A hidden danger identification method and device for a power transmission channel area based on a cross-section scene, an electronic device, and a storage medium

Abstract

The application discloses a kind of based on the hidden danger identification method, device, electronic equipment and storage medium of cross-section scene transmission channel area, belong to transmission line operation monitoring technical field, the method comprises: obtaining remote sensing image data;Remote sensing image data is input to cross-section target detection model, determine whether there is cross-section area in the transmission channel area to be measured;In the case where there is cross-section area in the transmission channel area to be measured, remote sensing image data is input to cross-section type identification model, determine the cross-section type of transmission channel area to be measured;According to cross-section type, determine hidden danger identification task;Based on hidden danger identification task, according to remote sensing image data, determine hidden danger identification result;In the case where there is no cross-section area in the transmission channel area to be measured, determine that there is no hidden danger under cross-section scene in the transmission channel area to be measured. By implementing the present application, the problem of low efficiency of hidden danger identification of transmission channel area in the prior art can be solved.

Description

Technical Field

[0001] This invention relates to the field of transmission line operation and maintenance monitoring technology, specifically to a method, device, electronic equipment, and storage medium for identifying potential hazards in transmission channel areas based on cross-section scenarios. Background Technology

[0002] With the deep integration of power grid construction with modern transportation and natural resource systems, overhead transmission lines inevitably cross specific areas such as high-speed railways, highways, large water systems, and construction zones. These spatial environments involving specific crossings are commonly referred to as cross-section scenarios, and their corresponding geographical scope is the cross-section area. Cross-section areas are often located near transportation hubs or in special geographical locations with variable environments. Not only are the design standards and daily operation and maintenance requirements for transmission facilities far higher than in conventional areas, but they also face more severe risks such as external debris damage, tower foundation settlement, and extreme environmental disasters. Once a fault occurs in such an area, it can easily trigger a chain reaction of traffic disruptions or public safety incidents. Therefore, in wide-area transmission channel monitoring, the ability to quickly locate cross-section areas and conduct targeted hazard identification is of paramount importance for ensuring the stable operation of the power grid.

[0003] Existing technologies for identifying potential hazards in power transmission corridors are often inefficient. These technologies typically identify the entire transmission corridor area directly, lacking a prior assessment of the existence of cross-section scenarios. Even when no cross-section scenarios exist, identification continues, wasting significant computational resources and resulting in low efficiency in identifying potential hazards in power transmission corridor areas. Summary of the Invention

[0004] This invention provides a method, device, electronic device, and storage medium for identifying potential hazards in power transmission channel areas based on cross-section scenarios, which can solve the problem of low efficiency in identifying potential hazards in power transmission channel areas in the prior art.

[0005] One embodiment of the present invention provides a method for identifying potential hazards in power transmission channel areas based on cross-section scenarios, including: Acquire remote sensing image data of the power transmission channel area to be tested; The remote sensing image data is input into a preset cross-segment target detection model, so that the cross-segment target detection model can determine whether there is a cross-segment area in the power transmission channel area to be tested based on the remote sensing image data; wherein, the cross-segment target detection model is trained by several target detection training samples; each target detection training sample includes historical remote sensing image data and corresponding cross-segment existence status labels; When there are cross-segment areas in the transmission channel area to be tested, the remote sensing image data is input into a preset cross-segment type recognition model, so that the cross-segment type recognition model can determine the cross-segment type of the transmission channel area to be tested based on the remote sensing image data; wherein, the cross-segment type recognition model is trained by several type recognition training samples; each type recognition training sample includes historical remote sensing image data and corresponding type semantic labels; Based on the cross-section type, determine the hidden danger identification task for the transmission channel area to be tested; Based on the aforementioned hazard identification task, the hazard identification results for the transmission channel area to be tested are determined according to the remote sensing image data of the area. If there are no cross-section areas in the area of ​​the transmission channel to be tested, it is determined that there are no potential hazards in the cross-section scenario in the area of ​​the transmission channel to be tested.

[0006] Furthermore, remote sensing image data of the transmission channel area to be measured is acquired, including: Acquire raw remote sensing data of the power transmission channel area to be measured; wherein, the raw remote sensing data includes raw optical images and raw radar images; Perform optical image standardization correction on the original optical image to generate a standard optical image; The original radar image is standardized and corrected to obtain a standard radar image. Spatial registration and resampling are performed on standard optical images to generate optical images; Standard radar images are spatially registered and resampled to generate radar images. Optical and radar images were used as remote sensing image data for the power transmission channel area to be measured.

[0007] Furthermore, the pre-defined cross-segment target detection model is trained in the following manner: Obtain several object detection training samples; Each target detection training sample is sequentially input into the cross-segment target detection model to train the model until a preset number of target detection training iterations are reached. Each time a target detection training sample is received, the cross-segment target detection model outputs a predicted existence state result corresponding to that sample. Based on the predicted existence state result and the corresponding cross-segment existence state label, a target detection loss function value is calculated. The cross-segment target detection model is then updated based on the target detection loss function value.

[0008] Furthermore, the pre-defined cross-segment type recognition model is trained in the following manner: Obtain several types of recognition training samples; The training samples for each type of identification are sequentially input into the cross-segment type identification model to train the model until a preset number of type identification training iterations are reached. Each time a type identification training sample is received, the cross-segment type identification model outputs the predicted cross-segment type corresponding to that training sample. Based on the predicted cross-segment type and its corresponding semantic label, a type identification loss function value is calculated. The cross-segment type identification model is then updated based on the type identification loss function value.

[0009] Furthermore, based on the aforementioned cross-section type, the task of identifying potential hazards in the transmission channel area to be tested is determined, including: When the cross-section type is cross-transportation hub, the hidden danger identification task for the power transmission channel area to be tested is determined to be a geological deformation identification task; In the case where the cross-section type is cross-vegetation and water body, the task of identifying hidden dangers in the power transmission channel area to be tested is determined to be an environmental disaster identification task; When the cross-section type is cross-construction building, the task of identifying potential hazards in the power transmission channel area to be tested is determined to be an external intrusion identification task.

[0010] Furthermore, based on the aforementioned hazard identification task, and using remote sensing image data of the transmission channel area to be tested, the hazard identification results for the transmission channel area to be tested are determined, including: When the hazard identification task is a geological deformation identification task, phase interferometry processing is performed on the radar image in the remote sensing image data of the transmission channel area to be measured to determine the geological cumulative deformation component of the transmission channel area to be measured; the geological cumulative deformation component is used as the hazard identification result of the transmission channel area to be measured. When the hazard identification task is an environmental disaster identification task, spectral features are extracted from the optical images in the remote sensing image data of the transmission channel area to be tested to determine the disaster distribution status of the transmission channel area to be tested; the disaster distribution status is used as the hazard identification result of the transmission channel area to be tested. When the hazard identification task is an external intrusion identification task, target identification is performed on the remote sensing image data of the power transmission channel area to be tested, and the power transmission lines and foreign objects in the power transmission channel area to be tested are extracted; based on the power transmission lines and foreign objects in the power transmission channel area to be tested, the spatial clearance distance between the foreign objects and the power transmission lines is calculated; the spatial clearance distance is used as the hazard identification result of the power transmission channel area to be tested.

