Power transmission line hidden danger target tracking method and system based on spectral image processing

Abstract

The invention relates to the technical field of target tracking, in particular to a power transmission line hidden danger target tracking method and system based on spectral image processing, and the method comprises the following steps: collecting a multiband spectral image, extracting an initial spectral stable point group, obtaining a spectral reflection peak value and an absorption valley value, constructing a spectral point template, fitting a reflection spectrum curve, and calculating a feature difference. The method comprises the following steps: judging spectral shift through a feature difference, screening abnormal frames, calculating similarity, generating a trajectory matching score, obtaining a score abnormal region, predicting a target position based on a multi-frame velocity vector, and updating trajectory node information. According to the method, abnormal areas are extracted through multiband spectral image analysis, a spectral feature vector set is constructed, a support vector machine regression model is utilized to stabilize a spectral curve, target spectral feature extraction precision is enhanced, spectral offset is calculated, abnormal frames are screened, a matching trajectory is scored, interference is estimated through a velocity vector, and tracking precision is enhanced through a dynamic adjustment mechanism. And the influence of environmental interference on the tracking process is reduced.

Description

Technical Field

[0001] This invention relates to the field of target tracking technology, and in particular to a method and system for tracking potential hazards in power transmission lines based on spectral image processing. Background Technology

[0002] The field of target tracking technology encompasses target tracking techniques, which are crucial for monitoring, identifying, and tracking targets. Its applications span multiple areas, including security monitoring, intelligent transportation, military reconnaissance, and autonomous driving. The key to this technology is utilizing image or video information acquired by sensors for target detection, identification, and localization. Core components include target identification and classification, target motion state analysis, and target-environment interaction analysis. A major challenge in target tracking technology lies in handling scene changes, target occlusion, target identification and tracking under multiple lighting conditions, and interference between multiple targets. Therefore, improving the accuracy and robustness of target tracking, especially in dynamic environments, is a key research focus in this field.

[0003] The method and system for tracking potential hazards in power transmission lines based on spectral image processing refers to the use of spectral image processing technology to monitor and track potential hazards in power transmission lines. This patent keyly addresses how to extract target features associated with the power transmission line from spectral images and how to enhance the distinction between the target and the background using multiple bands of the spectral image, thereby enabling the tracking of hazard targets. Specifically, the method acquires spectral image data of the power transmission line, uses image processing algorithms to extract the structural features of the power line, and locates and tracks areas where hazards are predicted. To improve the accuracy of hazard detection, the system employs multispectral data fusion technology, identifying predicted fault points or anomalies by comparing changes in multiple spectral data.

[0004] Current target tracking technologies rely on image data in a single or limited number of bands, resulting in low target recognition accuracy in various environments. Because the reflectivity of a single band cannot reflect the difference between the target and the background, it is easily affected by factors such as illumination changes and background clutter, leading to mismatches or missed detections. For example, in environments with significant illumination variations, the predicted spectral features of the target may shift, and existing methods often cannot determine the degree of this shift, thus affecting the stability and accuracy of tracking. Furthermore, traditional target recognition and tracking methods use simple similarity metrics, lacking sufficient self-matching capabilities and the ability to dynamically adjust algorithm parameters to match multiple scenarios. Maintaining high accuracy in complex and dynamic monitoring environments is difficult, often leading to target misidentification and tracking loss. In some current applications, rapid changes in the target during movement, spectral interference, or data gaps can all cause a decrease in tracking accuracy, and these methods are unable to effectively address these situations. Summary of the Invention

[0005] To address the technical problems existing in the prior art, embodiments of the present invention provide a method and system for tracking potential hazards in power transmission lines based on spectral image processing. The technical solution is as follows: To achieve the above objectives, the present invention adopts the following technical solution: a method for tracking potential hazards in power transmission lines based on spectral image processing, comprising the following steps: S1: Acquire multi-band spectral images of the transmission line, obtain the reflectance distribution of pixels in the image under multiple bands, and extract the initial spectral stable point group through clustering operation; S2: Obtain the reflection peak and absorption valley values ​​of the initial stable spectral point group in multiple bands, construct a spectral point template and extract the corresponding reflection spectrum curve shape, fit the reflection spectrum curve shape through a support vector machine regression model and calculate the average feature difference of the spectral points. S3: Call the average feature difference of the spectral points to determine the degree of target spectral shift between adjacent frames, filter out spectral abnormal frames by Mahalanobis distance detection, and generate a state region to be confirmed. S4: Calculate the similarity between the region to be confirmed and the spectral point template using the cosine similarity algorithm, determine whether the similarity is greater than the set spectral similarity matching threshold, and generate a trajectory matching score; S5: Obtain the region where the trajectory matching score value drops abnormally and the corresponding multi-frame velocity vectors. Based on the multi-frame velocity vectors, predict the target position and compare the position with the similarity. Relocate the target position and update the target trajectory node information of the current frame.

[0006] As a further aspect of the present invention, the initial stable spectral point group is a high-reflection peak point group, a low-absorption valley point group, and a stable fluctuation point group; the average characteristic difference of the spectral points includes peak offset, valley displacement, and morphological distortion; the region to be confirmed includes dynamic interference region, shadow coverage region, and reflectance abrupt change region; the trajectory matching score refers to morphological overlap score, band ratio score, and temporal continuity coefficient score; and the target trajectory node information of the current frame includes velocity vector correction value, occlusion compensation parameter, and position offset tolerance.

