause to power facilities, achieving dynamic and quantitative assessment of the risk of power channel hazards, providing a scientific basis for the power operation and maintenance department to develop targeted prevention and control measures, and effectively improving the forward-looking and effectiveness of power channel risk control. Finally, when the dynamic risk assessment value is greater than or equal to the preset threshold, the target foreign matter is marked as a risk foreign matter, and a visual warning display is performed on the real-time remote sensing image, realizing the intuitive and visual presentation of the power channel anomaly, so that the power operation and maintenance personnel can quickly and clearly grasp the specific location, type and risk level of the risk foreign matter in the power channel, greatly improving the efficiency and accuracy of information transmission. Operation and maintenance personnel can make response decisions based on the visual warning results, reasonably allocate resources, and prioritize high-risk hazards, effectively shortening the hazard processing period and reducing the probability of power accidents, ensuring the safe and stable operation of the power system.

[0062] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent to those skilled in the art from the description, or can be learned by practice of the application. The objects and other advantages of the present application can be realized and obtained by the structure particularly pointed out in the application.

[0063] The technical solutions of the present application will be further described in detail below with the help of the drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0064] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0065] Figure 1 This is a schematic diagram of a power channel foreign object intelligent detection system based on the multi-scale Retinex algorithm;

[0066] Figure 2 This is a schematic diagram of the image processing module of an intelligent foreign object detection system for power channels based on the multi-scale Retinex algorithm.

[0067] Figure 3 This is a schematic diagram of the intelligent detection module of a power channel foreign object intelligent detection system based on the multi-scale Retinex algorithm.

[0068] Figure 4 This is a schematic diagram of the hazard risk assessment module of an intelligent foreign object detection system for power channels based on the multi-scale Retinex algorithm.

[0069] Figure 5 This is a schematic diagram of the early warning visualization module of an intelligent foreign object detection system for power channels based on the multi-scale Retinex algorithm.

[0070] Figure 6 This is a flowchart of a method for intelligent detection of foreign objects in power channels based on the multi-scale Retinex algorithm. Detailed Implementation

[0071] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0072] Example 1:

[0073] This invention provides an intelligent foreign object detection system for power channels based on the multi-scale Retinex algorithm, such as... Figure 1 As shown, it includes:

[0074] The image processing module is used to process real-time remote sensing images using the multi-scale Retinex algorithm to obtain enhanced images;

[0075] The intelligent detection module is used to detect enhanced images using a pre-trained foreign object detection model and output the spatial coordinates of the detected target foreign object and its corresponding hazard type.

[0076] The hidden danger risk assessment module is configured to generate a dynamic risk assessment value of the target foreign matter to the power channel based on the spatial coordinates of the target foreign matter, the corresponding hidden danger type of the target foreign matter, and the space-time distribution rule of the power safety hidden danger.

[0077] The early warning visualization module is configured to mark the target foreign matter as a risk foreign matter when the dynamic risk assessment value is greater than or equal to a preset threshold, and to visually display a warning mark of the risk foreign matter on the real-time remote sensing image.

[0078] In this embodiment, the building image of the target detection area can be collected by the unmanned aerial vehicle carrying a shooting device in addition to the real-time remote sensing image.

[0079] The beneficial effects of the above technical solutions are: the image processing module uses a multi-scale Retinex algorithm to process real-time remote sensing images, effectively eliminating the influence of factors such as complex lighting conditions, atmospheric interference, and the like on image quality, significantly enhancing the contrast between foreign objects and the background in the image, and improving the clarity of image details. Compared with traditional methods, this algorithm can more accurately preserve subtle features in the power channel scene, allowing key information such as the outline and texture of foreign objects to be clearly presented, providing high-quality data support for the accurate identification of subsequent intelligent detection modules, and significantly improving the accuracy of foreign object detection. Then, the intelligent detection module uses a pre-trained foreign object detection model to detect the enhanced image, which can quickly and accurately identify various foreign objects in the power channel and output their spatial coordinates and corresponding hazard types, efficiently processing massive real-time remote sensing image data, and accurately classifying different types of foreign objects (such as temporary construction equipment, greenhouse film, illegal buildings, etc.), providing an efficient and reliable technical means for power channel hazard investigation. Subsequently, the hazard risk assessment module analyzes and calculates a dynamic risk assessment value based on the spatial coordinates of the target foreign object, the hazard type, and the spatiotemporal distribution of power safety hazards. This evaluation process takes into account the risk characteristics of the power channel in different time and spatial dimensions, as well as the severity of the damage that various foreign objects may cause to power facilities, enabling dynamic and quantitative assessment of power channel hazard risks, providing a scientific basis for the power operation and maintenance department to develop targeted prevention and control measures, and effectively improving the forward-looking and effectiveness of power channel risk management. Finally, the early warning visualization module marks the target foreign object as a risk foreign object when the dynamic risk assessment value is greater than or equal to a preset threshold, and displays a visual warning on the real-time remote sensing image, enabling intuitive and visual presentation of power channel abnormalities, allowing power operation and maintenance personnel to quickly and clearly understand the specific location, type, and risk level of risk foreign objects in the power channel, greatly improving the efficiency and accuracy of information transmission. Operation and maintenance personnel can respond quickly based on the visual warning results, reasonably allocate resources, and prioritize high-risk hazards, effectively shortening the hazard processing cycle and reducing the probability of power accidents, ensuring the safe and stable operation of the power system.

