A high-voltage disconnecting switch state recognition method based on gradient image fusion
By using gradient image fusion technology, the accuracy and stability issues of high-voltage disconnector status recognition methods have been resolved, achieving efficient recognition in different environments, overcoming the shortcomings of single image methods, and providing all-weather status recognition capabilities.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-11-16
- Publication Date
- 2026-03-17
AI Technical Summary
Existing methods for identifying the status of high-voltage disconnect switches suffer from problems such as inaccurate identification, susceptibility to external interference, and inability to operate around the clock. In particular, temperature-based detection is susceptible to damage from high voltage, infrared images have low pixel counts, and visible light images are easily affected by obstructions.
A gradient image fusion method is adopted, which decomposes infrared and visible light images into high-frequency and low-frequency sub-band images through image preprocessing, registration, decomposition and fusion techniques. Local and global fusion is performed using pulse-coupled neural networks and visual saliency segmentation algorithms to establish a gradient image fusion model, and the state is identified by pixel integral projection algorithm.
It achieves accurate identification of the status of high-voltage disconnect switches, overcomes the shortcomings of single image methods, can work effectively in various environments, and improves the accuracy and stability of identification.
Smart Images
Figure CN115909028B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image recognition, and more particularly to a method for recognizing the status of high-voltage disconnect switches based on gradient image fusion. Background Technology
[0002] High-voltage disconnect switches play a crucial role in power supply systems, serving as vital equipment to protect maintenance personnel and electrical equipment. However, due to their long-term outdoor operation, they are susceptible to mechanical failures caused by external weather and strong electric fields, leading to malfunctions in opening and closing mechanisms and threatening the safe operation of the power grid.
[0003] Methods for identifying the open / closed status of disconnecting switches mainly include temperature detection, image recognition, stress-strain detection, and motor current detection. Temperature detection determines the open / closed status by detecting the temperature of the moving and stationary contacts. This method is a contact-based measurement, and the sensor is easily damaged by high voltage and strong current. Infrared image-based disconnecting switch status identification is a non-contact measurement method with strong penetration and anti-interference capabilities. However, its low pixel count makes it prone to losing image details and is sensitive to temperature. Visible light image recognition acquires and processes images of the disconnecting switch, using intelligent algorithms to identify the open / closed status. Visible light images have high pixel counts, capturing more image details and are less sensitive to external temperature. However, it is easily affected by obstructions and is significantly affected by external interference. Stress-strain detection determines the open / closed status by detecting the strain of the operating lever. Since the relative position of the moving and stationary contacts is crucial to accurately reflect the disconnecting switch's open / closed status, this method suffers from inaccurate identification. The principle of motor current detection is to determine the opening and closing status by analyzing the peak changes of the drive motor current under different conditions. Simulations were performed on the drive motor torque, current, and acceleration under the opening and closing states of the disconnector switch, and the opening and closing status was determined by the current commutation time. However, this method cannot directly detect the relative position of the moving and stationary contacts of the disconnector switch, and therefore cannot guarantee the accuracy of opening and closing identification. Summary of the Invention
[0004] To address the aforementioned problems, this invention aims to provide a high-voltage disconnector switch status recognition method based on gradient image fusion. First, infrared and visible light images of the high-voltage disconnector switch are acquired. These images undergo preprocessing and registration. A non-subsampled shear wave transform algorithm is used to decompose the registered infrared and visible light images into high-frequency and low-frequency sub-band images, respectively. A pulse-coupled neural network algorithm is then used to fuse the high-frequency infrared and visible light sub-band images. Finally, a visual saliency feature-based segmentation algorithm is used to fuse the low-frequency infrared and visible light sub-band images, achieving local image fusion. Then, a non-subsampled shear wave inverse transform algorithm is used to fuse the locally fused images again, achieving global fusion. This yields a fused image of the high-voltage disconnector switch's infrared and visible light images, thus realizing a gradient image fusion model. An image fusion quality evaluation scheme is established to compare the effectiveness of the gradient image fusion algorithm with common image fusion schemes. Finally, the fused image is processed using a pixel integral projection algorithm to identify the opening and closing status of the high-voltage disconnector switch. The results of high-voltage disconnector switch opening and closing status recognition were verified through experimental simulation. This method achieves complementary advantages between infrared and visible light images of high-voltage disconnectors, addressing issues such as low pixel count, loss of image detail, and temperature sensitivity inherent in single infrared images, as well as susceptibility to obstructions, external interference, and inability to operate in all weather conditions associated with single visible light images.
[0005] To achieve the above objectives, the present invention provides a high-voltage disconnector state recognition method based on gradient image fusion, implemented as follows:
[0006] A high-voltage disconnector switch status recognition method based on gradient image fusion includes an image preprocessing scheme, an image registration scheme, a gradient image fusion scheme, a fused image evaluation scheme, a high-voltage disconnector switch status recognition scheme, and an experimental simulation scheme. The image preprocessing scheme converts the acquired infrared and visible light images of the high-voltage disconnector switch into grayscale images and performs filtering and noise reduction on both types of images to improve image quality and processing speed. The image registration scheme aligns the infrared and visible light images of the high-voltage disconnector switch in space, finding the mapping relationship between them to enable better spatial fusion. The gradient image fusion scheme decomposes the registered infrared and visible light images of the high-voltage disconnector switch into high-frequency and low-frequency sub-band images. Local image fusion is achieved by fusing the high-frequency and low-frequency sub-band images of the infrared and visible light images of the high-voltage disconnector switch, respectively. Then, global fusion is achieved by fusing the separately fused infrared low-frequency and high-frequency sub-band images. A gradient image fusion model is formed by combining the local and global fusion of the infrared and visible light images of the high-voltage disconnector switch. The fused image evaluation scheme is used to evaluate the quality of the gradient image fusion model after fusing the infrared and visible light images of the high-voltage disconnector switch. The high-voltage disconnector switch status recognition scheme is used to recognize the image after fusing the infrared and visible light images of the high-voltage disconnector switch to determine whether the high-voltage disconnector switch is in the open or closed state. The experimental simulation scheme is used to compare the recognition accuracy of the fused high-voltage disconnector switch image with that of a single high-voltage disconnector switch visible light image or high-voltage disconnector switch infrared image.
