An image evaluation and analysis method and device for insulator detection in ice and snow environment

Abstract

The invention relates to the field of insulator detection, and discloses an image evaluation and analysis method and device for insulator detection in an ice and snow environment. The method comprises the following steps: obtaining an original image of an insulator in an ice and snow environment; constructing an image snow removal model; the snow removal model comprises: a fusion filter module, a priori knowledge module and a deep learning module; processing the original image of the insulator using the model; constructing a system for extracting specific texture features of the insulator based on a grayscale co-occurrence matrix, and combining a color image similarity index algorithm to evaluate the similarity of images before and after processing from multiple dimensions to judge the image quality of the insulator in an ice and snow environment; constructing a relationship between image quality and snowflake noise and observation distance, and providing a theoretical basis for optimizing the acquisition and processing of insulator images in an ice and snow environment; the invention provides scientific and reliable technical support for the accurate evaluation of insulator video detection images in ice and snow environment areas, helps to improve the efficiency and accuracy of insulator fault detection, and ensures the stable operation of power systems.

Description

Technical Field

[0001] The present invention relates to the field of insulator detection, and in particular to an image evaluation and analysis method and device for insulator detection in an ice and snow environment. Background Art

[0002] With the change of global climate, extreme snowstorms are becoming more frequent in Northeast my country, and the increasing trend of heavy snowstorms has brought many problems to insulators in power systems. In a heavy snowstorm environment, the video image of the insulator is strongly interfered by snowflakes, and its key information such as outline and shed details are blurred, which greatly reduces the accuracy of image recognition and seriously hinders the monitoring of the operating status and fault diagnosis of insulators.

[0003] In recent years, unmanned aerial vehicle optical detection equipment has been widely used in insulator fault detection, but it is restricted by environmental factors. Under extreme meteorological conditions, rain, fog, and snow appear in combination or alternately, which has a serious negative impact on the visual effect of the optical detection front end, greatly increasing the difficulty of insulator optical detection image restoration. In the field of image processing, salt and pepper type feature video images are the focus of research, and natural interference factors such as snow are the focus of international research, because their morphology, snowfall level and visibility affect image judgment. However, there are currently few studies on image and video quality evaluation and screening in blizzard weather, and there is also a lack of effective evaluation methods to screen target images in the process of insulator fault detection, making it difficult to obtain clear background data for subsequent processing and analysis, which greatly limits the application and development of this technology in ice and snow environments. Summary of the invention

[0004] The purpose of the present invention is to propose an image evaluation and analysis method and device for insulator detection in an ice and snow environment, so as to fill the gap in the prior art in insulator image quality evaluation in blizzard weather.

[0005] The present invention provides an image evaluation and analysis method for insulator detection in an ice and snow environment, comprising the following steps: S1: Get the original image of the insulator in the ice and snow environment; S2: construct an image snow removal model; the snow removal model includes: a fusion filter module, a priori knowledge module and a deep learning module; S3: Use the image snow removal model to process the original image of the insulator to obtain the image after snow removal; S4: A system for extracting specific texture features of insulators is constructed based on the gray-level co-occurrence matrix, and the color image similarity index algorithm is used to evaluate the similarity of the images before and after the snow removal process from multiple dimensions to determine the image quality of insulators in ice and snow environments. S5: According to the insulator image quality, the relationship between image quality, snow noise and observation distance is constructed to provide a theoretical basis for the optimization of insulator image acquisition and processing in ice and snow environments.

[0006] A storage medium stores instructions and data for implementing an image evaluation and analysis method for insulator detection in an ice and snow environment.

[0007] An image evaluation and analysis device for insulator detection in an ice and snow environment comprises: a processor and a storage medium; the processor loads and executes instructions and data in the storage medium to implement an image evaluation and analysis method for insulator detection in an ice and snow environment.

[0008] The beneficial effects provided by the present invention are: 1. Construct an image snow removal model, which can effectively remove snow interference, make the insulator image clear, significantly improve the recognition accuracy, and assist in judging the operating status.