[0080] Embodiment 2:

[0081] Based on embodiment 1, the image processing module, as shown in Figure 2 , includes:

[0082] An image acquisition unit for acquiring a plurality of real-time remote sensing images of a target detection area;

[0083] An image denoising unit for denoising the plurality of real-time remote sensing images based on a preset filtering algorithm;

[0084] An image enhancement unit is configured to perform multi-scale decomposition on the denoised real-time remote sensing image to obtain image layers of multiple scales;

[0085] Based on the Retinex algorithm, each image layer is processed to obtain multiple light uniform image layers, and the multiple light uniform image layers are fused to obtain an enhanced image.

[0086] In this embodiment, during the fusion of the multiple light uniform image layers, weight adaptive allocation is performed according to the picture contribution degree of each layer.

[0087] The beneficial effects of the above technical solution are as follows: the image acquisition unit acquires multiple real-time remote sensing images of the target detection area to realize real-time acquisition of building information of the target detection area; the image denoising unit processes the real-time remote sensing image based on a preset filtering algorithm (for example, Gaussian filtering, median filtering, wavelet filtering, etc.), effectively avoiding misjudgment caused by noise and ensuring the reliability of the image data. Then, the image enhancement unit uses multi-scale decomposition combined with the Retinex algorithm to analyze and process the image from multiple scales, the low-scale image layer removes large-area light changes, the medium-scale image layer processes medium light changes and textures, and the high-scale image layer retains details. The multi-scale processing mode enables the system to adapt to various complex lighting conditions, effectively improving the image quality whether it is strong light direct, backlight or night scene. Then, the light uniform image layers are fused to fully integrate the information advantages under different scales, enhance the comprehensive performance of the image, more accurately retain and highlight the subtle features in the power channel scene, and clearly present the key information such as the outline and texture of the foreign matter, thereby providing high-quality data support for the accurate identification of the subsequent intelligent detection module.

[0088] Embodiment 3:

[0089] Based on the embodiment 1, the intelligent detection module, as shown in Figure 3 , includes:

[0090] A model training unit is configured to train a preset deep learning module based on a multi-scene image training set to generate a foreign matter detection model.

[0091] A foreign matter detection unit is configured to detect the enhanced image based on the foreign matter detection model, virtually mark the detected target foreign matter on the remote sensing image, and determine the foreign matter type of the marked target foreign matter.

[0092] Meanwhile, according to the virtual marking result, the spatial coordinates of the target foreign matter are output.

[0093] A list generation unit is configured to generate a foreign matter positioning list in the target detection area based on the spatial coordinates of the target foreign matter and the corresponding foreign matter type, and send the list to the power operation and maintenance center for display.

[0094] The beneficial effects of the above technical solutions are: the model training unit trains the preset deep learning module based on the multi-scene image training set, fully utilizes the image data under various abnormal and normal scenes of the power channel, covers diversified conditions such as complex illumination and different foreign object types, significantly improves the detection accuracy of the model for foreign objects in the power channel, and can effectively reduce missed detection and false detection. Then the foreign object detection unit detects the enhanced image by using the trained foreign object detection model, can quickly identify the target foreign object, and virtually marks on the remote sensing image, simultaneously determines the foreign object type and spatial coordinates, realizes the rapid intelligent detection of foreign objects in the power channel and its surrounding links, greatly shortens the time spent in the power channel inspection, and effectively improves the power channel operation and maintenance efficiency. Subsequently, the list generation unit generates a foreign object positioning list in the target detection area based on the spatial coordinates of the target foreign object and the corresponding foreign object type, and sends it to the power operation and maintenance center. The list presents the foreign object information in a clear and structured form, and the operation and maintenance personnel can intuitively understand the distribution, type and risk level of all foreign objects in the power channel and its surroundings, which provides a reliable reference for the operation and maintenance personnel to make more scientific and reasonable operation and maintenance plans and reasonably allocate resources, and helps to improve the scientificity and accuracy of power channel operation and maintenance management.

[0095] Embodiment 4:

[0096] Based on embodiment 3, the model training unit comprises:

[0097] The data acquisition subunit is configured to collect, based on big data acquisition technology, various abnormal scenes and normal scenes of the power channel;

[0098] The training set generation subunit is configured to, after cleaning the collected massive images, establish corresponding scene image data sets respectively;

[0099] According to the collection keywords and the collection image power channel position, the images in each scene image data set are marked with foreign objects, and combined with artificial quantitative inspection, a plurality of correctly labeled scene image data sets are obtained;

[0100] The images in the correctly labeled plurality of scene image data sets are cross combined to generate a plurality of multi-scene image sets, and based on the correctly labeled plurality of multi-scene image data sets and the plurality of multi-scene image sets, a plurality of corresponding multi-scene image training sets are generated;

[0101] The model training subunit is configured to continuously train the preset learning model according to each multi-scene image training set until the optimal solution is reached, and output the foreign object detection model.

[0102] In this embodiment, the acquisition of the keyword refers to the foreign matter keyword (for example, a name) triggered when an image is acquired in the image sample acquisition process by reusing big data.

[0103] In this embodiment, the image training set corresponding to the same foreign matter scene is first used for training in the training process, and then the multi-scene image set and the single-scene image data set are cross-trained.