[0007] As a further aspect of the present invention, the steps of acquiring multi-band spectral images of transmission lines, obtaining the reflectance distribution of pixels in the images under multiple bands, and extracting initial spectral stable point groups through clustering operations are as follows: S101: Acquire visible light, near-infrared and short-wave infrared spectral images of transmission lines, extract the reflectance of each pixel in multiple bands, calculate the deviation of the reflectance of the pixel in the three bands, and generate a set of pixel spectral fluctuation values. S102: Based on the set of pixel spectral fluctuation values, mark the coordinates of pixels whose three-band reflectance deviation exceeds 0.25, integrate the coordinate positions, reflectance values ​​and difference data, and construct an abnormal region feature set; S103: Call the feature set of the abnormal region, calculate the spatial distance and reflectance change gradient between feature data, use density clustering method to merge points with a distance of less than 1.2 and a reflectance change gradient of less than 0.12, and output the initial spectral stable point group.

[0008] As a further aspect of the present invention, the steps of obtaining the reflection peaks and absorption valleys of the initial spectral stable point group in multiple bands, constructing a spectral point template and extracting the corresponding reflection spectrum curve shape, fitting the reflection spectrum curve shape using a support vector machine regression model, and calculating the average feature difference of the spectral points are as follows: S201: Call the initial spectral stable point group, traverse the reflectance data of the stable points in the visible light, near infrared and short-wave infrared bands, identify the wavelength positions corresponding to the maximum and minimum reflectance values ​​of the multi-band, establish the mapping relationship between the band number and the wavelength position, and generate a multi-band spectral point template. S202: Based on the multi-band spectral point template, perform wavelength interval interpolation calculation for adjacent peaks and valleys in each band to generate a continuous reflectance variation curve and perform second derivative calculation, detect the distribution of inflection points and count the frequency of occurrence in the band, and construct a set of reflectance spectrum morphology parameters. S203: Using a support vector machine regression model, wavelength is taken as the input feature and reflectivity as the target variable from the set of reflectivity morphological parameters. The kernel function type is set to radial basis function. The model output reflectivity difference is calculated and all differences are averaged to generate the average feature difference of spectral points.

[0009] As a further aspect of the present invention, the steps of using the average feature difference of the spectral points to determine the degree of target spectral shift between adjacent frames, filtering spectrally abnormal frames by Mahalanobis distance detection, and generating a state region to be confirmed are as follows: S301: Call the average feature difference of the spectral points, traverse the reflectance data of the corresponding bands of adjacent frames, calculate the change in reflectance of each band between the current frame and the previous frame, and generate the inter-frame spectral offset. S302: Calculate the degree of deviation between the inter-frame spectral offset and the overall distribution using Mahalanobis distance and construct a multi-dimensional feature space. Calculate the covariance distance between each sample point and the center point in the feature space to generate a set of spectral anomaly frames. S303: Extract the coordinates of the abnormal frames in the set of spectral abnormal frames, calculate the Euclidean distance between adjacent coordinate points, perform spatial clustering to merge coordinate points whose distance is less than the reflectance sampling interval, count the coverage of the merged area, extract the average of the abnormal intensity recorded in the set, and output the region to be confirmed.

[0010] As a further aspect of the present invention, the deviation between the multi-band variation of each frame and the overall distribution is calculated using Mahalanobis distance. The formula used is: ; in, Representing the Frame multi-band change vector, Represents the vector of overall sample means. The inverse matrix representing the covariance matrix. Representing the Dynamic weighting factor of the band, ,in For the first Band sampling frequency, Representing the Adaptive adjustment coefficients for dimensional features. , The standard deviation of historical reflectance. The normalization compensation factor representing the covariance matrix. Representing the The squared Euclidean distance of the dimensional offset. and Represents the frame number and band dimension.

[0011] As a further aspect of the present invention, the steps of calculating the similarity between the region to be confirmed and the spectral point template using a cosine similarity algorithm, determining whether the similarity is greater than a set spectral similarity matching threshold, and generating a trajectory matching score are as follows: S401: Calculate the directional consistency between the spectral curve of the center point of the region to be confirmed and the reflectance spectral curve of the spectral point template by cosine similarity, extract the reflectance vectors of the corresponding bands of the two curves, calculate the ratio of the vector dot product to the modulus, and generate the region template similarity coefficient. S402: Based on the region template similarity coefficient, mark the region number whose coefficient is greater than the spectral similarity matching threshold, and generate a matching judgment flag bit; The spectral similarity matching threshold is set through statistical analysis based on transmission line spectral image data; S403: Integrate the region template similarity coefficient with the matching judgment flag, perform similarity coefficient normalization processing on regions that meet the threshold conditions, and generate trajectory matching scores.

[0012] As a further aspect of the present invention, the ratio of the calculated vector dot product to the magnitude is... The formula used is: ; in, Representing the The dynamic weighting factor of the band reflectivity vector. in For band numbering, Represents the region of unconfirmed status. Reflectivity of the band Representative spectral template number The reflectivity value of the band, Represents cloud transmittance. Represents the correction of the solar altitude angle. The noise suppression factor representing the denominator term, as tested... The false alarm rate decreased by 18.7%. This represents the total number of multispectral bands.

[0013] As a further aspect of the present invention, the steps of obtaining the region where the trajectory matching score shows an abnormal decrease and the corresponding multi-frame velocity vectors, predicting the target position based on the multi-frame velocity vectors, comparing the position with similarity, relocating the target position, and updating the target trajectory node information of the current frame are as follows: S501: Obtain the fluctuation information of the trajectory matching score, perform anomaly detection based on the trajectory matching score value of the current frame, identify the area where the score value drops abnormally and mark it as an abnormal area; S502: Obtain the multi-frame velocity vectors corresponding to the abnormal region for position prediction, compare the predicted target position with the current frame region, identify potential spectral interference effects and mark them as spectral interference regions; S503: Integrate the spectral interference region with the prediction results, relocate the target position and update the target trajectory nodes in the region, generate new target trajectory node information and update the trajectory status of the current frame.

