Method and device for drone inspection and identification of power transmission towers

Abstract

This invention provides a method and device for UAV inspection and identification of power transmission towers, belonging to the field of power facility health monitoring technology. The invention obtains the polar coordinates of the guy wire endpoints and calculates the guy wire length using UAV laser ranging to assess the degree of deformation anomaly; based on a 5-channel hyperspectral imaging system, it utilizes orthogonal flight path scanning imaging combined with a CNN algorithm to locate insulators in real time, and analyzes the insulator condition using UV band radiation values, discharge gradients, and temperature differences; it extracts reflectivity data in the 550-700nm band and calculates the corrosion area ratio using the band ratio method; and it comprehensively assesses the tower's health status by considering guy wire deformation, insulator anomalies, and corrosion degree. This method integrates multi-source data and intelligent algorithms to achieve high-precision detection of guy wire deformation, insulator defects, and corrosion area, solving the problems of low efficiency and single parameters in traditional inspections, and significantly improving the automation and reliability of power transmission tower condition assessment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power facility health monitoring, in particular to a UAV inspection and identification method and device for a power transmission tower. BACKGROUND

[0002] With the rapid development of the power industry, the power transmission tower as the key infrastructure of the power grid system, its health status directly affects the safety and stability of power transmission. Traditional inspection relies on manual climbing or ground observation, which has the disadvantages of low efficiency, high risk, single data, etc. In recent years, UAV technology has been gradually applied to power inspection, and non-contact detection is realized by carrying optical sensors, but due to the problems of insufficient multi-source data fusion, poor algorithm adaptability and limited detection accuracy, it is difficult to comprehensively evaluate the comprehensive state of the tower's stay wire deformation, insulator defects and metal corrosion, etc., leading to the lag of operation and maintenance decision-making, which has become a bottleneck restricting the development of intelligent inspection:

[0003] In the prior art, a typical UAV inspection scheme usually uses a single sensor, such as a visible light camera or a laser radar to obtain local data of the tower. For example, by hovering the UAV to take pictures of the tower, the threshold segmentation method is used to identify the position of the insulator, and the degree of contamination is judged in combination with artificial experience; or based on the laser range finder to measure the end point coordinates of the stay wire, and compare the historical data to estimate the deformation. In addition, some methods try to introduce infrared thermal imaging to detect insulator temperature anomalies, but rely on fixed threshold to determine local discharge, which is easily disturbed by the environment. In the aspect of metal corrosion detection, the existing technology mostly uses visible light band image to analyze the rust area, and extracts the corrosion area through color features, but it is significantly affected by light conditions and background interference, and the false detection rate is high.

[0004] The above prior art has the following technical defects: the existing methods mostly process data independently for a single detection target (such as deformation or local temperature), lack of multi-dimensional collaborative analysis of tower structure, electrical components and material state, resulting in insufficient integrity and accuracy of comprehensive health evaluation; sensor data is easily disturbed by light changes, weather conditions and background noise, for example, optical detection fails in backlight or shadow scenes, and thermal imaging is affected by environmental temperature fluctuations; the existing scheme mostly relies on static threshold or offline data analysis, and cannot fuse multi-modal data in real time.

[0005] The above information disclosed in the background section is only used to strengthen the understanding of the background of the present disclosure, so it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0006] The purpose of the present application is to provide a UAV inspection and identification method and device for a power transmission tower to solve the problems raised in the background.

[0007] To achieve the above object, the present application provides the following technical solutions:

[0008] The unmanned aerial vehicle inspection and identification method for power transmission towers specifically includes the following steps:

[0009] Step 1: A reference point is set on the ground in advance, and the unmanned aerial vehicle hovers above the end of the stay wire of the power transmission tower, identifies the height difference of the ground reference point through the laser ranging function, calculates the polar coordinates of the stay wire end point, and then calculates the length of the stay wire based on the polar coordinates of the stay wire end point to evaluate the abnormality degree of the stay wire deformation;

[0010] Step 2: The unmanned aerial vehicle is equipped with a 5-channel hyperspectral imaging system to perform orthogonal flight line scanning high-precision imaging, extract the radiation value of the 300-900nm band, and locate the insulator in real time through the CNN algorithm, and judge the state of the insulator by calculating the average radiation value of the UV band, the local discharge area gradient and the temperature difference;

[0011] Step 3: Extract the radiation value of the 550-700nm band, convert it into reflectivity after shielding the insulator area, and calculate the corrosion area ratio based on the reflectivity data using the band ratio method;

[0012] Step 4: Integrate the corrosion area ratio, abnormal state insulator and stay wire deformation abnormality degree to evaluate the comprehensive health status of the power transmission tower;

[0013] Further, the laser ranging function is used to identify the height difference of the ground reference point, and the polar coordinates of the stay wire end point are calculated, including the following steps:

[0014] The reference point is set on the ground in advance, and the coordinates of the ground reference point measured by the total station are The unmanned aerial vehicle hovers above the end of the stay wire, the laser ranging function is turned on, and the height difference of the ranging point to the ground reference point is output The measured polar coordinates of the stay wire end point are , wherein represents the distance of the stay wire end point obtained by laser ranging, represents the angle of the laser ranging direction in the horizontal plane, represents the angle of the ranging direction in the vertical plane.

[0015] Further, the method for calculating the length of the stay wire based on the polar coordinates of the stay wire end point and evaluating the abnormality degree of the stay wire deformation is as follows:

[0016] The measured polar coordinates of the stay wire end point are converted into three-dimensional coordinates:

[0017]

[0018]

[0019]

[0020] The length of the pull line is calculated as:

[0021]

[0022] In the formula, represents the length of the pull line;

[0023] The deformation anomaly degree formula is constructed as:

[0024]

[0025] In the formula, represents the initial length of the pull line, represents the deformation anomaly degree.