[0104] The above technical solution has the following beneficial effects: The data acquisition subunit of the present application collects various abnormal scenes and normal scenes of the power channel by using big data acquisition technology, which can widely cover complex scenes such as temporary construction, greenhouse planting, illegal buildings, and tree shading, and also covers power channel image data under different weather and lighting conditions, thereby providing a rich variety of samples for model training, so that the trained foreign matter detection model has the ability to handle complex actual situations, and can effectively avoid the problem of missing detection or misjudgment of foreign matters in a specific scene due to insufficient data coverage. Then, the training set generation subunit cleans the collected massive images first to remove invalid images such as blurred and damaged images, effectively ensures the quality of the images used for training, and labels the images with foreign matters by using the acquisition keyword and the image power channel position information, and combines artificial quantitative quality inspection to double-ensure the accuracy of the labeling, which is conducive to guiding the model to learn more accurate foreign matter features and significantly improving the accuracy of model recognition, thereby improving the reliability of the entire foreign matter detection system. Thereafter, the correctly labeled scene image data set is cross-combined to generate a multi-scene image set and a corresponding training set, which simulates the complex situation of the mixed appearance of various factors in the actual scene, so that the model can learn the commonalities and differences between different scenes, effectively improves the generalization ability of the model in various complex and mixed scenes, and enables the model to accurately detect foreign matters even in unknown new scenes, which is conducive to expanding the application range of the model. Finally, the model training subunit continuously trains the preset learning model according to each multi-scene image training set until the optimal solution is reached, effectively avoids the model from falling into a local optimal solution, and provides a reliable foundation for realizing efficient and accurate foreign matter detection.

[0105] Embodiment 5:

[0106] On the basis of embodiment 3, the foreign matter detection unit further comprises:

[0107] The channel foreign matter detection subunit is configured to use the foreign matter detection model to focus on detecting the power channel on the enhanced image, and when detecting that there is a foreign matter on the power channel or the power channel is in contact with the foreign matter, quickly positioning the position of the foreign matter or the contact position based on the power pole code, obtaining the operation and maintenance coordinates, and sending the operation and maintenance coordinates to the early warning visualization module for visual early warning.

[0108] A contact foreign matter tracking subunit is configured to continuously focus track the location of the foreign matter or the contact location before an operation and maintenance personnel arrives at the scene when the presence of the foreign matter on the power channel or the contact between the power channel and the foreign matter is detected.

[0109] In this embodiment, the operation and maintenance coordinates refer to coordinate information generated after the location of the foreign matter on the power channel or the contact location between the foreign matter and the power channel is quickly positioned based on the positioning marks such as the pole code.

[0110] The technical scheme has the beneficial effects that: the foreign matter detection subunit of the channel detects the power channel on the enhanced image based on the foreign matter detection model, can more efficiently lock the power channel region, accurately identifies the foreign matter on the channel or the contact between the foreign matter and the channel, and quickly positions the location of the foreign matter or the contact location based on the pole code to generate accurate operation and maintenance coordinates, greatly shortens the fault location confirmation time, enables the operation and maintenance personnel to quickly arrive at the scene to handle the problem, and effectively improves the response efficiency of the power failure. The contact foreign matter tracking subunit focuses tracks the location of the foreign matter or the contact location before the operation and maintenance personnel arrives at the scene after detecting the foreign matter or the contact. By monitoring the state changes of the foreign matter in real time, such as the movement of the foreign matter due to wind force and other factors, the aggravation of the contact with the power channel, and the like, potential risk escalation situations can be discovered in a timely manner. The operation and maintenance personnel can comprehensively understand the foreign matter based on these real-time dynamic data, thereby formulating a more scientific and reasonable handling scheme. For example, based on the movement trend of the foreign matter, the lines can be planned to be isolated in advance, the repair equipment can be allocated, and the like, to avoid resource waste or improper handling caused by blind processing. Once the risk is aggravated, the operation and maintenance personnel can be immediately notified to take emergency measures to avoid serious accidents such as line short circuit, tripping, and even fire caused by the foreign matter, reduce the probability of accidents caused by the foreign matter on the power channel, and ensure the safe and stable operation of the power system. If the foreign matter is removed due to wind force or other factors during the arrival of the operation and maintenance personnel, the operation and maintenance personnel can be quickly notified to avoid the operation and maintenance personnel from going on a wild goose chase.

[0111] Embodiment 6:

[0112] Based on the embodiment 1, the hidden danger risk assessment module, as shown in Figure 4 includes:

[0113] A distribution rule analysis unit is configured to collect foreign matter-channel influence data of foreign matters in multiple scenes, perform feature extraction on the foreign matter-channel influence data, and obtain spatiotemporal distribution features corresponding to foreign matters in different scenes.

[0114] Based on the spatiotemporal distribution features, the harm severity of foreign matters in different scenes to the power channel under different solar terms and climates is determined, and a time weight factor of the corresponding scene foreign matter is generated.

[0115] The position distribution determination unit is configured to determine, based on the spatial coordinates of the target foreign matter and in combination with the channel position coordinate cluster of the power channel in the target detection area, the power line distribution characteristics in the region around the target foreign matter and the index coefficient of the corresponding line importance degree;

[0116] The first evaluation unit is configured to determine, based on the power line distribution characteristics, the relative distance between the target foreign matter and each azimuth power channel, and obtain the distance coefficient of the target foreign matter in multiple azimuths according to the relative distance and the minimum suitable distance.

[0117] After calculating the quotient value between the maximum relative distance and the relative distance and performing normalization processing, the risk weight factor of each azimuth is obtained, and the distance risk coefficient of the target foreign matter is obtained according to the distance coefficient of each azimuth and the corresponding risk weight factor.