[0011] Based on the above method embodiments, the present invention provides corresponding apparatus embodiments.

[0012] An embodiment of the present invention provides a hazard identification device for power transmission channel areas in cross-section scenarios, comprising: a data acquisition module, a target detection module, and a hazard identification module; The data acquisition module is used to acquire remote sensing image data of the power transmission channel area to be tested; The target detection module is used to input the remote sensing image data into a preset cross-segment target detection model, so that the cross-segment target detection model can determine whether there is a cross-segment area in the power transmission channel area to be tested based on the remote sensing image data; wherein, the cross-segment target detection model is trained by several target detection training samples; each target detection training sample includes historical remote sensing image data and corresponding cross-segment existence status labels; The hazard identification module is used to input the remote sensing image data into a preset cross-segment type identification model when there are cross-segment areas in the transmission channel area to be tested, so that the cross-segment type identification model can determine the cross-segment type of the transmission channel area to be tested based on the remote sensing image data; wherein, the cross-segment type identification model is trained by several type identification training samples; each type identification training sample includes historical remote sensing image data and corresponding type semantic labels; based on the cross-segment type, a hazard identification task for the transmission channel area to be tested is determined; based on the hazard identification task, the hazard identification result for the transmission channel area to be tested is determined based on the remote sensing image data of the transmission channel area to be tested; when there are no cross-segment areas in the transmission channel area to be tested, it is determined that there are no hazards in the cross-segment scenario in the transmission channel area to be tested.

[0013] Furthermore, the data acquisition module acquires remote sensing image data of the transmission channel area to be tested, including: Acquire raw remote sensing data of the power transmission channel area to be measured; wherein, the raw remote sensing data includes raw optical images and raw radar images; Perform optical image standardization correction on the original optical image to generate a standard optical image; The original radar image is standardized and corrected to obtain a standard radar image. Spatial registration and resampling are performed on standard optical images to generate optical images; Standard radar images are spatially registered and resampled to generate radar images. Optical and radar images were used as remote sensing image data for the power transmission channel area to be measured.

[0014] Based on the above method embodiments, the present invention provides corresponding electronic device embodiments.

[0015] An embodiment of the present invention provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the method for identifying hidden dangers in power transmission channel areas based on cross-segment scenarios described in any of the above-described method embodiments.

[0016] Based on the above method embodiments, the present invention provides corresponding storage medium embodiments.

[0017] One embodiment of the present invention provides a storage medium storing a computer program thereon, wherein, when the computer program is running, it controls the device where the storage medium is located to execute any of the above-described method embodiments of the method for identifying potential hazards in power transmission channel areas based on cross-segment scenarios.

[0018] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a method, apparatus, electronic device, and storage medium for identifying potential hazards in power transmission corridor areas based on cross-section scenarios. The method acquires remote sensing image data of the power transmission corridor area to be tested; inputs the remote sensing image data into a preset cross-section target detection model to determine whether a cross-section area exists in the power transmission corridor area to be tested; if a cross-section area exists in the power transmission corridor area to be tested, inputs the remote sensing image data into a preset cross-section type identification model to determine the cross-section type of the power transmission corridor area to be tested; based on the cross-section type, determines the hazard identification task for the power transmission corridor area to be tested; based on the hazard identification task and the remote sensing image data of the power transmission corridor area to be tested, determines the hazard identification result for the power transmission corridor area to be tested; if no cross-section area exists in the power transmission corridor area to be tested, it is determined that there are no potential hazards in the cross-section scenario in the power transmission corridor area to be tested.

[0019] This application utilizes a cross-segment target detection model to pre-determine whether cross-segment scenarios exist within the area to be tested. If the determination result is that no such scenarios exist, the subsequent identification process is terminated promptly, directly solving the problem of wasted computational resources caused by the lack of pre-judgment in the prior art. Furthermore, by further identifying specific cross-segment types and accurately matching associated hazard identification tasks, the efficiency of hazard identification in power transmission channel areas is significantly improved. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a method for identifying potential hazards in a power transmission channel area based on a cross-section scenario, provided by an embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of the structure of a hidden danger identification device for a power transmission channel area based on a cross-section scenario, provided by an embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] like Figure 1 As shown, to address the problem of low efficiency in identifying potential hazards in power transmission corridor areas in existing technologies, an embodiment of the present invention provides a method for identifying potential hazards in power transmission corridor areas based on cross-section scenarios, comprising at least the following steps: Step S1: Obtain remote sensing image data of the power transmission channel area to be tested.

[0024] In a preferred embodiment, acquiring remote sensing image data of the transmission channel area to be measured includes: Acquire raw remote sensing data of the power transmission channel area to be measured; wherein, the raw remote sensing data includes raw optical images and raw radar images; Perform optical image standardization correction on the original optical image to generate a standard optical image; The original radar image is standardized and corrected to obtain a standard radar image. Spatial registration and resampling are performed on standard optical images to generate optical images; Standard radar images are spatially registered and resampled to generate radar images. Optical and radar images were used as remote sensing image data for the power transmission channel area to be measured.

[0025] It should be noted that the area of ​​the transmission channel to be tested refers to a strip-shaped geographical area extending outwards from the overhead transmission line at a predetermined distance. The value of the predetermined distance is usually determined based on the line voltage level and the complexity of the surrounding environment. Remote sensing image data refers to a two-dimensional digital image generated after non-contact sensors detect the electromagnetic wave characteristics of target objects in the area of ​​the transmission channel to be tested and perform imaging processing.

[0026] Specifically, the first step is to acquire raw remote sensing data of the transmission corridor area to be measured. Raw remote sensing data includes raw optical images and raw radar images. Raw optical images typically refer to raw digital level images acquired by optical sensors that have not undergone geometric and radiometric calibration; raw radar images refer to raw intensity images acquired by synthetic aperture radar sensors that record the ground backscattering characteristics of the transmission corridor area to be measured.

[0027] The original optical image undergoes optical image standardization correction to generate a standard optical image. Optical image standardization correction refers to the process of eliminating the influence of optical sensor errors, differences in solar altitude angle, and atmospheric scattering and absorption on image quality, restoring the original digital level values ​​to the true spectral reflectance or radiance of the target object. The process of optical image standardization correction involves radiometric calibration.

[0028] After completing the radiometric calibration, atmospheric correction processing is required on the optical images of the power transmission channel area to be tested. Atmospheric correction processing aims to eliminate the radiation errors caused by the absorption and scattering of light by the atmosphere, and finally generate standard optical images.

[0029] The original radar image is subjected to radar image standardization correction to obtain a standard radar image. Radar image standardization correction refers to the process of converting the intensity values ​​of the radar echo signal into physically meaningful radar backscattering coefficients and suppressing the unique speckle noise in the radar image. The radar image standardization correction process involves radar calibration processing.

[0030] After obtaining the radar backscattering coefficient, the original radar image needs to be processed sequentially by multi-view processing and speckle filtering. The speckle filtering process uses a statistical filter to smooth the image pixel values, thereby suppressing multiplicative noise in the radar image and finally obtaining a standard radar image.

