A method, device and system for detecting faults in electric power equipment

Abstract

The present invention relates to the field of infrared image enhancement technology, and in particular to a method, device and system for detecting faults in power equipment. In the process of sampling and interpolating an infrared grayscale image, the method obtains the pixel value of the interpolated pixel through a weight model constructed by weight sensitivity coefficient and distance, and obtains the initial interpolation quality in combination with the distribution of local suspected defective pixel points and the deviation from the visible light grayscale image; when the initial interpolation quality is less than expected, iteration is performed, and a new weight sensitivity coefficient is obtained according to the difference between the initial interpolation quality and the expected value and the local fuzzy difference between the interpolated pixel point and the visible light grayscale image at the corresponding position; a new initial interpolation quality is obtained based on the new weight sensitivity coefficient, until the iteration stops to obtain a high-resolution infrared image for fault detection. The present invention adjusts the interpolation strength of each pixel point of the infrared grayscale image adaptively according to the visible light image, and obtains a high-resolution infrared image more efficiently, making subsequent fault detection more accurate.

Description

Technical Field

[0001] The present invention relates to the field of infrared image enhancement technology, and in particular to a method, device and system for detecting faults of electric power equipment. Background Art

[0002] The safe operation of power equipment has a direct impact on the stability of the power system, so it is of great significance for the equipment to maintain its safe and stable operating state. When an abnormality occurs in the equipment, the staff can quickly receive an alarm, thereby determining the equipment defect or an operating condition problem of the equipment and taking appropriate measures to eliminate the defect in a timely manner, which can improve the power system's early warning ability for faults and defects, avoid the occurrence of corresponding accidents, and reduce the economic losses and energy waste caused by unnecessary power outages.

[0003] As a non-contact detection technology, infrared detection can detect the distribution of target temperature fields of several millimeters. Due to the advantages of strong safety, accurate detection, high sensitivity, and easy operation, infrared detection technology has become an important part of many equipment status detection and fault diagnosis, and is widely used in power, petrochemical, metallurgy, railway and other fields.

[0004] However, when infrared images are usually used for fault detection, since infrared images have unique infrared data and can capture more detailed infrared spectral information, their spatial resolution is usually low, which greatly limits the application of hyperspectral images in many fields. Therefore, it is necessary to upsample and interpolate the infrared images to improve the resolution. However, the commonly used interpolation methods do not consider the distribution of abnormal pixels in the image, and the accuracy improvement for defective parts is the same as that for normal parts. There is still a situation where the accuracy improvement for defective parts is insufficient. In addition, the commonly used bicubic interpolation method cannot guarantee the interpolation efficiency while improving the interpolation accuracy, which affects the efficient detection of fault detection. Summary of the invention

[0005] In order to solve the technical problem that the interpolation method commonly used in the prior art still has the defective part of insufficient improvement accuracy, cannot guarantee the interpolation efficiency while improving the accuracy, and affects the efficient detection of fault detection, the purpose of the present invention is to provide a method, device and system for detecting faults of power equipment, and the technical scheme adopted is as follows:

[0006] The present invention provides a method for detecting a fault of an electric power device, the method comprising:

[0007] Obtain infrared grayscale images and visible light grayscale images of the power equipment to be tested; filter out suspected defective pixels based on the deviation of pixels in the infrared grayscale image;

[0008] In the upsampling interpolation process of the infrared grayscale image, a weight model is constructed based on the weight sensitivity coefficient and the distance; the pixel value of each interpolated pixel is obtained by combining the distribution of local non-interpolated pixels with the weight model; the initial interpolation quality is obtained based on the distribution of local suspected defect pixels of the interpolated pixels and the deviation of the pixel values ​​between the interpolated pixels and the visible light grayscale image;

[0009] When the initial interpolation quality does not meet the expected quality, the weight sensitivity coefficient of each interpolation pixel is iteratively adjusted; in each iteration, based on the difference between the initial interpolation quality and the expected quality, according to the difference in the local blur between each interpolation pixel and the corresponding position in the visible light grayscale image, and the weight sensitivity coefficient, a new weight sensitivity coefficient of each interpolation pixel is obtained; a new weight model is obtained through the new weight sensitivity coefficient, and a new pixel value of each interpolation pixel is obtained through the new weight model, and a new initial interpolation quality is obtained based on the new pixel value;

[0010] Until the initial interpolation quality does not meet the expected quality or reaches the termination condition, a high-resolution infrared image after interpolation is obtained; and fault detection is performed based on the high-resolution infrared image.

[0011] Furthermore, the weight model constructed based on the weight sensitivity coefficient and the distance includes:

[0012] Taking the i-th local non-interpolation pixel of the a-th interpolation pixel as an example, the expression of the weight model is:

[0013] Wherein, H(x) represents the weight model of the i-th local non-interpolation pixel in the horizontal direction, d represents the distance between the a-th interpolation pixel and the i-th local non-interpolation pixel in the horizontal direction, H(y) represents the weight model of the i-th local non-interpolation pixel in the vertical direction, d′ represents the distance between the a-th interpolation pixel and the i-th local non-interpolation pixel in the vertical direction, and k represents the weight sensitivity coefficient.

[0014] Furthermore, the method for obtaining the pixel value of each interpolation pixel point includes:

[0015] For any interpolation pixel point, based on the weight model, the horizontal axis weight and the vertical axis weight of each local non-interpolation pixel point are obtained through the distance between each local non-interpolation pixel point and the interpolation pixel point;

[0016] Calculate the product of the pixel value of each local non-interpolation pixel point of the interpolation pixel point, the horizontal axis weight, and the vertical axis weight to obtain the pixel influence value of each local non-interpolation pixel point of the interpolation pixel point;

[0017] The sum of the pixel influence values ​​of all local non-interpolation pixel points of the interpolation pixel point is used as the pixel value of the interpolation pixel point.