[0026] Further, the unmanned aerial vehicle carrying the 5-channel hyperspectral imaging system is based on a spectral range of 300-900 nm, and the wavebands are divided into 300-400 nm, 400-500 nm, 550-700 nm, 700-800 nm, and 800-900 nm, and correspond to the 1st, 2nd, 3rd, 4th, and 5th channels, respectively.

[0027] Further, the method for real-time positioning of insulators through the CNN algorithm is:

[0028] The model for positioning insulators is pre-trained as:

[0029] At least 1000 hyperspectral images of power transmission towers are collected, and the size is uniformly modified under the condition of maintaining the aspect ratio as: The boundary box of the overall outline of the insulator is labeled through the LabelImg tool: wherein, represents the column position on the left side of the boundary box, represents the row position at the top of the boundary box, represents the column position on the right side of the boundary box, represents the row position at the bottom of the boundary box, and the class label is set as insulator,

[0030] After that, it is converted into normalized center point coordinates:

[0031]

[0032] In the formula, represents the center point coordinates, represents the center coordinates of the insulator, represents the width of the insulator, represents the height of the insulator, represents the width of the corresponding image, represents the height of the corresponding image;

[0033] The hyperspectral image, its corresponding center point coordinates, and class label are input into the ResNet-18 model. The transformed center point coordinates are used as the ground truth labels, and the output dimension of the last fully connected layer is modified to the corresponding center point coordinates. and confidence score Output a 5-dimensional vector dataset:

[0034]

[0035] In the formula, This indicates that the output is a 5-dimensional vector dataset, where the confidence score is calculated using the following formula:

[0036]

[0037] In the formula, This represents the raw confidence score of the fully connected layer output;

[0038] The center point coordinates output by the ResNet-18 model are used as the predicted results. The error between the predicted and actual center point coordinates is calculated using a localization loss function. The localization loss function used is as follows:

[0039]

[0040] In the formula, This represents the localization loss function. Indicates the first The predicted values ​​of each parameter, Indicates the first The true values ​​of each parameter, among which... This indicates the index of the parameter in the center point coordinates. These correspond to the x-coordinate of the center point, the y-coordinate of the center point, the width of the bounding box, and the height of the bounding box, respectively. Let represent the loss function, and satisfy:

[0041]

[0042] Construct the classification loss formula:

[0043]

[0044] In the formula, Indicates classification loss;

[0045] Construct the total loss formula:

[0046]

[0047] In the formula, Indicates the total loss;

[0048] Set training parameters:

[0049] Learning rate: Batch size: Training rounds: ;

[0050] The real-time transmission tower image taken by the unmanned aerial vehicle is unified to a fixed size, and the hyperspectral image is calibrated to The 5-dimensional vector dataset is output by inputting the model pre-trained to locate the insulator:

[0051]

[0052] Set the detection threshold , and Only the bounding box of is retained, and the location of the insulator is completed.

[0053] Further, the method for judging the state of the insulator by calculating the average radiation value of the UV band, the partial discharge area gradient and the temperature difference is:

[0054] The output 5-dimensional vector dataset is subjected to inverse normalization pixel coordinates, and the insulator area is extracted from the hyperspectral image based on the inverse normalized pixel coordinates, and the average radiation value is calculated by extracting the radiation value of 5 channels:

[0055]

[0056] In the formula, represents the total number of pixels in the insulator area, represents the radiation value of the th channel of the th pixel in the insulator area, wherein , , and the unit of the radiation value is unified to: ;

[0057] The partial discharge area gradient is calculated as:

[0058]

[0059] In the formula, represents the partial discharge area gradient, represents the average radiation value of the 1st channel, represents the average radiation value of the 2nd channel;

[0060] The temperature difference is calculated based on the radiation values of the 4th and 5th channels. First, the Planck's law formula is constructed: ​

[0061]

[0062] In the formula, represents the center wavelength of the wave band, represents the temperature, , , represents the radiation value at the wavelength ;

[0063] The temperature at the wavelength is obtained by inversion , The temperature at the wavelength ;

[0064] The temperature difference is calculated by the temperature at the wavelength and the temperature at the wavelength ;

[0065]

[0066] In the formula, represents the temperature difference;

[0067] An abnormality judgment formula is constructed:

[0068]

[0069] In the formula, represents the abnormality judgment formula, when , it is judged that the insulator is in an abnormal state, and when , it is judged that the insulator is in a normal state.

[0070] Further, the area ratio of the corrosion area is calculated based on the reflectivity data by using the wave band ratio method after shielding the insulator area:

[0071] All insulator areas in the hyperspectral image of

[0072]

[0073]

[0074] In the formula, represents the radiation value at the wavelength 550 nm in the background area, represents the radiation value at the wavelength 550 nm of the reference white board, represents the standard reflectivity of the white board, and is , represents the radiation value of the background area at a wavelength of 680 nm, represents the radiation value of the reference white board at a wavelength of 680 nm, represents the reflectivity at a wavelength of 550 nm in the background area, represents the reflectivity at a wavelength of 680 nm in the background area;

[0075] The band ratio method is used to construct a ratio formula:

[0076]

[0077] In the formula, represents the ratio formula, i.e. the reflectivity at a wavelength of 680 nm to the reflectivity at a wavelength of 550 nm in the background area;

[0078] The ratio threshold is set to , and the corrosion area ratio is calculated:

[0079]

[0080] In the formula, represents the band ratio of the th pixel in the background area, wherein, , represents the total number of pixels in the background area, represents an indicator function, when , when , , represents the corrosion area ratio.