[0009] 2. Image texture feature extraction and multi-dimensional evaluation of CISI algorithm provide quantitative results and provide accurate basis for image screening and fault detection.

[0010] 3. Based on the model and relationship, select the distance reasonably during acquisition, effectively reduce noise during processing, optimize the acquisition and processing process, and improve image quality.

[0011] 4. Solve the problem of ice and snow environment detection, promote the development of insulator detection technology, and ensure the efficient operation and maintenance of insulators in the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a schematic flow chart of the method of the present invention; Figure 2 It is a schematic diagram of the processing results of various snow removal methods under different conditions; Figure 3 It is a relationship diagram between snowfall level and CISI evaluation value of a video image evaluation and analysis method for insulator detection in an ice and snow environment area in an embodiment of the present application; Figure 4 This is a relationship diagram between shooting distance and CISI evaluation value of a method for evaluating and analyzing video images of insulator detection in an ice and snow environment area in an embodiment of the present application; Figure 5 This is a photo of a typical substation in snow and wind; Figure 6 It is a working schematic diagram of the hardware device of the present invention. DETAILED DESCRIPTION

[0013] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0014] Before formally describing the present invention, the scheme of the present invention is first generally described for easy understanding.

[0015] Please refer to Figure 1 , Figure 1 It is a schematic diagram of the process of the present invention.

[0016] The present invention provides an image evaluation and analysis method for insulator detection in an ice and snow environment, comprising the following steps: S1: Get the original image of the insulator in the ice and snow environment; S2: construct an image snow removal model; the snow removal model includes: a fusion filter module, a priori knowledge module and a deep learning module; It should be noted that the fusion filter module (Filter Method, FI) designs a filtering algorithm according to the geometric information and optical information of snowflakes in the outdoor image to achieve image snow removal.

[0017] As an embodiment, a filtering algorithm is designed based on the geometric and optical information of snowflakes in outdoor images. Iterative median filtering, histogram filtering, etc. are used. By analyzing the spatial distribution and other characteristics of snowflakes in the image, a suitable filter is designed to remove snowflakes from the image.

[0018] It should be noted that the prior knowledge method (KN) uses the Narasimhan model method for removing snow interference to remove snow.

[0019] As an example, the color, appearance and frequency of snowflakes are exploited to construct a snow removal algorithm. The Narasimhan model method with snowflake removal is particularly prominent, through the following equation: , in, k is the atmospheric scattering coefficient, is the input image before processing, is the undisturbed image, is the scene depth, is the transmission distribution, A is the ambient light parameter, S is the noise value caused by snowflakes.

[0020] It should be noted that the deep learning method (DL) module trains the convolutional neural network (CNN) by using snowy images and their corresponding snow-free images to obtain a mapping relationship in order to remove snow from the image.

[0021] S3: Use the image snow removal model to process the original image of the insulator to obtain the image after snow removal; In the present invention, the image snow removal model can be expressed as follows: , in, It is a fusion filter module. It is a prior knowledge module. is a deep learning module, and the subscripts 1 to n are a specific method. In order to obtain the best effect, a superposition method can be used to remove snow from video images. Of course, in some other embodiments, a specific method can also be used.

[0022] S4: A system for extracting specific texture features of insulators is constructed based on the gray-level co-occurrence matrix, and the color image similarity index algorithm is used to evaluate the similarity of the images before and after the snow removal process from multiple dimensions to determine the image quality of insulators in ice and snow environments. It should be noted that the texture features of the insulator shed details and the snow / ice-covered state are extracted based on the grayscale co-occurrence matrix. The texture features include image texture contrast, image energy distribution and homogeneity. The local changes of the image, the uniformity of grayscale distribution, and the clarity and regularity of the texture are measured through a specific calculation method.

[0023] That is, the system of insulator specific texture features in step S4 includes: image texture contrast features, image energy distribution features and homogeneity features.

[0024] As an embodiment, the image texture contrast is calculated, which is related to the corresponding value of the coordinates in the GLCM and is calculated by a specific formula to measure the local changes of the image and reflect the degree of contrast between the depth of the umbrella skirt structure. The stronger the contrast, the clearer the texture. The image texture contrast C is expressed by the following formula: .

