Insulator image enhancement method based on unmanned aerial vehicle infrared image denoising

Abstract

The present application relates to the technical field of image processing, in particular to an insulator image enhancement method based on infrared image denoising of a UAV. The original infrared image collected by the UAV is preprocessed, and the preprocessed image is decomposed in multiple scales by wavelet transform to obtain high-frequency detail coefficients and low-frequency approximation coefficients; the high-frequency detail coefficients are processed by adaptive thresholding, while the low-frequency approximation coefficients are subjected to Wiener filtering, and after wavelet reconstruction and guided filtering, a denoised image is obtained; based on the denoised image, an edge detection operator is used for preliminary edge extraction, edge information is obtained by combining non-maximum suppression and double thresholding, and the edge gap is connected by using morphological edge reconstruction technology; through adaptive contrast enhancement based on local temperature distribution and image fusion processing, the final image with enhanced features is generated. The present application can realize efficient denoising and accurate feature enhancement of the insulator infrared image collected by the UAV.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to an insulator image enhancement method based on unmanned aerial vehicle infrared image denoising. BACKGROUND

[0002] Insulator inspection in power systems is very important for the safe operation of power systems, and the traditional manual inspection method is low in efficiency and greatly affected by the environment. Therefore, unmanned aerial vehicle infrared inspection is widely used due to its flexibility and wide range, but the existing unmanned aerial vehicle infrared inspection method still has problems: the unmanned aerial vehicle infrared image is greatly disturbed by flight jitter and environmental interference, the image contrast is low, the features are not prominent, and the image quality is unstable under different scene parameters, which affects the identification and fault diagnosis of insulators. SUMMARY

[0003] The present application provides an insulator image enhancement method based on unmanned aerial vehicle infrared image denoising to solve the problems of low image contrast, non-prominent features, and unstable image quality under different scene parameters in the prior art.

[0004] The technical solution of the present application to solve the above technical problems is as follows: in a first aspect, the present application provides an insulator image enhancement method based on unmanned aerial vehicle infrared image denoising, comprising: preprocessing the original infrared image collected by the unmanned aerial vehicle, and using wavelet transform to perform multi-scale decomposition on the preprocessed image to obtain high-frequency detail coefficients and low-frequency approximation coefficients; the high-frequency detail coefficients are processed using an adaptive threshold, while the low-frequency approximation coefficients are subjected to Wiener filtering, and after wavelet reconstruction and guided filtering processing, a denoising image is obtained; based on the denoising image, an edge detection operator is used for preliminary edge extraction, edge information is obtained by combining non-maximum suppression and double threshold processing, and morphological edge reconstruction technology is used to connect edge gaps; through adaptive contrast enhancement based on local temperature distribution and image fusion processing, a final image with enhanced features is generated.

[0005] Optionally, the original infrared image collected by the unmanned aerial vehicle is preprocessed, including: obtaining the original infrared image data collected by the infrared thermal imager carried by the unmanned aerial vehicle; adjusting the size of the original infrared image to a fixed resolution of 640x480, while maintaining the image proportion using a bilinear interpolation algorithm; converting the size-adjusted image from the RGB color space to the HSV color space, and extracting the V component in the HSV color space as the luminance component.

[0006] Optionally, the preprocessed image is subjected to multi-scale decomposition using wavelet transform to obtain high-frequency detail coefficients and low-frequency approximation coefficients, including: selecting a Daubechies db4 wavelet basis function as a transform basis, and performing a three-layer two-dimensional discrete wavelet transform on the preprocessed luminance component image; in the first layer decomposition, the input image is decomposed into a low-frequency approximation subband LL1 and three high-frequency detail subbands, wherein the three high-frequency detail subbands include a horizontal direction LH1, a vertical direction HL1 and a diagonal direction HH1; in the second layer decomposition, the LL1 subband is further decomposed to obtain a low-frequency approximation subband LL2 and corresponding three high-frequency detail subbands LH2, HL2 and HH2; in the third layer decomposition, the LL2 subband is decomposed to obtain a final low-frequency approximation subband LL3 and three high-frequency detail subbands LH3, HL3 and HH3; the LL3 is defined as the low-frequency approximation coefficient C LL3 , and the LH1, HL1, HH1, LH2, HL2, HH2, LH3, HL3 and HH3 are collectively defined as the high-frequency detail coefficient.

[0007] Optionally, the high-frequency detail coefficients are processed using an adaptive threshold, and the low-frequency approximation coefficients are subjected to Wiener filtering, including: for each layer of high-frequency detail coefficient subbands, a standard deviation σ is calculated, and an adaptive threshold T is determined based on a Bayesian estimation principle, wherein T = σ (2lnMxN), and M and N are the sizes of the high-frequency detail coefficient subbands; a first soft threshold function is applied to process the high-frequency detail coefficients of the horizontal direction LH and the vertical direction HL, and coefficients with absolute values greater than the adaptive threshold T are retained; a second soft threshold function is applied to process the high-frequency detail coefficients of the diagonal direction HH, wherein the threshold of the second soft threshold function is adjusted to 0.7T; for the low-frequency approximation coefficient C LL3 , the mean and variance in the corresponding local neighborhood are calculated, and a Wiener filter is constructed based on the mean and variance; the low-frequency approximation coefficient C LL3 is subjected to filtering processing using the Wiener filter to obtain a denoised low-frequency approximation coefficient.

[0008] Optionally, for the low-frequency approximation coefficient C LL3 , the mean and variance in the corresponding local neighborhood are calculated, and a Wiener filter is constructed based on the mean and variance, including: setting a 7x7 pixel sliding window as a local neighborhood; in the local neighborhood, the local mean μ and the local variance σ LL3 of the low-frequency approximation coefficient C local are calculated, and the overall noise variance σ 2 of the low-frequency approximation coefficient C LL3 is obtained. noise 2 ;

[0009] According to the local mean μ, the local variance σ local 2 and the overall noise variance σ noise 2 , a transfer function of a Wiener filter is constructed:

[0010]

[0011] wherein (x, y) represents the position coordinates of the current coefficient in the low-frequency approximation coefficient C LL3 ;

[0012] The low-frequency approximation coefficient C LL3 is filtered by using the Wiener filter to obtain a denoised low-frequency approximation coefficient.

[0013] Optionally, after wavelet reconstruction and guided filtering processing, a denoised image is obtained, including: using the denoised low-frequency approximation coefficient and the processed high-frequency detail coefficient of each layer, wavelet reconstruction is performed by using an inverse wavelet transform algorithm to generate a preliminary denoised image; the preliminary denoised image is taken as a guide image, the luminance component of the original infrared image is taken as an input image, and guided filtering processing is performed by using a guided filter to eliminate the pseudo Gibbs phenomenon in the preliminary denoised image to obtain a final denoised image.