[0014] On the other hand, a transmission line hazard target tracking system based on spectral image processing is provided. This system is applied to the transmission line hazard target tracking method based on spectral image processing, and includes: The initial spectral screening module acquires multi-band spectral images of the transmission line, obtains the reflectance distribution of pixels in the image under multiple bands, extracts the initial spectral stable point group through clustering operation, and passes it to the spectral difference calculation module. The spectral difference calculation module obtains the reflection peak and absorption valley values ​​of the initial spectral stable point group in multiple bands, constructs a spectral point template and extracts the corresponding reflection spectrum curve shape, fits the reflection spectrum curve shape through a support vector machine regression model and calculates the average feature difference of the spectral points and transmits it to the offset detection module. The offset detection module calls the average feature difference of the spectral points to determine the degree of target spectral offset between adjacent frames, filters spectral abnormal frames by Mahalanobis distance detection, generates a state region to be confirmed and passes it to the trajectory matching module. The trajectory matching module calculates the similarity between the region to be confirmed and the spectral point template using a cosine similarity algorithm, determines whether the similarity is greater than a set spectral similarity matching threshold, generates a trajectory matching score, and transmits it to the trajectory correction module. The trajectory correction module acquires the regions where the trajectory matching score has abnormally decreased and the corresponding multi-frame velocity vectors. Based on the multi-frame velocity vectors, it predicts the target position and compares the position with similarity to relocate the target position and update the target trajectory node information of the current frame.

[0015] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: By acquiring and analyzing multi-band spectral images, anomalous regions with significant spectral differences are first extracted from the images, and a set of spectral feature vectors is constructed. Stable point groups are then extracted through clustering. This process ensures the identification of the target region, laying a solid foundation for trajectory matching. During the construction of the spectral point template, the shape of the reflectance curve is stabilized and reliable through the fitting of a support vector machine regression model, thereby improving the accuracy of target spectral feature extraction and avoiding common noise interference and data inconsistency problems in traditional methods. The degree of spectral offset between adjacent frames is determined by calculating the feature difference. Mahalanobis distance is used to filter spectrally anomalous frames, reducing false positive errors and ensuring stability and efficiency in the target tracking process. Spectral similarity judgment using a cosine similarity algorithm can match trajectory scores and reposition the predicted spectral interference using the velocity vector of the previous frame, avoiding positional deviations caused by spectral errors during target tracking. These innovative steps not only enhance the extraction and matching capabilities of spectral features but also improve tracking accuracy and reduce the impact of environmental interference by dynamically adjusting to accommodate multiple environmental changes. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the workflow of the present invention; Figure 2 This is a system flowchart of the present invention. Detailed Implementation

[0018] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0019] In embodiments of the present invention, words such as "exemplarily" and "comprising" are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0020] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0021] In embodiments of the present invention, sometimes the subscript is such as W 1预估 When written in a non-subscript form such as W1, the meaning they convey is the same unless the distinction is emphasized.

[0022] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0023] Please see Figure 1 This invention provides a technical solution: a method for tracking potential hazards in power transmission lines based on spectral image processing, comprising the following steps: S1: Acquire multi-band spectral images of the transmission line, obtain the reflectance distribution of pixels in the image under multiple bands, and extract the initial spectral stable point group through clustering operation; S2: Obtain the reflection peaks and absorption valleys of the initial stable spectral point group in multiple bands, construct spectral point templates and extract the corresponding reflection spectrum curve shapes, fit the reflection spectrum curve shapes through a support vector machine regression model and calculate the average feature difference of the spectral points. S3: Call the average feature difference of spectral points to determine the degree of target spectral shift between adjacent frames, filter spectral abnormal frames by Mahalanobis distance detection, and generate a state region to be confirmed. S4: Calculate the similarity between the state region to be confirmed and the spectral point template using the cosine similarity algorithm, determine whether the similarity is greater than the set spectral similarity matching threshold, and generate a trajectory matching score; S5: Obtain the regions where the trajectory matching score drops abnormally and the corresponding multi-frame velocity vectors. Based on the multi-frame velocity vectors, predict the target position and compare the position with similarity. Relocate the target position and update the target trajectory node information of the current frame.

[0024] The initial stable spectral point group consists of high reflectance peaks, low absorption valleys, and stable fluctuations. The average characteristic difference of the spectral points includes peak offset, valley displacement, and morphological distortion. The regions to be confirmed include dynamic interference regions, shadowed regions, and regions with abrupt changes in reflectance. The trajectory matching score refers to the morphological overlap score, band ratio score, and temporal continuity coefficient score. The target trajectory node information in the current frame includes velocity vector correction values, occlusion compensation parameters, and position offset tolerance.

[0025] Please see Figure 1 The specific steps for acquiring multi-band spectral images of transmission lines, obtaining the reflectance distribution of pixels in the images under multiple bands, and extracting the initial spectral stable point group through clustering operations are as follows: S101: Acquire visible light, near-infrared and short-wave infrared spectral images of transmission lines, extract the reflectance of each pixel in multiple bands, calculate the deviation of the reflectance of the pixel in the three bands, and generate a set of pixel spectral fluctuation values. In the visible light band image acquisition stage, operators maneuvered drones along the power transmission line to... Flying at cruising speed, the multispectral camera... Shooting at a frequency of [frequency], obtaining a resolution of [resolution]. The original image of the pixels. When using a tower insulator as the monitoring object, select a pixel area on its surface ( This will be analyzed in detail. During the radiometric correction phase, the original digital quantization values ​​in the visible light band will be analyzed. Converted to reflectivity, the result can be calculated similarly. , The mean was calculated using... The standard deviation calculation is broken down into three steps: first, calculate the square of the multi-band deviation. , , Then summation yields Finally, divide the band number by the square root of 3 to get the standard deviation. Detected When an anomaly is detected, an automatic anomaly marking mechanism is triggered, with the threshold set based on: [the following information is missing from the original text]. Analysis of a group of normal insulator samples showed that The maximum value is Therefore, take As a safety margin boundary value.