[0081] 8. The unmanned aerial vehicle inspection and identification method for a power transmission tower according to claim 1, wherein the method for evaluating the comprehensive health status of the power transmission tower is:

[0082] The corrosion area ratio threshold is set to , and the guy wire deformation threshold is ;

[0083] When the corrosion area ratio , the number of abnormal state insulators , and the guy wire deformation abnormality , the evaluation result of the power transmission tower is normal state;

[0084] When any one of the corrosion area ratio , the number of abnormal state insulators , or the guy wire deformation abnormality is met, the evaluation result of the power transmission tower is general alarm state.

[0085] When any two or more of the following conditions are met: the corrosion area ratio , the number of insulators in abnormal state , or the abnormality degree of the stay deformation , the evaluation result of the power transmission tower is an urgent warning state.

[0086] The application further provides a UAV inspection and identification device for a power transmission tower, which is used to execute the UAV inspection and identification method for a power transmission tower as described above, and comprises:

[0087] A stay deformation monitoring module is configured to set a reference point on the ground in advance, and the UAV hovers above the end of the stay of the power transmission tower, identifies the height difference of the reference point on the ground through the laser ranging function, calculates the polar coordinates of the stay end point, and then calculates the stay length based on the polar coordinates of the stay end point to evaluate the abnormality degree of the stay deformation.

[0088] An insulator detection module is configured to carry a 5-channel hyperspectral imaging system on the UAV, execute orthogonal flight line scanning high-precision imaging, extract the radiation value in the 300-900nm band, locate the insulator in real time through the CNN algorithm, and judge the state of the insulator by calculating the average radiation value in the UV band, the gradient of the local discharge area, and the temperature difference.

[0089] A corrosion evaluation module is configured to extract the radiation value in the 550-700nm band, convert the reflectivity after shielding the insulator area, and calculate the corrosion area ratio based on the reflectivity data using the band ratio method.

[0090] A health evaluation module is configured to comprehensively evaluate the overall health state of the power transmission tower based on the corrosion area ratio, the abnormal state insulator, and the abnormality degree of the stay deformation.

[0091] Compared with the prior art, the application has the following advantages:

[0092] The application solves the limitations of the single parameter evaluation of the traditional method by fusing laser ranging to evaluate the stay deformation, hyperspectral imaging to monitor the state of the insulator, and detection of the corrosion area, and constructing a multi-data fusion model. The improved CNN algorithm is used to realize real-time positioning of the insulator, and the positioning accuracy reaches the pixel level. The state of the insulator is judged by the average radiation value in the UV band, the gradient of the local discharge area, and the temperature difference, which avoids the dependence on experience for manual identification of the quality of the insulator, and has higher accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0093] Figure 1 The figure is a schematic diagram of the overall method of the application.

[0094] Figure 2 The figure is a statistical diagram of the detection threshold and the recall rate of the application.

[0095] Figure 3 The detection threshold and accuracy statistical chart of the present application;

[0096] Figure 4 The detection threshold and accuracy statistical chart of the present application;

[0097] Figure 5 The detection threshold and accuracy statistical chart of the present application;

[0098] Figure 6 The corrosion area ratio statistical chart of the present application;

[0099] Figure 7 The abnormal insulator quantity statistical chart of the present application;

[0100] Figure 8 The pull wire deformation anomaly degree statistical chart of the present application;

[0101] Figure 9 The overall device module schematic diagram of the present application. DETAILED DESCRIPTION

[0102] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application is further described in detail below in combination with specific embodiments.

[0103] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present application should be understood as the usual meaning understood by those skilled in the art to which the present application belongs. The "first", "second" and similar words used in the present application do not represent any order, quantity or importance, but are only used to distinguish different components. "Include" or "contain" and similar words mean that the elements or objects before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connected" or "connected" and similar words are not limited to physical or mechanical connection, but can include electrical connection, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to represent relative positional relationship, when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0104] Embodiment:

[0105] Please refer to Figures 1 to 8 The present application provides a technical scheme:

[0106] The unmanned aerial vehicle inspection identification method for power transmission towers, the specific steps include:

[0107] Step 1: Set a reference point on the ground in advance, and the UAV hovers above the end of the stay wire of the power transmission tower. The height difference between the reference point on the ground and the UAV is identified by the laser ranging function, and the polar coordinates of the stay wire end point are calculated. Then, the length of the stay wire is calculated based on the polar coordinates of the stay wire end point, and the abnormality degree of the stay wire deformation is evaluated.

[0108] A reference point is set on the ground in advance, and a crosshair is engraved. This is because the reference point on the ground is the conversion reference of the UAV laser coordinates and the geodetic coordinate system, and it ensures that the length of the stay wire can be calculated by three-dimensional coordinates. Therefore, the coordinates of the ground reference point measured by a total station such as Leica TS60 are , which provides accurate ground reference point coordinate data. The UAV hovers above the connection point of the stay wire, and the crosshair is aligned with the center of the connection point through the zoom of the visible light camera. The laser ranging function is turned on, and the angle deviation caused by the inclination of the laser beam is avoided. The height difference between the ranging point and the ground reference point is output . The measured polar coordinates of the stay wire end point are , where represents the distance of the stay wire end point obtained by laser ranging, represents the angle of the laser ranging direction in the horizontal plane, represents the angle of the ranging direction in the vertical plane.