[0007] The image preprocessing scheme of the present invention is as follows:
[0008] S1. Image Acquisition:
[0009] The high-voltage disconnect switch was set to open and closed states respectively, and images of the high-voltage disconnect switch under different states were collected at the same location using a visible light camera and an infrared thermal imager.
[0010] S2. Image grayscale conversion
[0011] The acquired infrared and visible light images are converted to grayscale to improve the image processing speed. Grayscale conversion is achieved using the following formula:
[0012] Gray(i,j)=0.299×R(i,j)+0.587×G(i,j)+0.114×B(i,j) (1)
[0013] In the formula, Gray(i,j) represents the gray value of the pixel at coordinate (i,j), R(i,j) represents the red component of the image, G(i,j) represents the green component of the image, and B(i,j) represents the blue component of the image.
[0014] S3. Image Filtering and Denoising
[0015] Two-dimensional median filtering is used to remove noise from the image. The pixel values in the two-dimensional sliding template are arranged from largest to smallest to obtain a two-dimensional data sequence. Let the filtered image be g(x,y), then:
[0016] g(x,y)=med{f(xk),yl},(k,l∈A) (2)
[0017] In the formula, f(xk,yl) represents the original image of the disconnecting switch, A represents the two-dimensional sliding template, and the template area is selected as a 3×3 template.
[0018] The image registration scheme of the present invention is as follows:
[0019] The acquired infrared and visible light images were normalized using the maximum-minimum method:
[0020]
[0021] In the formula, norm is the normalized value, and x i Let x be the value of a pixel in the image, and max(x) and min(x) represent the maximum and minimum pixel values in the image, respectively.
[0022] The resolution of the infrared and visible light images of the high-voltage disconnector was adjusted to 2306*2658. The visible light image of the high-voltage disconnector was mapped onto its infrared image using an image feature transformation model. First, the Canny operator was used to perform edge detection on the infrared and visible light images to extract the image contours. Then, the SURF algorithm was used to detect and extract the corner points of the image contours. Finally, the Hessian matrix was used to extract the feature points of the image contours.
[0023]
[0024] In the formula, L xx (x,σ) represents the convolution of the second-order Gaussian derivative at scale σ with a point I(x,y) in the image, calculated as follows:
[0025]
[0026] G(x,y) represents the image; similarly, L can be calculated. xy (x,σ) and L yy (x,σ);
[0027] Next, the Euler distance formula is used to match the feature points of the image. Finally, the least squares method is used to fit these feature points, exchange parameter estimates, and obtain the optimal transformation model. Finally, the registered infrared and visible light images of the high-voltage disconnector are obtained.
[0028] The gradient image fusion scheme of the present invention is as follows:
[0029] The registered infrared and visible light images of the high-voltage disconnector are decomposed into high-frequency and low-frequency sub-band images using a non-subsampled shear wave transform algorithm. A pulse-coupled neural network algorithm is then used to fuse the infrared and visible high-frequency sub-band images. A visual saliency feature segmentation algorithm is employed to fuse the infrared and visible low-frequency sub-band images, achieving local image fusion. Finally, a non-subsampled shear wave inverse transform algorithm is used to fuse the locally fused images again, achieving global fusion. This yields the fused infrared and visible light images of the high-voltage disconnector, thus realizing a gradient image fusion model. The specific steps are as follows:
[0030] S1. Image Decomposition
[0031] The non-subsampled shear wave transform decomposition comprises two parts: multi-scale decomposition and multi-directional decomposition. At the multi-scale decomposition level, a non-subsampled pyramid filter is used to decompose the infrared and visible light images of the high-voltage disconnector into one low-frequency sub-band image and multiple high-frequency sub-band images, respectively. Then, a non-subsampled shear wave filter is used to decompose the decomposed high-frequency sub-band images in multiple directions. A Meyer window function is constructed, and a convolution algorithm is used to convolve the high-frequency sub-band images with the Meyer window function to obtain the high-frequency sub-band coefficients in different directions. This enables the decomposition of the infrared and visible light images of the high-voltage disconnector in different directions.
[0032] S2. Gradient Image Fusion Design
[0033] S21. Image Local Fusion
[0034] S211. High-frequency sub-band diagram fusion
[0035] Spatial frequency is used as the input to the pulse-coupled neural network (PLN) algorithm, and Laplacian energy is used as the connection strength of the PLA. The ignition frequency after n iterations of the PLA is calculated, and the high-frequency sub-band image fusion is achieved based on the ignition frequency. Let the spatial frequency of the image be MSF, then:
[0036]
[0037] In the formula, RF and CF represent the row frequency and column frequency of the image, respectively;
[0038] In a pulse-coupled neural network model, the degree of correlation between neurons is represented by connection strength, which is expressed as:
[0039]
[0040] LP(i,j)=|2C(i,j)-C(i-step,j)-C(i+step,j)|+|2C(i,j)-C(i,j-step)-C(i,j+step)| (8)
[0041] In equations (7) and (8), step is the distance between pixels, with a value of 1, C(i,j) represents the high-frequency subband coefficient, ω(i,j) is the weight of the high-frequency subband coefficient, and MSLR(i,j) is the sum of the Laplacian energy of the image at coordinates (i,j).
[0042] Ignition frequency is expressed as:
[0043]
[0044] In the formula, T R (n) and T V (n) represents the ignition frequency of the high-frequency sub-band image after non-subsampled shear wave transform decomposition of the infrared and visible light images of the high-voltage disconnector, respectively. Thus, the infrared high-frequency sub-band image and the visible light high-frequency sub-band image of the high-voltage disconnector are fused together to obtain the high-frequency sub-band fused image.