[0026] Table 1 Multispectral Reflectance Correction Parameters Parameter name Visible light band Near-infrared band Shortwave infrared band Gain coefficient kk 1.02 1.05 0.98 Offset bb -0.15 -0.12 -0.18 wavelength range 450-650 780-1100 1300-2500 As shown in Table 1, the multi-band sensor parameters were obtained through laboratory calibration, and the corrected reflectivity data was stored as a 32-bit floating-point matrix. When traversing all pixels, parallel computing techniques were employed to divide the image into... Pixel blocks are assigned to GPU threads for processing, and processing a single pixel takes time. Total processing time for the entire image .

[0027] S102: Based on the set of pixel spectral fluctuation values, mark the coordinates of pixels with a deviation of more than 0.25 in the three-band reflectance, integrate the coordinate position, reflectance value and difference data, and construct a set of abnormal region features; In the anomaly feature extraction stage, the coordinates are read ( ) Three-band reflectivity data at point ) calculate absolute deviation and Threshold The setup process includes three stages: first, data collection... Group of corona discharge samples, measured mean Standard deviation Secondly, an analysis. Group of normal fever samples, The maximum value is ; Determined through ROC curve analysis The optimal classification threshold corresponds to the true positive rate. False positive rate When constructing the feature vector, 128 bytes of storage space are allocated for each outlier. The data structure includes: 32-bit integer coordinates. (Converted to floating-point storage), 32-bit floating-point maximum reflectivity 32-bit floating-point deviation and , with point ( For example, the storage format is: (Corresponding decimal values: 1024.0, 768.0, 0.41, 0.35).

[0028] S103: Call the feature set of the abnormal region, calculate the spatial distance and reflectance change gradient between feature data, use density clustering method to merge points with a distance of less than 1.2 and a reflectance change gradient of less than 0.12, and output the initial spectral stable point group; Create a KD-tree index for each outlier, with the search radius initially set to... Pixel. Using point A ( Centered on point B, neighboring point B was retrieved. When calculating the Euclidean distance components... , ,have to Reflectivity gradient The average reflectance of pixel a across the three bands The average reflectance of pixel B across the three bands Therefore When both conditions are met and Cluster merging is performed at that time, in this example because... Trigger radius adaptive adjustment mechanism: expand the search radius to Then recalculate, at this time point E ( ) and point F ( distance ,gradient The merging conditions are met. Record the coordinates of the merged cluster center. Update the maximum reflectivity of the cluster. Of the 2145 groups generated, there are The group consists of 3-5 member points, with a spatial distribution standard deviation. Pixels Pixel.

[0029] Please see Figure 1 The specific steps for obtaining the reflection peaks and absorption valleys of the initial stable spectral point group in multiple bands, constructing spectral point templates and extracting the corresponding reflection spectrum curve shapes, fitting the reflection spectrum curve shapes using a support vector machine regression model, and calculating the average feature difference of the spectral points are as follows: S201: Call the initial stable spectral point group, traverse the reflectance data of the stable points in the visible light, near infrared and short-wave infrared bands, identify the wavelength positions corresponding to the maximum and minimum reflectance values ​​of the multi-band, establish the mapping relationship between the band number and the wavelength position, and generate a multi-band spectral point template. Using the stable point P-1024 from the initial spectral stable point group, extract visible light (band number B1, wavelength). ), near-infrared (B2, ), shortwave infrared (B3, Reflectance data for three bands , , Compare the three reflectivity values ​​and determine the maximum value. Corresponding band number B2, minimum value Corresponding to B3, a mapping table is established, where the wavelength positions are obtained through spectrometer calibration data, including wavelength B2. The calibration process is as follows: In the laboratory, using the standard spectral lines of a mercury lamp (546.1 nm, 871.6 nm), the spectrometer gain is adjusted so that the peak of the 871.6 nm spectral line is aligned with the center wavelength of B2, and the wavelength offset at this point is recorded. ,Sure For the current wavelength of B2, after traversing all 2145 stable points, a spectral point template database with 6 columns is generated and stored in CSV format, with each point data occupying 42 bytes.

[0030] Table 2 Spectral point template data Point number B1 wavelength B1 reflectivity B2 wavelength B2 reflectivity B3 wavelength B3 reflectivity P-1024 550 0.38 850 0.42 1650 0.35 P-2048 548 0.41 852 0.39 1648 0.33 As shown in Table 2, the extreme wavelength bands are determined by comparing the reflectance values. The maximum value of point P-1024 is B2 reflectance of 0.42. The wavelength calibration error is controlled within ±2nm.

[0031] S202: Based on a multi-band spectral point template, interpolation calculation of adjacent peak and valley wavelength intervals is performed for each band to generate a continuous reflectivity variation curve and perform second derivative calculation. The distribution of inflection points is detected and the frequency of occurrence in the band is counted to construct a set of reflectivity spectral morphology parameters. Adjacent peak analysis was performed on the B2 band (850nm) at point P-1024, and a valley point on the left was detected. (Reflectivity 0.38), Right side valley point (Reflectivity 0.39), interpolate in the interval [830nm, 870nm] with a step size of 5nm to generate the reflectivity sequence: 830nm (0.38), 835nm (0.39), 840nm (0.40), 845nm (0.41), 850nm (0.42), 855nm (0.41), 860nm (0.40), 865nm (0.39), 870nm (0.39). When calculating the second derivative, use the three-point central difference formula, including the first derivative at 845nm. Second derivative After traversing all interpolation points, the inflection point was detected at 840nm. ) and 860nm ( When counting the inflection point frequency of all bands, a threshold is set. To identify effective inflection points, 12,308 inflection points were detected out of 2,145 points. A set of morphological parameters was constructed, including three columns of data: inflection point wavelength, second derivative value, and band number.