[0109] The measured polar coordinates of the stay wire end point are converted to three-dimensional coordinates:

[0110]

[0111]

[0112]

[0113] The length of the stay wire is calculated:

[0114]

[0115] In the formula, represents the length of the stay wire,

[0116] The deformation abnormality formula is constructed:

[0117]

[0118] In the formula, represents the initial length of the stay wire, which is preferably searched from the tower drawing or measured during operation as the initial length of the stay wire, represents the deformation abnormality degree. When the stay wire is deformed due to corrosion, fatigue, overload, etc., its actual length will deviate from the initial design value. By monitoring ​The mechanical property degradation of the stay wire can be found in advance, and the tower tilt or collapse caused by insufficient tension can be avoided. According to the industry standard, the maximum allowable deviation of the allowable stay wire length is only Therefore, in the subsequent setting of the stay wire deformation threshold , .

[0119] Step 2: The UAV carries a 5-channel hyperspectral imaging system, performs orthogonal flight line scanning high-precision imaging, extracts 300-900nm band radiation values, and locates the insulator in real time through the CNN algorithm. The insulator state is judged by calculating the average radiation value of the UV band, the local discharge area gradient and the temperature difference;

[0120] In the transmission tower inspection, the insulator is a key component, and its state directly affects the safe operation of the transmission line. However, in a complex transmission tower scene, the insulator may be blocked or confused with the background. Through the positioning model, the location of the insulator in the image can be quickly and accurately found, providing a basis for subsequent evaluation of its state, so as to improve the inspection efficiency and reduce the workload of manual identification of the insulator.

[0121] Collect at least 1000 hyperspectral images of transmission towers. Hyperspectral images contain rich spectral information, and different substances have unique spectral characteristics in different bands. Among them, channels 1-5 correspond to 300-400nm, 400-500nm, 550-700nm, 700-800nm, and 800-900nm, respectively. Using these features, the material and state of the insulator can be more accurately identified. The included RGB image is a common color image that can provide intuitive visual information. Combining the two can fully utilize the advantages of different types of images, improve the generalization ability and recognition accuracy of the model, and maintain the aspect ratio to modify the size to: Uniform image size can ensure that the model can normally process input data and better adapt to deep learning models. Using the rectangular frame drawing tool provided by the LabelImg tool, the overall outline of the insulator is selected, and the boundary box of the overall outline of the insulator is labeled as: , where represents the column position on the left side of the boundary box, represents the row position at the top of the boundary box, represents the column position on the right side of the boundary box, represents the row position at the bottom of the boundary box, and the class label is insulator. Save it as XML format;

[0122] Then, it is converted into normalized center point coordinates:

[0123]

[0124] wherein, represents the center point coordinate, represents the insulator center coordinate, represents the width of the insulator, represents the height of the insulator, represents the width of the corresponding image, represents the height of the corresponding image, the normalized center point coordinate can eliminate the influence of image size, so that the model can learn and predict uniformly on images of different sizes. At the same time, the center point coordinate and the width and height information are more in line with the input and processing method of the model, which is convenient for the model to calculate and learn;

[0125] The processed hyperspectral image is used as input data, and the corresponding center point coordinate is used as the true label. The class label is used for classification task. These data are input into the ResNet-18 model in batches according to a certain batch size, for example, 32. For the first convolutional layer, the input channel number is changed from 3 to 5 to match the 5 channel number of the hyperspectral image, and the weights are randomly initialized. The output dimension of the last fully connected layer is modified to the corresponding center point coordinate and the confidence score , output a 5-dimensional vector data set:

[0126]

[0127] wherein, represents the output 5-dimensional vector data set, and ResNet-18 is a pre-trained deep convolutional neural network with strong feature extraction capability. By inputting data into the model for training, the pre-trained weights and network structure can be used to quickly learn the features of the insulator, improve the training efficiency and accuracy of the model. The confidence score can represent the credibility of the model's prediction results, which is helpful for subsequent screening and judgment of the prediction results;

[0128] wherein, the calculation formula of the confidence score is:

[0129]

[0130] wherein, represents the original confidence score output by the fully connected layer;

[0131] The center point coordinate output by the ResNet-18 model is used as the prediction result, and the error between the actual center point coordinate and the prediction result is calculated by using the positioning loss function. The positioning loss function is:

[0132]

[0133] wherein, represents a positioning loss function, represents the predicted value of the th parameter, represents the true value of the th parameter, wherein, represents the index of the parameter in the center point coordinates, , respectively corresponding to the center point horizontal coordinate, the center point vertical coordinate, the bounding box width and the bounding box height, used to calculate the error between the center point coordinates predicted by the model and the true center point coordinates, for example, represents the predicted coordinate, represents the true coordinate, using The loss function can use square loss when the error is small, so that the model converges faster; when the error is large, use linear loss to avoid gradient explosion, represents the loss function, and satisfies:

[0134]

[0135] The classification loss formula is constructed as follows:

[0136]

[0137] In the formula, represents the classification loss, This term is a sigmoid function, which is used to map the output confidence score to the probability interval , This term further takes the logarithm of the probability. When the confidence score is higher, the value of the classification loss is closer to 1, indicating smaller loss. Conversely, when the confidence score is lower, the value of the classification loss is closer to 0, indicating larger loss. After a large number of experiments and research, it is proved that this method is suitable for binary classification problems, can reduce the error classification, and promote the model to quickly correct. It is used to calculate the classification error of the model for the insulator class, and can measure the difference between the class probability predicted by the model and the true class;

[0138] The total loss formula is constructed as follows:

[0139]

[0140] In the formula, represents the total loss, which is the sum of the positioning loss and the classification loss, realizing the linear addition of the positioning loss and the classification loss, so that the model optimizes the positioning and classification tasks at the same time in the training process, guarantees the accuracy of classification and positioning, and improves the overall performance of the model;

[0141] Set the training parameters:

[0142] Learning rate: , make the model more stable in the training process, avoid jumping over the optimal solution because the step is too large;

[0143] Batch size: , the batch size determines the number of data samples input into the model each time, which can improve training efficiency and avoid leading the model to a local optimal solution;

[0144] Training rounds: , the model learns the characteristics of the data sufficiently, and also avoids overfitting of the model;

[0145] The power transmission tower image captured by the unmanned aerial vehicle in real time is unified to a fixed size of , and the hyperspectral image with a size of is input into the pre-trained model for locating the insulator, and a 5-dimensional vector data set is output:

[0146]

[0147] Set the detection threshold , and , only keep the bounding box with , complete the location of the insulator, since the unmanned aerial vehicle needs to ensure the detection speed and the continuity of the results during real-time inspection, a too high detection threshold may cause the detection box to lose the target, affecting the stability of the positioning, according to the tolerance and consequences of the missed detection, the detection threshold is set to , which may cause serious accidents.