[0045] S212. Low-frequency subband map fusion
[0046] This invention uses a hypercomplex Fourier transform algorithm to obtain salient features (such as contours and background information) in low-frequency subband images, dividing the low-frequency subband image into two regions: salient and non-salient. Let the hypercomplex matrix be:
[0047] f(m,n)=ω1f1i+ω2f2j+ω3f3k (10)
[0048] In the formula, f(m,n) represents a hypercomplex matrix, and i, j, k satisfy i 2 =j 2 =k 2 =ijk=-1, ω1~ω3 represent the weights of the image's brightness, orientation, and texture information, respectively, with values of 0.5, 0.25, and 0.25. f1~f3 represent the feature matrices of these three dimensions. Performing a hypercomplex Fourier transform on f(m,n) yields:
[0049] F H [μ,ν]=||F H [μ,ν]||eμφ(μ,ν) (11)
[0050] In the formula, ||F H [μ,ν]|| denotes the first-order norm of f(m,n) in the frequency domain;
[0051] Applying a Gaussian kernel function to ||F H [μ,ν]|| is smoothed to obtain ||F H A multi-scale map of [μ,ν]|| is used, and then the salient image is obtained using the inverse hypercomplex Fourier transform. The salient and non-salient regions of the image are segmented in the image space based on the minimum entropy value. An image fusion method based on visual saliency feature segmentation is used to divide the image features into salient and non-salient regions for fusion. The similarity between two salient maps is compared to determine whether the images are similar, as shown below:
[0052]
[0053] SIM A,B (i,j) represents the significant similarity, SM A (i,j) and SM B (i,j) represent the significance values of the infrared and visible light images of the high-voltage disconnector, respectively. A,B The larger (i,j) is, the more similar the infrared and visible light images of the high-voltage disconnector are. The fused output is:
[0054]
[0055] In the formula, ω represents the weighting value of fusion, T represents the similarity threshold (ranging from 0.5 to 1), IR represents the infrared image, and VI represents the visible light image;
[0056] The image fusion process uses the magnitude of information entropy as a criterion, fusing images with higher information entropy. The criteria for fusion are as follows:
[0057]
[0058] In the formula, F(i,j) represents the fused image, LR is the information entropy ratio, and LR = LR IR / LR VI IR and VI represent infrared image and visible light image, respectively. When LR>1, the information entropy of the infrared image in this region is higher than that of the visible light image. Therefore, the infrared image is used as the fused image, and vice versa. Thus, the infrared low-frequency sub-band image and the visible light low-frequency sub-band image of the high-voltage disconnector are fused together to obtain the low-frequency sub-band fused image.
[0059] S22. Global Image Fusion
[0060] The high-frequency subband fusion image and low-frequency subband fusion image of the high-voltage disconnector are reconstructed by using the non-subsampled shear wave inverse transform algorithm to obtain the final fused image of the infrared image and visible light image of the high-voltage disconnector, thus realizing global image fusion. The process of decomposing the infrared image and visible light image of the high-voltage disconnector into local image fusion and global image fusion is called gradient image fusion model.
[0061] The image fusion evaluation scheme of the present invention is as follows:
[0062] To evaluate the quality of gradient image fusion model algorithms compared with common image fusion methods such as color model-based algorithms, spatial transformation algorithms, weighted average methods, ratio transformation methods, wavelet transform methods, and principal component analysis methods in fusing infrared and visible light images of high-voltage disconnecting switches, the edge information preservation rate (Q) of the fused image was assessed. AB / F The image fusion model algorithm is evaluated using six dimensions: entropy (EN), mutual information (MI), spatial frequency (MSF), standard deviation (STD), and structural similarity (SSIM). These six quality evaluation dimensions are used to compare the advantages of the gradient image fusion model algorithm with six common image fusion methods. The six quality evaluation dimensions are represented as follows:
[0063] (1)Q AB / F This reflects the degree to which image details are preserved before and after fusion. A larger value indicates more complete preservation of edge information. It is represented as:
[0064]
[0065] In the formula, Q AF (i,j) represents the intensity at pixel (i,j), Q BF The direction value at pixel (i,j), ω A (i,j), ω B (i,j) represent the weights of intensity and direction, respectively;
[0066] (2) EN corresponds to the information richness of an image. The quality of an image is directly proportional to its information entropy value, expressed as:
[0067]
[0068] In the formula, L represents the gray level of the image, with a value of 255, and P... i This indicates the probability of the corresponding gray level appearing;
[0069] (3) MI reflects the amount of information before and after image fusion, and is expressed as:
[0070] MI = MI A,F +MI B,F(17)
[0071] In the formula MI A,F and MI B,F These represent the intersection of information content in the infrared image and the visible light image of the high-voltage disconnector, respectively. The larger the MI, the better the fusion between the two images.
[0072] (4) MSF reflects the spatial frequency information of an image. The higher the frequency, the more contour and edge information the image will have, as shown below:
[0073]
[0074] In the formula, RF and CF represent the row frequency and column frequency of the image, respectively;
[0075] (5) STD reflects the contrast of an image. The larger the standard deviation, the more detail information is retained in the image, and the higher the effect and quality of the fusion. It is expressed as:
[0076]
[0077] (6) SSIM reflects the similarity between the original images before and after fusion. The larger the value, the higher the similarity between the two, expressed as:
[0078]
[0079] In the formula, μ A μ B μ F These represent the mean values of the infrared image, the visible light image, and the fused image, respectively. Let σ represent the variances of these three factors respectively. AF σ BF These represent the joint variances of the infrared image, the visible light image, and the fused image, respectively.