[0031] Standard optical images are spatially registered and resampled to generate optical images; similarly, standard radar images are spatially registered and resampled to generate radar images. Spatial registration refers to mapping multi-source remote sensing images acquired by different sensors and at different times to a unified geographic coordinate system, ensuring precise correspondence between pixel coordinates of the same geographical location on different images. Resampling refers to the process of re-estimating and filling the pixel grayscale values ​​of the power transmission channel area under test during image coordinate transformation using specific mathematical interpolation algorithms. Specific resampling algorithms include bilinear interpolation, nearest neighbor interpolation, or cubic convolution interpolation.

[0032] Optical and radar images were used as remote sensing image data for the power transmission channel area to be measured.

[0033] By performing the aforementioned standardization correction, spatial registration, and resampling processing on optical and radar images, errors between multiple sensor sources are effectively eliminated.

[0034] In a specific embodiment, the process of performing optical image standardization correction on the original optical image and generating a standard optical image is as follows: First, radiometric calibration is performed on the raw optical image. Radiometric calibration refers to the process of establishing a quantitative conversion relationship between the digital signal values ​​observed by the remote sensing sensor and the physical radiance values ​​of the corresponding target objects within the field of view. The pixel grayscale values ​​of the raw optical image only represent the recorded level intensity. Through radiometric calibration, physically meaningless pixel grayscale values ​​can be converted into physically meaningful radiance values. The specific calculation process for radiometric calibration uses the following radiometric calibration formula: In the formula, This represents the radiance value received by the sensor; This represents the radiation calibration gain coefficient of the sensor in a specific wavelength band. This represents the grayscale value of the corresponding pixel in the original optical image; This indicates the radiometric calibration offset of the sensor in a specific wavelength band.

[0035] After obtaining the radiance value, in order to further eliminate the radiation deviation caused by different imaging times, different solar altitude angles, and variations in the Earth-Sun distance, the radiance value needs to be converted into apparent reflectance. Apparent reflectance refers to the reflectance observed when the sensor is at the top of the atmosphere. The conversion process uses the following apparent reflectance conversion formula: In the formula, This represents the dimensionless apparent reflectance obtained after conversion; This indicates the astronomical distance between the Sun and the Earth at the moment of imaging; This represents the average solar spectral irradiance value in the upper atmosphere; The solar zenith angle at the moment of imaging.

[0036] After obtaining the apparent reflectance, atmospheric correction is performed on the optical image. Atmospheric correction refers to eliminating radiation errors caused by the absorption and scattering of electromagnetic waves by atmospheric components such as aerosols, water vapor, and carbon dioxide as they enter the atmosphere, thereby extracting the true ground reflectance of the target object. The specific operation of atmospheric correction involves calculating atmospheric environmental parameters such as atmospheric transmittance, atmospheric radiation, and spherical albedo using a pre-set atmospheric radiative transfer model. Subsequently, the atmospheric influence is eliminated using the following ground reflectance extraction formula to obtain a standard optical image: In the formula, This represents the true ground reflectance of the target object, i.e., the pixel value of a standard optical image; This represents the path reflectivity produced by atmospheric scattering; This represents the atmospheric transmittance from the sun to the ground. This represents the atmospheric transmittance layer by layer from the ground to the sensor. It represents the spherical albedo of the atmosphere.

[0037] After extracting the ground reflectance, the pixel values ​​of all bands are normalized and damaged pixels are repaired to generate a standard optical image. Through the radiometric calibration, apparent reflectance conversion, and atmospheric correction processes described above, external environmental interference and sensor measurement errors are effectively removed.

[0038] In a specific embodiment, the process of performing radar image standardization correction on the original radar image to obtain a standard radar image is as follows: First, the raw radar imagery undergoes radiometric calibration. Radiometric calibration refers to the process of converting the raw echo signal intensity acquired by the radar sensor into a physically meaningful radar backscattering coefficient. The raw radar imagery records the reflected energy of electromagnetic waves from the target object; however, due to factors such as observation distance and sensor gain, raw values ​​acquired at different times or locations cannot be directly compared. Radiometric calibration calculates the radar backscattering coefficient using the following radar calibration formula: In the formula, This represents the calculated radar backscattering coefficient, expressed in decibels. This represents the amplitude value of the corresponding pixel in the original radar image; This represents the preset radar calibration constant.

[0039] After radiometric calibration, multi-look processing is performed on the image. Multi-look processing refers to the process of reducing speckle noise intensity in radar images by averaging them along the range or azimuth direction. Speckle noise refers to granular random noise generated during coherent imaging due to the interference of echo signals from multiple scatterers within a spatial resolution cell. Multi-look processing involves merging multiple adjacent pixels of the original radar image into a new pixel. The intensity value of the new pixel is determined using the following multi-look averaging formula: In the formula, This represents the image intensity value generated after multi-view processing; Indicates the number of pixels merged in the azimuth direction; Indicates the number of pixels merged along the distance direction; This indicates that the coordinates in the original radar image are The power value corresponding to each pixel.

[0040] After multi-view processing, speckle filtering is required to smooth the image pixels. This speckle filtering process involves using a specific statistical filter to smooth the image pixels, further suppressing noise generated by the radar coherent imaging mechanism. The speckle filtering process uses the following local mean filtering formula to reconstruct the pixels: In the formula, Indicates the coordinates after filtering. Reconstructed pixel values ​​at the location; Represented by coordinates The average intensity value of all pixels within the central filtering window; Indicates the coordinates before filtering The original pixel intensity value at that location; This represents the weighting factor calculated based on the pixel variance and noise variance within the filtering window.

[0041] After completing the speckle filtering process, a standard radar image is finally generated. Through the above-mentioned radiometric calibration, multi-view processing, and speckle filtering, the inherent measurement errors of the radar sensor are eliminated.

[0042] In a specific embodiment, the process of spatially registering and resampling a standard optical image to generate an optical image is as follows: First, spatial registration is performed. Spatial registration refers to mapping the image to be corrected onto the coordinate system of a reference image, so that corresponding ground target points in the two images completely overlap geographically. The spatial registration process first requires selecting a reference image that covers the area of ​​the power transmission channel to be measured and has high-precision geographic information, and establishing a set of corresponding control points between the reference image and the standard optical image. The set of corresponding control points is extracted using a feature matching algorithm and includes multiple pixel coordinate pairs with the same ground feature attributes in both the reference image and the standard optical image. A spatial coordinate transformation model is then constructed using the set of corresponding control points. This model typically employs an affine transformation, specifically calculating the target coordinates of each pixel in the standard optical image in the reference image coordinate system using the following affine transformation formula: In the formula, This represents the horizontal target coordinates after a pixel of a standard optical image is mapped to the reference image coordinate system. This represents the vertical target coordinates after a pixel of a standard optical image is mapped to the reference image coordinate system. This represents the horizontal source coordinates of the standard optical image in the original image coordinate system; This represents the vertical source coordinates of the standard optical image in the original image coordinate system; , , , , , All represent the affine transformation coefficients obtained by fitting using the least squares method.