[0148] The normalized center point coordinates need to be converted back to the pixel coordinates of the original image, so that the insulator area can be accurately extracted from the image, therefore, the 5-dimensional vector data set is subjected to inverse normalization pixel coordinates, and is output, based on the inverse normalization pixel coordinates, the insulator area is extracted from the hyperspectral image with a size of , and the average radiation value is calculated by extracting the radiation values of 5 channels:

[0149]

[0150] In the formula, represents the total number of pixels in the insulator area, represents the radiation value of the th pixel in the th channel in the insulator area, wherein , ;

[0151] The radiation value of corona discharge in the UV band is several times that of blue light. Therefore, the radiation difference between UV (300-400nm) and blue light (400-500nm) is extracted to calculate the gradient of the partial discharge region.

[0152]

[0153] In the formula, This represents the gradient of the partial discharge region. This represents the average radiation value of the first channel. This represents the average radiation value of the second channel, and 100 is the difference between the center wavelengths of the first and second channels. Eliminating the influence of dimensions makes it closer to the physical characteristics of discharge.

[0154] The temperature difference is calculated based on the radiance values ​​of channels 4 and 5. These two channels, corresponding to wavelengths of 750nm and 850nm, represent an ideal combination for dual-band infrared thermometry. Planck's law indicates that monochromatic radiance has a non-linear relationship with temperature, but a single wavelength cannot eliminate the influence of emissivity. Selecting two neighboring wavelengths to simultaneously invert the temperature can eliminate interference. First, the Planck's law formula is constructed:

[0155]

[0156] In the formula, Indicates the center wavelength of the band. Indicates temperature. , , Indicates wavelength The radiation value at that location;

[0157] Through inversion Temperature at wavelength is , Temperature at wavelength is ;

[0158] pass Temperature at wavelength and Temperature at wavelength Calculate the temperature difference:

[0159]

[0160] In the formula, Indicates temperature difference, temperature and temperature The combination of the two can capture the tiny temperature gradient on the surface of the insulator, that is, the temperature difference between the partial discharge point and the normal area;

[0161] Construct an anomaly detection formula:

[0162]

[0163] wherein, represents an abnormality judgment formula, when the insulator is judged to be in an abnormal state, and when the insulator is judged to be in a normal state.

[0164] According to the statistics of 1000 power transmission towers in the past 3 years, the average UV band radiation value of the normal insulator is 130, so the insulator is determined to be abnormal. According to the experiment, the gradient of the partial discharge area is According to the recommended standard of the power industry, the temperature difference of the insulator higher than 10℃ is a serious defect, and higher than 5℃ is an early warning threshold. For example, the morning dew reflection may cause The pollution on the surface of the insulator may reach The large day and night temperature difference in the region may cause Therefore, when the output is 1, indicating that the insulator is abnormal, otherwise the output is 0, indicating that the insulator is normal. Table 1 shows that after collecting 40 groups of hyperspectral images of different power transmission tower images and processing the UV radiation value, the gradient of the partial discharge area and the temperature difference according to the above calculation formula, the insulator state judgment result is output by comparing with the set threshold value. Finally, the UV radiation value, the gradient of the partial discharge area, the temperature difference and the insulator state judgment result are summarized to form a data report.

[0165]

[0166] Table 1 Insulator state judgment data report

[0167] As shown in Figures 2-5 , based on the detection threshold setting of the insulator positioning model, the experiment verification is carried out based on the detection threshold setting method of the insulator positioning model for a sample set containing 100 real insulator hyperspectral images and 100 non-insulator hyperspectral images. The recall rate, precision rate, accuracy rate and average precision under different threshold values are output, and the number of insulators correctly identified, the number of insulators not detected, the number of non-insulators not misdetected and the number of non-insulators misjudged as insulators are recorded.

[0168] The recall rate of the calculation model is:

[0169]

[0170] wherein, represents the number of insulators correctly identified, represents the number of insulators not detected, represents the recall rate, which is used to measure the ability of the model to correctly detect all insulators, the higher the value, the fewer the missed insulators, as the detection threshold gradually rises, the model only retains the detection boxes with higher confidence, so the recall rate continues to decrease;

[0171] The accuracy of the calculation model is:

[0172]

[0173] In the formula, represents the number of non-insulators that are not misdetected, represents the number of non-insulators misjudged as insulators, represents the accuracy, the size of the accuracy represents the overall classification ability of the model to all samples, that is, the proportion of overall correct prediction;

[0174] The precision of the calculation model is:

[0175]

[0176] In the formula, represents the precision, which reflects how many of the results detected by the model as insulators are truly correct, and is used to evaluate the reliability of the model's detection results. High precision can reduce the workload of manual review and avoid dealing with a large number of false insulators that are misdetected;