[0118] Based on the distance risk coefficient and the index coefficient, the basic risk index of the target foreign matter is calculated, and the basic risk index of the target foreign matter in the current time period is obtained in combination with the time weight factor.

[0119] The intelligent setting unit is configured to mark the target foreign matter as a key detection foreign matter in the current time period when the basic risk index is greater than a preset value.

[0120] Otherwise, the target foreign matter is marked as a common detection foreign matter in the current time period.

[0121] And based on the key detection foreign matter and the common detection foreign matter marking, the detection frequency of each target foreign matter in the current time period is set.

[0122] In this embodiment, the foreign matter-channel influence data refers to various data collected from multiple scenes that foreign matters interfere with the operation of power channels, including foreign matter types (for example, branches, plastic cloth, kites, etc.), spatial position relationships between foreign matters and power channels (for example, vertical distance from the conductor, horizontal distance), electrical parameter changes caused by foreign matters (for example, short circuit risk, grounding fault probability), influences on channel operation and maintenance (for example, increased maintenance difficulty, power outage risk), etc. The data sources cover different sub-areas (for example, cities, mountains, farmlands, etc.) in the target detection area (for example, a city, a province, a district, etc.), and are used to analyze the actual harm degree of foreign matters to power channels.

[0123] In this embodiment, the spatio-temporal distribution characteristics refer to the distribution rules in the time and space dimensions extracted based on the foreign matter-channel influence data:

[0124] The time dimension includes the frequency of foreign matter occurrence, the change rule of harm degree (for example, the discharge risk caused by the growth of branches in spring, the kites winding around the conductor in summer) under different seasons, solar terms and climates (for example, strong wind, heavy rain, high temperature, etc.).

[0125] The spatial dimension includes the distribution of foreign matters in different geographical areas (for example, trees are prone to falling in mountainous areas, and construction foreign matters are prone to occurring in suburban areas), and the distribution of the positions of the power channel along the line (for example, a certain section of the line is prone to gathering floating matters due to the terrain).

[0126] In this embodiment, the channel position coordinate cluster refers to a set of position coordinates of the power channel related facilities in the target detection area, including the coordinates of the transmission line tower, the coordinates of the wire direction, and the coordinates of the substation position.

[0127] In this embodiment, the power line distribution feature refers to the spatial layout attribute of the power line around each target foreign matter, including the line direction (for example, east-west, north-south), the distribution density (the number of lines per unit length), the voltage level (high / low voltage line distribution), and the importance (for example, main line, branch line).

[0128] In this embodiment, the index coefficient refers to a quantitative parameter reflecting the importance of the power channel line, which is calculated in advance based on the line voltage level, the power supply range (for example, whether it is an important load power supply), the fault influence range (for example, the population density and industrial value of the power outage area), and other factors. The more important the line is, the larger the index coefficient is.

[0129] In this embodiment, the minimum suitable distance refers to the minimum allowed distance between the foreign matter and the power channel to maintain safe operation, which is a threshold value set according to the power safety specification. For example, the minimum safe distance between a tree branch and a high-voltage wire is 5 meters, and below this distance, wind deflection may cause discharge; the minimum distance between a floating matter and a wire needs to be greater than 10 meters to avoid short circuit caused by static adsorption.

[0130] In this embodiment, the distance coefficient refers to the ratio between the minimum suitable distance and the relative distance.

[0131] In this embodiment, the risk weight factor refers to the weight value obtained by normalizing the quotient of the maximum relative distance (the maximum value of the relative distances corresponding to multiple directions of the same target foreign matter) distribution and the relative distance of each direction.

[0132] In this embodiment, the distance risk coefficient refers to the product of the distance coefficient and the risk weight factor.

[0133] In this embodiment, the basic risk index refers to the sum of the product of the distance risk coefficient and the index coefficient and the basic risk index, multiplied by the time weight factor corresponding to the current period.

[0134] In this embodiment, the length of the current period can be flexibly set according to the detection requirements, for example, a quarter, a month, a solar term, etc.

[0135] The beneficial effects of the above technical solutions are: the distribution rule analysis unit collects foreign matter-channel influence data of multiple scene foreign matters, and extracts space-time distribution characteristics, can deeply mine the internal law of different scene foreign matters under different solar terms and climates, and generates a time weight factor based on this, realizes the quantification of the influence of seasonal changes, weather factors and the like on foreign matter risks, for example, in the season when the wind blows frequently, the plastic film foreign matter which is easy to be blown up and entangled with the line is given a higher time weight factor, which greatly improves the accuracy and scientificity of risk assessment. Then, the position distribution determination unit and the first evaluation unit combine the spatial coordinates of the target foreign matter, the distribution characteristics of the power channel and the importance of the line to refine the risk assessment from the spatial dimension, determine the relative distance between the target foreign matter and the power channel, calculate the distance coefficient and the risk weight factor, and comprehensively obtain the distance risk coefficient and the basic risk index, which fully considers the spatial position relationship between the foreign matter and the power channel and the importance of the line itself, and for the foreign matter close to the key power transmission line, even if its own danger is small, it will be evaluated as high risk due to the high index coefficient of the line importance, so that the final evaluation is more in line with the actual distribution of the power channel and the foreign matter, and provides a more reliable basis for risk control. Then, the intelligent setting unit marks the target foreign matter according to the basic risk index, and differentially sets the detection frequency, realizes the dynamic grading management of the foreign matter of the power channel: for the key detection foreign matter with high basic risk index, the detection frequency is improved to ensure that the potential risk change is discovered in time; for the ordinary detection foreign matter, the detection frequency is reasonably reduced to avoid resource waste. In the case of limited inspection resources, the resources are preferentially allocated to the key detection foreign matter for high-frequency monitoring, so that the operation and maintenance personnel can focus their efforts on the high-risk area, effectively optimizing the operation and maintenance resource allocation and improving the resource utilization efficiency.

