Composite insulator multispectral feature extraction method and device considering corona characteristics

Abstract

The application relates to the technical field of image feature extraction, in particular to a composite insulator multispectral feature extraction method and device considering corona characteristics, which comprises the following steps: collecting a multispectral image of a composite insulator and performing registration and fusion to obtain a composite image; extracting an insulator image in the composite image; for any connected domain in the insulator image, according to the brightness difference, texture features and time sequence change features influenced by sunlight irradiation of the same connected domain in the HSV space of visible light, infrared and ultraviolet images, analyzing whether the connected domain is changed by corona and whether it is influenced by sunlight irradiation, further analyzing the key feature region of the composite insulator, and extracting the key feature region in the insulator image which is greatly influenced by the corona phenomenon. The application aims to effectively exclude the interference of environmental factors such as sunlight irradiation, so as to improve the accuracy and reliability of corona detection.

Description

Technical Field

[0001] This application relates to the field of image feature extraction technology, specifically to a method and apparatus for multispectral feature extraction of composite insulators considering corona characteristics. Background Technology

[0002] Composite insulators are widely used in power systems, and their performance directly affects the safe operation of the power grid. During operation, when the electric field strength on the surface of a composite insulator reaches a certain value, corona discharge will occur. The high-energy charged particles generated by corona discharge bombard the insulator surface, damaging its molecular structure. At the same time, corrosive substances such as nitric acid produced by corona discharge will also cause chemical corrosion to the insulator surface, leading to a decrease in the insulator's hydrophobicity, surface cracks, holes, and other defects, thus affecting its insulation performance. Therefore, timely detection and evaluation of the corona characteristics of composite insulators are crucial for maintaining the reliability of the power system.

[0003] Existing technologies mainly use ultraviolet and infrared imaging to detect composite insulators. By extracting relevant characteristic parameters and performing data analysis and comparison, the aging and insulation deterioration of composite insulators can be assessed.

[0004] However, when existing technologies detect composite insulators using multispectral imaging methods, they only analyze and extract the features of the composite insulator in the ultraviolet and infrared images at the current moment, without deeply analyzing the differences in the effects of sunlight and corona characteristics on the composite insulator. This leads to deviations in the extracted features, resulting in false detections and affecting the accuracy of composite insulator detection. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides a method and apparatus for extracting multispectral features of composite insulators considering corona characteristics. The specific technical solution adopted is as follows:

[0006] In a first aspect, one embodiment of this application provides a method for extracting multispectral features of composite insulators considering corona characteristics, the method comprising the following steps:

[0007] Multispectral images of the composite insulator are acquired, registered, and fused to obtain a composite image; the multispectral images include visible light images, infrared images, and ultraviolet images;

[0008] The composite insulator region in the composite image is extracted by a neural network and converted into a grayscale image to obtain the insulator image;

[0009] For any connected region in an insulator image, based on the brightness differences of the same connected region in the HSV space of the visible light, infrared and ultraviolet images, a preliminary analysis is made as to whether the connected region undergoes corona change.

[0010] Based on the texture features of the connected component in the insulator image, further analysis is conducted to determine whether the connected component is affected by sunlight.

[0011] Based on the temporal variation characteristics of the insulator images obtained at different times from the same connected region under the influence of sunlight, the key feature regions of the composite insulator can be further analyzed.

[0012] In conjunction with a preliminary analysis of whether corona changes occur within the same connected region, key feature regions significantly affected by corona phenomena in insulator images are extracted.

[0013] Preferably, the method for preliminary analysis of whether the connected region has undergone corona changes includes:

[0014] Calculate the similarity of brightness values ​​of the same connected component in the HSV space of infrared and ultraviolet images, respectively;

[0015] Calculate the normalized value of the mean brightness of any connected component in the HSV space of any spectral image, and calculate the range of the normalized values ​​of the mean brightness of all connected components in the HSV space of any spectral image.

[0016] Using the similarity and the normalized value and range of the mean brightness of any connected component in all spectral images, the corona brightness confidence of any connected component is determined, and its value is used to preliminarily analyze whether the connected component has undergone corona changes.

[0017] Preferably, the method for determining the corona brightness confidence level of any connected component further includes:

[0018] Let the confidence score of the corona intensity of the v-th connected component be denoted as . The expression is: In the formula, Let v be the similarity of the v-th connected component between the infrared HSV image and the ultraviolet HSV image; , , respectively, are the normalized values ​​of the mean brightness of the v-th connected component in the infrared HSV image, ultraviolet HSV image, and visible light HSV image; , , representing the brightness range of the v-th connected component in the infrared HSV image, ultraviolet HSV image, and visible light HSV image, respectively.