[0014] Optionally, the guided filtering processing by taking the preliminary denoised image as a guide image and the luminance component of the original infrared image as an input image includes: setting the preliminary denoised image as a guide image, setting the luminance component of the original infrared image as an input image, and setting the filter window radius and the regularization parameter of the guided filter; a local linear model is established on the guide image to determine the linear relationship between the output image and the guide image; the coefficient parameters in the linear relationship are solved by minimizing the cost function; the pixel value of the output image is calculated according to the solved coefficient parameters to obtain a denoised image in which the pseudo Gibbs phenomenon is eliminated.

[0015] Optionally, based on the denoised image, preliminary edge extraction is performed by using an edge detection operator, edge information is obtained by combining non-maximum suppression and double-threshold processing, and edge gaps are connected by using a morphological edge reconstruction technique, including: the denoised image is convolved and calculated by using a temperature gradient weighted edge detection operator to obtain a preliminary edge response map; the non-maximum suppression processing is performed on the preliminary edge response map to thin the edge response; the double-threshold processing is performed on the non-maximum suppressed edge response to divide the edge pixels into strong edge pixels, weak edge pixels and non-edge pixels; the 3*3 circular structural element is used to perform morphological closing operation on the strong edge pixels and the weak edge pixels to connect the edge gaps and form a complete edge contour.

[0016] The non-maximum suppression processing is performed on the preliminary edge response map to thin the edge response, including: calculating the gradient amplitude and the gradient direction of each pixel point in the preliminary edge response map; quantizing the gradient direction of each pixel point into four main directions of horizontal, vertical, positive diagonal line and negative diagonal line according to the gradient direction; comparing the gradient amplitude of each pixel point with the gradient amplitudes of two adjacent pixels in the gradient direction; when the gradient amplitude of the pixel point is greater than or equal to the gradient amplitudes of the two adjacent pixels, the edge response of the current pixel point is retained; when the gradient amplitude of the pixel point is less than the gradient amplitude of any adjacent pixel, the edge response of the current pixel point is suppressed.

[0017] Optionally, the final image with enhanced features is generated through adaptive contrast enhancement based on local temperature distribution and image fusion processing, including: dividing the edge-thinned denoising image into local processing blocks of 8x8 pixels, and dynamically setting a contrast enhancement coefficient based on the local temperature distribution features of each processing block; performing contrast enhancement processing on each processing block based on the contrast enhancement coefficient, to obtain a contrast-enhanced image, by combining a limited-contrast adaptive histogram equalization algorithm; fusing the edge-thinned denoising image and the contrast-enhanced image according to a preset fusion ratio to obtain a preliminary fusion image; and performing multi-scale fusion on the preliminary fusion image through a Laplacian pyramid fusion algorithm to generate the final image with enhanced features.

[0018] By implementing the present application, the original infrared image collected by the unmanned aerial vehicle can be preprocessed, and the preprocessed image can be multi-scale decomposed by using wavelet transform to obtain high-frequency detail coefficients and low-frequency approximation coefficients, and the size standardization eliminates the problem of inconsistent image sizes under different flight parameters, providing standard input for subsequent unified processing, and the bilinear interpolation can reduce image distortion when adjusting the size and ensure the integrity of the basic structure of the image.

[0019] By implementing the present application, the high-frequency detail coefficients can be processed by using adaptive thresholding, and the low-frequency approximation coefficients can be processed by using Wiener filtering, and after wavelet reconstruction and guided filtering processing, a denoising image is obtained, and the Wiener filtering combines local image information and overall noise characteristics, and the filtering of the low-frequency approximation coefficients can effectively suppress low-frequency noise and well preserve the overall temperature distribution and main structure of the image, thereby ensuring the overall quality of the image.

[0020] By implementing the present application, based on the denoised image, preliminary edge extraction is carried out by using an edge detection operator, edge information is obtained by combining non-maximum suppression and double threshold processing, and morphological edge reconstruction technology is used to connect edge gaps, non-maximum suppression processing effectively refines the edge, removes redundant pixels around the edge, makes the edge clearer and more coherent, highlights the contour features of the insulator, and provides clear edge basis for positioning and identification of the insulator in subsequent fault diagnosis.

[0021] By implementing the present application, through adaptive contrast enhancement based on local temperature distribution and image fusion processing, a final image with enhanced features is generated, the adaptive contrast histogram equalization algorithm effectively prevents the local overexposure phenomenon that may occur in the histogram equalization process, and ensures the gray level and detail richness of the enhanced image.

[0022] In summary, by implementing the present application, efficient denoising and accurate feature enhancement of the insulator infrared image collected by the unmanned aerial vehicle can be realized. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 The flowchart of the insulator image enhancement method based on unmanned aerial vehicle infrared image denoising provided by the present application is shown. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0025] In the description of the present application, the terms "first" and "second" are used only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined with "first" and "second" can explicitly or implicitly include one or more features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0026] In the description of the present application, the term "for example" is used to mean "serving as an example, instance, or illustration." Any embodiment described as "for example" in the present application is not necessarily to be construed as preferred or advantageous over other embodiments. The following description is presented to enable any person skilled in the art to make and use the application. In the following description, for the purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the present application. It will be apparent, however, to one skilled in the art that the present application can be practiced without using these specific details. In other instances, well-known structures and processes are not elaborated upon in order to avoid unnecessary detail, which can obscure the description of the present application. Thus, the present application is not intended to be limited by the embodiments shown, but is to be accorded the widest scope consistent with the principles and features disclosed herein.

[0027] Embodiment one, as shown in the present application, provides an insulator image enhancement method based on unmanned aerial vehicle infrared image denoising, comprising: Figure 1

[0028] S100: preprocessing the original infrared image collected by the unmanned aerial vehicle, and using wavelet transform to perform multi-scale decomposition on the preprocessed image to obtain high-frequency detail coefficients and low-frequency approximation coefficients;

[0029] S200: using adaptive threshold to process the high-frequency detail coefficients, while implementing Wiener filtering on the low-frequency approximation coefficients, and through wavelet reconstruction and guided filtering processing, obtaining a denoising image;

[0030] S300: based on the denoising image, using an edge detection operator to perform preliminary edge extraction, combining non-maximum suppression and double threshold processing to obtain edge information, and using morphological edge reconstruction technology to connect edge gaps;

[0031] S400: through adaptive contrast enhancement and image fusion processing based on local temperature distribution, generating a final image with enhanced features.