[0177] The average precision is the area under the precision-recall curve, which combines recall and precision, and is the core indicator for measuring the performance of a single insulator detection model in the target detection task. In the actual engineering scenario of unmanned aerial vehicle inspection, the dynamic adjustment of the detection threshold needs to be closely combined with business needs to achieve a balance between leak detection risk control and misdetected cost optimization. Selecting a low detection threshold, between 0.5 and 0.6, can ensure that insulators can be detected, but the number of non-insulators that are not misdetected increases, requiring secondary verification. For example, to avoid the risk of missing insulator detection and causing safety hazards, a lower detection threshold, such as 0.5, can be selected. Selecting a high detection threshold, such as between 0.8 and 0.95, can significantly reduce the number of non-insulators misjudged as insulators, avoiding the increase in costs caused by misdirecting maintenance personnel, but some insulators may be missed. It is recommended to select a detection threshold based on the peak value of the average precision, so that most real insulators can be detected, and the number of non-insulators misjudged as insulators can be controlled within an acceptable range. When the average precision is between 0.7 and 0.75, the recall rate is between 75% and 80%, and the precision is between 90% and 93%. Figure 2 Figure 3

[0178] ​​Step 3: Extract the radiation value of the 550-700 nm band, shield the insulator area, and convert it to reflectivity. Based on the reflectivity data, use the band ratio method to calculate the corrosion area ratio.

[0179] In the hyperspectral image, all insulator areas are shielded, and the remaining areas are marked as background areas. This is because corrosion mainly occurs in metal components such as tower main materials and guy anchorages, while insulators are non-metallic and their spectral characteristics are unrelated to metal corrosion. After shielding, the background metal structure can be focused on 100%, avoiding interference from insulator skirts and steel legs.

[0180] Through radiation calibration, the radiation values of the background area at 550 nm and 680 nm are converted to reflectivity:

[0181]

[0182]

[0183] where, R550 represents the radiation value at wavelength 550 nm in the background area, R550_ref represents the radiation value of the reference white board at wavelength 550 nm, R550_ref_default represents the standard reflectivity of the white board, which is , R680 represents the radiation value at wavelength 680 nm in the background area, R680_ref represents the radiation value of the reference white board at wavelength 680 nm, R550 represents the reflectivity at wavelength 550 nm in the background area, R680 represents the reflectivity at wavelength 680 nm in the background area. Iron oxide has strong absorption at 550 nm band (green light area), and the reflectivity is significantly reduced. At 680 nm (red light area), the reflectivity is significantly increased. The characteristics of low reflection in green light and high reflection in red light make the corrosion area 680 nm and 550 nm reflectivity ratio significantly higher than the uncorroded area, with a ratio between 1.5 and 2.

[0184] Therefore, the band ratio method is used to construct the ratio formula:

[0185]

[0186] where, R represents the ratio formula, which is the ratio of the reflectivity at wavelength 680 nm to the reflectivity at wavelength 550 nm in the background area.

[0187] Set the ratio threshold value When , it indicates that iron oxide begins to accumulate, and the corrosion area ratio is calculated:

[0188]

[0189] In the formula, Indicates the first in the background area The band ratio of each pixel, where... , This represents the total number of pixels in the background area. Indicates an indicator function, when hour, ,when hour, This method of converting to binary variables of 0 / 1 masks noise fluctuations and avoids misjudging a certain area as corrosion. This indicates the percentage of the corroded area.

[0190] Step 4: Assess the overall health status of the transmission tower by considering the proportion of corrosion area, the degree of abnormal deformation of insulators and guy wires under abnormal conditions;

[0191] Set the threshold for the percentage of corrosion area as follows: According to the power industry recommended standards, the corrosion depth of the main materials of the iron tower... The strength of steel will decrease by more than Therefore, set A 50% safety redundancy is reserved as a threshold.

[0192] The wire deformation threshold is This is 2%, which is set according to the industry standard that the deviation of the guy wire length is within ±2%. If the guy wire is tightened, the problem can be resolved. Otherwise, the risk of the tower tilting will increase significantly.

[0193] When the corrosion area accounts for Number of insulators in abnormal condition And the anomaly of the wire deformation At that time, the assessment result of the transmission tower was normal.

[0194] When the corrosion area ratio is satisfied Number of insulators in abnormal condition Or the degree of anomaly in wire deformation If any of these conditions are met, the assessment result of the transmission tower is a general alarm state. In this case, it should be included in the routine maintenance plan to avoid overreaction.

[0195] When the corrosion area ratio is satisfied Number of insulators in abnormal condition Or the degree of anomaly in wire deformation When any two or more of them, this is a high-risk scenario of fault coupling, the evaluation result of the power transmission tower is an emergency alarm state, to avoid serious consequences. Table 2 shows that after collecting the hyperspectral images of 40 different power transmission towers, the corrosion area ratio, the number of abnormal insulators and the abnormal degree of stay wire deformation are obtained by the above calculation formula processing, and compared with the set threshold to output the evaluation result, and the data report is generated.

[0196]

[0197] Table 2 Evaluation results of power transmission tower

[0198] As Figures 6-8 shown, the number of abnormal insulators, the abnormal degree of stay wire deformation and the corrosion area ratio of each sample are shown. By observing whether each index exceeds the threshold value, the state of the tower can be quickly judged. The number of abnormal insulators of most samples is 0, and a certain number of abnormal insulators exist in some samples. For power transmission towers with abnormal insulators, attention should be paid. The abnormal degree of stay wire deformation reflects the stability of the stay wire. For some samples that exceed the threshold value, the risk of tower tilt is significantly increased, and timely repair is required to avoid the instability of the tower structure caused by the stay wire problem, which may lead to safety accidents. The corrosion area ratio directly reflects the corrosion condition of the metal part. When the threshold value is exceeded, the strength of the steel decreases, affecting the safety of the tower structure. By comprehensively considering the indexes of the three dimensions, the output evaluation result can be directly observed. According to the evaluation results of different levels, the selection of the routine maintenance plan or immediate treatment can be made to avoid serious consequences.