[0136] Embodiment 7:

[0137] Based on the embodiment 6, the hidden danger risk assessment module, as shown in Figure 4 , further comprises:

[0138] The second evaluation unit is configured to determine the power hidden danger breeding path of each target foreign matter based on the corresponding hidden danger type of the target foreign matter and in combination with the preset foreign matter database.

[0139] The current presentation state of the target foreign matter is obtained.

[0140] When the similarity between the current presentation state and any link on the power hidden danger breeding path thereof is greater than or equal to a similarity threshold, a dynamic risk assessment value of the target foreign matter to the power channel is generated in combination with the stage sequence of the current presentation state on the corresponding power hidden danger breeding path and the corresponding basic risk index thereof.

[0141] In this embodiment, the power hazard breeding path refers to a complete evolution process chain of the foreign matter from appearing to possibly causing a power failure, which is preset based on the type of foreign matter and historical data, and is divided into multiple risk development stages with time or state as the axis, for example:

[0142] Plastic film foreign matter: falls near the power channel (initial stage) → is blown close to the conductor by the wind (development stage) → winding around the conductor causes discharge (failure stage);

[0143] Tree growth foreign matter: tree sprouts (initial stage) → branches and leaves approach the conductor (development stage) → touches the conductor in windy and rainy weather (failure stage);

[0144] The later the sequence is, the higher the corresponding basic risk index is, for example: 0.5 for the initial stage, 0.7 for the development stage, and 0.9 for the failure stage, which can be flexibly set.

[0145] In this embodiment, the current presentation state refers to the actual existence state of the target foreign matter at the detection time and the interaction relationship with the power channel.

[0146] Advantages of the above technical solution: the second evaluation unit of the present application can determine the power hazard breeding path based on the corresponding hazard type of the target foreign matter and in combination with the preset foreign matter database, which can deeply analyze the complete evolution process of different foreign matters from appearing to causing a power failure. For example, for a temporary construction tower crane near the power channel, the potential path of gradually approaching the line due to improper operation and finally causing short circuit can be predicted. By obtaining the current presentation state of the target foreign matter and comparing it with the breeding path link, when the similarity reaches a threshold value, the risk evolution trend can be captured in advance, which greatly improves the forward-looking nature of risk warning compared with only relying on current state evaluation, so that the operation and maintenance personnel have more sufficient time to take preventive measures to avoid accidents. In combination with the stage sequence of the current presentation state on the hazard breeding path and the basic risk index, a dynamic risk evaluation value is generated, so that the risk evaluation is not fixed, but is updated in real time with the change of the state of the foreign matter, for example, when the plastic film gradually approaches the line from the initial state of falling near the power channel, the stage of the plastic film on the breeding path changes, and the dynamic risk evaluation value also increases. This dynamic adjustment mechanism makes the risk control of the power channel more in line with the actual situation of the foreign matter, and enhances the flexibility and timeliness of the hazard control of the power channel.

[0147] Embodiment 8:

[0148] On the basis of embodiment 6, the distribution rule analysis unit comprises:

[0149] The weight factor generation subunit is configured to collect annual historical power channel interference data of various hazard types in the target detection area, analyze the annual historical power channel interference data, and determine a comprehensive power hazard score corresponding to different scene foreign matters in different regions, a hazard score of each time period, an annual incident frequency, and an incident frequency of each time period, respectively.

[0150] Based on the hazard score of different scene foreign matters in the current time period and the comprehensive power hazard score, a first hazard severity corresponding to various scene foreign matters in the current time period is calculated.

[0151] Based on the incident frequency of different scene foreign matters in the current time period and the annual incident frequency, a second hazard severity corresponding to various scene foreign matters in the current time period is calculated.

[0152] According to the annual historical power channel interference data, a regional incident proportion of each sub-region in the target detection area corresponding to different scene foreign matters is determined.

[0153] Based on the regional incident proportion, the first hazard severity, the second hazard severity, and the weighted weight corresponding to the first hazard severity and the second hazard severity, a time weight factor of the corresponding scene foreign matter in different sub-regions is obtained.

[0154] In this embodiment, the first hazard severity refers to a ratio between the hazard score of the current time period and the comprehensive power hazard score.

[0155] In this embodiment, the second hazard severity refers to a ratio between the incident frequency of the current time period and the annual incident frequency.