[0032] In step 100 of the present application, the original infrared image collected by the unmanned aerial vehicle is preprocessed, comprising:

[0033] Obtaining the original infrared image data collected by the infrared thermal imager carried by the unmanned aerial vehicle;

[0034] Standardizing the size of the original infrared image to a fixed resolution of 640x480, while using a bilinear interpolation algorithm to maintain the image proportion;

[0035] Converting the image after size adjustment from RGB color space to HSV color space, and extracting the V component in the HSV color space as the luminance component.

[0036] ​In the embodiments of the present application, step 100 is the basic link of the infrared image enhancement of insulators, and the core objective is to unify the image specifications, strip invalid information, and focus on key features, thereby providing standardized and high-quality input data for subsequent core processing such as wavelet denoising and edge enhancement.

[0037] To achieve the above-mentioned object, first, the original infrared image data collected by the infrared thermal imager carried by the unmanned aerial vehicle needs to be obtained. Here, the original infrared data of insulators collected directly by the unmanned aerial vehicle infrared thermal imager is obtained, avoiding the influence of subsequent processing due to data transmission loss, format conversion distortion and other problems, thereby providing real and complete initial materials for the entire enhancement process.

[0038] Specifically, the original infrared image data can be collected by a professional inspection unmanned aerial vehicle carrying a high-resolution infrared thermal imager, ensuring that the infrared thermal imager meets at least 30Hz thermal imaging frame rate and ±2℃ temperature accuracy. When collecting, a stable environment period is selected, for example, 2 hours after sunrise / 2 hours before sunset, an environmental temperature difference of ≤3℃, and a wind speed of <5m / s. The corresponding metadata of the image, such as GPS coordinates, environmental temperature and humidity, etc., are synchronously collected. At least 10 effective thermal images are obtained for each base tower, ensuring that the key parts such as the entire string of insulators and the connecting hardware are covered, thereby providing sufficient effective samples for subsequent processing.

[0039] Then, the size of the original infrared image needs to be standardized and adjusted to a fixed resolution of 640×480, while maintaining the image proportion by using a bilinear interpolation algorithm.

[0040] Due to the different flight heights and camera parameter settings when the unmanned aerial vehicle is shooting, the original image resolution may be inconsistent, such as 640×480, 1280×960, etc. Here, the size is unified to 640×480, which can ensure the universality of the algorithm parameters such as wavelet decomposition and filtering window in subsequent processing, thereby avoiding unstable processing results due to differences in image size; at the same time, the image proportion is maintained by bilinear interpolation to prevent stretching and deformation of the insulator outline, thereby ensuring the accuracy of the structural features. The resolution unification can use image processing tools such as OpenCV and MATLAB to call the size adjustment interface to forcibly set all original infrared images to a fixed resolution of 640 pixels×480 pixels. The bilinear interpolation algorithm is a prior art, which will not be described here.

[0041] Next, the image after size adjustment needs to be converted from the RGB color space to the HSV color space, and the V component in the HSV color space is extracted as the luminance component.

[0042] The core value of the infrared image is to reflect the temperature distribution of the insulator, and the temperature difference corresponds to a fault hidden danger. In the RGB color space, the red, green and blue three channels mix redundant information such as brightness and hue, and cannot be directly associated with temperature characteristics. After the image is converted to the HSV color space, the V component, i.e. the brightness component, is directly linked to the infrared thermal radiation intensity, and can accurately map the temperature distribution of the insulator. The higher the temperature, the larger the V component value. Extracting the V component can strip off invalid information such as H (hue) and S (saturation), reduce the data volume, and let the subsequent processing only target the key features related to temperature, thereby improving the processing efficiency and accuracy. The specific method for extracting the V component value is known to those skilled in the art, and will not be described here.

[0043] In step S100 of the embodiment of the present application, the preprocessed image is subjected to multi-scale decomposition by wavelet transform to obtain high-frequency detail coefficients and low-frequency approximation coefficients, including:

[0044] A Daubechies db4 wavelet basis function is selected as the transform basis, and a three-layer two-dimensional discrete wavelet transform is performed on the preprocessed luminance component image;

[0045] In the first layer decomposition, the input image is decomposed into a low-frequency approximation subband LL1 and three high-frequency detail subbands, wherein the three high-frequency detail subbands include a horizontal direction LH1, a vertical direction HL1 and a diagonal direction HH1;

[0046] In the second layer decomposition, the LL1 subband is further decomposed to obtain a low-frequency approximation subband LL2 and corresponding three high-frequency detail subbands LH2, HL2 and HH2;

[0047] In the third layer decomposition, the LL2 subband is decomposed to obtain a final low-frequency approximation subband LL3 and three high-frequency detail subbands LH3, HL3 and HH3;

[0048] LL3 is defined as the low-frequency approximation coefficient C LL3 LH1, HL1, HH1, LH2, HL2, HH2, LH3, HL3 and HH3 are collectively defined as the high-frequency detail coefficient.

[0049] In the embodiment of the present application, the purpose of the above steps is to perform multi-scale decomposition on the preprocessed image by wavelet transform, separate the low-frequency basic information and high-frequency detail information of the image, and lay a foundation for subsequent targeted denoising and enhancement.

[0050] To achieve the above purpose, first need to select Daubechiesdb4 wavelet basis function as the transform base, and the preprocessed luminance component image is executed three layer two-dimensional discrete wavelet transform. Because db4 wavelet base has 4 order vanishing moment, the low frequency information fitting of smooth area such as normal surface of insulator is good, and the sharpness retention ability of high frequency area such as insulator defect edge is strong, so it is suitable for the feature distribution of smooth background + sudden defect in infrared image. Specifically, the decomposition filter of db4 wavelet base can be loaded through the wfilters function of MATLAB or the cv:dwt interface of OpenCV.

[0051] Then, three layer two-dimensional discrete wavelet transform needs to be executed, that is, the image is filtered along the row direction with h0(low pass) and h1(high pass) first, and then repeated filtering along the column direction, to obtain low frequency approximation subband LL and three high frequency subbands, including horizontal LH, vertical HL and diagonal HH high frequency subbands.

[0052] The specific process is as follows:

[0053] The first layer: the preprocessed V component image with a resolution of 640x480 is directly executed two-dimensional DWT to obtain one low frequency approximation subband LL1 with a resolution of 320x240, and three high frequency subbands with a resolution of 320x240, wherein LH1 corresponds to horizontal direction details such as umbrella skirt horizontal edge, HL1 corresponds to vertical direction details such as insulator string vertical arrangement, and HH1 corresponds to diagonal direction details such as corner defect.