[0025] The image energy distribution is calculated to measure the grayscale uniformity of the shed. According to the corresponding formula, the more uniform the distribution of the shed texture is, the larger the value is, and the rougher the texture is.

[0026] The image energy distribution E is expressed by the following formula: .

[0027] Calculate the homogeneity of image quality, which reflects the regularity of the umbrella skirt texture and structure. It is calculated by a specific formula. The larger the value is, the clearer and more regular the texture is.

[0028] The homogeneity I can be expressed by the following formula: , Among the above three formulas, , is the number of rows and columns of GLCM, Represents the value corresponding to the coordinate (p, q) in the GLCM.

[0029] Next, the color image similarity index algorithm is used to evaluate the similarity between the processed image and the original image from three dimensions of brightness, chroma and hue. The algorithm calculates the comparison values ​​of texture contrast, energy distribution value and quality homogeneity value of the three-dimensional parameter images of brightness, chroma and hue, and obtains a comprehensive evaluation result based on the weight coefficient determined by visual sensitivity. The weight coefficients are 1.0 for brightness, 0.45 for chroma and 0.45 for hue.

[0030] As an embodiment, in step S4, the color image similarity index algorithm is expressed by the following formula: , , , , Among them, C L , C C and C H are the comparison values ​​of the texture contrast of the three-dimensional parameters of brightness, chroma and hue between the image after snow removal and the original image; E L 、E C and E H are the comparison values ​​of the energy distribution values ​​of the three-dimensional parameter images of brightness, chroma and hue between the image after snow removal and the original image; E L 、E C and E H are the comparison values ​​of the image quality homogeneity values ​​of the three-dimensional parameters of brightness, chroma and hue between the processed image and the original image respectively; n Indicates the scale number of the M-SSIM algorithm.

[0031] S5: According to the insulator image quality, the relationship between image quality, snow noise and observation distance is constructed to provide a theoretical basis for the optimization of insulator image acquisition and processing in ice and snow environments.

[0032] It should be noted that the present invention proposes a Narasimhan model method containing snowflake interference removal and uses it to explain that the image quality is in a negative linear relationship with the snowflake noise and in a negative exponential relationship with the observation distance, which are respectively a negative linear relationship and a negative exponential relationship.

[0033] The present application provides an embodiment as follows, which analyzes the relationship between image quality, snowflake noise and observation distance based on the Narasimhan model method including snowflake interference removal. When the observation distance is fixed, the image quality is in a negative linear relationship with the snowflake noise, that is, the greater the snowflake noise, the worse the image quality; when the snowflake noise is fixed, the image quality is in a negative exponential relationship with the observation distance, and the longer the observation distance, the faster the image quality decreases.

[0034] The present invention takes a 220kV line in Hulunbuir area in 2023 as an example, and further explains as follows: First, the wind and snow image was taken from a 220kV line tower head image in Hulunbuir area in 2023, 30m away from the high-voltage end of the mid-phase insulator, and the image was obtained at three snowfall levels: moderate snow, heavy snow, and blizzard. The image was processed according to the method of the present invention, and the processing results are shown in Figure 2 shown.

[0035] Figure 2 Among them, the FI method adopts the common Gaussian filter; the KN method adopts the Narasimhan model method; the DL method adopts the DBTNet fine-grained image classification model to connect the DesnowNet, HDCWNet and SMGARN snow removal models.

[0036] use Figure 1 The CISI evaluation system was used to evaluate the original image and the processed image. For comparison, the structural similarity index measure (SSIM) and peak signal-to-noise ratio (PSNR) were used for evaluation. The results are shown in Table 1. Combined with Table 1, the DL method is relatively better than the FI method and the KN method. This method automatically learns and optimizes the snow removal model through training data. Due to the mapping relationship in principle, the contour details of the insulator shed and the snow covering condition of the insulator are retained, which can provide corresponding important image data for the diagnosis of extreme wind and snow weather.