[0080] The high-voltage disconnector status identification scheme of the present invention is as follows:
[0081] After segmenting the high-voltage disconnector image after global image fusion using the Qtsu threshold segmentation algorithm, the high-voltage disconnector is extracted from the background image. The segmented high-voltage disconnector image is then processed using pixel integral projection, performing horizontal and vertical integral projections. The ratio between the horizontal and vertical integral projections is calculated to determine the disconnector's state. When the disconnector is in the open state, the ratio of the vertical integral to the horizontal integral is 0.842; when closed, the ratio is 0.237. Based on the requirement that the switch angle of the high-voltage disconnector should not be less than 65° when opening or closing, the integral projection ratio of the fused image is 0.826 when the switch angle is 65°. Therefore, a pixel integral projection ratio greater than 0.826 indicates an open state, less than 0.237 indicates a closed state, and values between 0.237 and 0.826 indicate unsuccessful opening or closing.
[0082] The experimental simulation scheme of this invention is as follows:
[0083] One hundred infrared and visible light images of a high-voltage disconnector were acquired and identified to determine the accuracy of different image recognition methods. Different scenarios were set during image acquisition. For example, in cases where obstacles were present near the disconnector or visibility was low, the infrared image could capture an image of the disconnector, but the camera was obstructed and could not capture a complete image, or the captured image was blurry. Secondly, when the power was disconnected, no current flowed through the disconnector, and no heat was generated; therefore, the infrared image could not capture a clear image of the disconnector, while the camera could. The recognition accuracy of the high-voltage disconnector image fused using the gradient image fusion model algorithm proposed in this invention was compared with that of the high-voltage disconnector's infrared and visible light images, thus demonstrating the advantages of the proposed gradient image fusion model algorithm.
[0084] Because this invention uses a gradient image fusion algorithm to fuse infrared and visible light images of high-voltage disconnect switches, and uses a pixel integral projection algorithm to identify the state of high-voltage disconnect switches, the following beneficial effects can be achieved:
[0085] 1. By using an image registration scheme, the infrared and visible light images of the high-voltage disconnector are aligned in space. The mapping relationship between the infrared and visible light images of the high-voltage disconnector is found, so that the infrared and visible light images of the high-voltage disconnector can be well fused in space.
[0086] 2. A gradient image fusion scheme is used to decompose the registered infrared and visible light images of the high-voltage disconnector into high-frequency and low-frequency sub-band images. These sub-band images are then fused separately to achieve local image fusion. Finally, the fused infrared low-frequency and high-frequency sub-band images are fused to achieve global fusion. This local and global fusion of the infrared and visible light images forms a gradient image fusion model. A high-voltage disconnector status recognition scheme is then used to identify the fused image to determine whether the high-voltage disconnector is in an open or closed state. This allows the infrared and visible light images of the high-voltage disconnector to complement each other, enabling all-weather operation. Attached Figure Description
[0087] Figure 1 This is a flowchart illustrating the overall scheme of a high-voltage disconnector state recognition method based on gradient image fusion according to the present invention.
[0088] Figure 2 The image shown is a grayscale and filtered image of a high-voltage disconnector state recognition method based on gradient image fusion according to the present invention.
[0089] Figure 3 This is a flowchart of the image registration process for a high-voltage disconnector state recognition method based on gradient image fusion according to the present invention.
[0090] Figure 4 This is a gradient image fusion model diagram of a high-voltage disconnector state recognition method based on gradient image fusion according to the present invention;
[0091] Figure 5 This is a pixel integral projection calculation model diagram of a high-voltage disconnector state recognition method based on gradient image fusion according to the present invention;
[0092] Figure 6 This is an image fused from an infrared image and a visible light image of a high-voltage disconnector switch, as described in the present invention, which is a method for recognizing the state of a high-voltage disconnector switch based on gradient image fusion.
[0093] Figure 7 This is a comparison of the fused image effects under different image fusion methods for the high-voltage disconnector state recognition method based on gradient image fusion of the present invention;
[0094] Figure 8 This is a fusion image state recognition result diagram of a high-voltage disconnector state recognition method based on gradient image fusion according to the present invention. Detailed Implementation
[0095] The present invention will now be described in further detail with reference to the embodiments and the accompanying drawings.
[0096] Please see Figures 1 to 8 The present invention illustrates a high-voltage disconnector state recognition method based on gradient image fusion, comprising an image preprocessing scheme, an image registration scheme, a gradient image fusion scheme, a fused image evaluation scheme, a high-voltage disconnector state recognition scheme, and an experimental simulation scheme.
[0097] like Figure 1 As shown, the image preprocessing scheme converts the acquired infrared and visible light images of the high-voltage disconnector into grayscale images and performs filtering and noise reduction on both images to improve image quality and processing speed. The image registration scheme aligns the infrared and visible light images of the high-voltage disconnector in space, finding the mapping relationship between them to ensure good spatial fusion. The gradient image fusion scheme decomposes the registered infrared and visible light images of the high-voltage disconnector into high-frequency and low-frequency sub-band images, and fuses these sub-band images separately to achieve [fuse / merge]. Local image fusion is performed, followed by fusion of the separately fused low-frequency and high-frequency infrared sub-band images to achieve global fusion. A gradient image fusion model is formed by the local and global fusion of infrared and visible light images of the high-voltage disconnector. The fused image evaluation scheme is used to evaluate the quality of the gradient image fusion model after fusing the infrared and visible light images of the high-voltage disconnector. The high-voltage disconnector status recognition scheme is used to recognize the status of the fused infrared and visible light images of the high-voltage disconnector to determine whether the high-voltage disconnector is in the open or closed state. The experimental simulation scheme is used to compare the recognition accuracy of the fused high-voltage disconnector image with that of a single high-voltage disconnector visible light image or infrared image.
[0098] The image preprocessing scheme is as follows:
[0099] S1. Image Acquisition:
[0100] The high-voltage disconnect switch was set to open and closed states respectively, and images of the high-voltage disconnect switch under different states were collected at the same location using a visible light camera and an infrared thermal imager.