[0199] Please refer to Figure 9 , the application further provides a UAV inspection and identification device for a power transmission tower, which is used to execute the UAV inspection and identification method for a power transmission tower as described above, comprising:

[0200] A stay wire deformation monitoring module is used to pre-set a reference point on the ground, and the UAV hovers above the end of the stay wire of the power transmission tower. The height difference of the ground reference point is identified by the laser ranging function, the polar coordinates of the stay wire end point are calculated, and the length of the stay wire is calculated based on the polar coordinates of the stay wire end point to evaluate the abnormal degree of stay wire deformation.

[0201] An insulator detection module is used to carry a 5-channel hyperspectral imaging system on the UAV to perform orthogonal flight line scanning high-precision imaging, extract the radiation value of the 300-900nm band, and locate the insulator in real time through the CNN algorithm. The state of the insulator is judged by calculating the average radiation value of the UV band, the local discharge area gradient and the temperature difference.

[0202] The corrosion evaluation module is used for extracting the radiation value in the wave band of 550-700 nm, converting the radiation value into reflectivity after shielding the insulator area, and calculating the corrosion area ratio based on the reflectivity data by using the wave band ratio method;

[0203] The health evaluation module is used for comprehensively evaluating the comprehensive health state of the power transmission tower based on the corrosion area ratio, the abnormal state insulator and the abnormal degree of the stay deformation.

[0204] The above formulas are all dimensionless values, and the formulas are obtained by software simulation based on a large amount of data to obtain a formula of the most recent real situation, and the preset parameters in the formula are set by the person skilled in the art according to the actual situation.

[0205] The above embodiments can be realized by software, hardware, firmware or any combination thereof, in whole or in part. When realized by software, the above embodiments can be realized in the form of a computer program product in whole or in part. Those skilled in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized by hardware or software methods depends on the specific application and design constraints of the technical solutions.

[0206] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, which can be located in one place or distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiments.

[0207] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for identifying the inspection of unmanned aerial vehicles for power transmission towers, characterized by, The specific steps include: Step 1: A reference point is set on the ground in advance, and the unmanned aerial vehicle hovers above the end of the stay wire of the power transmission tower, identifies the height difference of the ground reference point through the laser ranging function, calculates the polar coordinates of the stay wire end point, and then calculates the length of the stay wire based on the polar coordinates of the stay wire end point to evaluate the abnormality degree of the stay wire deformation; Step 2: The unmanned aerial vehicle is equipped with a 5-channel hyperspectral imaging system to perform orthogonal flight line scanning high-precision imaging, extract the radiation value of the 300-900nm band, and use the CNN algorithm to locate the insulator in real time, and judge the state of the insulator by calculating the average radiation value of the UV band, the local discharge area gradient and the temperature difference; Step 3: Extract the radiation value of the 550-700nm band, convert it to reflectivity after shielding the insulator area, and calculate the corrosion area ratio based on the reflectivity data using the band ratio method; Step 4: Integrate the corrosion area ratio, abnormal state insulator and stay wire deformation abnormality degree to evaluate the comprehensive health status of the power transmission tower; The unmanned aerial vehicle is equipped with a 5-channel hyperspectral imaging system based on a spectral range of 300-900nm, which is divided into 300-400nm, 400-500nm, 550-700nm, 700-800nm and 800-900nm, and corresponds to channels 1, 2, 3, 4 and 5 respectively; The model for locating the insulator is pre-trained: Collect at least 1000 hyperspectral images of power transmission towers, uniformly modify the size under the condition of maintaining the aspect ratio to: Through the LabelImg tool, mark the boundary box of the overall outline of the insulator: , wherein, represents the column position of the left side of the boundary box, represents the row position of the top of the boundary box, represents the column position of the right side of the boundary box, represents the row position of the bottom of the boundary box, and makes its category label an insulator, Then, it is converted into normalized center point coordinates: wherein, represents the center point coordinate, represents the insulator center coordinate, represents the insulator width, represents the insulator height, represents the corresponding image width, represents the corresponding image height; The hyperspectral image, the corresponding center point coordinates and the category label are input into the ResNet-18 model, the converted center point coordinates are taken as the real label, and the output dimension of the last full connection layer is modified to the corresponding center point coordinates and the confidence score , and a 5-dimensional vector data set is output: wherein represents outputting a 5-dimensional vector dataset, wherein the formula for calculating the confidence score is: In the formula, denotes the original confidence score of the fully connected layer output; The center point coordinates output by the ResNet-18 model are used as the prediction results, and the error between the two is calculated using a positioning loss function based on the actual center point coordinates, and the positioning loss function is: wherein, represents a localization loss function, represents a predicted value of the th parameter, represents a true value of the th parameter, wherein, represents an index of the parameter in the center point coordinates, , respectively corresponding to the center point horizontal coordinate, the center point vertical coordinate, the bounding box width, and the bounding box height, represents a loss function, and satisfies: A classification loss formula is constructed: In the formula, denotes the classification loss; A total loss formula is constructed: In the formulae, represents the total loss; The training parameters are set: learning rate: , batch size: , training epochs: ; The power transmission tower images taken by the unmanned aerial vehicle in real time are unified to a fixed size, calibrated to hyperspectral images, input to a model trained in advance to locate insulators, and output a 5-dimensional vector data set: Setting a detection threshold , and , only the bounding box of is retained, completing the positioning of the insulator; Outputting a 5-dimensional vector dataset performing inverse normalization of the pixel coordinates, extracting the insulator region from the hyperspectral image based on the inverse normalized pixel coordinates extracting the insulator region from the hyperspectral image based on the inverse normalized pixel coordinates, calculating the average radiation value from the radiation values of the 5 channels wherein represents the total number of pixels of the insulator region, represents the radiation value of the i-th pixel of the j-th channel within the insulator region, wherein, and the units of the radiation values are unified as: ;​​​ The local discharge area gradient is calculated: In the formula, represents a partial discharge region gradient, represents an average radiation value of the first channel, represents an average radiation value of the second channel; The temperature difference is calculated based on the radiation values of channels 4 and 5. First, the Planck's law formula is constructed: wherein denotes the center wavelength of the wavelength band, denotes the temperature, , , denotes the radiation value at the wavelength . The temperature at the wavelength of 1 1 1 1 nm is The temperature at the wavelength of 1 1 1 1 nm is , The temperature at the wavelength of 1 1 1 1 nm is ; By temperature at a wavelength and temperature at a wavelength temperature difference is calculated: In the formula, represents the temperature difference; An abnormality judgment formula is constructed: In the formula, represents an abnormality determination formula, when the insulator is determined to be in an abnormal state, and when the insulator is determined to be in a normal state.