[0054] The second layer: two-dimensional DWT is executed again with LL1 as input to obtain low frequency approximation subband LL2 with a resolution of 160x120, and three high frequency subbands with a resolution of 160x120, focusing on more coarse scale basic features.

[0055] The third layer: two-dimensional DWT is executed with LL2 as input to obtain final low frequency approximation subband LL3 with a resolution of 80x60, and three high frequency subbands with a resolution of 80x60, at this time LL3 has contained the most core overall trend information of the image.

[0056] Finally, the LL3 subband data obtained by the third layer decomposition is directly defined as the low frequency approximation coefficient C LL3 , which reflects the overall temperature distribution and contour structure of the insulator.

[0057] The data of all high frequency subbands, i.e. LH1, HL1, HH1, LH2, HL2, HH2, LH3, HL3, HH3, are collectively used as high frequency detail coefficients, which cover edge, noise and other detail information from fine to coarse, and provide clear processing objects for subsequent steps.

[0058] In step S200 of the embodiment, the high-frequency detail coefficients are processed by using the adaptive threshold, and the low-frequency approximation coefficients are processed by using the Wiener filter, including:

[0059] For each layer of high-frequency detail coefficient subband, the standard deviation σ is calculated, and the adaptive threshold T is determined based on the Bayesian estimation principle, wherein, M and N are the sizes of the high-frequency detail coefficient subband, respectively.

[0060] The high-frequency detail coefficients in the horizontal direction LH and the vertical direction HL are processed by using the first soft threshold function, and the coefficients with the absolute value greater than the adaptive threshold T are reserved.

[0061] The high-frequency detail coefficients in the diagonal direction HH are processed by using the second soft threshold function, wherein the threshold of the second soft threshold function is adjusted to 0.7T.

[0062] For the low-frequency approximation coefficient C LL3 , the mean and variance in the corresponding local neighborhood are calculated, and the Wiener filter is constructed based on the mean and variance.

[0063] The low-frequency approximation coefficient C LL3 is filtered by using the Wiener filter to obtain the denoised low-frequency approximation coefficient.

[0064] In the embodiment, the purpose of step S200 is to balance the noise removal and effective feature reservation by using the adaptive threshold denoising of the high-frequency detail coefficients and the Wiener filtering of the low-frequency approximation coefficients, so as to provide high-quality image data for subsequent insulator defect identification.

[0065] To achieve the above purpose, first, for each layer of high-frequency detail coefficient subband, the standard deviation σ is calculated, and the adaptive threshold T is determined based on the Bayesian estimation principle, wherein, M and N are the sizes of the high-frequency detail coefficient subband, respectively.

[0066] Specifically, for each layer of LH, HL, and HH high-frequency subband, the standard deviation σ of the pixel value is calculated, and then the adaptive threshold T is determined based on the Bayesian estimation principle according to the above formula. The threshold can adapt to the noise intensity of different subbands and ensure the pertinence of denoising.

[0067] Then, the high-frequency detail coefficients in the horizontal direction LH and the vertical direction HL are processed by using the first soft threshold function, and the coefficients with the absolute value greater than the adaptive threshold T are reserved.

[0068] That is, for the high frequency detail coefficients in the horizontal direction LH and the vertical direction HL, a soft threshold function rule is followed, if the absolute value of the high frequency detail coefficient |x|>T, the high frequency detail coefficient is reserved, and the result is sign(x)×(|x|-T); if |x|≤T, the high frequency detail coefficient is suppressed to 0. In this way, the noise in LH and HL is removed, while the useful details such as the edge of the insulator and the local temperature anomaly are retained.

[0069] Then, the high frequency detail coefficients in the diagonal direction HH need to be processed by applying a second soft threshold function, wherein the threshold of the second soft threshold function is adjusted to 0.7T.

[0070] That is, for the high frequency detail coefficients in the diagonal direction HH, the adjusted threshold is 0.7T. If the absolute value of the high frequency detail coefficient |x|>0.7T, the high frequency detail coefficient is reserved, and the result is sign(x)×(|x|-0.7T); if |x|≤0.7T, the high frequency detail coefficient is suppressed to 0. The diagonal direction high frequency noise is removed, while the local details in the diagonal direction of the insulator are retained.

[0071] Further, for the low frequency approximation coefficient C LL3 , the mean and variance in the corresponding local neighborhood are calculated, and a Wiener filter is constructed based on the mean and variance.

[0072] In step S200 of the embodiment, for the low frequency approximation coefficient C LL3 , the mean and variance in the corresponding local neighborhood are calculated, and a Wiener filter is constructed based on the mean and variance, including:

[0073] A sliding window of 7×7 pixels is set as the local neighborhood;

[0074] In the local neighborhood, the local mean μ and the local variance σ LL3 of the low frequency approximation coefficient C local 2 are calculated, and the overall noise variance σ LL3 of the low frequency approximation coefficient C noise 2 is obtained.

[0075] According to the local mean μ, the local variance σ local 2 and the overall noise variance σ noise 2 , a transfer function of a Wiener filter is constructed:

[0076]

[0077] Wherein, (x, y) represents the current coefficient in the low frequency approximation coefficient C LL3The position coordinates in the image correspond to the size (640*480) of the original image after preprocessing.

[0078] applying the Wiener filter to the low-frequency approximation coefficient C LL3 to obtain the low-frequency approximation coefficient after denoising.

[0079] In the embodiment of the present application, this step is aimed at the Wiener filtering of the low-frequency approximation coefficient C LL3 , and the core target is to remove the residual noise in the low-frequency region while retaining the overall temperature distribution trend of the insulator, so as to provide a smooth and accurate basic feature for subsequent image reconstruction and defect identification.

[0080] To achieve the above-mentioned purpose, a 7*7 pixel sliding window is first set as a local neighborhood. The 7*7 window size takes into account both local detail capture and calculation efficiency. A too small window is easily disturbed by single-point noise, and a too large window may blur the local structure of the insulator shed. Specifically, the sliding window can be defined by an image processing tool, and the window traverses the low-frequency approximation coefficient C LL3 in a pixel-by-pixel step.

[0081] Then, the local mean μ and the local variance σ of the low-frequency approximation coefficient C LL3 in the local neighborhood are calculated. local 2 and the overall noise variance σ of the low-frequency approximation coefficient C LL3 is obtained. noise 2 The local mean μ is the arithmetic mean of all low-frequency coefficient values in the current 7*7 window. The local variance σ local 2 is the deviation of the coefficients in the window from the mean value. The overall noise variance σ local 2 is then calculated by the median absolute deviation method, that is, the deviation of all coefficients in LL3 from the median value is calculated, the median is taken and multiplied by 1.4826 to obtain the noise standard deviation, and then squared to obtain the variance, which represents the overall noise level of the image.