[0043] After determining the target coordinates, since the calculated target coordinates usually do not fall within an integer multiple of the reference image's pixel grid, resampling is required. Resampling refers to the process of estimating and filling the pixel grayscale values ​​at newly generated regular grid points using mathematical interpolation methods after geometric transformation of the image. The resampling process employs a bilinear interpolation algorithm, which considers the influence of the grayscale values ​​of the four neighboring pixels of the pixel to be calculated on the target pixel value. Specifically, the grayscale value of each pixel in the optical image is determined using the following bilinear interpolation formula: In the formula, This indicates the coordinates of the optical image generated after resampling. Output cell value at; This represents the normalized distance offset of the pixel to be calculated relative to its left-adjacent pixel in the horizontal direction. This represents the normalized distance offset of the pixel to be calculated relative to its upper neighboring pixel in the vertical direction. , , , These represent the grayscale values ​​of the four original pixels adjacent to the pixel to be calculated in the standard optical image.

[0044] After the above resampling process is completed, the generated image pixels will be aligned with the reference geographic coordinate grid to finally generate an optical image. By implementing spatial registration and resampling, geometric distortion caused by sensor imaging angle and aerial photography attitude deviation is eliminated, ensuring that the generated optical image has extremely high positioning accuracy in the geographic spatial dimension, providing a unified coordinate reference for the accurate identification of potential hazards in power transmission channels.

[0045] In a specific embodiment, the process of spatially registering and resampling standard radar images to generate radar images is as follows: First, spatial registration is performed. Spatial registration refers to transforming the geometric coordinate system of the standard radar image to a geographic coordinate system completely consistent with that of the optical image using a mathematical transformation model, thus aligning the pixel positions of different source images. The spatial registration process first selects the generated optical image as a geographic reference and extracts multiple corresponding control points with the same ground features between the standard radar image and the optical image. These corresponding control points include the row and column coordinates of pixels in the standard radar image and their corresponding geographic spatial coordinates in the optical image. Using multiple sets of corresponding control points, a quadratic polynomial transformation model is obtained through least squares fitting. The pixel coordinate transformation process in the standard radar image uses the following quadratic polynomial transformation formula: In the formula, This represents the converted geographic longitude coordinates or horizontal projected coordinates. This represents the converted geographic latitude coordinates or vertical projection coordinates; This represents the horizontal pixel coordinates of a standard radar image in the original radar coordinate system. This represents the vertical pixel coordinates of a standard radar image in the original radar coordinate system. to as well as to All of these represent transformation coefficients determined by fitting with control points of the same name.

[0046] After completing the spatial coordinate transformation, since the positions generated by the coordinate mapping are usually located at non-integer pixels, resampling is required. Resampling refers to the process of calculating and filling the observed values ​​of new pixels in the geometrically transformed coordinate grid using an interpolation function based on the intensity values ​​of adjacent original pixels. The resampling process employs a cubic convolution interpolation algorithm. The cubic convolution interpolation algorithm considers the surrounding area of ​​the point to be calculated... The grayscale contributions from a total of 16 pixels in the neighborhood are used to construct a cubic polynomial interpolation kernel. The calculation process of the cubic convolution interpolation algorithm adopts the following cubic convolution formula: In the formula, This indicates the coordinates of the radar image generated after resampling. Pixel intensity value at; Indicates the first element involved in the calculation of the standard radar image. The original intensity values ​​of the adjacent pixels; This represents the interpolation weight function value in the horizontal direction; This represents the interpolation weight function value in the vertical direction; as well as These represent the normalized distances between the pixel to be calculated and the original pixels used in the calculation.

[0047] After the above resampling process is completed, the generated image will completely eliminate the geometric distortion caused by radar side-view imaging, and finally generate radar image. By implementing the above spatial registration and resampling process, the spatial displacement error and scale difference between optical image and radar image are eliminated, and the pixel-level accurate fusion and alignment of multi-source remote sensing data is achieved.

[0048] Step S2: Input the remote sensing image data into a preset cross-segment target detection model so that the cross-segment target detection model can determine whether there is a cross-segment area in the power transmission channel to be tested based on the remote sensing image data; wherein, the cross-segment target detection model is trained by several target detection training samples; each target detection training sample includes historical remote sensing image data and corresponding cross-segment existence status labels.

[0049] In a preferred embodiment, a pre-defined cross-segment target detection model is trained in the following manner: Obtain several object detection training samples; Each target detection training sample is sequentially input into the cross-segment target detection model to train the model until a preset number of target detection training iterations are reached. Each time a target detection training sample is received, the cross-segment target detection model outputs a predicted existence state result corresponding to that sample. Based on the predicted existence state result and the corresponding cross-segment existence state label, a target detection loss function value is calculated. The cross-segment target detection model is then updated based on the target detection loss function value.

[0050] It should be noted that the remote sensing image data input to the cross-segment target detection model is not a single large-area map, but exists in the form of a multi-channel fused tile sequence. The multi-channel fused tile sequence is a tensor set formed by dividing the preprocessed optical image and radar image into a series of local image patches with overlapping areas according to a preset geographical interval, and then superimposing the local image patches in the channel dimension.

[0051] Specifically, each data unit input to the cross-segment target detection model is a four-dimensional tensor, and the structure of the tensor is defined by the following tensor description formula: In the formula, This represents the complete set of slices input to the cross-segment target detection model; Indicates the first region in the transmission channel under test Fusion image slices corresponding to each geographic segment; This indicates the total number of segments into which the power transmission channel under test is divided.

[0052] Each fused image slice Each is composed of spatial and channel information of a specific dimension, specifically represented by the following slice structure formula: In the formula, This represents the pixel height value of the merged image slice; This represents the pixel width value of the merged image slice; This represents the total number of channels in the fused image slice, which is determined by the sum of the optical channels and the radar channels.

[0053] In constructing the aforementioned fused image slices, the three channels of red, green, and blue visible light information and one channel of near-infrared information provided by the optical image are used as the basic channels, while the backscattering intensity information and coherence information provided by the radar image are used as supplementary channels. Through channel stitching, the four channels of the optical image and the two channels of the radar image are stacked in the depth direction to form a fused image slice with a total of six channels. Each fused image slice corresponds to a specific geographic coordinate span within the area of ​​the power transmission channel to be tested. A preset overlap ratio is set between adjacent fused image slices to ensure that continuous potential hazards crossing the slice boundaries are not lost during the detection process.

[0054] The above-mentioned organization of multi-channel fused slice sequences as input can effectively reduce the memory occupation of ultra-large area images of power transmission channels under test, and enhance the ability of cross-segment target detection models to identify small hidden targets in complex backgrounds by fusing optical texture features and radar scattering features in the same data structure.

[0055] In one specific embodiment, a pre-defined cross-segment target detection model is trained in the following manner: Several target detection training samples were obtained. These training samples are supervised learning sets constructed from remote sensing data collected over historical time periods, showing known hazard distribution. To ensure the cross-segment target detection model can handle real-world scenarios, the training samples are sourced from a historical remote sensing database. This database stores historical optical and radar images of the transmission channel area under test, along with corresponding historical hazard inspection records, from past operating cycles. Each target detection training sample contains a set of registered historical multi-source image blocks and a corresponding cross-segment presence status label. The cross-segment presence status label is a category and location annotation of hazard targets present in the historical multi-source image blocks, based on historical hazard inspection records.