2. The method for identifying the inspection of the power transmission tower by the UAV according to claim 1, characterized in that: The laser ranging function is used to identify the height difference of the ground reference point, and the polar coordinates of the stay wire end point are calculated, including the following steps: A reference point is set on the ground in advance, and the coordinates of the reference point set on the ground are measured by a total station The unmanned aerial vehicle hovers above the end of the guy wire, the laser ranging function is turned on, and the height difference between the ranging point and the ground reference point is output The measured polar coordinates of the guy wire end point are wherein, represents the distance of the guy wire end point obtained by laser ranging, represents the angle of the laser ranging direction in the horizontal plane, represents the angle of the ranging direction in the vertical plane.

3. The method for identifying the inspection of the power transmission tower by the UAV according to claim 2, characterized in that: Based on the polar coordinates of the stay wire end point, the length of the stay wire is calculated, and the method for evaluating the abnormality degree of the stay wire deformation is: The measured pull line end point polar coordinates are converted to three-dimensional coordinates: The length of the stay wire is calculated: In the formula, represents the length of the guy line; A deformation abnormality degree formula is constructed: In the formula, represents the initial length of the guy wire, represents the abnormality degree of deformation.

4. The method for identifying the inspection of the power transmission tower by the UAV according to claim 1, characterized in that: After shielding the insulator area, the reflectivity is converted, and the method for calculating the corrosion area ratio based on the reflectivity data using the band ratio method is: The hyperspectral image of Figure 5 is masked to exclude all insulator regions, leaving the remaining regions labeled as background regions. The radiance values of the background regions at 550 nm and 680 nm are converted to reflectance by the method of radiometric calibration: wherein, represents the radiance value at wavelength 550 nm in the background area, represents the radiance value at wavelength 550 nm of the reference white board, represents the white board standard reflectance, by default , represents the radiance value at wavelength 680 nm in the background area, represents the radiance value at wavelength 680 nm of the reference white board, represents the reflectance at wavelength 550 nm in the background area, represents the reflectance at wavelength 680 nm in the background area; The band ratio method is used to construct a ratio formula: wherein represents a ratio formula, i.e. the ratio of the reflectance at a wavelength of 680 nm to the reflectance at a wavelength of 550 nm in the background region; Setting a ratio threshold , calculating the corrosion area ratio: wherein, represents the band ratio of the i-th pixel in the background region, wherein, represents the total number of pixels in the background region, represents an indicator function when when , represents the corrosion area ratio.​​​​ 5. The method for identifying the inspection of the power transmission tower by the UAV according to claim 1, wherein: The method for evaluating the comprehensive health status of the power transmission tower is: The corrosion area proportion threshold is set as , and the pull wire deformation threshold is ; When the corrosion area proportion , the abnormal state insulator quantity , and the abnormal degree of the stay deformation are normal, the evaluation result of the power transmission tower is a normal state. When any one of the following conditions is met: a corrosion area ratio , an abnormal state insulator number , or a stay deformation abnormality , the evaluation result of the power transmission tower is a general warning state. When any two or more of the following conditions are met, the evaluation result of the power transmission tower is an emergency warning state: the corrosion area ratio , the number of abnormal state insulators , or the abnormal degree of stay deformation ​ 6. An unmanned aerial vehicle inspection identification device for transmission towers, characterized in that: The device is used for the unmanned aerial vehicle inspection and identification method for the power transmission tower according to any one of claims 1-5: The stay wire deformation monitoring module is used to set a reference point on the ground in advance, and the unmanned aerial vehicle hovers above the end of the stay wire of the power transmission tower, identifies the height difference of the ground reference point through the laser ranging function, calculates the polar coordinates of the stay wire end point, and then calculates the length of the stay wire based on the polar coordinates of the stay wire end point to evaluate the abnormality degree of the stay wire deformation; The insulator detection module is used for the unmanned aerial vehicle equipped with a 5-channel hyperspectral imaging system to perform orthogonal flight line scanning high-precision imaging, extract the radiation value of the 300-900nm band, and use the CNN algorithm to locate the insulator in real time, and judge the state of the insulator by calculating the average radiation value of the UV band, the local discharge area gradient and the temperature difference; The corrosion evaluation module is used for extracting the radiation value in the wave band of 550-700 nm, converting the radiation value into reflectivity after shielding the insulator area, and calculating the corrosion area proportion based on the reflectivity data by using the wave band ratio method; The health evaluation module is used for comprehensively evaluating the comprehensive health state of the power transmission tower based on the corrosion area proportion, the abnormal state insulator and the abnormality degree of the stay deformation.

Images