[0082] Then, the above values are substituted into the transfer function of the Wiener filter to construct the filter. When the local variance is much larger than the noise variance, the filtered value is close to the original value, retaining the details. When the local variance is close to the noise variance, the filtered value is close to the local mean, achieving noise smoothing.

[0083] Further, the Wiener filter is applied to the low-frequency approximation coefficient C LL3 for filtering processing to obtain the low-frequency approximation coefficient after denoising.

[0084] That is, for each pixel (x, y) of LL3, substitute μ and σ of its corresponding window local and the overall σ noise , calculate the filtered value through the transfer function, generate the low-frequency approximation coefficient C after denoising LL3 '. C LL3 ' removes the local fluctuations caused by low-frequency noise and retains the overall temperature distribution and contour features of the insulator, providing a high-quality low-frequency basis for subsequent wavelet inverse transform.

[0085] In step S200 of the embodiment of the present application, after wavelet reconstruction and guided filtering processing, a denoised image is obtained, including:

[0086] The low-frequency approximation coefficient after denoising and the processed high-frequency detail coefficient of each layer are used to perform wavelet reconstruction by using an inverse wavelet transform algorithm to generate a preliminary denoised image.

[0087] The preliminary denoised image is used as a guide image, and the luminance component of the original infrared image is used as an input image. Guided filtering is applied to eliminate the pseudo-Gibbs phenomenon in the preliminary denoised image and obtain a final denoised image.

[0088] In the embodiment of the present application, the core goal of the above subdivision step of step S200 is to generate a high-quality denoised image, which integrates the high and low frequency features after denoising to restore the main information of the image, and eliminates the artifacts that may be introduced by wavelet transform, finally realizing the optimal balance between noise suppression and detail preservation.

[0089] To realize the above steps, first, the low-frequency approximation coefficient after denoising and the processed high-frequency detail coefficient of each layer are used to perform wavelet reconstruction by using an inverse wavelet transform algorithm to generate a preliminary denoised image.

[0090] Specifically, the Daubechies db4 wavelet basis function consistent with the preceding decomposition stage can be used to ensure the reversibility of the transform; and three layers of inverse transform are set, corresponding to the decomposition layers.

[0091] First, for the third layer of inverse transform, the denoised LL3 and the processed LH3, HL3, and HH3 are combined through inverse two-dimensional discrete wavelet transform (IDWT) to obtain LL2', which is the second layer of low-frequency approximation subband after denoising.

[0092] The second layer of inverse transform is to combine LL2' and the processed LH2, HL2, and HH2 to obtain LL1', which is the first layer of low-frequency approximation subband after denoising.

[0093] The first layer inverse transform is to combine LL1' with the processed LH1, HL1 and HH1 to generate a preliminary denoised image with a size of 640*480, i.e., a single-channel luminance image, which has integrated the high-frequency and low-frequency features after denoising, but may have edge ringing artifacts.

[0094] Further, a guided filter is applied for guided filtering processing with the preliminary denoised image as a guide image and the luminance component of the original infrared image as an input image to eliminate the pseudo Gibbs phenomenon in the preliminary denoised image and obtain a final denoised image.

[0095] In step S200 of the embodiment, the guided filtering processing is performed with the preliminary denoised image as a guide image and the luminance component of the original infrared image as an input image, including:

[0096] The preliminary denoised image is set as the guide image, the luminance component of the original infrared image is set as the input image, and the filter window radius and the regularization parameter of the guided filter are set;

[0097] A local linear model is established on the guide image to determine the linear relationship between the output image and the guide image;

[0098] The coefficient parameter in the linear relationship is solved by minimizing a cost function;

[0099] The pixel value of the output image is calculated according to the solved coefficient parameter to obtain the denoised image after the pseudo Gibbs phenomenon is eliminated.

[0100] First, the preliminary denoised image is set as the guide image, the luminance component of the original infrared image is set as the input image, and the filter window radius and the regularization parameter of the guided filter are set.

[0101] The guide image selects the preliminary denoised image as a guide, denoted as G, which contains clear structural information such as insulator contour and temperature distribution trend, and can guide the filtering process to retain these structures.

[0102] The input image selects the luminance component of the original infrared image, denoted as I, i.e., the V component extracted in the preprocessing stage, which retains the original details without excessive smoothing and avoids the guide image artifacts being strengthened.

[0103] For the filter parameters, the window radius of the guided filter is set to 3, corresponding to a 7*7 pixel window, which takes into account the smoothing effect and detail retention, and the regularization parameter is set to ε=0.01 to control the smoothing degree and avoid excessive blurring.

[0104] Then, a local linear model needs to be established on the guide image to determine the linear relationship between the output image and the guide image, and the coefficient parameters in the linear relationship are solved by minimizing the cost function.

[0105] Specifically, all pixels are traversed in a 7x7 window on the guide image (G), and for each local region in the window, it is assumed that the pixel value of the output image (O) and the pixel value of the guide image satisfy a linear relationship, that is, y=a x g+b. Wherein, y is the pixel value of the output image, g is the corresponding pixel value of the guide image, a (proportion coefficient) and b (offset coefficient) are window parameters to be solved.

[0106] Then, the pixel value of the input image (I) is taken as a reference to construct a cost function. Specifically, the cost function is constructed by first calculating the error between the output value of the local linear model and the pixel value of the input image, and amplifying the significant error by squaring form to ensure that the local linear model has a good fitting to the true details of the input image; secondly, a regularization constraint is added to limit the excessive change of the proportion coefficient in the linear relationship, to prevent the filtering process from over-emphasizing local noise and blurring effective details; finally, by minimizing the cost function which combines error fitting and parameter constraint, the proportion coefficient (a) and the offset coefficient (b) of the linear local linear model in the current window are solved.

[0107] Then, for each 7x7 window, the output value of all pixels in the window is calculated according to y=a x g+b using the solved a and b, for example, g=190, a=0.92, b=8.5, y=183.3. The output value eliminates the shock and retains the original brightness trend.

[0108] Then, for the pixels in the image covered by multiple windows, the average value of all window calculation results is taken as the final pixel value to avoid the splicing traces of the window edges. Finally, the final denoised image which eliminates the pseudo Gibbs phenomenon is generated, which has clear edges, no shock artifacts, and completely retains the temperature details of the insulator.