[0101] S2. Image grayscale conversion
[0102] like Figure 2As shown, since both the infrared and visible light images of the high-voltage disconnector are color images, the color information in these images has little impact on the recognition results. To reduce the computational load and improve efficiency, both images can be converted to grayscale. The grayscale conversion of the acquired infrared and visible light images is achieved using the following formula to improve the image processing speed:
[0103] Gray(i,j)=0.299×R(i,j)+0.587×G(i,j)+0.114×B(i,j) (1)
[0104] In the formula, Gray(i,j) represents the gray value of the pixel at coordinate (i,j), R(i,j) represents the red component of the image, G(i,j) represents the green component of the image, and B(i,j) represents the blue component of the image.
[0105] S3. Image Filtering and Denoising
[0106] like Figure 2 As shown, images are easily affected by external environmental factors and their own instability during acquisition, resulting in noise in the acquired images. This noise significantly impacts the accuracy of recognition. Two-dimensional median filtering is used to remove image noise. The pixel values in the two-dimensional sliding template are arranged from largest to smallest to obtain a two-dimensional data sequence. Let the filtered image be g(x,y), then:
[0107] g(x,y)=med{f(xk),yl},(k,l∈A) (2)
[0108] In the formula, f(xk,yl) represents the original image of the disconnecting switch, A represents the two-dimensional sliding template, and the template area is selected as a 3×3 template.
[0109] like Figure 3 As shown, the image registration scheme is as follows:
[0110] Because the target image cannot be aligned in space due to differences in shooting angle and object distance during the image acquisition process, the infrared and visible light images cannot find a mapping relationship during the fusion process, and cannot be fused well in space. Therefore, it is necessary to register the two different images.
[0111] The acquired infrared and visible light images were normalized using the maximum-minimum method:
[0112]
[0113] In the formula, norm is the normalized value, and x iLet x be the value of a pixel in the image, and max(x) and min(x) represent the maximum and minimum pixel values in the image, respectively.
[0114] The resolution of the infrared and visible light images of the high-voltage disconnector was adjusted to 2306*2658. The visible light image of the high-voltage disconnector was mapped onto its infrared image using an image feature transformation model. First, the Canny operator was used to perform edge detection on the infrared and visible light images to extract the image contours. Then, the SURF algorithm was used to detect and extract the corner points of the image contours. Finally, the Hessian matrix was used to extract the feature points of the image contours.
[0115]
[0116] In the formula, L xx (x,σ) represents the convolution of the second-order Gaussian derivative at scale σ with a point I(x,y) in the image, calculated as follows:
[0117]
[0118] G(x,y) represents the image; similarly, L can be calculated. xy (x,σ) and L yy (x,σ);
[0119] Next, the Euler distance formula is used to match the feature points of the image. Finally, the least squares method is used to fit these feature points, exchange parameter estimates, and obtain the optimal transformation model. Finally, the registered infrared and visible light images of the high-voltage disconnector are obtained.
[0120] like Figure 4 As shown, the gradient image fusion scheme is as follows:
[0121] The registered infrared and visible light images of the high-voltage disconnector are decomposed into high-frequency and low-frequency sub-band images using a non-subsampled shear wave transform algorithm. A pulse-coupled neural network algorithm is then used to fuse the infrared and visible high-frequency sub-band images. A visual saliency feature segmentation algorithm is employed to fuse the infrared and visible low-frequency sub-band images, achieving local image fusion. Finally, a non-subsampled shear wave inverse transform algorithm is used to fuse the locally fused images again, achieving global fusion. This yields the fused infrared and visible light images of the high-voltage disconnector, thus realizing a gradient image fusion model. The specific steps are as follows:
[0122] S1. Image Decomposition
[0123] The non-subsampled shear wave transform decomposition comprises two parts: multi-scale decomposition and multi-directional decomposition. At the multi-scale decomposition level, a non-subsampled pyramid filter is used to decompose the infrared and visible light images of the high-voltage disconnector into one low-frequency sub-band image and multiple high-frequency sub-band images, respectively. Then, a non-subsampled shear wave filter is used to decompose the decomposed high-frequency sub-band images in multiple directions. A Meyer window function is constructed, and a convolution algorithm is used to convolve the high-frequency sub-band images with the Meyer window function to obtain the high-frequency sub-band coefficients in different directions. This enables the decomposition of the infrared and visible light images of the high-voltage disconnector in different directions.
[0124] S2. Gradient Image Fusion Design
[0125] S21. Image Local Fusion
[0126] S211. High-frequency sub-band diagram fusion
[0127] Spatial frequency is used as the input to the pulse-coupled neural network (PLN) algorithm, and Laplacian energy is used as the connection strength of the PLA. The ignition frequency after n iterations of the PLA is calculated, and the high-frequency sub-band image fusion is achieved based on the ignition frequency. Let the spatial frequency of the image be MSF, then:
[0128]
[0129] In the formula, RF and CF represent the row frequency and column frequency of the image, respectively;
[0130] In a pulse-coupled neural network model, the degree of correlation between neurons is represented by connection strength, which is expressed as:
[0131]
[0132] LP(i,j)=|2C(i,j)-C(i-step,j)-C(i+step,j)|+|2C(i,j)-C(i,j-step)-C(i,j+step)| (8)
[0133] In equations (7) and (8), step is the distance between pixels, with a value of 1, C(i,j) represents the high-frequency subband coefficient, ω(i,j) is the weight of the high-frequency subband coefficient, and MSLR(i,j) is the sum of the Laplacian energy of the image at coordinates (i,j).
[0134] Ignition frequency is expressed as:
[0135]
[0136] In the formula, T R (n) and T V(n) represents the ignition frequency of the high-frequency sub-band image after non-subsampled shear wave transform decomposition of the infrared and visible light images of the high-voltage disconnector, respectively. Thus, the infrared high-frequency sub-band image and the visible light high-frequency sub-band image of the high-voltage disconnector are fused together to obtain the high-frequency sub-band fused image.