[0037] Table 1 Evaluation values ​​of various processing methods under ice and snow environment

[0038] Please refer to Figure 3 , Figure 3 It is a relationship diagram between snowfall level and CISI evaluation value of a video image evaluation and analysis method for insulator detection in an ice and snow environment area in an embodiment of the present application. The relationship diagram includes: comparison between the original image and the FI processed image, comparison between the original image and the KN processed image, and comparison between the original image and the DL processed image.

[0039] By comparing the original image with the results of the three processing methods, FI, KN and DL, the evaluation results of moderate snow, heavy snow and blizzard (corresponding to snowfall levels of 3, 4 and 5, respectively) are given. Figure 3 As shown, from Figure 3 It can be concluded that the evaluation values ​​of the medium snow environment and the heavy snow environment are relatively close, and the evaluation value of the blizzard environment is relatively small. Figure 2From Table 1, it can be seen that in the deep learning method snow removal processing image, in the medium snow environment and heavy snow environment (distance 30m), the texture features of the insulator image are retained, the insulator shed retains certain details, and the snow on the shed surface is clearer. CISI The value is greater than 0.6; however, in a blizzard environment, the image texture features are blurred due to the interference of snowflakes, and the image effectiveness is poor. CISI The value is less than 0.6, which has a certain impact on the insulator image detection, while the FI method and KN method are relatively poor.

[0040] image L ( p , q )and CISI ( p , q ) values ​​have a linear relationship, CISI 0( p , q ), CISI b( p , q ) value is the input image before processing L ( p , q ), undisturbed image L0 ( p , q ) can be approximately expressed by the following formula: , , R is the evaluation coefficient, M 0 and Mb To evaluate the compensation value.

[0041] Here, CISI Value Replacement L ( p , q )value, Figure 3 The three processing results can be further explained by the image texture contrast equation. In the equation, k , A The value can be considered as a constant, then L ( p , q ), L0 ( p , q ) is determined by two parameters, namely, snowflake noise S and observation distance d Fixed observation distance d , then only with the snowflake noise S Related, L0 ( p, q ) e -kd(p,q) is a constant, set V ,set up t 1- e -kd(p,q), the equation for image texture contrast can be simplified to , From this equation, we can see that L ( p , q ) value and S This is due to the decreasing trend Figure 3 The trend is the same. Relatively speaking, the deep learning method has the best snow removal effect.

[0042] Please refer to Figure 4 , Figure 4 It is a relationship diagram between shooting distance and CISI evaluation value of a method for evaluating and analyzing video images of insulator detection in an ice and snow environment area in an embodiment of the present application, and the relationship diagram includes: comparison between the original image and the FI processed image, comparison between the original image and the KN processed image, and comparison between the original image and the DL processed image; from Figure 4 It can be seen from the figure that the CISI evaluation values ​​of the three image processing results are all inversely saturated with the observation distance. S Under the conditions, t The value is approximately 1, then the equation for image texture contrast is: , L ( p , q ) value and d(p,q) The two are in a negative exponential relationship with a decreasing trend, which is a typical anti-saturation relationship, that is, the greater the distance, the faster the texture features of the insulator image blur. Figure 5 This is a photo of a typical substation in the snow. The observation distance and snowfall level are unknown. There are 5 post insulators from left to right. Although the observation distance and snowfall level are unknown, the distances of the insulators from near to far are No. 2, No. 4, No. 5, No. 1 and No. 3. Starting from No. 5 insulator, the details of the post insulators are quickly blurred. The images of No. 1 and No. 3 insulators are basically invalid views. This picture further explains Figure 4 relationship.

[0043] Finally, the overall application method is as follows: First, a snow removal model is constructed, which integrates the fusion filter module, prior knowledge module and deep learning module.

[0044] The fusion filter module is designed based on the geometric and optical information of snowflakes. For example, it uses specific methods such as iterative median filtering to design filters to remove snowflakes by analyzing the spatial distribution of snowflakes and other characteristics.