[0019] Preferably, the method for further analyzing whether the connected component is affected by sunlight includes:

[0020] Obtain the degree of dispersion of the texture feature values ​​of all pixels in the insulator image for this connected component;

[0021] The gray values ​​of all pixels between the two pixels with the highest and lowest gray values ​​in the connected component are formed into a sequence; the proportion of negative elements after differencing the sequence is denoted as the gray-level decrease index of the connected component.

[0022] Obtain the mean gradient value of all edge pixels within the connected component;

[0023] By using the mean of the calculated dispersion, gray level decrease index, and gradient value within the connected component, the illumination influence coefficient of the connected component is determined, and its value is used to further analyze whether the connected component is affected by sunlight.

[0024] Preferably, the method for determining the illumination influence coefficient of the connected domain further includes:

[0025] Let the illumination influence coefficient of the v-th connected component be denoted as... The expression is: In the formula, Let be the gray-level decreasing index of the v-th connected component; The degree of discreteness of all texture feature values ​​in the v-th connected component; It is the mean of the gradient values ​​of all edge pixels in the v-th connected region.

[0026] Preferably, the method for further analyzing the key feature regions of the composite insulator includes:

[0027] For all insulator images, generate connected components, and then register all corresponding connected components of the connected component in other insulator images as the time-separated connected components of the connected component.

[0028] Calculate the degree of dispersion of the illumination influence coefficients of the connected component and all its asynchronous connected components;

[0029] The Hu moment value between the connected component and each of its different time-separated connected components is calculated using the Hu moment function; the product of the mean of all calculated Hu moment values ​​and the degree of dispersion of all Hu moment values ​​is denoted as the contour change index of the connected component.

[0030] By using the illumination influence coefficient of the connected region, the dispersion of the calculated illumination influence coefficient, and the contour variation coefficient, the temporal variation coefficient of the connected region is determined. The magnitude of the coefficient is then used to further analyze whether the connected region is a key characteristic region of the composite insulator.

[0031] Preferably, the method for determining the temporal variation coefficients of the connected component further includes:

[0032] Let the time-series variation coefficient of the v-th connected component be denoted as The expression is: In the formula, Let be the illumination influence coefficient of the v-th connected component; The degree of dispersion of the illumination influence coefficients of the v-th connected component and all its asynchronous connected components; Let be the contour change index of the v-th connected component.

[0033] Preferably, the method for extracting key feature regions in the insulator image that are significantly affected by corona discharge is as follows:

[0034] The corona confidence score of the same connected domain is obtained by inversely fusing the corona brightness confidence score with the temporal variation coefficient.

[0035] The segmentation thresholds for the corona confidence of all connected components in the insulator image are obtained. Connected components with corona confidence values ​​greater than the segmentation threshold are taken as key feature regions in the insulator image that are significantly affected by the corona phenomenon.

[0036] Preferably, the method for determining the corona confidence of the connected region is further determined by the ratio of the corona brightness confidence to the temporal variation coefficient of the same connected region.

[0037] Secondly, another embodiment of this application provides a multispectral feature extraction device for composite insulators considering corona characteristics, implementing the multispectral feature extraction method for composite insulators considering corona characteristics described in any one of the above claims, the device comprising:

[0038] The image processing module is used to perform registration, fusion, and extraction of composite insulator regions on the acquired multispectral images, and inputs the obtained insulator images into the feature extraction module.

[0039] The feature extraction module is used to analyze the brightness, texture and temporal variation features of the insulator image to take into account the halo effect, the effect of sunlight exposure and the temporal variation features of the composite insulator.

[0040] The corona region extraction module is used to extract key feature regions of composite insulators from insulator images by analyzing various features from the feature extraction module.

[0041] This application has at least the following beneficial effects:

[0042] This application first constructs a corona brightness confidence level based on the corona characteristics of composite insulators, thereby initially identifying areas that may be affected by corona. Then, considering that sunlight may cause false detection, an illumination influence coefficient is constructed to further filter out the error caused by sunlight. Finally, considering the temporal variation characteristics of corona characteristics and sunlight, a temporal variation coefficient is constructed to accurately locate key areas affected by corona characteristics.