[0156] The beneficial effects of the above technical solutions are: the weight factor generation subunit collects annual historical power channel interference data, and analyzes from multiple dimensions such as comprehensive power hazard score, period hazard score, annual incident frequency, period incident frequency, and regional incident proportion, fully considers the differences of different scene foreign matters in time, space and hazard degree, avoids the deviation caused by single data or one-sided analysis, and then calculates the first hazard severity and the second hazard severity: the first hazard severity combines the current period hazard score and the comprehensive power hazard score, measures the relationship between the current hazard and the overall hazard; the second hazard severity is based on the current period incident frequency and the annual incident frequency, and reflects the probability change of the current risk occurrence. The two complement each other, quantify the hazard degree of different scene foreign matters in the current period, and also improve the accuracy of the evaluation of the hazard degree of foreign matters, which can more accurately evaluate the risk of different scene foreign matters in a specific period, thereby improving the accuracy of the entire risk assessment system. And considering the regional incident proportion of different scene foreign matters in each sub-region in the target detection region, the generation of the time weight factor has regional characteristics. Due to the differences in geographical environment, economic activities and other factors of different sub-regions, the distribution and risk of power channel foreign matters also differ. For example, the risk of plastic film foreign matters in agricultural planting areas is relatively high, while the risk of temporary construction equipment foreign matters in urban construction areas is relatively large. The time weight factor generated based on the regional incident proportion is conducive to improving the accuracy of risk assessment of different scene foreign matters, and also helps operation and maintenance personnel to develop more targeted detection and prevention strategies according to the characteristics of different regions, thereby effectively improving the efficiency and effect of regional power channel management.

[0157] Embodiment 9:

[0158] On the basis of embodiment 1, the early warning visualization module, as shown in Figure 5 , includes:

[0159] The early warning unit is configured to generate an early warning signal according to the hazard type and spatial coordinates or operation and maintenance coordinates of the risk foreign matter, and send the early warning signal to the power operation and maintenance center and the corresponding regional power special person in charge;

[0160] The visualization unit is configured to highlight mark the risk position or operation and maintenance position based on the early warning signal, and display the risk position or operation and maintenance position on the monitoring end or personal end;

[0161] The tracking operation and maintenance unit is configured to detect the to-be-operated position, update the operation and maintenance state of the visualized operation and maintenance position after confirming the arrival of the operation and maintenance personnel, and record the arrival time of the operation and maintenance personnel;

[0162] The remote communication unit is configured to communicate online with the operation and maintenance personnel who arrive at the to-be-operated position.

[0163] The beneficial effects of the above technical solutions are: the early warning unit generates accurate early warning signals according to the risk type and space / operation and maintenance coordinates of the risk foreign matter, and directly sends the early warning signals to the power operation and maintenance center and the special person in charge of the corresponding area, realizes directional early warning of power channel detection, effectively avoids the problem of invalid diffusion of early warning information, and also ensures that key personnel can obtain accurate risk information at the first time. For example, when it is detected that there is a risk of foreign matter in a certain area of the power channel due to illegal construction, the early warning signal can be immediately sent to the operation and maintenance personnel in charge of the area, greatly shortening the information transmission time, enabling the operation and maintenance personnel to respond quickly and take timely measures to eliminate the hidden danger, and effectively improving the timeliness of power failure response. And through the visualization unit, the risk position is highlighted and displayed based on the early warning signal at the monitoring end or the personal end. Whether it is to macroscopically check the overall risk distribution of the power channel on the monitoring large screen or to check the hidden danger details of the area responsible by the individual on the mobile terminal, the visual display of the highlighted mark can display the risk position through intuitive visualization, which can help the operation and maintenance personnel to quickly locate the target, save a lot of time for on-site investigation and processing, and effectively improve the efficiency of hidden danger disposal. Through the tracking operation unit, the to-be-operated position is continuously monitored, and when it is determined that the operation personnel have arrived, the state of the visual operation position is updated in time, and the arrival time is recorded, realizing real-time tracking of the operation process, ensuring that the hidden danger is traceable from discovery, dispatching to completion of processing, and the operation and management personnel can check the operation state to master the work progress in real time, avoiding delay or omission in processing. For example, if a hidden danger is in an untreated state for a long time, the management personnel can intervene in coordination in time to ensure that the power channel hidden danger is processed in time and effectively, forming a complete operation and management closed loop. At the same time, the remote communication unit supports online communication with the operation personnel who have arrived on the scene, so that the operation and maintenance center and the on-site personnel can communicate in real time. When the on-site personnel are processing complex hidden dangers, they can obtain expert guidance, allocate resources or coordinate other departments for assistance in time through remote communication. For example, when processing the foreign matter of the power channel caused by large construction equipment, the on-site personnel can show the on-site situation to the technical expert through video call to obtain professional demolition scheme suggestions, avoiding improper processing due to lack of experience or insufficient information. This remote communication and cooperation mechanism effectively enhances the processing capacity of the on-site personnel and improves the efficiency of solving complex hidden dangers, while reducing the safety risks caused by information asymmetry.

[0164] Embodiment 10:

[0165] The application provides an intelligent detection method for foreign matter in a power channel based on a multi-scale Retinex algorithm, as shown in the following formula (1): Figure 6 The method comprises the following steps:

[0166] Step 1: processing real-time remote sensing images by a multi-scale Retinex algorithm to obtain enhanced images;

[0167] Step 2: Detect the enhanced image using the foreign object detection model, and output the spatial coordinates of the detected target foreign object and its corresponding hazard type;

[0168] Step 3: Based on the spatial coordinates of the target foreign object and its corresponding hazard type, and the spatiotemporal distribution rule of power safety hazards, generate a dynamic risk assessment value of the target foreign object to the power channel;

[0169] Step 4: When the dynamic risk assessment value is greater than or equal to the preset threshold, mark the target foreign object as a risk foreign object, and visually display the risk foreign object mark on the real-time remote sensing image for early warning.