[0056] In constructing training samples for object detection, to improve the model's adaptability to environmental features across different time periods, it is necessary to perform sample equalization on historical remote sensing data. The sample equalization process involves statistical analysis of the historical sample distribution, and the specific calculation logic is expressed by the following sample distribution weight formula: In the formula, This represents the loss weight compensation value for a specific category of potential hazards during the training process; This represents the total number of all historical image slices included in the object detection training samples; This indicates the total number of hazard target categories covered in the historical hazard inspection records; This indicates the number of historical image slices belonging to a specific category.

[0057] Each target detection training sample is sequentially input into the cross-segment target detection model for training until a preset number of training iterations are reached. Upon receiving each target detection training sample, the cross-segment target detection model outputs a predicted existence state result corresponding to that sample. Based on the predicted existence state result and the corresponding cross-segment existence state label, a target detection loss function value is calculated. This loss function value measures the difference between the cross-segment target detection model's predicted values ​​and actual historical data.

[0058] In calculating the target detection loss function, a segment consistency constraint term is introduced to enhance the model's learning of cross-segment spatial correlation features. The specific loss calculation process is expressed by the following comprehensive loss formula: In the formula, This represents the final calculated value of the target detection loss function; This indicates the classification loss component resulting from the identification of historical hidden danger categories; This represents the proportional adjustment coefficient corresponding to the classification loss. This represents the coordinate regression loss component resulting from the prediction of historical hazard locations; This represents the proportional adjustment coefficient corresponding to the coordinate regression loss; The constraint loss component represents the logical association of hidden danger characteristics between adjacent sections; This represents the proportional adjustment coefficient corresponding to the constraint loss.

[0059] The cross-segment object detection model is updated based on the object detection loss function value. The model update process involves using the gradient descent algorithm to fine-tune the neuron connection weights within the cross-segment object detection model based on the gradient signal generated by the object detection loss function value. The direction of the weight update aims to reduce the prediction error of the cross-segment object detection model on historical data.

[0060] It should be noted that the cross-segment target detection model employs a deep convolutional neural network architecture, consisting of a feature extraction backbone network, a cross-segment spatial feature fusion module, and a target detection head. The feature extraction backbone network extracts deep semantic features from the input multi-source remote sensing image data slices. The cross-segment spatial feature fusion module, located after the feature extraction backbone network, establishes long-range dependencies between image features of adjacent geographic segments through an attention mechanism, thereby capturing continuous hazard target features that cross segment boundaries. The target detection head, based on the fused feature vectors, regresses the location coordinates of the hazard targets and determines their category probability. The hierarchical connections of the cross-segment target detection model follow the principle of progressive feature layer integration, fusing shallow spatial information with deep semantic information through skip connections.

[0061] Regarding the input data settings for the cross-segment target detection model, the input data consists of a multi-channel fusion tensor composed of preprocessed optical and radar images. Optical images provide high-resolution texture information to characterize the apparent shape of potential targets; radar images provide backscattering features of ground features to assist in identifying physical entities in environments with optical obstruction or insufficient lighting. To reflect the inherent correlation between the input data and the power transmission channel scene, the cross-segment target detection model aligns the visible light and near-infrared bands of the optical images with the polarization intensity band of the radar images on the depth axis. Specifically, the spatial dimension of the multi-source remote sensing image data slices corresponds to the actual geographical scale of the power transmission channel, ensuring that the cross-segment target detection model can learn the comprehensive representation of targets such as transmission towers, conductors, surrounding vegetation, and construction machinery under multi-source data.

[0062] In the training steps of the cross-segment target detection model, the internal weight parameters of the model are first initialized. Then, training samples containing historical hazard information are input into the model in batches. For each training sample, the model performs forward propagation to generate a predicted state. The predicted state is calculated using the following prediction mapping formula: In the formula, This indicates the predicted state result output by the cross-segment target detection model; This represents the input multichannel fusion tensor; This represents the weight matrix of the feature extraction backbone network; This represents the nonlinear mapping function corresponding to the feature extraction backbone network; This represents the distribution of attention weights in the cross-segment spatial feature fusion module; This represents the feature aggregation function corresponding to the cross-segment spatial feature fusion module; This represents the parameter vector of the target detection head; This represents the classification and regression mapping function corresponding to the target detection head.

[0063] After obtaining the predicted state, the deviation between the predicted value and the true label in the historical inspection record is calculated using a preset object detection loss function. The object detection loss function includes positional deviation loss and category determination loss. Based on the object detection loss function value, the cross-segment object detection model's weights are updated using a stochastic gradient descent algorithm. The weight update process involves adjusting training parameters such as the learning rate, momentum factor, and weight decay coefficient. The training parameter settings should ensure that the cross-segment object detection model reaches convergence within a preset number of training iterations. The logic of the weight update is expressed by the following gradient update formula: In the formula, This represents the updated model weight parameters; This represents the current model weight parameters before the update; This represents the preset training learning rate parameter; This represents the calculated value of the comprehensive target detection loss function; This represents the vector of partial derivatives of the loss function with respect to the model parameters.

[0064] Through the aforementioned clear model architecture design, input and output data settings, and refined training steps, the cross-section target detection model can fully learn the spatiotemporal distribution patterns of potential hazards in power transmission channels, achieving deep integration and accurate identification of multi-source remote sensing information in specific power inspection scenarios, and significantly improving the automatic detection efficiency of potential hazards in complex cross-section environments.

[0065] Step S3: When there are cross-segment areas in the transmission channel area to be tested, the remote sensing image data is input into a preset cross-segment type recognition model so that the cross-segment type recognition model can determine the cross-segment type of the transmission channel area to be tested based on the remote sensing image data; wherein, the cross-segment type recognition model is trained by several type recognition training samples; each type recognition training sample includes historical remote sensing image data and corresponding type semantic labels.

[0066] In a preferred embodiment, a pre-defined cross-segment type recognition model is trained in the following manner: Obtain several types of recognition training samples; The training samples for each type of identification are sequentially input into the cross-segment type identification model to train the model until a preset number of type identification training iterations are reached. Each time a type identification training sample is received, the cross-segment type identification model outputs the predicted cross-segment type corresponding to that training sample. Based on the predicted cross-segment type and its corresponding semantic label, a type identification loss function value is calculated. The cross-segment type identification model is then updated based on the type identification loss function value.

[0067] Specifically, several type recognition training samples are acquired. These type recognition training samples refer to a multi-source remote sensing image dataset collected within a historical time period, covering known environmental features within the power transmission channel area. The type recognition training samples include historical optical images, historical radar images, and type semantic labels corresponding to the historical optical and radar images. The type semantic labels are textual or numerical supervisory information pre-annotated on historical image data to clearly indicate specific land cover categories or risk hazard types within the power transmission channel. During the acquisition of type recognition training samples, to improve the generalization ability of the cross-segment type recognition model to features of different geographical areas, sample equalization processing of the historical image data is required. The calculation process for sample equalization involves the following sample weight distribution formula: In the formula, This represents the sampling probability weight value assigned to a specific category of historical image slice during training; This represents the total number of all historical image slices used in the training process; This indicates the total number of categories included in the preset type semantic tags; This indicates the actual number of historical image slices belonging to the current specific category.