[0043] This application addresses the problem that existing technologies fail to adequately analyze the differences in the impact of sunlight and corona characteristics on composite insulators, resulting in poor feature extraction. By constructing a corona confidence level, this application effectively eliminates interference from environmental factors such as sunlight, thereby improving the accuracy and reliability of corona detection. This allows for better extraction of key features based on the corona characteristics of composite insulators, ensuring the safe and stable operation of high-voltage lines. Attached Figure Description

[0044] Figure 1 Flowchart of the method for extracting key feature regions in an image that are significantly affected by corona discharge, as provided in this application;

[0045] Figure 2 The flowchart illustrates the steps of the multispectral feature extraction method for composite insulators considering corona characteristics provided in this application. Detailed Implementation

[0046] This application addresses the lack of in-depth analysis of the differences in the impact of sunlight and corona characteristics on composite insulators, and provides a flowchart of a method for extracting key feature regions in images that are significantly affected by corona phenomena, as shown in the attached diagram. Figure 1 As shown, the method includes the following steps:

[0047] Multispectral images of the composite insulator are acquired, registered, and fused to obtain a composite image; the multispectral images include visible light images, infrared images, and ultraviolet images;

[0048] The composite insulator region in the composite image is extracted by a neural network and converted into a grayscale image to obtain the insulator image;

[0049] For any connected region in an insulator image, based on the brightness differences of the same connected region in the HSV space of the visible light, infrared and ultraviolet images, a preliminary analysis is made as to whether the connected region undergoes corona change.

[0050] Based on the texture features of the connected component in the insulator image, further analysis is conducted to determine whether the connected component is affected by sunlight.

[0051] Based on the temporal variation characteristics of the insulator images obtained at different times from the same connected region under the influence of sunlight, the key feature regions of the composite insulator can be further analyzed.

[0052] In conjunction with a preliminary analysis of whether corona changes occur within the same connected region, key feature regions significantly affected by corona phenomena in insulator images are extracted.

[0053] Specifically, this application provides a flowchart of the steps for a multispectral feature extraction method for composite insulators considering corona characteristics, as attached. Figure 2As shown, the method includes the following steps:

[0054] Step 1: Obtain relevant images of the composite insulator and perform preprocessing.

[0055] This application utilizes a sensor integration method to integrate an ultraviolet-sensitive CMOS sensor, a near-infrared thermal imaging sensor, and a standard visible light sensor into the same device, and uses a drone to load the integrated device for photographing a 500kV ultra-high voltage line.

[0056] Subsequently, these multispectral images of different bands are registered and fused to obtain a fused composite image. Image registration algorithms include SIFT and SURF, while image fusion algorithms include cascaded feature fusion and weighted fusion algorithms. This application uses SIFT and cascaded feature fusion algorithms for image configuration and fusion. The multispectral images include visible light images, infrared images, and ultraviolet images.

[0057] This application acquires one composite image every t minutes, for a total of N images. The acquisition interval t and the number of images N can be set by the implementer according to actual conditions. In this embodiment, the value of t is limited to 20-30, specifically 25, and the value of N is 10.

[0058] Based on the image processing module in the multispectral feature extraction device for composite insulators, all composite images acquired by the integrated device are denoised. The denoising algorithms include, but are not limited to, Gaussian filtering, median filtering, bilateral filtering, etc. In this embodiment, Gaussian filtering is used for denoising. After denoising, all composite images are converted into grayscale images.

[0059] Since drone photography inevitably captures surrounding environmental factors, this application uses a neural network to extract the composite insulator region from the composite image and convert it into a grayscale image to obtain an insulator image.

[0060] In a preferred embodiment, this application uses one-hot encoding of the composite image and label data before feeding them into a neural network for ROI region extraction. The label data is manually labeled, with composite insulator regions and non-composite insulator regions represented by the numbers 0 and 1, respectively. The output of the neural network is the composite insulator region in the composite image, denoted as the insulator image. The neural network can be ResNet18, CNN, YOLO, etc.; this embodiment uses a ResNet18 network. The training of the neural network is a well-known technique, and the specific process will not be described in detail.

[0061] After preprocessing the composite image, the insulator image is further analyzed and its features are extracted using the feature extraction module in the composite insulator multispectral feature extraction device, as detailed below:

[0062] Step 2: Perform preliminary analysis of the acquired images based on brightness.

[0063] Around high-voltage conductors, air molecules are easily ionized due to the strong electric field, triggering a corona discharge. When a composite insulator experiences corona discharge, the heat generated causes a local temperature increase. In infrared images, higher temperatures result in increased brightness. Simultaneously, the ultraviolet light produced by the corona discharge also increases the brightness of corresponding areas in the ultraviolet image, creating a degree of synchronicity between the brightness changes of pixels in the infrared and ultraviolet images. However, in visible light images, the brightness changes in the corona region are not significant and do not exhibit synchronous characteristics.