[0170] The beneficial effects of the above technical solution are: Firstly, the real-time remote sensing image is processed using the multi-scale Retinex algorithm, which effectively eliminates the influence of factors such as complex lighting conditions and atmospheric interference on image quality, significantly enhances the contrast between foreign objects and the background in the image, and improves the clarity of image details. Compared with traditional methods, this algorithm can more accurately preserve the subtle features in the power channel scene, making the key information such as the outline and texture of the foreign object clearly presented, providing high-quality data support for the accurate identification of subsequent intelligent detection modules, and greatly improving the accuracy of foreign object detection. Then, a pre-trained foreign object detection model is used to detect the enhanced image, which can quickly and accurately identify various foreign objects in the power channel and output their spatial coordinates and corresponding hazard types. Not only can it efficiently process massive real-time remote sensing image data, but it can also accurately classify different types of foreign objects (such as temporary construction equipment, greenhouse film, illegal buildings, etc.), providing an efficient and reliable technical means for power channel hazard investigation. Subsequently, based on the spatial coordinates of the target foreign object, the hazard type, and the spatiotemporal distribution rule of power safety hazards, a dynamic risk assessment value is generated through comprehensive analysis and calculation. This evaluation process fully considers the risk characteristics of the power channel in different time and space dimensions, as well as the severity of the damage that various foreign objects may cause to power facilities, achieving dynamic and quantitative evaluation of the risk of power channel hazards, providing a scientific basis for the power operation and maintenance department to develop targeted prevention and control measures, and effectively improving the forward-looking and effectiveness of power channel risk management. Finally, when the dynamic risk assessment value is greater than or equal to the preset threshold, the target foreign object is marked as a risk foreign object, and a visual early warning display is performed on the real-time remote sensing image, realizing the intuitive and visual presentation of power channel abnormalities. This allows power operation and maintenance personnel to quickly and clearly grasp the specific location, type, and risk level of risk foreign objects in the power channel, greatly improving the efficiency and accuracy of information transmission. Based on the visual early warning results, operation and maintenance personnel can quickly make response decisions, reasonably allocate resources, and prioritize high-risk hazards, effectively shortening the hazard processing period and reducing the probability of power accidents, ensuring the safe and stable operation of the power system.

[0171] Obviously, many modifications and variations of the present application are possible in light of the above teachings. It is, therefore, to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. A smart foreign object detection system for power channels based on a multi-scale Retinex algorithm, characterized in that, include: The image processing module is used to process real-time remote sensing images using the multi-scale Retinex algorithm to obtain enhanced images; The intelligent detection module is used to detect enhanced images using a foreign object detection model and output the spatial coordinates of the detected target foreign object and its corresponding hazard type. The hazard risk assessment module is used to generate dynamic risk assessment values ​​of the target foreign object on the power channel based on the spatial coordinates of the target foreign object, its corresponding hazard type, and the spatiotemporal distribution pattern of power safety hazards. The early warning visualization module is used to mark target foreign objects as risky foreign objects when the dynamic risk assessment value is greater than or equal to a preset threshold, and to visualize and display the risky foreign object markings on real-time remote sensing images.

2. The intelligent foreign object detection system for power channels based on the multi-scale Retinex algorithm according to claim 1, characterized in that, Image processing module, including: The image acquisition unit is used to acquire multiple real-time remote sensing images of the target detection area; The image denoising unit is used to perform denoising processing on multiple real-time remote sensing images based on a preset filtering algorithm. The image enhancement unit is used to perform multi-scale decomposition on the denoised real-time remote sensing image to obtain image layers of multiple scales. Based on the Retinex algorithm, each image layer is processed separately to obtain multiple uniform light image layers, and these multiple uniform light image layers are then fused to obtain an enhanced image.

3. The intelligent foreign object detection system for power channels based on the multi-scale Retinex algorithm according to claim 1, characterized in that, The intelligent detection module includes: The model training unit is used to train a preset deep learning module based on a multi-scene image training set to generate a foreign object detection model. The foreign object detection unit is used to detect foreign objects in enhanced images based on a foreign object detection model, virtually mark the detected target foreign objects on the remote sensing images, and determine the type of foreign object of the marked target foreign object. Simultaneously, based on the virtual marking results, the spatial coordinates of the target foreign object are output; The list generation unit is used to generate a list of foreign object locations within the target detection area based on the spatial coordinates of the target foreign object and its corresponding foreign object type, and send it to the power operation and maintenance center for display.

4. The intelligent foreign object detection system for power channels based on the multi-scale Retinex algorithm according to claim 3, characterized in that, The model training unit includes: The data acquisition subunit is used to collect data based on big data acquisition technology for various abnormal and normal scenarios of power channels. The training set generation sub-unit is used to clean the massive amount of images collected and then establish corresponding scene image datasets. Based on the collected keywords and the location of the power channel in the collected images, foreign objects were marked in the images of each scene image dataset. Combined with manual quantitative quality inspection, multiple scene image datasets with correct annotations were obtained. Images from multiple correctly labeled scene image datasets are cross-combined to generate multiple multi-scene image sets, and corresponding multi-scene image training sets are generated based on the multiple correctly labeled multi-scene image datasets and multiple multi-scene image sets respectively. The model training subunit is used to continuously train the preset learning model based on each multi-scene image training set until the optimal solution is reached, and output the foreign object detection model.