[0068] In constructing the cross-segment type recognition model, the model employs a neural network architecture consisting of a convolutional feature extraction layer based on deep residual connections, a cross-segment attention fusion layer, and a fully connected classification layer. The convolutional feature extraction layer extracts spatial texture features from the input historical optical imagery and ground feature scattering features from the input historical radar imagery. The cross-segment attention fusion layer, located after the convolutional feature extraction layer, aggregates local features into a global feature vector that characterizes the cross-segment environmental relationships by calculating the correlation weights between feature maps of different geographical segments. The fully connected classification layer then maps the global feature vector to a predefined category space.

[0069] Training samples for each type of recognition are sequentially input into the cross-segment type recognition model to train the model until a preset number of training iterations are reached. Upon receiving a type recognition training sample, the cross-segment type recognition model outputs the predicted cross-segment type corresponding to that training sample. The predicted cross-segment type refers to the probability prediction set for each preset category calculated by the cross-segment type recognition model based on the currently input multi-source image features. Based on the predicted cross-segment type and its corresponding semantic label, a type recognition loss function value is calculated. The calculation process for the type recognition loss function value involves the following cross-entropy discriminant formula: In the formula, This represents the calculated type recognition loss function value; This represents the total number of categories that the cross-segment type recognition model can identify; The first in the type semantic tag The true probability of each category is 1 if the class belongs to that category, and 0 otherwise. This indicates that the output of the cross-segment type recognition model is for the first segment. The predicted probability values ​​for each category.

[0070] The cross-segment type recognition model is updated based on the type recognition loss function value. The model update process involves calculating the partial derivative of the type recognition loss function value with respect to the weights of the neurons within the cross-segment type recognition model, and then adjusting the internal weight matrix using a pre-defined optimization algorithm to reduce the bias between the predicted cross-segment type and the type semantic label. The specific logic for weight adjustment follows the gradient iteration formula: In the formula, This represents the internal weight parameters of the cross-segment type recognition model generated after the update; This indicates the current internal weight parameters of the cross-segment type recognition model before the update; This represents the preset learning rate update step size factor; This represents the gradient vector generated by the type recognition loss function value with respect to the internal weight parameters.

[0071] The process of sample input, loss calculation, and weight adjustment is repeated until the classification accuracy of the cross-segment type recognition model on the validation set reaches a preset threshold or the preset number of type recognition training iterations are completed. By using training samples containing historical multi-source remote sensing features and their associated semantic labels for iterative training, the cross-segment type recognition model can accurately capture the coupling patterns of complex ground features within the transmission channel in both optical and radar dimensions, significantly improving the accuracy and reliability of risk type identification under complex cross-segment conditions.

[0072] Step S4: Based on the cross-section type, determine the hidden danger identification task for the transmission channel area to be tested.

[0073] In a preferred embodiment, determining the hazard identification task for the transmission channel area to be tested, based on the cross-segment type, includes: When the cross-section type is cross-transportation hub, the hidden danger identification task for the power transmission channel area to be tested is determined to be a geological deformation identification task; In the case where the cross-section type is cross-vegetation and water body, the task of identifying hidden dangers in the power transmission channel area to be tested is determined to be an environmental disaster identification task; When the cross-section type is cross-construction building, the task of identifying potential hazards in the power transmission channel area to be tested is determined to be an external intrusion identification task.

[0074] It should be noted that the hazard identification task for the transmission channel area under test is determined based on the cross-segment type. The cross-segment type refers to the classification result output by the cross-segment type identification model, which characterizes the geographical environment category and spatial distribution characteristics of the transmission channel area under test. The hazard identification task refers to a pre-defined image detection operation used to detect power operation safety risks within the specific geographical environment of the transmission channel area under test.

[0075] Specifically, the operational process for identifying potential hazards in the transmission channel area to be tested, based on the type of cross-section, includes: When the cross-section type is crossing a transportation hub, the task of identifying potential hazards in the area of ​​the power transmission channel to be tested is determined to be a geological deformation identification task. Crossing a transportation hub refers to areas where the power transmission channel crosses or is close to railways, highways, expressways, or large stations. The geological deformation identification task refers to using remote sensing imagery to detect whether there are physical displacement characteristics such as settlement or slippage in the transmission tower foundations, slopes, and roadbeds.

[0076] In cases where the transmission line crosses vegetation and water bodies, the task of identifying potential hazards in the area to be tested as an environmental disaster identification task. "Crossing vegetation and water bodies" refers to areas where the transmission line crosses forests, shrublands, rivers, lakes, or wetlands. The environmental disaster identification task refers to automated image recognition operations targeting risks such as wildfire smoke, excessively high discharge rates from vegetation, and tower foundation damage caused by flooding.

[0077] When the cross-section type is cross-construction building, the task of identifying potential hazards in the area of ​​the transmission channel to be tested is defined as an external intrusion identification task. Cross-construction building refers to areas surrounding the transmission channel where large construction machinery, newly built sheds, or civil engineering works are present. The external intrusion identification task refers to the identification and ranging operations for risks of third-party damage, such as large cranes encroaching on safety distances, illegal buildings occupying roads, or foreign objects being attached to safety nets.

[0078] By classifying the hazard identification task into geological deformation identification, environmental disaster identification, or external intrusion identification tasks according to different cross-section types, precise safety monitoring for different geographical backgrounds has been achieved, significantly improving the targeting and identification efficiency of hazard investigation in power transmission channels.

[0079] Step S5: Based on the hazard identification task, determine the hazard identification result of the transmission channel area to be tested according to the remote sensing image data of the transmission channel area to be tested.

[0080] In a preferred embodiment, based on the hazard identification task, the hazard identification result of the transmission channel area to be tested is determined according to the remote sensing image data of the area, including: When the hazard identification task is a geological deformation identification task, phase interferometry processing is performed on the radar image in the remote sensing image data of the transmission channel area to be measured to determine the geological cumulative deformation component of the transmission channel area to be measured; the geological cumulative deformation component is used as the hazard identification result of the transmission channel area to be measured. When the hazard identification task is an environmental disaster identification task, spectral features are extracted from the optical images in the remote sensing image data of the transmission channel area to be tested to determine the disaster distribution status of the transmission channel area to be tested; the disaster distribution status is used as the hazard identification result of the transmission channel area to be tested. When the hazard identification task is an external intrusion identification task, target identification is performed on the remote sensing image data of the power transmission channel area to be tested, and the power transmission lines and foreign objects in the power transmission channel area to be tested are extracted; based on the power transmission lines and foreign objects in the power transmission channel area to be tested, the spatial clearance distance between the foreign objects and the power transmission lines is calculated; the spatial clearance distance is used as the hazard identification result of the power transmission channel area to be tested.

[0081] Specifically, when the hazard identification task is a geological deformation identification task, phase interferometry is performed on radar images in the remote sensing image data of the transmission channel area to be measured to determine the cumulative geological deformation components of the transmission channel area. Phase interferometry refers to a signal processing method that uses the phase difference information between two or more radar images acquired at different times to invert the characteristics of minute surface displacements. The cumulative geological deformation components refer to the total displacement values ​​occurring on the surface within the transmission channel area to be measured within a continuous time period. During the phase interferometry process, the cumulative geological deformation components are determined using the following deformation inversion formula: In the formula, This represents the calculated cumulative geological deformation component; This indicates the wavelength of the electromagnetic waves used by the radar sensor when acquiring the radar images; This indicates the angle of incidence of the radar beam relative to the surface being measured. This represents the interference phase value after flat-ground effect elimination and atmospheric phase correction.