[0032] S203: Using a support vector machine regression model, wavelength is taken as the input feature and reflectivity as the target variable in the set of reflectance spectrum morphology parameters. The kernel function type is set to radial basis function. The model output reflectivity difference is calculated and all differences are averaged to generate the average feature difference of spectral points. The support vector machine regression model consists of an input feature space, a kernel function mapping layer, and a regression hyperplane; The B2 band inflection point wavelengths (840nm, 860nm) of point P-1024 are extracted from the morphological parameter set as input features, and the corresponding reflectivities of 0.40 and 0.40 are used as target variables. The radial basis function kernel parameters are set. Based on the feature dimension experiment, it was determined that when the input feature is 2-dimensional, the test... This leads to overfitting. This resulted in underfitting. Therefore, 0.1 was chosen as the cross-validation accuracy, and a regression model was trained using the LIBSVM library. The predicted reflectance at 840nm was 0.398, and at 860nm it was 0.402. The difference was then calculated. , Average feature difference The model hyperplane parameters are updated using a sequential minimum optimization algorithm, with 1000 iterations and a learning rate of [missing information]. The average feature difference distribution interval for all points is [0.0015, 0.0043].

[0033] Please see Figure 1 The specific steps for determining the degree of target spectral shift between adjacent frames by calling the average feature difference of spectral points, filtering out spectrally abnormal frames by Mahalanobis distance detection, and generating the state region to be confirmed are as follows: S301: Call the average feature difference of spectral points, traverse the reflectance data of corresponding bands in adjacent frames, calculate the change in reflectance of each band between the current frame and the previous frame, and generate the inter-frame spectral offset. Call the average characteristic difference of the spectral point numbered F-210 traverse adjacent frames (the first frame) Frame and Reflectance data of the B1 (550nm), B2 (850nm), and B3 (1650nm) bands of the frame were extracted. Frame reflectivity , , ,as well as Frame data , , Calculate multi-band variation , , Generate inter-frame spectral offset vectors After traversing all 2145 stable points, a set of 3200 offset vectors is generated.

[0034] Table 3 Examples of Inter-Frame Spectral Offset Frame pair number Change in B1 Change in B2 Change in B3 F-210 0.03 0.03 -0.02 F-311 0.01 -0.05 0.04 As shown in Table 3, the change in band reflectivity is calculated by frame-by-frame subtraction. A negative change in band B3 indicates a decrease in reflectivity.

[0035] S302: Calculate the deviation of the inter-frame spectral offset from the overall distribution using Mahalanobis distance and construct a multi-dimensional feature space. Calculate the covariance distance between each sample point and the center point in the feature space to generate a set of spectral anomaly frames. Sample F-210 was randomly selected from 3200 offset vectors. and F-311 Calculate the overall mean vector Sum all changes in the B1 band Set 3, sum of 200 points), mean Similarly, calculate , ,have to covariance matrix B1 variance (through 3200) (Value calculation), covariance of B1 and B2 , Calculate the dynamic weighting factor Set the B1 sampling frequency ,but B2 ( )have to B3 ( )have to Calculate the adaptive adjustment coefficient B1 standard deviation ,but ; Substitute the F-210 samples to calculate the Mahalanobis distance, and the difference between the offset vector and the mean: Calculate the numerator: denominator: Degree of deviation Set an abnormal threshold ,when The frame was identified as abnormal, and the F-210 was included in the abnormal set because 3.18 > 2.5.

[0036] S303: Extract the coordinates of abnormal frames in the spectral abnormal frame set, calculate the Euclidean distance between adjacent coordinate points, perform spatial clustering to merge coordinate points whose distance is less than the reflectance sampling interval, count the coverage of the merged area, extract the average of the abnormal intensity recorded in the set, and output the region to be confirmed. Extracting the coordinates of F-210 frames from the anomaly set and F-311 ( ), calculate Euclidean distance Comparison of reflectivity sampling interval Pixels (based on drone speed) With shooting frequency The calculation shows that: (Current sampling interval is 1 pixel), because The merging criteria are not met, continue searching for the next adjacent point F-415. ),calculate If the clusters are still not merged, adjust the cluster radius to... After pixel division, F-210, F-311, and F-415 are merged into group G-15. The group coverage is calculated as follows: minimum x=1023, maximum x=1025, span 3 pixels; minimum y=767, maximum y=769, span 3 pixels; coverage area. Pixels, corresponding to the current ground area Statistical analysis of the average anomaly intensity across all groups Output a list of regions to be verified, including three columns of data: coordinate range, coverage area, and average anomaly intensity.

[0037] Please see Figure 1 The specific steps for calculating the similarity between the region to be confirmed and the spectral point template using the cosine similarity algorithm, determining whether the similarity is greater than a set spectral similarity matching threshold, and generating a trajectory matching score are as follows: S401: Calculate the directional consistency between the spectral curve of the center point of the region to be confirmed and the reflectance spectral curve of the spectral point template by cosine similarity, extract the reflectance vectors of the corresponding bands of the two curves, calculate the ratio of the vector dot product to the modulus, and generate the region template similarity coefficient. Call the center point (coordinates) of the area to be confirmed, G-15 From the spectral curve data, the reflectance of the B1 (550nm), B2 (850nm), and B3 (1650nm) bands was extracted. , , Read the reflectance of the corresponding point P-1024 from the spectral template. , , Calculate the dynamic weighting factor (k represents band number 1-3), thus... , , Obtain cloud transmittance (Through meteorological satellite data interpolation) , Solar altitude angle correction (Noon time is 1.0, which is increased by 20% here due to the tilt of the shooting angle). , Substitute into the formula to calculate the numerator: Calculate the denominator: , Add noise suppression factor Then, the ratio of the vector dot product to the magnitude. ; Table 4. Spectral Curve Parameters band Regional reflectivity Template reflectivity Cloud transmittance Sun Correction B1 0.41 0.38 0.85 1.2 B2 0.45 0.42 0.78 1.1 B3 0.33 0.35 0.92 1.3 As shown in Table 4, the B3 band is affected by the dynamic weighting factor. Cloud transmittance was not included in molecular calculations and is updated hourly via meteorological data API.