[0064] Therefore, in infrared and ultraviolet images, there will be a significant difference in pixel brightness values ​​between the feature regions where corona discharge has occurred and the normal regions, while in visible light images, the brightness difference is not significant. Thus, the significant differences in brightness of the same region in infrared, ultraviolet, and visible light images can be used to preliminarily analyze whether corona discharge has occurred in that region.

[0065] Accordingly, this application preliminarily analyzes whether corona changes occur in any connected region in an insulator image based on the brightness differences of the same connected region in the HSV space of the visible light, infrared, and ultraviolet images.

[0066] This application takes the i-th insulator image as an example: the connected components in the i-th insulator image are obtained by a connected component extraction algorithm. The connected component extraction algorithm includes, but is not limited to, the region growing algorithm and the seed filling algorithm. This embodiment adopts the region growing algorithm, which is a well-known technology and will not be described in detail here.

[0067] The composite image containing i insulator images before ROI region extraction is denoted as the i-th composite image. The infrared, ultraviolet, and visible light images of the i-th composite image before image fusion are obtained and HSV image conversion is performed on each.

[0068] Based on the region range of each connected component inside the i-th insulator image in the i-th composite image, the corresponding connected component regions are extracted from the infrared HSV image, ultraviolet HSV image, and visible light HSV image corresponding to the i-th composite image, respectively.

[0069] Furthermore, as a preferred embodiment, this application calculates the similarity of brightness values ​​of the same connected component in the HSV space of infrared and ultraviolet images, respectively.

[0070] In this embodiment, taking the v-th connected component as an example: the brightness value sequences of the v-th connected component in the infrared HSV image and the ultraviolet HSV image are obtained respectively. The elements in the two brightness value sequences obtained in the two images are arranged in the same order, and each element in a corresponding position represents the brightness value of the pixel at the same location. The absolute value of the correlation coefficient between the two brightness value sequences is calculated and denoted as the similarity of the v-th connected component between the infrared HSV image and the ultraviolet HSV image. The calculation of the correlation coefficient includes, but is not limited to, Pearson similarity coefficient, Spearman similarity coefficient, cosine similarity, etc. This embodiment uses the Spearman similarity coefficient. The similarity reflects whether the brightness changes of the v-th connected component in the infrared and ultraviolet images are synchronous due to the corona phenomenon.

[0071] Furthermore, as a preferred embodiment, this application calculates the normalized value of the mean brightness of any connected component in the HSV space of any spectral image, and calculates the range of the normalized values ​​of the mean brightness of all connected components in the HSV space of any spectral image.

[0072] In this embodiment, taking an infrared HSV image in HSV space as an example, the ratio between the mean brightness of the v-th connected component and the maximum mean brightness of all connected components is calculated, denoted as the normalized value of the mean brightness of the v-th connected component in the infrared HSV image. The normalized value of the mean brightness reflects the prominence of the v-th connected component in the infrared HSV image. The range between the normalized values ​​of the mean brightness of all connected components in the infrared HSV image is calculated, denoted as the brightness range of the infrared HSV image. The normalized value of the mean brightness of the v-th connected component in the infrared HSV image can also be obtained by normalizing the mean brightness of the v-th connected component in the infrared HSV image using a known normalization method. Specifically, max-min normalization, linear normalization, etc., can be used.

[0073] It should be noted that the normalized value of the mean brightness and the brightness range of the v-th connected component in the ultraviolet HSV image and the visible light HSV image are calculated using the same method as the normalized value of the mean brightness and the brightness range of the v-th connected component in the infrared HSV image.

[0074] Furthermore, as a preferred embodiment, this application uses the similarity and the normalized value and range of the mean brightness of any connected region calculated in all spectral images to determine the corona brightness confidence of any connected region, and uses its numerical value to preliminarily analyze whether the connected region has undergone corona changes.

[0075] In this embodiment, the corona brightness confidence score of the v-th connected component is denoted as... The expression is: In the formula, Let v be the similarity of the v-th connected component between the infrared HSV image and the ultraviolet HSV image; , , respectively, are the normalized values ​​of the mean brightness of the v-th connected component in the infrared HSV image, ultraviolet HSV image, and visible light HSV image; , , representing the brightness range of the v-th connected component in the infrared HSV image, ultraviolet HSV image, and visible light HSV image, respectively.