[0109] In step S300 of the embodiment of the present application, based on the denoised image, an edge detection operator is used for preliminary edge extraction, edge information is obtained by combining non-maximum suppression and double threshold processing, and a morphological edge reconstruction technique is used to connect edge gaps, including:

[0110] The denoised image is convolved and calculated by using a temperature gradient weighted edge detection operator to obtain a preliminary edge response map;

[0111] The non-maximum suppression processing is performed on the preliminary edge response map to thin the edge response;

[0112] The edge response after non-maximum suppression is double-threshold processed, and the edge pixels are divided into strong edge pixels, weak edge pixels and non-edge pixels;

[0113] The strong edge pixels and the weak edge pixels are morphologically closed operated using a 3*3 circular structure element, and edge gaps are connected to form complete edge contours.

[0114] In the embodiment of the application, the purpose of step S300 is to accurately extract the complete edge contour of the insulator, so as to provide clear structural features for subsequent defect positioning.

[0115] To achieve the above purpose, firstly, a temperature gradient weighted edge detection operator is used to perform convolution calculation on the denoising image to obtain a preliminary edge response map.

[0116] The edge detection operator can be based on a traditional Sobel or Canny operator, and a temperature gradient weight is introduced. For regions with sharp changes in pixel gray value in the denoising image, the operator convolution weight is increased; for temperature gentle regions, the weight is reduced, and the response of the temperature related edge is strengthened.

[0117] Specifically, a designed operator such as a 3*3 convolution kernel is used to perform full-image convolution on the denoising image to calculate the edge response value of each pixel. The greater the edge response value, the higher the edge possibility. A preliminary edge response map is generated, and the edge response map is a gray image, and the brightness represents the response intensity.

[0118] Then, the preliminary edge response map is subjected to non-maximum suppression processing to refine the edge response.

[0119] In step S300 of the embodiment of the application, the preliminary edge response map is subjected to non-maximum suppression processing to refine the edge response, including:

[0120] The gradient amplitude and the gradient direction of each pixel point in the preliminary edge response map are calculated;

[0121] According to the gradient direction, the gradient direction of each pixel point is quantized into four main directions of horizontal, vertical, positive diagonal and negative diagonal;

[0122] The gradient amplitude of each pixel point is compared with the gradient amplitudes of two adjacent pixels in the gradient direction;

[0123] When the gradient amplitude of the pixel point is greater than or equal to the gradient amplitudes of the two adjacent pixels, the edge response of the current pixel point is retained;

[0124] When the gradient amplitude of the pixel point is less than the gradient amplitude of any adjacent pixel, the edge response of the current pixel point is suppressed.

[0125] Firstly, the gradient magnitude and gradient direction of each pixel in the preliminary edge response map need to be calculated, that is, for each pixel in the preliminary edge response map, based on the gray level change of its neighborhood pixels, the gradient magnitude (M) reflecting the edge intensity and the gradient direction (θ) reflecting the edge direction are solved. Specifically, the gradient components Gx, Gy in the x and y directions can be calculated by the Sobel operator.

[0126] wherein the gradient magnitude M = sqrt(Gx2+ Gy2) ; and the gradient direction: θ = arctan(Gy / Gx), θ ranges from 0° to 180°, representing the tangent direction of the edge.

[0127] Then, the gradient direction of each pixel needs to be quantized into four main directions of horizontal, vertical, positive diagonal and negative diagonal according to the gradient direction. Among them, the horizontal direction (0°) is the gradient direction close to the left and right horizontal (θ≈0° or 180°); the vertical direction (90°) is the gradient direction close to the up and down vertical (θ≈90°); the positive diagonal direction (45°) is the gradient direction close to from the lower left to the upper right (θ≈45°); the negative diagonal direction (135°) is the gradient direction close to from the upper left to the lower right (θ≈135°).

[0128] Then, the gradient magnitude of each pixel needs to be compared with the gradient magnitudes of two adjacent pixels in the gradient direction. That is, for each pixel, according to its quantized gradient direction, two adjacent pixels in that direction are determined, such as left and right adjacent pixels in the horizontal direction, and up and down adjacent pixels in the vertical direction.

[0129] Comparing the gradient magnitude M of the current pixel with the gradient magnitudes of the two adjacent pixels, specifically, if the M of the current pixel is greater than or equal to the M of the two adjacent pixels, it means that it is in the peak position of the edge, and its edge response is retained; if the M of the current pixel is less than the M of any adjacent pixel, it means that it is in the non-peak region of the edge, and its edge response is suppressed, such as setting the magnitude to 0.

[0130] After the above processing, the edge region only retains the single-pixel-width peak response, and the redundant response of the non-edge region is eliminated, obtaining a refined edge response map with clear edges and accurate positioning, laying a foundation for subsequent double-threshold processing and edge connection.

[0131] Further, the non-maximum suppression edge response needs to be processed by double-threshold, and the edge pixels are divided into strong edge pixels, weak edge pixels and non-edge pixels.

[0132] Specifically, according to the gray distribution of the denoised image, a high threshold and a low threshold are set, wherein the high threshold is, for example, the top 20% of the response value, and the low threshold is, for example, 1 / 2 of the high threshold. Then, when the response value is higher than the high threshold, it is marked as a strong edge pixel, that is, determined as a real edge; when the response value is between the high threshold and the low threshold, it is marked as a weak edge pixel, that is, possibly an extension of a real edge; and when the response value is lower than the low threshold, it is marked as a non-edge pixel, that is, needs to be removed.

[0133] Further, a morphological closing operation is performed on the strong edge pixels and the weak edge pixels using a 3*3 circular structural element to connect edge gaps and form a complete edge profile.

[0134] The 3*3 circular structural element is more suitable for the edge shape of the arc-shaped shed of the insulator than a square, and ensures that the natural trend of the edge is not damaged when the gaps are connected.

[0135] The closing operation is a morphological closing operation performed on a binary image containing strong and weak edge pixels, that is, first inflation and then corrosion. The inflation is to fill the edge gaps by the structural element to connect adjacent weak edge segments, and the corrosion is to restore the original width of the edge to avoid the edge being too thick due to the inflation. Finally, a complete insulator edge profile image is obtained, the edge is continuous and has no gaps, and can be directly used for subsequent defect structure feature analysis.