[0038] S402: Based on the region template similarity coefficient, mark the region number whose coefficient is greater than the spectral similarity matching threshold, and generate a matching judgment flag bit; The spectral similarity matching threshold was set through statistical analysis based on transmission line spectral image data; Set spectral similarity matching threshold This threshold was determined by analyzing 500 groups of normal insulator samples (mean similarity). Standard deviation ) and 200 sets of fault samples (mean similarity) Standard deviation After plotting the ROC curve, the optimal classification point was determined to be 0.3, corresponding to a true positive rate of 85% and a false positive rate of 12%. All regions to be confirmed were traversed, and the G-15 similarity coefficient was labeled. Generate flag bits (Mismatch), while region G-18 ( Then mark Generate a decision list consisting of two columns: region number and flag. Store the list as a binary file, with each record occupying 4 bytes (int32 type number + bool type flag).

[0039] S403: Integrate the region template similarity coefficient and the matching judgment flag, perform similarity coefficient normalization processing on regions that meet the threshold conditions, and generate trajectory matching scores; For flag bits The regions are normalized, and the similarity coefficients of G-18 are extracted. The coefficient range of all matching regions [0.301, 0.417] was calculated, and the minimum-maximum normalization formula was used: Spatial coordinate data were merged to generate a trajectory scoring matrix, including three columns: region center coordinates (1024, 768), coverage area (9 pixels), and score (0.198). This data was then written to a geographic information database with field types POINT, INT, and FLOAT. Index construction took [time]. .

[0040] Please see Figure 1 The specific steps for obtaining regions where trajectory matching scores show abnormal drops and their corresponding multi-frame velocity vectors, predicting target positions based on multi-frame velocity vectors, comparing positions using similarity, relocating target positions, and updating target trajectory node information in the current frame are as follows: S501: Obtain fluctuation information of trajectory matching score, perform anomaly detection based on trajectory matching score value of the current frame, identify areas where score value drops abnormally and mark them as abnormal areas; Using trajectory score data for region G-18 (scores for the last 5 frames: 0.32, 0.31, 0.29, 0.28, 0.19), calculate the sliding window mean. with standard deviation Set the abnormal drop threshold as The current frame score is 0.19. Since 0.19 > 0.182, no anomaly marker is triggered. The analysis continues with the score sequence of region G-25 (0.41, 0.40, 0.38, 0.35, 0.22), and the calculation is performed. , threshold The current frame score of 0.22 > 0.200 is still not abnormal. Finally, the G-07 sequence (0.28, 0.27, 0.26, 0.25, 0.12) is examined, and the results are as follows: , The threshold is 0.112. The current score is 0.12 < 0.112. It is marked as an abnormal region. Its center coordinates (x=768, y=256) and coverage area of ​​15 pixels are recorded. The abnormal region list is updated to [G-07].

[0041] Table 5 Examples of Trajectory Score Fluctuation Detection Area code Rating 1 Rating 2 Rating 3 Rating 4 Current rating threshold Judgment Result G-18 0.32 0.31 0.29 0.28 0.19 0.182 normal G-25 0.41 0.40 0.38 0.35 0.22 0.200 normal G-07 0.28 0.27 0.26 0.25 0.12 0.112 abnormal As shown in Table 5, the system calculates statistics through a 5-frame sliding window. When the score is lower than the dynamic threshold, a flag is triggered. The threshold is adaptively adjusted as the data fluctuates.

[0042] S502: Obtain multi-frame velocity vectors corresponding to the abnormal region for position prediction. By comparing the predicted target position with the current frame region, identify potential spectral interference effects and mark them as spectral interference regions. Extracting the velocity vectors of the first three frames of the anomaly region G-07: Frame position (760, 250) Frame (764, 253) Frame (768, 256), calculate Frame rate ,,predict Frame position , , The frame detected position G-07 as (780, 265). Calculate the offset distance. Comparison tolerance threshold (Based on the maximum displacement speed of the drone) (Multiple safety factor), since 10>8, it is determined to be spectral interference. The coordinates of the interference area (780, 265) and the deviation vector (8, 9) are recorded and added to the interference list.

[0043] S503: Integrate the spectral interference region with the prediction results, relocate the target position and update the target trajectory nodes in the region, generate new target trajectory node information and update the trajectory status of the current frame. Position correction was performed on the interference area (780, 265), and the positions of the previous 3 frames (772, 259), (776, 262), and (780, 265) were obtained. The average velocity was then calculated. , Predict the position of the next frame , Simultaneously, the frame position (785, 270) is collected, and the corrected position is calculated as a weighted average: , The trajectory node is updated to (784.4, 268.8). The original node sequence [(772, 259), (776, 262), (780, 265)] is expanded to [(772, 259), (776, 262), (780, 265), (784.4, 268.8)] and written to the trajectory database. The field type is DOUBLE precision floating-point number, and the storage occupies 16 bytes / node.

[0044] Please see Figure 2 A transmission line hazard target tracking system based on spectral image processing is used to execute the aforementioned transmission line hazard target tracking method based on spectral image processing. The system includes: The initial spectral screening module acquires multi-band spectral images of the transmission line, obtains the reflectance distribution of pixels in the image under multiple bands, extracts the initial spectral stable point group through clustering operation, and passes it to the spectral difference calculation module. The spectral difference calculation module obtains the reflection peaks and absorption valleys of the initial stable spectral point group in multiple bands, constructs spectral point templates and extracts the corresponding reflection spectrum curve shapes, fits the reflection spectrum curve shapes through a support vector machine regression model, calculates the average feature difference of spectral points and transmits it to the offset detection module. The offset detection module calls the average feature difference of spectral points to determine the degree of target spectral offset between adjacent frames, filters spectral abnormal frames by Mahalanobis distance detection, generates a state region to be confirmed and passes it to the trajectory matching module. The trajectory matching module calculates the similarity between the state region to be confirmed and the spectral point template using the cosine similarity algorithm, determines whether the similarity is greater than the set spectral similarity matching threshold, generates a trajectory matching score, and transmits it to the trajectory correction module. The trajectory correction module obtains the regions where the trajectory matching score has dropped abnormally and the corresponding multi-frame velocity vectors. Based on the multi-frame velocity vectors, it predicts the target position and compares the position with similarity to relocate the target position and update the target trajectory node information of the current frame.