[0137] S212. Low-frequency subband map fusion
[0138] This invention uses a hypercomplex Fourier transform algorithm to obtain salient features (such as contours and background information) in low-frequency subband images, dividing the low-frequency subband image into two regions: salient and non-salient. Let the hypercomplex matrix be:
[0139] f(m,n)=ω1f1i+ω2f2j+ω3f3k (10)
[0140] In the formula, f(m,n) represents a hypercomplex matrix, and i, j, k satisfy i 2 =j 2 =k 2 =ijk=-1, ω1~ω3 represent the weights of the image's brightness, orientation, and texture information, respectively, with values of 0.5, 0.25, and 0.25. f1~f3 represent the feature matrices of these three dimensions. Performing a hypercomplex Fourier transform on f(m,n) yields:
[0141] F H [μ,ν]=||F H [μ,ν]||e μφ(μ,ν) (11)
[0142] In the formula, ||F H [μ,ν]|| denotes the first-order norm of f(m,n) in the frequency domain;
[0143] Applying a Gaussian kernel function to ||F H [μ,ν]|| is smoothed to obtain ||F H A multi-scale map of [μ,ν]|| is used, and then the salient image is obtained using the inverse hypercomplex Fourier transform. The salient and non-salient regions of the image are segmented in the image space based on the minimum entropy value. An image fusion method based on visual saliency segmentation is used to divide the image features into salient and non-salient regions for fusion. For salient regions, the probability of target appearance is directly proportional to the saliency value. Therefore, the similarity between two salient maps is compared to determine whether the images are similar, expressed as:
[0144]
[0145] SIM A,B (i,j) represents the significant similarity, SM A (i,j) and SM B(i,j) represent the significance values of the infrared and visible light images of the high-voltage disconnector, respectively. A,B The larger (i,j) is, the more similar the infrared and visible light images of the high-voltage disconnector are. The fused output is:
[0146]
[0147] In the formula, ω represents the weighting value of fusion, T represents the similarity threshold (ranging from 0.5 to 1), IR represents the infrared image, and VI represents the visible light image;
[0148] Non-salient regions are generally background images. The higher the information entropy of an image, the greater its information content. Therefore, comparing the magnitude of information entropy is used as a criterion for image fusion, and images with higher information entropy are fused. The criteria are as follows:
[0149]
[0150] In the formula, F(i,j) represents the fused image, LR is the information entropy ratio, and LR = LR IR / LR VI IR and VI represent infrared image and visible light image, respectively. When LR>1, the information entropy of the infrared image in this region is higher than that of the visible light image. Therefore, the infrared image is used as the fused image, and vice versa. Thus, the infrared low-frequency sub-band image and the visible light low-frequency sub-band image of the high-voltage disconnector are fused together to obtain the low-frequency sub-band fused image.
[0151] S22. Global Image Fusion
[0152] like Figure 6 As shown, the high-frequency subband fusion image and low-frequency subband fusion image of the high-voltage disconnector are reconstructed using the non-subsampled shear wave inverse transform algorithm to obtain the final fused image of the infrared image and visible light image of the high-voltage disconnector, thus achieving global image fusion. The process of decomposing the infrared image and visible light image of the high-voltage disconnector into local image fusion and global image fusion is called the gradient image fusion model.
[0153] The image fusion evaluation scheme of the present invention is as follows:
[0154] like Figure 7 As shown, to evaluate the quality of the gradient image fusion model algorithm compared with common image fusion methods such as color model-based algorithms, spatial transformation algorithms, weighted average methods, ratio transformation methods, wavelet transform methods, and principal component analysis methods in fusing infrared and visible light images of high-voltage disconnecting switches, the edge information preservation degree (Q) of the fused image is considered. AB / FThe image fusion model algorithm is evaluated using six dimensions: entropy (EN), mutual information (MI), spatial frequency (MSF), standard deviation (STD), and structural similarity (SSIM). These six quality evaluation dimensions are used to compare the advantages of the gradient image fusion model algorithm with six common image fusion methods. The six quality evaluation dimensions are represented as follows:
[0155] (1)Q AB / F This reflects the degree to which image details are preserved before and after fusion. A larger value indicates more complete preservation of edge information. It is represented as:
[0156]
[0157] In the formula, Q AF (i,j) represents the intensity at pixel (i,j), Q BF The direction value at pixel (i,j), ω A (i,j), ω B (i,j) represent the weights of intensity and direction, respectively;
[0158] (2) EN corresponds to the information richness of an image. The quality of an image is directly proportional to its information entropy value, expressed as:
[0159]
[0160] In the formula, L represents the gray level of the image, with a value of 255, and P... i This indicates the probability of the corresponding gray level appearing;
[0161] (3) MI reflects the amount of information before and after image fusion, and is expressed as:
[0162] MI = MI A,F +MI B,F (17)
[0163] In the formula MI A,F and MI B,F These represent the intersection of information content in the infrared image and the visible light image of the high-voltage disconnector, respectively. The larger the MI, the better the fusion between the two images.
[0164] (5) MSF reflects the spatial frequency information of an image. The higher the frequency, the more contour and edge information the image will have, as shown below:
[0165]
[0166] In the formula, RF and CF represent the row frequency and column frequency of the image, respectively;
[0167] (5) STD reflects the contrast of an image. The larger the standard deviation, the more detail information is retained in the image, and the higher the effect and quality of the fusion. It is expressed as:
[0168]
[0169] (6) SSIM reflects the similarity between the original images before and after fusion. The larger the value, the higher the similarity between the two, expressed as:
[0170]
[0171] In the formula, μ A μ B μ F These represent the mean values of the infrared image, the visible light image, and the fused image, respectively. Let σ represent the variances of these three factors respectively. AF σ BF These represent the joint variances of the infrared image, the visible light image, and the fused image, respectively.