[0136] In step S400 of the embodiment, a final image with enhanced features is generated through adaptive contrast enhancement based on local temperature distribution and image fusion processing, including:

[0137] The denoised image subjected to edge thinning is divided into local processing blocks of 8*8 pixels, and a contrast enhancement coefficient is dynamically set based on the local temperature distribution characteristics of each processing block;

[0138] Based on the contrast enhancement coefficient, a contrast enhancement processing is performed on each processing block by combining a limited contrast adaptive histogram equalization algorithm, to obtain a contrast enhancement image;

[0139] The denoised image subjected to edge thinning and the contrast enhancement image are fused according to a preset fusion ratio to obtain a preliminary fusion image;

[0140] A multi-scale fusion is performed on the preliminary fusion image through a Laplacian pyramid fusion algorithm to generate a final image with enhanced features.

[0141] In the embodiment, the core target of step S400D1 is to enhance the temperature difference features and edge structure features in the infrared image of the insulator, and to improve the recognition degree of the defect regions such as local overheating and structural abnormalities.

[0142] To achieve the above object, first of all, the edge-thinned denoised image needs to be divided into 8x8 pixel local processing blocks, and based on the local temperature distribution characteristics of each processing block, the contrast enhancement coefficient is dynamically set.

[0143] Specifically, the 640x480 edge-thinned denoised image is uniformly divided into 80x60 local processing blocks of 8x8 pixels, each local processing block corresponds to a local area of the insulator, such as a small part of the shed skirt or a connecting fitting, ensuring that the enhanced granularity matches the typical size of the defect. Then calculate the temperature distribution characteristics of each local processing block: such as the standard deviation and mean value of the pixel gray value in the block. Then set the contrast enhancement coefficient rule, for the local processing block with large standard deviation, i.e. significant temperature fluctuation, give a higher contrast enhancement coefficient, such as 1.2-1.5, to strengthen the difference between the defect and the surrounding; for the local processing block with small standard deviation, i.e. smooth temperature, give a lower contrast enhancement coefficient, such as 0.6-0.9, to avoid over-enhancement and amplify noise.

[0144] Then, based on the contrast enhancement coefficient, the contrast enhancement processing is performed on each processing block by combining the limited contrast adaptive histogram equalization algorithm to obtain a contrast-enhanced image. This step can be realized by enhancing the local contrast based on the CLAHE algorithm.

[0145] Specifically, the 8x8 local processing block can be used as the processing unit, the contrast limit threshold (clipLimit) is set to 2.0-3.0, and the number of histogram bins is 256. Then, based on the set enhancement coefficient, the output gain of CLAHE is adjusted for each local processing block, the local contrast is improved by equalizing the histogram in the block, the gray scale range is stretched for the area with large temperature gradient, such as the defect edge, to highlight the difference; for the smooth temperature area, the details are preserved by moderate equalization, and finally the contrast-enhanced image is generated.

[0146] Then, the edge-thinned denoised image and the contrast-enhanced image are fused according to a preset fusion ratio to obtain a preliminary fused image.

[0147] For example, first of all, the fusion ratio, i.e. the weight, needs to be set. The weight of the edge-thinned denoised image containing structural information can be set to 0.3-0.4 to ensure that the structural outline is not weakened; the weight of the contrast-enhanced image containing temperature characteristics can be set to 0.6-0.7 to highlight the temperature difference of the defect. Then, for each pixel of the two types of images, the fusion value is calculated according to the ratio to obtain a preliminary fused image.

[0148] Further, the preliminary fused image needs to be multi-scale fused by the Laplacian pyramid fusion algorithm to generate a final image with enhanced features.

[0149] Specifically, for the preliminary fusion image, a multi-layer pyramid is constructed by Gaussian blur and down-sampling, such as 5 layers, each layer retains different scale features, in which the low layer is the detail and the high layer is the outline. Then for each layer of the pyramid, an edge-based fusion strategy is adopted, that is, more structural image features are retained in the structural edge area, and more enhanced image features are retained in the temperature anomaly area, and the transition is smoothed by weighted average. Finally, the reconstructed image after feature enhancement is generated by up-sampling and superposition of each layer of the fused pyramid. The image not only retains the complete structural outline of the insulator, but also significantly enhances the temperature features of local overheating defects, which can be directly used for defect identification and positioning.

[0150] Through the above processing, the final image achieves an optimal balance between structural integrity and defect recognition, providing high-quality visual input for the automatic detection of insulator faults.

[0151] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0152] Those skilled in the art should understand that the embodiments of the present application can provide a method, a system or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0153] The present application is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams can be implemented by computer program instructions, and the combination of flows and / or blocks in the flowcharts and / or block diagrams. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing devices to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices produce a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 The means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 The means for implementing the functions specified in the flowcharts and / or block diagrams.

[0154] These computer program instructions can also be stored in a computer readable storage medium which can guide the computer or other programmable data processing devices to work in a specific way, so that the instructions stored in the computer readable storage medium produce a product including instruction means, which implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The means for implementing the functions specified in the flowcharts and / or block diagrams.Figure 1 the function specified in the one or more blocks.

[0155] These computer program instructions can also be loaded into computer or other programmable data processing devices, so that a series of operation steps are performed on the computer or other programmable data processing devices to generate computer-implemented processes, thus the instructions executed on the computer or other programmable data processing devices provide a process for implementing the flowchart Figure 1 the flowchart or flowcharts and / or a block Figure 1 the steps of the function specified in the one or more blocks.

[0156] Although preferred embodiments of the application have been described, those skilled in the art will be able to make additional modifications and variations to these embodiments without departing from the spirit and scope of the application.