[0045] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist: A and / or B, which can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also be interpreted as an "and / or" relationship; please refer to the context for a more detailed understanding.

[0046] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. "At least one of a, b, or c" can be represented as: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0047] It should be understood that, in various embodiments of the present invention, the order of the sequence numbers of the multiple processes does not imply the order of execution. The execution order of the multiple processes should be determined by their functions and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0048] Those skilled in the art will recognize that the various example units and algorithm steps described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use multiple methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0049] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0050] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. This includes the fact that the apparatus embodiments described above are merely illustrative, and that the division of units is only a logical functional division; in actual implementation, there may be other division methods, including the possibility that multiple units or components can be combined or integrated into another device, or that some features can be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed can be through some interface; the indirect coupling or communication connection of devices or units can be electrical, mechanical, or other forms.

[0051] The units described as separate components may or may not be physically separate. The components shown as units 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 units can be selected to achieve the purpose of this embodiment according to actual needs.

[0052] In addition, the multifunctional unit in various embodiments of the present invention can be integrated into one processing unit, or multiple units can exist physically separately, or two or more units can be integrated into one unit.

[0053] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0054] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for tracking potential hazards in power transmission lines based on spectral image processing, characterized in that, The method includes: S1: Acquire multi-band spectral images of the transmission line, obtain the reflectance distribution of pixels in the image under multiple bands, and extract the initial spectral stable point group through clustering operation; S2: Obtain the reflection peak and absorption valley values ​​of the initial stable spectral point group in multiple bands, construct a spectral point template and extract the corresponding reflection spectrum curve shape, fit the reflection spectrum curve shape through a support vector machine regression model and calculate the average feature difference of the spectral points. S3: Call the average feature difference of the spectral points to determine the degree of target spectral shift between adjacent frames, filter out spectral abnormal frames by Mahalanobis distance detection, and generate a state region to be confirmed. S4: Calculate the similarity between the region to be confirmed and the spectral point template using the cosine similarity algorithm, determine whether the similarity is greater than the set spectral similarity matching threshold, and generate a trajectory matching score; S5: Obtain the region where the trajectory matching score value drops abnormally and the corresponding multi-frame velocity vectors. Based on the multi-frame velocity vectors, predict the target position and compare the position with the similarity. Relocate the target position and update the target trajectory node information of the current frame.

2. The method for tracking potential hazards in transmission lines based on spectral image processing according to claim 1, characterized in that, The initial stable spectral point group consists of a high-reflectance peak point group, a low-absorption valley point group, and a stable fluctuation point group. The average characteristic difference of the spectral points includes peak offset, valley displacement, and morphological distortion. The region to be confirmed includes dynamic interference region, shadow coverage region, and reflectance abrupt change region. The trajectory matching score refers to morphological overlap score, band ratio score, and temporal continuity coefficient score. The target trajectory node information of the current frame includes velocity vector correction value, occlusion compensation parameter, and position offset tolerance.

3. The method for tracking potential hazards in transmission lines based on spectral image processing according to claim 1, characterized in that, The specific steps for acquiring multi-band spectral images of transmission lines, obtaining the reflectance distribution of pixels in the images across multiple bands, and extracting initial spectral stable point groups through clustering operations are as follows: S101: Acquire visible light, near-infrared and short-wave infrared spectral images of transmission lines, extract the reflectance of each pixel in multiple bands, calculate the deviation of the reflectance of the pixel in the three bands, and generate a set of pixel spectral fluctuation values. S102: Based on the set of pixel spectral fluctuation values, mark the coordinates of pixels whose three-band reflectance deviation exceeds 0.25, integrate the coordinate positions, reflectance values ​​and difference data, and construct an abnormal region feature set; S103: Call the feature set of the abnormal region, calculate the spatial distance and reflectance change gradient between feature data, use density clustering method to merge points with a distance of less than 1.2 and a reflectance change gradient of less than 0.12, and output the initial spectral stable point group.

4. The method for tracking potential hazards in transmission lines based on spectral image processing according to claim 1, characterized in that, The specific steps for obtaining the reflection peaks and absorption valleys of the initial stable spectral point group in multiple bands, constructing a spectral point template and extracting the corresponding reflection spectrum curve shape, fitting the reflection spectrum curve shape using a support vector machine regression model, and calculating the average feature difference of the spectral points are as follows: S201: Call the initial spectral stable point group, traverse the reflectance data of the stable points in the visible light, near infrared and short-wave infrared bands, identify the wavelength positions corresponding to the maximum and minimum reflectance values ​​of the multi-band, establish the mapping relationship between the band number and the wavelength position, and generate a multi-band spectral point template. S202: Based on the multi-band spectral point template, perform wavelength interval interpolation calculation for adjacent peaks and valleys in each band to generate a continuous reflectance variation curve and perform second derivative calculation, detect the distribution of inflection points and count the frequency of occurrence in the band, and construct a set of reflectance spectrum morphology parameters. S203: Using a support vector machine regression model, wavelength is taken as the input feature and reflectivity as the target variable in the set of reflectivity morphology parameters. The kernel function type is set to radial basis function. The model output reflectivity difference is calculated and all differences are averaged to generate the average feature difference of spectral points. The support vector machine regression model consists of an input feature space, a kernel function mapping layer, and a regression hyperplane.