[0076] It should be understood that the corona brightness confidence score comprehensively considers the synchronicity of brightness changes and the degree of brightness prominence of connected components in infrared, ultraviolet, and visible light images. A larger value indicates better brightness synchronization of the v-th connected component in infrared and ultraviolet images. and The larger the value, the greater the brightness of the v-th connected component in infrared and ultraviolet images compared to the visible light image, and the greater the overall brightness variation.

[0077] Step 3: Further analyze the acquired images based on texture features.

[0078] Furthermore, since composite insulators are very susceptible to the influence of the external environment, the temperature difference between the sun-facing and shaded sides of the insulator under sunlight is large, and the surface of the composite insulator will reflect light. The brightness difference of visible light may also be large. Therefore, if only grayscale values ​​are used to extract the features of the insulator, there may be deviations. It is necessary to further combine corona characteristics and sunlight irradiation characteristics to further screen the feature extraction area.

[0079] The brightness of areas illuminated by sunlight exhibits textural variations, with grayscale values ​​fluctuating smoothly and gradually. Corona discharge, on the other hand, is a localized discharge phenomenon caused by a localized high electric field. Areas affected by corona discharge typically appear in images as having distinct edges and rapidly changing textures.

[0080] Accordingly, this application further analyzes whether the connected region is affected by sunlight based on the texture features of the connected region in the insulator image.

[0081] In a preferred embodiment, this application obtains the degree of dispersion of the texture feature values ​​of all pixels in the insulator image of the connected domain, which is used to reflect the texture differences between each pixel in the connected domain.

[0082] In this embodiment, all pixels in the v-th connected component of the insulator image are used as input. A texture analysis algorithm is employed to obtain the texture feature value of each pixel within the connected component, and the dispersion between all feature values ​​is calculated. Texture analysis algorithms include, but are not limited to, LBP and MLBP algorithms. This application uses the MLBP algorithm, and the calculated neighborhood size is a*a. In this embodiment, the value of a ranges from 3 to 5, and is 3. The dispersion can be calculated using variance, standard deviation, or root mean square deviation; this application uses variance. The dispersion between texture feature values ​​reflects the texture differences between pixels in the v-th connected component; a higher dispersion indicates that the v-th connected component is more likely to be affected by corona discharge, resulting in higher brightness.

[0083] Furthermore, since the area illuminated by sunlight has a smooth gradient, its grayscale value changes in a certain regularity and continuity; therefore, from the pixel with the highest brightness to the pixel with the lowest brightness, its grayscale value must continuously decrease, while the corona region caused by the corona characteristic does not have this characteristic.

[0084] In a preferred embodiment, this application constructs a sequence of gray values ​​of all pixels between the two pixels with the largest and smallest gray values ​​within the connected region; the proportion of negative elements after differencing the sequence is recorded as the gray-level decrease index of the connected region, which reflects the smoothness of gray-level value changes within the connected region.

[0085] In this embodiment, in the v-th connected region, a straight line with a width of 1 pixel is constructed between the two pixels with the largest and smallest gray values. Starting from the pixel with the largest gray value and ending at the pixel with the smallest gray value, a pixel value sequence is constructed based on the gray values ​​of all pixels on the straight line.

[0086] It should be noted that if there are multiple pixels with the largest or smallest gray values ​​in the v-th connected component, the pixel with the largest or smallest gray value that has the closest Euclidean distance to each other is selected to construct the straight line.

[0087] Obtain the first-order difference sequence of the pixel value sequence, and denote the ratio of the number of all negative elements in the first-order difference sequence to the length of the first-order difference sequence as the gray-level decrease index of the v-th connected component. The gray-level decrease index reflects the smoothness of the gray-level value change within the v-th connected component, further indicating whether the gray-level gradient feature in this area is caused by sunlight or influenced by other factors.

[0088] Furthermore, as a preferred embodiment, this application obtains the average gradient value of all edge pixels within the connected region to reflect the salience of the edges of the connected region.

[0089] In this embodiment, all edge pixels in the v-th connected component are obtained using an edge detection algorithm, and the average gradient value of all edge pixels is obtained using the Sobel operator to reflect the saliency of the edge in the v-th connected component. Edge detection algorithms include, but are not limited to, Canny, Prewitt, and Laplacian; this application uses Canny edge detection.

[0090] Furthermore, as a preferred embodiment, this application uses the mean of the calculated dispersion, gray level decrease index, and gradient value within the connected domain to determine the illumination influence coefficient of the connected domain, and uses its numerical value to further analyze whether the connected domain is affected by sunlight.