[0157] Obviously, numerous modifications and variations of the present application are possible in light of the above teachings. It is therefore to be understood that within the scope of the application and its equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. An insulator image enhancement method based on drone infrared image denoising, characterized in that, The method comprises: The raw infrared image collected by the unmanned aerial vehicle is preprocessed, and the preprocessed image is decomposed by wavelet transform to obtain high-frequency detail coefficients and low-frequency approximation coefficients; The high-frequency detail coefficients are processed by adaptive threshold value, and the low-frequency approximation coefficients are processed by Wiener filter, and after wavelet reconstruction and guided filter processing, a denoising image is obtained; Based on the denoising image, an edge detection operator is used for preliminary edge extraction, edge information is obtained by combining non-maximum suppression and double threshold processing, and a morphological edge reconstruction technique is used to connect edge gaps; Through adaptive contrast enhancement and image fusion processing based on local temperature distribution, a final image with enhanced features is generated; The raw infrared image collected by the unmanned aerial vehicle is preprocessed, including: Obtaining raw infrared image data collected by an infrared thermal imager carried by the unmanned aerial vehicle; The size of the raw infrared image is standardized to a fixed resolution of 640x480, and the bilinear interpolation algorithm is used to maintain the image proportion; The image after size adjustment is converted from RGB color space to HSV color space, and the V component in the HSV color space is extracted as the luminance component; The preprocessed image is decomposed by wavelet transform to obtain high-frequency detail coefficients and low-frequency approximation coefficients, including: Selecting Daubechies db4 wavelet basis function as the transform basis, and performing three-layer two-dimensional discrete wavelet transform on the preprocessed luminance component image; In the first layer decomposition, the input image is decomposed into one low-frequency approximation subband LL1 and three high-frequency detail subbands, wherein the three high-frequency detail subbands include horizontal direction LH1, vertical direction HL1 and diagonal direction HH1; In the second layer decomposition, the LL1 subband is further decomposed to obtain a low-frequency approximation subband LL2 and corresponding three high-frequency detail subbands LH2, HL2 and HH2; In the third layer decomposition, the LL2 subband is decomposed to obtain the final low-frequency approximation subband LL3 and three high-frequency detail subbands LH3, HL3 and HH3; LL3 is defined as the low frequency approximation coefficient C LL3 LH1, HL1, HH1, LH2, HL2, HH2, LH3, HL3, HH3 are collectively defined as the high frequency detail coefficients; The high-frequency detail coefficients are processed by adaptive threshold value, and the low-frequency approximation coefficients are processed by Wiener filter, including: For each high frequency detail coefficient subband, a standard deviation σ is calculated, and an adaptive threshold T is determined based on the Bayesian estimation principle, wherein, M and N are the size of the high frequency detail coefficient subband, respectively. The first soft threshold function is applied to the high-frequency detail coefficients of the horizontal direction LH and the vertical direction HL for processing, and the coefficients with absolute value greater than the adaptive threshold value T are retained; The second soft threshold function is applied to the high-frequency detail coefficients of the diagonal direction HH for processing, wherein the threshold value of the second soft threshold function is adjusted to 0.7T; For low frequency approximation coefficients C LL3 , the mean and variance within the corresponding local neighborhood are computed, and a Wiener filter is constructed based on the mean and variance; applying the wiener filter to the low frequency approximation coefficients C LL3 performing filtering to obtain de-noised low frequency approximation coefficients.

2. The insulator image enhancement method based on drone infrared image denoising according to claim 1, characterized in that, For low frequency approximation coefficients C LL3 , computing the mean and variance within the corresponding local neighborhood, constructing a Wiener filter based on the mean and variance, comprising: A sliding window of 7x7 pixels is set as a local neighborhood; Within said local neighborhood, a local mean μ and a local variance σ of said low frequency approximation coefficients C LL3 are calculated local 2 and an overall noise variance σ LL3 of said low frequency approximation coefficients C noise 2 is obtained According to the local mean μ, the local variance σ local 2 and the overall noise variance σ noise 2 the transfer function of the Wiener filter is constructed: where (x, y) denotes the position coordinates of the current coefficient in the low frequency approximation coefficients C LL3 ; applying the wiener filter to the low frequency approximation coefficients C LL3 performing filtering to obtain de-noised low frequency approximation coefficients.

3. The insulator image enhancement method based on drone infrared image denoising according to claim 1, characterized in that, After wavelet reconstruction and guided filter processing, a denoising image is obtained, including: The denoised low-frequency approximation coefficients and the processed high-frequency detail coefficients of each layer are used to perform wavelet reconstruction by using the inverse wavelet transform algorithm to generate a preliminary denoising image; The preliminary denoising image is used as a guide image, and the luminance component of the raw infrared image is used as an input image, and a guided filter is applied for guided filter processing to eliminate the pseudo Gibbs phenomenon in the preliminary denoising image, and a final denoising image is obtained.

4. The insulator image enhancement method based on UAV infrared image denoising according to claim 3, characterized in that, Applying a guided filter to perform guided filtering processing on the preliminary denoising image as a guide image and the luminance component of the original infrared image as an input image, including: Setting the preliminary denoising image as a guide image, setting the luminance component of the original infrared image as an input image, and setting a filter window radius and a regularization parameter of the guided filter; Establishing a local linear model on the guide image to determine a linear relationship between an output image and the guide image; Solving a coefficient parameter in the linear relationship by minimizing a cost function; According to the solved coefficient parameter, calculating a pixel value of the output image to obtain a denoising image after eliminating the pseudo-Gibbs phenomenon.

5. The insulator image enhancement method based on drone infrared image denoising according to claim 1, characterized in that, Based on the denoising image, performing preliminary edge extraction using an edge detection operator, combining non-maximum suppression and double-threshold processing to obtain edge information, and connecting edge gaps using a morphological edge reconstruction technique, including: Performing convolution calculation on the denoising image using a temperature gradient weighted edge detection operator to obtain a preliminary edge response map; Performing non-maximum suppression processing on the preliminary edge response map to refine the edge response; Performing double-threshold processing on the edge response after non-maximum suppression to divide the edge pixels into strong edge pixels, weak edge pixels, and non-edge pixels; Using a 3*3 circular structural element to perform morphological closing operation on the strong edge pixels and the weak edge pixels to connect the edge gaps and form complete edge contours.

6. The insulator image enhancement method based on UAV infrared image denoising according to claim 5, characterized in that, Performing non-maximum suppression processing on the preliminary edge response map to refine the edge response, including: Calculating the gradient amplitude and gradient direction of each pixel point in the preliminary edge response map; Quantizing the gradient direction of each pixel point into four main directions of horizontal, vertical, positive diagonal, and negative diagonal according to the gradient direction; Comparing the gradient amplitude of each pixel point with the gradient amplitudes of two adjacent pixels in the gradient direction; When the gradient amplitude of the pixel point is greater than or equal to the gradient amplitudes of the two adjacent pixels, retaining the edge response of the current pixel point; When the gradient amplitude of the pixel point is less than the gradient amplitude of any adjacent pixel, suppressing the edge response of the current pixel point.

7. The insulator image enhancement method based on UAV infrared image denoising according to claim 1, characterized in that, Generating a final image with enhanced features through adaptive contrast enhancement based on local temperature distribution and image fusion processing, including: Dividing the denoising image after edge refinement into 8*8 pixel local processing blocks, and dynamically setting a contrast enhancement coefficient based on the local temperature distribution features of each processing block; Based on the contrast enhancement coefficient, combining a limited contrast adaptive histogram equalization algorithm to perform contrast enhancement processing on each processing block to obtain a contrast enhancement image; Fusing the denoising image after edge refinement and the contrast enhancement image according to a preset fusion ratio to obtain a preliminary fusion image; Performing multi-scale fusion on the preliminary fusion image through a Laplacian pyramid fusion algorithm to generate a final image with enhanced features.

Images