5. The method for tracking potential hazards in transmission lines based on spectral image processing according to claim 1, characterized in that, The steps for determining the degree of target spectral shift between adjacent frames by calling the average feature difference of the spectral points, filtering out spectrally abnormal frames by Mahalanobis distance detection, and generating the state region to be confirmed are as follows: S301: Call the average feature difference of the spectral points, traverse the reflectance data of the corresponding bands of adjacent frames, calculate the change in reflectance of each band between the current frame and the previous frame, and generate the inter-frame spectral offset. S302: Calculate the degree of deviation between the inter-frame spectral offset and the overall distribution using Mahalanobis distance and construct a multi-dimensional feature space. Calculate the covariance distance between each sample point and the center point in the feature space to generate a set of spectral anomaly frames. S303: Extract the coordinates of the abnormal frames in the set of spectral abnormal frames, calculate the Euclidean distance between adjacent coordinate points, perform spatial clustering to merge coordinate points whose distance is less than the reflectance sampling interval, count the coverage of the merged area, extract the average of the abnormal intensity recorded in the set, and output the region to be confirmed.

6. The method for tracking potential hazards in transmission lines based on spectral image processing according to claim 5, characterized in that, The deviation of the multi-band variation in each frame from the overall distribution is calculated using Mahalanobis distance. The formula used is: ； in, Representing the Frame multi-band change vector, Represents the vector of overall sample means. The inverse matrix representing the covariance matrix. Representing the Dynamic weighting factor of the band, ,in For the first Band sampling frequency, Representing the Adaptive adjustment coefficients for dimensional features. , The standard deviation of historical reflectance. The normalization compensation factor representing the covariance matrix. Representing the The squared Euclidean distance of the dimensional offset. and Represents the frame number and band dimension.

7. The method for tracking potential hazards in transmission lines based on spectral image processing according to claim 1, characterized in that, The steps for calculating the similarity between the region to be confirmed and the spectral point template using the cosine similarity algorithm, determining whether the similarity is greater than a set spectral similarity matching threshold, and generating a trajectory matching score are as follows: S401: Calculate the directional consistency between the spectral curve of the center point of the region to be confirmed and the reflectance spectral curve of the spectral point template by cosine similarity, extract the reflectance vectors of the corresponding bands of the two curves, calculate the ratio of the vector dot product to the modulus, and generate the region template similarity coefficient. S402: Based on the region template similarity coefficient, mark the region number whose coefficient is greater than the spectral similarity matching threshold, and generate a matching judgment flag bit; The spectral similarity matching threshold is set through statistical analysis based on transmission line spectral image data; S403: Integrate the region template similarity coefficient with the matching judgment flag, perform similarity coefficient normalization processing on regions that meet the threshold conditions, and generate trajectory matching scores.

8. The method for tracking hidden danger targets in transmission lines based on spectral image processing according to claim 7, characterized in that, The ratio of the calculated vector dot product to the magnitude. The formula used is: ； in, Representing the The dynamic weighting factor of the band reflectivity vector. in For band numbering, Represents the region of unconfirmed status. Reflectivity of the band Representative spectral template number The reflectivity value of the band, Represents cloud transmittance. Represents the correction of the solar altitude angle. The noise suppression factor representing the denominator term, as tested... The false alarm rate decreased by 18.7%. This represents the total number of multispectral bands.

9. The method for tracking potential hazards in transmission lines based on spectral image processing according to claim 1, characterized in that, The specific steps for obtaining the regions where the trajectory matching score shows an abnormal decrease and the corresponding multi-frame velocity vectors, predicting the target position based on the multi-frame velocity vectors, comparing the position using similarity, relocating the target position, and updating the target trajectory node information of the current frame are as follows: S501: Obtain the fluctuation information of the trajectory matching score, perform anomaly detection based on the trajectory matching score value of the current frame, identify the area where the score value drops abnormally and mark it as an abnormal area; S502: Obtain the multi-frame velocity vectors corresponding to the abnormal region for position prediction, compare the predicted target position with the current frame region, identify potential spectral interference effects and mark them as spectral interference regions; S503: Integrate the spectral interference region with the prediction results, relocate the target position and update the target trajectory nodes in the region, generate new target trajectory node information and update the trajectory status of the current frame.

10. A transmission line hazard target tracking system based on spectral image processing, characterized in that, The system is used to implement the transmission line hazard target tracking method based on spectral image processing as described in any one of claims 1-9, and the system includes: The initial spectral screening module acquires multi-band spectral images of the transmission line, obtains the reflectance distribution of pixels in the image under multiple bands, extracts the initial spectral stable point group through clustering operation, and passes it to the spectral difference calculation module. The spectral difference calculation module obtains the reflection peak and absorption valley values ​​of the initial spectral stable point group in multiple bands, constructs a spectral point template and extracts the corresponding reflection spectrum curve shape, fits the reflection spectrum curve shape through a support vector machine regression model and calculates the average feature difference of the spectral points and transmits it to the offset detection module. The offset detection module calls the average feature difference of the spectral points to determine the degree of target spectral offset between adjacent frames, filters spectral abnormal frames by Mahalanobis distance detection, generates a state region to be confirmed and passes it to the trajectory matching module. The trajectory matching module calculates the similarity between the region to be confirmed and the spectral point template using a cosine similarity algorithm, determines whether the similarity is greater than a set spectral similarity matching threshold, generates a trajectory matching score, and transmits it to the trajectory correction module. The trajectory correction module acquires the regions where the trajectory matching score has abnormally decreased and the corresponding multi-frame velocity vectors. Based on the multi-frame velocity vectors, it predicts the target position and compares the position with similarity to relocate the target position and update the target trajectory node information of the current frame.

Images