[0172] like Figure 5 , Figure 8 As shown, the high-voltage disconnector status identification scheme is as follows:
[0173] After segmenting the high-voltage disconnector image after global image fusion using the Qtsu threshold segmentation algorithm, the high-voltage disconnector is extracted from the background image. The segmented high-voltage disconnector image is then processed using pixel integral projection, performing horizontal and vertical integral projections. The ratio between the horizontal and vertical integral projections is calculated to determine the disconnector's state. When the disconnector is in the open state, the ratio of the vertical integral to the horizontal integral is 0.842; when closed, the ratio is 0.237. Based on the requirement that the switch angle of the high-voltage disconnector should not be less than 65° when opening or closing, the integral projection ratio of the fused image is 0.826 when the switch angle is 65°. Therefore, a pixel integral projection ratio greater than 0.826 indicates an open state, less than 0.237 indicates a closed state, and values between 0.237 and 0.826 indicate unsuccessful opening or closing.
[0174] The experimental simulation scheme of this invention is as follows:
[0175] One hundred infrared and visible light images of a high-voltage disconnector were acquired and identified to determine the accuracy of different image recognition methods. Different scenarios were set during image acquisition. For example, in cases where obstacles were present near the disconnector or visibility was low, the infrared image could capture an image of the disconnector, but the camera was obstructed and could not capture a complete image, or the captured image was blurry. Secondly, when the power was disconnected, no current flowed through the disconnector, and no heat was generated; therefore, the infrared image could not capture a clear image of the disconnector, while the camera could. The recognition accuracy of the high-voltage disconnector image fused using the gradient image fusion model algorithm proposed in this invention was compared with that of the high-voltage disconnector's infrared and visible light images, thus demonstrating the advantages of the proposed gradient image fusion model algorithm.
[0176] The working principle and process of this invention are as follows:
[0177] First, infrared and visible light images of the high-voltage disconnector switch are acquired. Image preprocessing and registration are then performed on these images. A non-subsampled shear wave transform algorithm is used to decompose the registered infrared and visible light images into high-frequency and low-frequency sub-band images, respectively. A pulse-coupled neural network algorithm is then used to fuse the high-frequency infrared and visible light sub-band images. A visual saliency feature-based segmentation algorithm is used to fuse the low-frequency infrared and visible light sub-band images, achieving local image fusion. Finally, a non-subsampled shear wave inverse transform algorithm is used to fuse the locally fused images again, achieving global fusion. This yields the fused image of the high-voltage disconnector switch, thus realizing a gradient image fusion model. An image fusion quality evaluation scheme is established to compare the effectiveness of the gradient image fusion algorithm with common image fusion schemes. The fused image is then processed using a pixel integral projection algorithm to identify the opening and closing status of the high-voltage disconnector switch. Experimental simulations are used to verify the identification results of the high-voltage disconnector switch's opening and closing status.
Claims
1. A high-voltage disconnecting switch state recognition method based on gradient image fusion, characterized in that: The image preprocessing scheme, the image registration scheme, the gradient image fusion scheme, the fused image evaluation scheme, the high-voltage disconnecting switch state recognition scheme and the experimental simulation scheme are included. The maximum and minimum value method is used for normalizing the collected infrared and visible light images. where norm is the normalized value, x i is the value of the image pixel, max(x) and min(x) represent the maximum and minimum pixel values in the image, respectively. The resolution of the high-voltage disconnecting switch infrared image and the visible light image is adjusted to 2306*2658, the visible light image of the high-voltage disconnecting switch is mapped to the infrared image of the high-voltage disconnecting switch through the image feature transformation model, the Canny operator is used for edge detection of the high-voltage disconnecting switch infrared image and the visible light image, the image contour is extracted, the SURF algorithm is used for detecting and extracting the corner points of the image contour, and the Hess matrix is used for extracting the feature points of the image contour. where L xx (x,σ) denotes the convolution of the Gaussian second derivative at scale σ with a point I(x,y) in the image, and is calculated as G(x,y) represents the image, and L xy (x,σ) and L yy (x,σ); Then, the Euler distance formula is used for matching the feature points of the image, finally, the least square method is used for fitting the feature points, the parameter estimation is exchanged, and the optimal transformation model is obtained, and finally, the registered high-voltage disconnecting switch infrared image and the visible light image are obtained. The gradient image fusion scheme is used for decomposing the registered high-voltage disconnecting switch infrared image and the visible light image into high-frequency sub-band images and low-frequency sub-band images, respectively, and fusing the high-frequency sub-band images and the low-frequency sub-band images of the high-voltage disconnecting switch infrared image and the visible light image, realizing local image fusion, and then fusing the respectively fused infrared low-frequency sub-band images and high-frequency sub-band images, realizing global fusion, and finally, the gradient image fusion model is composed through local fusion and global fusion of the high-voltage disconnecting switch infrared image and the visible light image, the fused image evaluation scheme is used for evaluating the quality of the fused high-voltage disconnecting switch infrared image and the visible light image, the high-voltage disconnecting switch state recognition scheme is used for recognizing the fused image of the high-voltage disconnecting switch infrared image and the visible light image, so as to judge whether the high-voltage disconnecting switch is in the open state or the closed state, and the experimental simulation scheme is used for comparing the recognition accuracy of the fused high-voltage disconnecting switch image and the single high-voltage disconnecting switch visible light image or the high-voltage disconnecting switch infrared image.