[0082] The geological cumulative deformation component is used as the result of hazard identification in the power transmission channel area to be tested.

[0083] When the hazard identification task is an environmental disaster identification task, spectral features are extracted from the optical images of the remote sensing imagery of the transmission line area to be monitored to determine the disaster distribution status of the area. Spectral feature extraction refers to the process of analyzing the differences in reflectance intensity across different spectral frequency bands in the optical images and combining this with the spectral characteristic curves of ground features to determine the physical attributes of the targets. The disaster distribution status refers to the determined geographic boundary coordinates, disaster center location, and disaster area value of areas within the transmission line area to be monitored, including areas affected by wildfire smoke, vegetation invasion, or flooding. During the spectral feature extraction process, the environmental disaster assessment index is calculated using the following disaster sensitivity evaluation formula: In the formula, This represents the calculated environmental disaster assessment index; This represents the normalized reflectance value corresponding to the near-infrared band in the optical image; This represents the normalized reflectance value corresponding to the visible red light band in the optical image; This represents a preset deviation correction constant for a specific environmental context.

[0084] The disaster distribution status is determined by comparing the environmental disaster assessment index with a preset disaster determination threshold. This disaster distribution status is then used as the hazard identification result for the power transmission channel area under test.

[0085] When the hazard identification task is an external intrusion identification task, target identification is performed on the remote sensing image data of the transmission line area to be tested, extracting the transmission lines and foreign objects within the area. Target identification refers to the process of using deep learning operators to perform feature mapping and regression on pixels in the remote sensing image, thereby locating and classifying specific ground objects. Foreign objects refer to non-power facility objects appearing within the transmission line area to be tested that may endanger the safe operation of the line; specifically, these include cranes, tower cranes, construction machinery, and scaffolding. Based on the transmission lines and foreign objects within the transmission line area to be tested, the spatial clearance distance between the foreign objects and the transmission lines is calculated. This spatial clearance distance is the shortest straight line length between the outermost edge contour point of the foreign object and the three-dimensional geometric center line of the transmission line. After performing target identification and determining the coordinates, the spatial clearance distance is determined using the following three-dimensional Euclidean distance formula: In the formula, This represents the calculated spatial clearance distance; These represent the first, second, and third coordinate components of the outermost edge point of the foreign object target in a spatial rectangular coordinate system, respectively. These represent the first, second, and third coordinate components of the point on the power transmission line closest to the foreign object target in a spatial rectangular coordinate system.

[0086] The aforementioned spatial clearance distance is used as the result of hazard identification in the power transmission channel area to be tested.

[0087] By performing targeted mathematical analysis and spatial distance calculations based on geological deformation identification tasks, environmental disaster identification tasks, or external intrusion identification tasks, a quantitative characterization of multi-source risks and hidden dangers in power transmission channels was achieved, significantly improving the objectivity of inspection and identification results and the accuracy of identification in multiple environments.

[0088] Step S6: If there are no cross-section areas in the area of ​​the power transmission channel to be tested, determine that there are no hidden dangers in the cross-section scenario in the area of ​​the power transmission channel to be tested.

[0089] By determining that the area of ​​the power transmission channel under test does not have cross-section characteristics, the conclusion that there are no hidden dangers in the cross-section scenario in the area of ​​the power transmission channel under test is directly drawn, which effectively eliminates invalid analysis objects and significantly improves the overall processing efficiency of hidden danger screening for large-scale remote sensing image data.

[0090] Based on the above method embodiments, the present invention provides corresponding apparatus embodiments.

[0091] like Figure 2As shown, an embodiment of the present invention provides a hazard identification device for power transmission channel areas based on cross-section scenarios, including: a data acquisition module, a target detection module, and a hazard identification module; The data acquisition module is used to acquire remote sensing image data of the power transmission channel area to be tested; The target detection module is used to input the remote sensing image data into a preset cross-segment target detection model, so that the cross-segment target detection model can determine whether there is a cross-segment area in the power transmission channel area to be tested based on the remote sensing image data; wherein, the cross-segment target detection model is trained by several target detection training samples; each target detection training sample includes historical remote sensing image data and corresponding cross-segment existence status labels; The hazard identification module is used to input the remote sensing image data into a preset cross-segment type identification model when there are cross-segment areas in the transmission channel area to be tested, so that the cross-segment type identification model can determine the cross-segment type of the transmission channel area to be tested based on the remote sensing image data; wherein, the cross-segment type identification model is trained by several type identification training samples; each type identification training sample includes historical remote sensing image data and corresponding type semantic labels; based on the cross-segment type, a hazard identification task for the transmission channel area to be tested is determined; based on the hazard identification task, the hazard identification result for the transmission channel area to be tested is determined based on the remote sensing image data of the transmission channel area to be tested; when there are no cross-segment areas in the transmission channel area to be tested, it is determined that there are no hazards in the cross-segment scenario in the transmission channel area to be tested.

[0092] In a preferred embodiment, the data acquisition module acquires remote sensing image data of the power transmission channel area to be measured, including: Acquire raw remote sensing data of the power transmission channel area to be measured; wherein, the raw remote sensing data includes raw optical images and raw radar images; Perform optical image standardization correction on the original optical image to generate a standard optical image; The original radar image is standardized and corrected to obtain a standard radar image. Spatial registration and resampling are performed on standard optical images to generate optical images; Standard radar images are spatially registered and resampled to generate radar images. Optical and radar images were used as remote sensing image data for the power transmission channel area to be measured.

[0093] It should be noted that the embodiments of the device described above correspond to the embodiments of the present invention described above, and can realize the method for identifying hidden dangers in power transmission channel areas based on cross-section scenarios as described above. Furthermore, the embodiments of the device described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Additionally, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between modules indicates that they have a communication connection, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without creative effort.

[0094] Based on the above-described method embodiments of the present invention, a corresponding embodiment of an electronic device is provided.

[0095] An embodiment of the present invention provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the method for identifying hidden dangers in power transmission channel areas based on cross-segment scenarios according to any one of the present invention, or, when the processor executes the computer program, it implements the functions of each module in the above-described device embodiments.

[0096] For example, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to perform the present invention. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the terminal device.

[0097] The terminal device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and memory.

[0098] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0099] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, applications required for at least one function, etc.; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital card (SD card), flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0100] Based on the above method embodiments, the present invention provides corresponding storage medium embodiments; Another embodiment of the present invention provides a storage medium including a stored computer program, wherein, when the computer program is running, the device where the storage medium is located executes any of the above-described methods for identifying potential hazards in power transmission channel areas based on cross-segment scenarios.