[0091] In this embodiment, the illumination influence coefficient of the v-th connected component is denoted as... The expression is: In the formula, Let be the gray-level decreasing index of the v-th connected component; The degree of discreteness of all texture feature values ​​in the v-th connected component; It is the mean of the gradient values ​​of all edge pixels in the v-th connected region.

[0092] It should be understood that the illumination influence coefficient can reflect the degree of influence of sunlight on the v-th connected domain. The larger the illumination influence coefficient, the greater the possibility that the brightness change of the v-th connected domain is mainly affected by sunlight, rather than by temperature rise or electric field discharge caused by corona discharge.

[0093] Step 4: Perform an overall analysis of the acquired images based on temporal changes.

[0094] Furthermore, by analyzing the temporal differences in corona characteristics and sunlight factors, more precise feature region extraction is performed, as follows:

[0095] Although corona discharge causes irreversible aging in composite insulators, the temperature and area of ​​the corona region remain relatively stable over a short period due to the long-term nature of this aging process. However, changes in cloud cover and solar angle lead to variations in light intensity, resulting in significant fluctuations in the area and grayscale value of the sunlit region. This, in turn, causes fluctuations in the illumination influence coefficient across different insulator images. Therefore, the key feature regions of composite insulators can be further segmented based on the temporal changes in images at different times.

[0096] Accordingly, this application uses the temporal variation characteristics of insulator images obtained at different times from the same connected region under the influence of sunlight to further analyze the key feature regions of composite insulators.

[0097] In a preferred embodiment, this application generates connected components for all insulator images, and records all corresponding connected components of the connected component after registration in other insulator images as the time-series connected components of the connected component.

[0098] This embodiment still takes the v-th connected component in the i-th insulator image as an example. Connected components are generated for all insulator images according to the parameters of the aforementioned region growing algorithm. Then, the SIFT image matching algorithm is used to perform image registration on all insulator images, thereby obtaining the connected components in other insulator images corresponding to the v-th connected component in the i-th insulator image, denoted as the time-separated connected component of the v-th connected component in the i-th insulator image.

[0099] Furthermore, this application calculates the degree of dispersion of the illumination influence coefficients of the connected component and all its asynchronous connected components.

[0100] In this embodiment, following the calculation principle of the illumination influence coefficient of the v-th connected component, the illumination influence coefficients of all time-varying connected components are obtained, and the dispersion among all illumination influence coefficients is calculated. This embodiment uses variance to calculate the dispersion. The dispersion among the illumination influence coefficients can reflect the differences in texture features of the v-th connected component at different acquisition times.

[0101] Subsequently, this application calculates the Hu moment values ​​between the connected component and each of its different time-separated connected components using the Hu moment function; the product of the mean of all calculated Hu moment values ​​and the degree of dispersion of all Hu moment values ​​is denoted as the contour variation index of the connected component.

[0102] In this embodiment, the Hu moment value between the v-th connected component and its asynchronous connected component in the i-th insulator image is calculated using the Hu moment function. The mean and dispersion of all Hu moment values ​​are calculated, and the product of the mean and dispersion is denoted as the contour change index of the v-th connected component. The Hu moment value reflects the contour difference between two connected components, while the contour change index reflects the degree of temporal change in the contour of the v-th connected component. The larger the contour change index, the greater the change in the contour shape of the v-th connected component over time, and the less consistent it is with the change characteristics caused by the corona phenomenon.

[0103] Furthermore, as a preferred embodiment, this application uses the illumination influence coefficient of the connected domain, the dispersion of the calculated illumination influence coefficient, and the contour change coefficient to determine the temporal change coefficient of the connected domain. The magnitude of the coefficient is then used to further analyze whether the connected domain is a key characteristic region of the composite insulator, reflecting the stability of the connected domain and judging whether the connected domain conforms to the key characteristic region of changes caused by corona discharge.

[0104] In this embodiment, the time-series variation coefficient of the v-th connected component is denoted as... The expression is: In the formula, Let be the illumination influence coefficient of the v-th connected component; The degree of dispersion of the illumination influence coefficients of the v-th connected component and all its asynchronous connected components; Let be the contour change index of the v-th connected component.

[0105] It should be understood that the temporal variation coefficient can reflect the degree of change in the illumination effect and contour shape of the v-th connected domain at different times. The smaller the temporal variation coefficient, the better it reflects the stability of the v-th connected domain and the more it conforms to the change characteristics caused by corona discharge.