2. The high-voltage disconnecting switch state recognition method based on gradient image fusion according to claim 1, characterized in that: The gradient image fusion scheme is: The registered infrared image and visible light image of the high-voltage disconnecting switch are decomposed into high-frequency sub-band images and low-frequency sub-band images by using the non-subsampled shearlet transform algorithm, the infrared high-frequency sub-band image and the visible light high-frequency sub-band image of the high-voltage disconnecting switch are fused together by using the pulse coupled neural network algorithm, the infrared low-frequency sub-band image and the visible light low-frequency sub-band image of the high-voltage disconnecting switch are fused together by using the visual saliency segmentation algorithm, local image fusion is realized, and then the local fused images are fused again by using the non-subsampled shearlet inverse transform algorithm, global fusion is realized, the fused image of the infrared image and the visible light image of the high-voltage disconnecting switch is obtained, and thus the gradient image fusion model is realized, and the specific steps are as follows: S1. Image decomposition The non-subsampled shearlet transform decomposition includes multi-scale decomposition and multi-direction decomposition, in the aspect of multi-scale decomposition, the infrared image and the visible light image of the high-voltage disconnecting switch are decomposed into one low-frequency sub-band image and multiple high-frequency sub-band images by using a non-subsampled pyramid filter, then the decomposed high-frequency sub-band images are decomposed in multiple directions by using a non-subsampled shearlet filter, a Meyer window function is constructed, the high-frequency sub-band images are convolved with the Meyer window function through a convolution algorithm, high-frequency sub-band coefficients in different directions are obtained, and thus the decomposition of the infrared image and the visible light image of the high-voltage disconnecting switch in different directions is realized; S2. Gradient image fusion design S21. Local image fusion S211. High-frequency sub-band image fusion The spatial frequency is used as the input of the pulse coupled neural network algorithm, and the Laplace energy is used as the connection strength of the pulse coupled neural network; The firing rate of the pulse coupled neural network after n iterations is calculated, the fusion of the high-frequency sub-band image is realized according to the firing rate, the spatial frequency of the image is MSF, and there is: In the formula, RF and CF represent the row frequency and the column frequency of the image respectively; In the pulse coupled neural network model, the correlation between neurons is represented by the connection strength, and the connection strength is represented as: LP(i,j) = |2C(i,j)-C(i-step,j)-C(i+step,j)|+|2C(i,j)-C(i,j-step)-C(i,j+step)| (8) In formula (7) and (8), step is the distance between pixels and pixels, and the value is 1, C(i,j) represents the high-frequency sub-band coefficient, ω(i,j) is the weight value of the high-frequency sub-band coefficient, and MSLR(i,j) is the Laplace energy sum of the image at the coordinates (i,j); The firing rate is represented as: In the formula, T R (n) and T V (n) are the ignition frequencies of the high-frequency sub-band images of the infrared image and the visible light image of the high-voltage disconnector after non-subsampled shearlet transform decomposition, so as to fuse the infrared high-frequency sub-band image and the visible light high-frequency sub-band image of the high-voltage disconnector to obtain a high-frequency sub-band fusion image. S212. Low-frequency sub-band image fusion In the formula, ω represents the weighted value of fusion, T represents the similarity threshold (the value range is 0.5-1), IR represents the infrared image, and VI represents the visible light image. In the formula, f(m, n) represents a hypercomplex matrix, i, j, k satisfy i 2 = j 2 = k 2 = ijk = -1, ω1 ~ ω3 respectively represent weight values of brightness, azimuth, and texture information of an image, and take values of 0.5, 0.25, and 0.25 respectively, f1 ~ f3 respectively represent feature matrices of the three dimensions, and the hypercomplex Fourier transform of f(m, n) is obtained as follows: F H [μ,ν] = ||F H [μ,ν]||e μφ(μ,ν) (11) where ||F H [μ,ν]|| denotes the first order norm of the frequency domain of f(m,n); The Gaussian kernel function is used to smooth ||F H [μ,ν]||, and then a multi-scale map of ||F H [μ,ν]|| is obtained, and an inverse hypercomplex Fourier transform is used to obtain a saliency image. The salient region and the non-salient region of the image are segmented in the image space according to the minimum entropy value of the image. An image fusion method based on visual saliency segmentation is used to divide the image features into the salient region and the non-salient region for fusion. Whether the images are similar is determined by comparing the similarity of two saliency maps, which is represented as: SIM A,B (i,j) represents the saliency similarity, SM A (i,j) and SM B (i,j) respectively represent the saliency values of the infrared image and the visible light image of the high-voltage disconnector, SIM A,B The greater the SIM(i,j) is, the more similar the infrared image and the visible light image of the high-voltage disconnector are, and the fusion output is: The judgment basis of image fusion is the size of information entropy, and the image with larger information entropy is fused, and the judgment basis is as follows: where F(i,j) represents the fused image, LR is the information entropy ratio, LR = LR IR / LR VI , IR and VI represent the infrared image and the visible light image respectively, when LR > 1, the information entropy of the infrared image is higher than that of the visible light image in the region, so the infrared image is taken as the fused image, and vice versa, the visible light image is taken as the fused image, thereby the infrared low-frequency sub-band image and the visible light low-frequency sub-band image of the high-voltage disconnecting switch are fused to obtain a low-frequency sub-band fused image; S22. image global fusion The high-frequency sub-band fusion image and the low-frequency sub-band fusion image of the high-voltage disconnecting switch are reconstructed by using the non-sub-sampling shear wave inverse transform algorithm to obtain the final fusion image of the infrared image and the visible light image of the high-voltage disconnecting switch, and the global image fusion is realized.
3. The high-voltage disconnecting switch state recognition method based on gradient image fusion according to claim 1, characterized in that: The high-voltage disconnecting switch state recognition scheme is: After the high-voltage disconnecting switch image fused by the global image fusion is segmented by the Qtsu threshold segmentation algorithm, the high-voltage disconnecting switch is extracted from the background image, the high-voltage disconnecting switch image segmented by the target is processed by horizontal integral projection and vertical integral projection through the pixel integral projection method, and the ratio between the horizontal integral projection and the vertical integral projection is calculated to judge the state of the disconnecting switch. When the disconnecting switch is in the open state, the ratio of the integral in the vertical direction to the integral in the horizontal direction is 0.842; when the disconnecting switch is in the closed state, the ratio of the two is 0.
237. According to the requirement that the angle of the disconnecting switch is not less than 65° when the high-voltage disconnecting switch is in the open or closed state, when the angle of the disconnecting switch is 65°, the integral projection ratio of the fusion image is 0.826, so the pixel integral projection ratio greater than 0.826 is set as the open state, the pixel integral projection ratio less than 0.237 is set as the closed state, and the pixel integral projection ratio between 0.237 and 0.826 indicates that the open or closed operation is unsuccessful.