[0045] The prior knowledge module mines the characteristics of snowflake color, appearance, etc. to build an algorithm. The Narasimhan model method considers the interference of snowflake noise through an equation that includes parameters such as atmospheric scattering coefficient, input image, undisturbed image, scene depth, transmittance distribution rate, ambient light parameters, and noise value caused by snowflakes. It can also analyze the impact of snowfall intensity and the distance between the lens and the device on the video image.

[0046] The degree learning module uses images with snow and without snow to train the convolutional neural network to obtain the mapping relationship to achieve snow removal. In practical applications, in order to achieve the best effect, according to the actual situation of the image, such as the distribution characteristics of snowflakes, snowfall intensity, etc., you can choose a suitable method or superimpose these methods to remove snow from the insulator image, remove snow interference, and improve image clarity.

[0047] Secondly, the texture features of the insulator image are extracted based on the gray-level co-occurrence matrix (GLCM). The image texture contrast is calculated, which is related to the corresponding value of the coordinates in the GLCM. It is calculated through a specific formula and is used to measure the local changes of the image and reflect the degree of contrast between the depth of the shed structure. The stronger the contrast, the clearer the texture. The image energy distribution is calculated to measure the uniformity of the shed grayscale. According to the corresponding formula, the more uniform the distribution of the shed texture, the larger the value, and the rougher the texture.

[0048] The image quality homogeneity is calculated to reflect the regularity of the umbrella skirt texture and structure. It is calculated by a specific formula. When the texture is clear and regular, the larger the value is. Then, the color image similarity index (CISI) algorithm is used to evaluate the similarity between the processed image and the original image from three dimensions: brightness, chroma and hue. By calculating the comparison values ​​of the texture contrast, energy distribution value and quality homogeneity value of the three-dimensional parameter image of brightness, chroma and hue, and determining the weight coefficient based on visual sensitivity (brightness 1.0, chroma 0.45, hue 0.45), a comprehensive evaluation result is finally obtained to judge the image quality and screen out valid images for subsequent analysis.

[0049] Finally, the Narasimhan model is proposed and used to describe the image formation process. The relationship between image quality, snowflake noise and observation distance is analyzed based on the model. When the observation distance is fixed, the image quality is in a negative linear relationship with the snowflake noise, that is, the greater the snowflake noise, the worse the image quality; when the snowflake noise is fixed, the image quality is in a negative exponential relationship with the observation distance, and the longer the observation distance, the faster the image quality decreases. In the image acquisition stage, the observation distance can be reasonably selected according to this relationship. For example, when using drones to collect insulator images, the appropriate flight altitude can be determined according to different snowfall conditions and equipment performance to avoid serious degradation of image quality due to excessive distance. In the image processing stage, according to the relationship between image quality and snowflake noise, when the image quality is poor and the observation distance is fixed, it can be inferred that the snowflake noise is large, and then targeted denoising measures can be taken, such as adjusting the snow removal algorithm parameters or replacing more suitable snow removal methods, so as to optimize the image acquisition and processing process and improve the accuracy of insulator detection.

[0050] See also Figure 6 , Figure 6 It is a schematic diagram of the working of the hardware device of an embodiment of the present invention, and the hardware device specifically includes: an image evaluation and analysis device 401 for insulator detection in an ice and snow environment, a processor 402 and a storage medium 403.

[0051] An image evaluation and analysis device 401 for detecting insulators in an ice and snow environment: The image evaluation and analysis device 401 for detecting insulators in an ice and snow environment implements the image evaluation and analysis method for detecting insulators in an ice and snow environment.

[0052] Processor 402: The processor 402 loads and executes the instructions and data in the storage medium 403 to implement the image evaluation and analysis method for insulator detection in an ice and snow environment.

[0053] Storage medium 403: The storage medium 403 stores instructions and data; the storage medium 403 is used to implement the image evaluation and analysis method for insulator detection in an ice and snow environment.

[0054] The beneficial effects of the present invention are: 1. Construct an image snow removal model, which can effectively remove snow interference, make the insulator image clear, significantly improve the recognition accuracy, and assist in judging the operating status.

[0055] 2. Image texture feature extraction and multi-dimensional evaluation of CISI algorithm provide quantitative results and provide accurate basis for image screening and fault detection.