[0106] Step 5: Based on the above analysis results, extract the key feature regions in the image that are significantly affected by the corona phenomenon.

[0107] Based on the above analysis, this application, in addition to step 4, further combines a preliminary analysis of whether corona changes occur in the same common area to extract key feature regions in the insulator image that are significantly affected by the corona phenomenon.

[0108] In a preferred embodiment, the corona confidence of the connected domain is obtained by inversely fusing the corona brightness confidence and the temporal variation coefficient of the same connected domain.

[0109] It is understood that reverse fusion refers to fusion methods such as subtraction and division between data. The specific reverse fusion method shall be determined by the implementer based on the actual situation, and this application does not impose any special restrictions.

[0110] In this embodiment, the ratio of the corona brightness confidence level to the temporal variation coefficient of the same connected domain is used as the corona confidence level of the connected domain.

[0111] Corona confidence of the v-th connected component For example, the expression is: In the formula, Let be the confidence level of the corona brightness of the v-th connected component; Let be the time-series variation coefficient of the v-th connected component; These are preset parameter tuning coefficients. To avoid the denominator being 0, values ​​are taken from the range [0.005, 0.01]. The value has no impact on the calculation, and the implementer can choose the value as needed.

[0112] Corona confidence level, by comprehensively considering factors such as brightness changes, sunlight exposure, and temporal variations, can more accurately identify key areas affected by corona phenomena.

[0113] Furthermore, this application obtains the segmentation threshold of the corona confidence of all connected components in the insulator image, and takes the connected components with corona confidence greater than the segmentation threshold as key feature regions in the insulator image that are greatly affected by the corona phenomenon.

[0114] In this embodiment, the corona confidence scores of all connected components in the i-th insulator image are obtained, and all corona confidence scores are used as input to the Otsu thresholding method, outputting a segmentation threshold. That is, the connected components are divided into two parts by the Otsu thresholding method, and the connected components with corona confidence scores greater than the segmentation threshold are recorded as corona connected components, which are considered as key feature regions in the composite insulator that are significantly affected by the corona phenomenon. By more closely monitoring the key feature regions, the safe and stable operation of the high-voltage line can be ensured.

[0115] Based on the same inventive concept as the above method, this application also provides a multispectral feature extraction device for composite insulators considering corona characteristics. The device implements the multispectral feature extraction method for composite insulators considering corona characteristics described in any one of the above claims. The device includes:

[0116] The image processing module is used to perform registration, fusion, and extraction of composite insulator regions on the acquired multispectral images, and inputs the obtained insulator images into the feature extraction module.

[0117] The feature extraction module is used to analyze the brightness, texture and temporal variation features of the insulator image to take into account the halo effect, the effect of sunlight exposure and the temporal variation features of the composite insulator.

[0118] The corona region extraction module is used to extract key feature regions of composite insulators from insulator images by analyzing various features from the feature extraction module.

[0119] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not invented in this application.

[0120] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A method for extracting multispectral features of composite insulators considering corona characteristics, characterized in that, The method includes: Multispectral images of the composite insulator are acquired, registered, and fused to obtain a composite image; the multispectral images include visible light images, infrared images, and ultraviolet images; The composite insulator region in the composite image is extracted by a neural network and converted into a grayscale image to obtain the insulator image; For any connected region in an insulator image, based on the brightness differences of the same connected region in the HSV space of the visible light, infrared and ultraviolet images, a preliminary analysis is made as to whether the connected region undergoes corona change. Based on the texture features of the connected component in the insulator image, further analysis is conducted to determine whether the connected component is affected by sunlight. Based on the temporal variation characteristics of the insulator images obtained at different times in the same connected region under the influence of sunlight, the key feature regions of the composite insulator can be further analyzed. In conjunction with a preliminary analysis of whether corona changes occur within the same connected region, key feature regions in the insulator image that are significantly affected by the corona phenomenon are extracted. The method for preliminary analysis of whether the connected component has undergone corona changes includes: Calculate the similarity of brightness values ​​of the same connected component in the HSV space of infrared and ultraviolet images, respectively; Calculate the normalized value of the mean brightness of any connected component in the HSV space of any spectral image, and calculate the range of the normalized values ​​of the mean brightness of all connected components in the HSV space of any spectral image. Using the similarity and the normalized value and range of the mean brightness of any connected component in all spectral images, the corona brightness confidence of any connected component is determined, and its value is used to preliminarily analyze whether the connected component has undergone corona changes.