5. The intelligent foreign object detection system for power channels based on the multi-scale Retinex algorithm according to claim 3, characterized in that, The foreign object detection unit also includes: The channel foreign object detection subunit is used to focus on the detection of power channels in the enhanced image using a foreign object detection model. When a foreign object is detected on the power channel or the power channel is in contact with a foreign object, the location of the foreign object or the contact location is quickly located based on the pole code to obtain the operation and maintenance coordinates, and the operation and maintenance coordinates are sent to the early warning visualization module for visualization early warning. The foreign object tracking subunit is used to continuously focus and track the location of the foreign object or the contact point before maintenance personnel arrive at the site when a foreign object is detected on the power channel or when the power channel is in contact with a foreign object.

6. The intelligent foreign object detection system for power channels based on the multi-scale Retinex algorithm according to claim 1, characterized in that, The hazard and risk assessment module includes: The distribution pattern analysis unit is used to collect foreign object-channel impact data of foreign objects in multiple scenarios, and to extract features from the foreign object-channel impact data to obtain the spatiotemporal distribution characteristics of foreign objects in different scenarios. Based on the aforementioned spatiotemporal distribution characteristics, the severity of the damage to power channels caused by foreign objects in different scenarios under different solar terms and climates is determined, and a time weighting factor for foreign objects in the corresponding scenario is generated. The location distribution determination unit is used to determine the distribution characteristics of power lines around the target object and the index coefficients of the corresponding line importance based on the spatial coordinates of the target object and the channel location coordinate cluster of the power channel within the target detection area. The first evaluation unit is used to determine the relative distance between the target foreign object and the power channels in various directions based on the distribution characteristics of the power lines, and to obtain the distance coefficients of the target foreign object in multiple directions based on the relative distance and the minimum suitable distance. After calculating the quotient between the maximum relative distance and the relative distance and performing normalization, risk weight factors for multiple directions are obtained. Based on the distance coefficients of multiple directions and their corresponding risk weight factors, the distance risk coefficient of the target foreign object is obtained. Based on the distance risk coefficient and the index coefficient, the basic risk index of the target foreign object is calculated. Combined with the time weight factor, the basic risk index of the target foreign object in the current time period is obtained. The intelligent setting unit is used to mark the target foreign object as a key foreign object to be detected in the current time period when the basic risk index is greater than a preset value; Otherwise, mark the target foreign object as a regular foreign object detected within the current time period; Based on the markers for key foreign objects and ordinary foreign objects, the detection frequency for each target foreign object is set within the current time period.

7. The intelligent foreign object detection system for power channels based on the multi-scale Retinex algorithm according to claim 6, characterized in that, The hazard risk assessment module also includes: The second assessment unit is used to determine the path of electrical hazards arising from each target foreign object based on the type of hazard corresponding to the target foreign object and in conjunction with a pre-set foreign object database. Obtain the current presentation state of the target foreign object; When the similarity between the current state and any link in the path of power hazard development is greater than or equal to the similarity threshold, a dynamic risk assessment value of the target foreign object on the power channel is generated by combining the current state with the stage sequence in the corresponding power hazard development path and its corresponding basic risk index.

8. The intelligent foreign object detection system for power channels based on the multi-scale Retinex algorithm according to claim 6, characterized in that, The distribution pattern analysis unit includes: The weighting factor generation subunit is used to collect historical power channel interference data for various types of hazards in the target detection area throughout the year. The historical power channel interference data throughout the year is analyzed to determine the comprehensive power hazard score, hazard score for each time period, annual incident frequency, and incident frequency for each time period for foreign objects in different scenarios in different regions. Based on the hazard scores of foreign objects in different scenarios at the current time period and the comprehensive power hazard score, the first hazard severity of foreign objects in various scenarios at the current time period is calculated. Based on the frequency of foreign objects in different scenarios during the current period and the annual frequency of foreign objects, the severity of the second hazard corresponding to foreign objects in various scenarios during the current period is calculated. Based on historical power channel interference data throughout the year, the proportion of foreign object incidents in each sub-region of the target detection area corresponding to different scenarios was determined. Based on the regional incident rate, the severity of the first hazard, the severity of the second hazard, and the weighted weights corresponding to the severity of the first hazard and the severity of the second hazard, the time weight factor of foreign objects in different sub-regions of the corresponding scenario is obtained.

9. The intelligent foreign object detection system for power channels based on the multi-scale Retinex algorithm according to claim 1, characterized in that, The early warning visualization module includes: The early warning unit is used to generate early warning signals based on the type of potential hazards and spatial or operational coordinates of foreign objects, and send them to the power operation and maintenance center and the power project manager in the corresponding area. The visualization unit is used to highlight risk locations or maintenance locations based on early warning signals and then display them on the monitoring terminal or personal terminal; The tracking and maintenance unit is used to detect the location to be maintained, and after confirming the arrival of maintenance personnel, it updates the maintenance status of the visualized maintenance location and records the arrival time of the maintenance personnel. The remote communication unit is used for online communication with maintenance personnel who have arrived at the site to be maintained.

10. A method for intelligent detection of foreign objects in power channels based on the multi-scale Retinex algorithm, characterized in that, include: Enhanced images are obtained by processing real-time remote sensing images using the multi-scale Retinex algorithm. The foreign object detection model is used to detect the enhanced image and output the spatial coordinates of the detected target foreign object and its corresponding hazard type. Based on the spatial coordinates of the target foreign object, its corresponding hazard type, and the spatiotemporal distribution pattern of power safety hazards, a dynamic risk assessment value of the target foreign object on the power channel is generated. When the dynamic risk assessment value is greater than or equal to the preset threshold, the target foreign object is marked as a risky foreign object, and the risky foreign object mark is displayed visually on the real-time remote sensing image.

Images