[0056] 3. Based on the model and relationship, select the distance reasonably during acquisition, effectively reduce noise during processing, optimize the acquisition and processing process, and improve image quality.

[0057] 4. Solve the problem of ice and snow environment detection, promote the development of insulator detection technology, and ensure the efficient operation and maintenance of insulators in the power system.

[0058] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for evaluating and analyzing insulator detection images in an ice and snow environment, characterized by: The method comprises the following steps: S1: Get the original image of the insulator in the ice and snow environment; S2: Build an image snow removal model; The snow removal model includes: a fusion filter module, a priori knowledge module and a deep learning module; S3: Use the image snow removal model to process the original image of the insulator to obtain the image after snow removal; S4: A system for extracting specific texture features of insulators is constructed based on the gray-level co-occurrence matrix, and the color image similarity index algorithm is used to evaluate the similarity of the images before and after the snow removal process from multiple dimensions to determine the image quality of insulators in ice and snow environments. S5: According to the insulator image quality, the relationship between image quality, snow noise and observation distance is constructed to provide a theoretical basis for the optimization of insulator image acquisition and processing in ice and snow environments.

2. The image evaluation and analysis method for insulator detection in an ice and snow environment according to claim 1, characterized in that: The fusion filtering module in step S2 specifically adopts: iterative median filtering and histogram filtering.

3. The image evaluation and analysis method for insulator detection in an ice and snow environment according to claim 1, characterized in that: In step S2, the prior knowledge module removes snow using the Narasimhan model method for removing snow interference.

4. The image evaluation and analysis method for insulator detection in an ice and snow environment according to claim 1, characterized in that: In step S2, the deep learning module uses the snow image and its corresponding snow-free image to train a convolutional neural network to obtain a mapping relationship between the two, thereby achieving image snow removal.

5. The method for evaluating and analyzing insulator detection images in an ice and snow environment according to claim 1, characterized in that: The system of insulator specific texture features in step S4 includes: image texture contrast features, image energy distribution features and homogeneity features.

6. The method for evaluating and analyzing insulator detection images in an ice and snow environment according to claim 5, characterized in that: The image texture contrast C is expressed by the following formula: ， The image energy distribution E is expressed by the following formula: ， The homogeneity I can be expressed by the following formula: ， Among the above three formulas, , is the number of rows and columns of GLCM, Represents the value corresponding to the coordinate (p, q) in the GLCM.

7. The image evaluation and analysis method for insulator detection in an ice and snow environment according to claim 6, characterized in that: In step S4, the color image similarity index algorithm is expressed by the following formula: ， ， ， ， Among them, C L , C C and C H are the comparison values ​​of the texture contrast of the three-dimensional parameters of brightness, chroma and hue between the image after snow removal and the original image; E L 、E C and E H are the comparison values ​​of the energy distribution values ​​of the three-dimensional parameter images of brightness, chroma and hue between the image after snow removal and the original image; E L 、E C and E H are the comparison values ​​of the image quality homogeneity values ​​of the three-dimensional parameters of brightness, chroma and hue between the processed image and the original image respectively; n Indicates the scale number of the M-SSIM algorithm.

8. The method for evaluating and analyzing insulator detection images in an ice and snow environment according to claim 1, characterized in that: In step S5, the relationship between image quality, snowflake noise and observation distance is explained according to the Narasimhan model. When the observation distance is fixed, the image quality is in a negative linear relationship with the snowflake noise; when the snowflake noise is fixed, the image quality is in a negative exponential relationship with the observation distance.

9. A storage medium, characterized in that: The storage medium stores instructions and data for implementing an image evaluation and analysis method for insulator detection in an ice and snow environment as described in any one of claims 1 to 8.

10. An image evaluation and analysis device for insulator detection in an ice and snow environment, characterized in that: include: A processor and a storage medium; the processor loads and executes instructions and data in the storage medium to implement an image evaluation and analysis method for insulator detection in an ice and snow environment as described in any one of claims 1 to 8.

Images