2. The method for extracting multispectral features of composite insulators considering corona characteristics according to claim 1, characterized in that, The method for determining the corona brightness confidence level of any connected component further includes: Let the confidence score of the corona intensity of the v-th connected component be denoted as . The expression is: In the formula, Let v be the similarity of the v-th connected component between the infrared HSV image and the ultraviolet HSV image; , , respectively, are the normalized values ​​of the mean brightness of the v-th connected component in the infrared HSV image, ultraviolet HSV image, and visible light HSV image; , , representing the brightness range of the v-th connected component in the infrared HSV image, ultraviolet HSV image, and visible light HSV image, respectively.

3. The method for extracting multispectral features of composite insulators considering corona characteristics according to claim 1, characterized in that, The methods for further analyzing whether the connected component is affected by sunlight include: Obtain the degree of dispersion of the texture feature values ​​of all pixels in the insulator image for this connected component; The gray values ​​of all pixels between the two pixels with the highest and lowest gray values ​​in the connected component are formed into a sequence; the proportion of negative elements after differencing the sequence is denoted as the gray-level decrease index of the connected component. Obtain the mean gradient value of all edge pixels within the connected component; By using the mean of the calculated dispersion, gray level decrease index, and gradient value within the connected component, the illumination influence coefficient of the connected component is determined, and its value is used to further analyze whether the connected component is affected by sunlight.

4. The method for extracting multispectral features of composite insulators considering corona characteristics according to claim 3, characterized in that, The method for determining the illumination influence coefficient of the connected domain further includes: Let the illumination influence coefficient of the v-th connected component be denoted as... The expression is: In the formula, Let be the gray-level decreasing index of the v-th connected component; The degree of discreteness of all texture feature values ​​in the v-th connected component; It is the mean of the gradient values ​​of all edge pixels in the v-th connected region.

5. The method for extracting multispectral features of composite insulators considering corona characteristics according to claim 3, characterized in that, The method for further analysis of key characteristic regions of composite insulators includes: For all insulator images, generate connected components, and then register all corresponding connected components of the connected component in other insulator images as the time-separated connected components of the connected component. Calculate the degree of dispersion of the illumination influence coefficients of the connected component and all its asynchronous connected components; The Hu moment value between the connected component and each of its different time-separated connected components is calculated using the Hu moment function; the product of the mean of all calculated Hu moment values ​​and the degree of dispersion of all Hu moment values ​​is denoted as the contour change index of the connected component. By using the illumination influence coefficient of the connected region, the dispersion of the calculated illumination influence coefficient, and the contour change index, the temporal variation coefficient of the connected region is determined. The magnitude of the coefficient is then used to further analyze whether the connected region is a key characteristic region of the composite insulator.

6. The method for extracting multispectral features of composite insulators considering corona characteristics according to claim 5, characterized in that, The method for determining the time-series variation coefficients of the connected component further includes: Let the time-series variation coefficient of the v-th connected component be denoted as The expression is: In the formula, Let be the illumination influence coefficient of the v-th connected component; The degree of dispersion of the illumination influence coefficients of the v-th connected component and all its asynchronous connected components; Let be the contour change index of the v-th connected component.

7. The method for extracting multispectral features of composite insulators considering corona characteristics according to claim 1, characterized in that, The method for extracting key feature regions that are significantly affected by corona discharge in insulator images is as follows: The corona confidence score of the same connected component is obtained by inversely fusing the corona brightness confidence score with the temporal variation coefficient. The segmentation thresholds for the corona confidence of all connected components in the insulator image are obtained. Connected components with corona confidence values ​​greater than the segmentation threshold are taken as key feature regions in the insulator image that are significantly affected by the corona phenomenon.

8. The method for extracting multispectral features of composite insulators considering corona characteristics according to claim 7, characterized in that, The method for determining the corona confidence of the connected region is further determined by the ratio of the corona brightness confidence to the temporal variation coefficient of the same connected region.

9. A multispectral feature extraction device for composite insulators considering corona characteristics, characterized in that, The apparatus for implementing the multispectral feature extraction method for composite insulators considering corona characteristics as described in any one of claims 1-8 includes: The image processing module is used to perform registration, fusion, and extraction of composite insulator regions on the acquired multispectral images, and inputs the obtained insulator images into the feature extraction module. The feature extraction module is used to analyze the brightness, texture and temporal variation features of the insulator image to take into account the effects of corona discharge, sunlight exposure and temporal variation of the composite insulator. The corona region extraction module is used to extract key feature regions of composite insulators from insulator images by analyzing various features from the feature extraction module.

Images