[0101] The aforementioned storage medium is a computer-readable storage medium, and the computer program includes computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0102] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0103] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for identifying potential hazards in power transmission corridor areas based on cross-section scenarios, characterized in that, include: Acquire remote sensing image data of the power transmission channel area to be tested; The remote sensing image data is input into a preset cross-segment target detection model, so that the cross-segment target detection model can determine whether there is a cross-segment area in the power transmission channel area to be tested based on the remote sensing image data; wherein, the cross-segment target detection model is trained by several target detection training samples; each target detection training sample includes historical remote sensing image data and corresponding cross-segment existence status labels; When there are cross-segment areas in the transmission channel area to be tested, the remote sensing image data is input into a preset cross-segment type recognition model, so that the cross-segment type recognition model can determine the cross-segment type of the transmission channel area to be tested based on the remote sensing image data; wherein, the cross-segment type recognition model is trained by several type recognition training samples; each type recognition training sample includes historical remote sensing image data and corresponding type semantic labels; Based on the cross-section type, determine the hidden danger identification task for the transmission channel area to be tested; Based on the aforementioned hazard identification task, the hazard identification results for the transmission channel area to be tested are determined according to the remote sensing image data of the area. If there are no cross-section areas in the area of ​​the transmission channel to be tested, it is determined that there are no potential hazards in the cross-section scenario in the area of ​​the transmission channel to be tested.

2. The method for identifying potential hazards in power transmission channel areas based on cross-section scenarios as described in claim 1, characterized in that, Acquire remote sensing image data of the power transmission corridor area to be tested, including: Acquire raw remote sensing data of the power transmission channel area to be tested; wherein, the raw remote sensing data includes raw optical images and raw radar images; Perform optical image standardization correction on the original optical image to generate a standard optical image; The original radar image is standardized and corrected to obtain a standard radar image. Spatial registration and resampling are performed on standard optical images to generate optical images; Standard radar images are spatially registered and resampled to generate radar images; Optical and radar images were used as remote sensing image data for the power transmission channel area to be measured.

3. The method for identifying potential hazards in power transmission channel areas based on cross-section scenarios as described in claim 2, characterized in that, The pre-defined cross-segment target detection model is trained using the following method: Obtain several object detection training samples; Each target detection training sample is sequentially input into the cross-segment target detection model to train the model until a preset number of target detection training iterations are reached. Each time a target detection training sample is received, the cross-segment target detection model outputs a predicted existence state result corresponding to that sample. Based on the predicted existence state result and the corresponding cross-segment existence state label, a target detection loss function value is calculated. The cross-segment target detection model is then updated based on the target detection loss function value.

4. The method for identifying potential hazards in power transmission channel areas based on cross-section scenarios as described in claim 3, characterized in that, The pre-defined cross-segment type recognition model is trained using the following method: Obtain several types of recognition training samples; The training samples for each type of identification are sequentially input into the cross-segment type identification model to train the model until a preset number of type identification training iterations are reached. Each time a type identification training sample is received, the cross-segment type identification model outputs the predicted cross-segment type corresponding to that training sample. Based on the predicted cross-segment type and its corresponding semantic label, a type identification loss function value is calculated. The cross-segment type identification model is then updated based on the type identification loss function value.

5. The method for identifying potential hazards in power transmission channel areas based on cross-section scenarios as described in claim 4, characterized in that, Based on the aforementioned cross-section type, the task of identifying potential hazards in the transmission channel area to be tested is determined, including: When the cross-section type is cross-transportation hub, the hidden danger identification task for the power transmission channel area to be tested is determined to be a geological deformation identification task; In the case where the cross-section type is cross-vegetation and water body, the task of identifying hidden dangers in the power transmission channel area to be tested is determined to be an environmental disaster identification task; When the cross-section type is cross-construction building, the task of identifying potential hazards in the power transmission channel area to be tested is determined to be an external intrusion identification task.

6. The method for identifying potential hazards in power transmission channel areas based on cross-section scenarios as described in claim 5, characterized in that, Based on the aforementioned hazard identification task, and using remote sensing image data of the transmission channel area to be tested, the hazard identification results for the transmission channel area to be tested are determined, including: When the hazard identification task is a geological deformation identification task, phase interferometry processing is performed on the radar image in the remote sensing image data of the transmission channel area to be measured to determine the geological cumulative deformation component of the transmission channel area to be measured; the geological cumulative deformation component is used as the hazard identification result of the transmission channel area to be measured. When the hazard identification task is an environmental disaster identification task, spectral features are extracted from the optical images in the remote sensing image data of the transmission channel area to be tested to determine the disaster distribution status of the transmission channel area to be tested; the disaster distribution status is used as the hazard identification result of the transmission channel area to be tested. When the hazard identification task is an external intrusion identification task, target identification is performed on the remote sensing image data of the power transmission channel area to be tested, and the power transmission lines and foreign objects in the power transmission channel area to be tested are extracted; based on the power transmission lines and foreign objects in the power transmission channel area to be tested, the spatial clearance distance between the foreign objects and the power transmission lines is calculated; the spatial clearance distance is used as the hazard identification result of the power transmission channel area to be tested.

7. A hazard identification device for power transmission channel areas in cross-section scenarios, characterized in that, Also includes: Data acquisition module, target detection module, and hazard identification module; The data acquisition module is used to acquire remote sensing image data of the power transmission channel area to be tested; The target detection module is used to input the remote sensing image data into a preset cross-segment target detection model, so that the cross-segment target detection model can determine whether there is a cross-segment area in the power transmission channel area to be tested based on the remote sensing image data; wherein, the cross-segment target detection model is trained by several target detection training samples; each target detection training sample includes historical remote sensing image data and corresponding cross-segment existence status labels; The hazard identification module is used to input the remote sensing image data into a preset cross-segment type identification model when there are cross-segment areas in the transmission channel area to be tested, so that the cross-segment type identification model can determine the cross-segment type of the transmission channel area to be tested based on the remote sensing image data; wherein, the cross-segment type identification model is trained by several type identification training samples; each type identification training sample includes historical remote sensing image data and corresponding type semantic labels; based on the cross-segment type, a hazard identification task for the transmission channel area to be tested is determined; based on the hazard identification task, the hazard identification result for the transmission channel area to be tested is determined based on the remote sensing image data of the transmission channel area to be tested; when there are no cross-segment areas in the transmission channel area to be tested, it is determined that there are no hazards in the cross-segment scenario in the transmission channel area to be tested.

8. The hazard identification device for power transmission channel areas based on cross-section scenarios as described in claim 7, characterized in that, The data acquisition module acquires remote sensing image data of the power transmission channel area to be measured, including: Acquire raw remote sensing data of the power transmission channel area to be tested; wherein, the raw remote sensing data includes raw optical images and raw radar images; Perform optical image standardization correction on the original optical image to generate a standard optical image; The original radar image is standardized and corrected to obtain a standard radar image. Spatial registration and resampling are performed on standard optical images to generate optical images; Standard radar images are spatially registered and resampled to generate radar images; Optical and radar images were used as remote sensing image data for the power transmission channel area to be measured.

9. An electronic device, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the method for identifying potential hazards in a power transmission channel area based on a cross-segment scenario as described in any one of claims 1 to 6.

10. A storage medium, characterized in that, The storage medium includes a stored computer program, wherein, when the computer program is running, it controls the device where the storage medium is located to execute the hazard identification method for power transmission channel areas based on cross-segment scenarios as described in any one of claims 1 to 6.

Images