[0018] Furthermore, the method for obtaining the initial interpolation quality includes:

[0019] For any interpolation pixel point, if there is no suspected defective pixel point within the preset neighborhood range of the interpolation pixel point, the attention weight of the interpolation pixel point is set to the preset weight; otherwise, the number of suspected defective pixels within the preset neighborhood range of the interpolation pixel point is counted and normalized to obtain the attention weight of the interpolation pixel point;

[0020] The difference between the pixel value of the interpolated pixel point and the pixel value of the pixel point at the corresponding position in the visible light grayscale image is used as the interpolation deviation value of the interpolated pixel point;

[0021] The interpolation deviation values ​​are weighted summed and normalized by the attention weights of the interpolation pixels to obtain the initial interpolation quality.

[0022] Furthermore, the method for obtaining the new weight sensitivity coefficient of each interpolation pixel point includes:

[0023] The difference between the initial interpolation quality and the expected quality is used as the adjustment range indicator for the current iteration;

[0024] For any interpolation pixel point, calculate the mean value of the gradient value of the interpolation pixel point within a preset local range to obtain the local blurriness of the interpolation pixel point; calculate the mean value of the gradient value of the interpolation pixel point within a preset local range at the corresponding position of the interpolation pixel point in the visible light grayscale image to obtain the reference blurriness of the interpolation pixel point;

[0025] The difference between the reference blurriness and the local blurriness of the interpolation pixel point is used as the adjustment coefficient of the interpolation pixel point;

[0026] The product of the adjustment coefficient of the interpolation pixel point and the adjustment range index is used as the adjustment amount of the interpolation pixel point; the sum of the adjustment amount of the interpolation pixel point and the weight sensitivity coefficient is calculated as the new weight sensitivity coefficient of the interpolation pixel point.

[0027] Furthermore, the fault detection based on the high-resolution infrared image includes:

[0028] The high-resolution infrared image of the power equipment to be detected is input into the trained network model, and the fault detection results are output.

[0029] Furthermore, the termination condition is that the number of iterations meets a preset number.

[0030] Furthermore, the method for obtaining the suspected defective pixel includes:

[0031] Arrange all pixel values ​​in the infrared grayscale image in ascending order to obtain a pixel sequence of the infrared grayscale image;

[0032] Calculate the pixel difference between every two adjacent pixel values ​​in the pixel sequence, and use the minimum pixel value of the two pixel values ​​corresponding to the maximum pixel difference as the partition pixel value;

[0033] The pixels in the infrared grayscale image whose pixel values ​​are greater than the divided pixel values ​​are regarded as suspected defective pixels.

[0034] The present invention also provides a power equipment fault detection device, comprising:

[0035] The data acquisition unit is used to obtain the infrared grayscale image and the visible light grayscale image of the power equipment to be tested; and to screen out the suspected defective pixels according to the deviation of the pixels in the infrared grayscale image;

[0036] The interpolation quality analysis unit is used to construct a weight model based on the weight sensitivity coefficient and the distance during the upsampling interpolation process of the infrared grayscale image; the pixel value of each interpolation pixel is obtained by combining the distribution of local non-interpolation pixels with the weight model; the initial interpolation quality is obtained according to the distribution of local suspected defective pixels of the interpolation pixel and the deviation of the pixel value between the interpolation pixel and the visible light grayscale image;

[0037] An iterative adjustment unit is used to iteratively adjust the weight sensitivity coefficient of each interpolation pixel when the initial interpolation quality does not meet the expected quality; in each iteration, based on the difference between the initial interpolation quality and the expected quality, according to the difference in the local blur between each interpolation pixel and the corresponding position in the visible light grayscale image, and the weight sensitivity coefficient, a new weight sensitivity coefficient of each interpolation pixel is obtained; a new weight model is obtained through the new weight sensitivity coefficient, and a new pixel value of each interpolation pixel is obtained through the new weight model, and a new initial interpolation quality is obtained according to the new pixel value;

[0038] The detection unit is used to obtain a high-resolution infrared image after interpolation until the initial interpolation quality does not meet the expected quality or reaches a termination condition; and perform fault detection based on the high-resolution infrared image.

[0039] The present invention also provides an electric power equipment fault detection system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any one of the above-mentioned electric power equipment fault detection methods when executing the computer program.

[0040] The present invention has the following beneficial effects:

[0041] In the upsampling interpolation process, the present invention obtains pixel values ​​of interpolated pixels through a weight model constructed by weight sensitivity coefficients and distances, and obtains initial interpolation quality based on the distribution of local suspected defective pixels of the interpolated pixels and the deviation from the visible light grayscale image. The strength of the weight model is adjusted through the weight sensitivity coefficient, and the interpolation evaluation is performed through the difference with the visible light grayscale image, so that the coefficient can be adaptively adjusted for each interpolated pixel point later. Further, when the initial interpolation quality is not satisfactory to the expected situation, each interpolation pixel point is subjected to adaptive coefficient iteration. In each iteration, the difference between the initial interpolation quality and the expected situation reflects the total strength of the adjustable coefficient. According to the local fuzzy difference and weight sensitivity coefficient of each interpolation pixel point at the corresponding position of the visible light, a new weight sensitivity coefficient is obtained. The weight sensitivity coefficient is adaptively adjusted for each interpolation pixel point, and a new pixel point is obtained based on a new weight model constructed based on the new weight sensitivity coefficient. The new initial interpolation quality is re-obtained to iterate and satisfy the evaluation conditions, so that each interpolation pixel point is subjected to adaptive interpolation strength acquisition, and finally a better high-resolution infrared image is obtained for fault detection. The present invention adjusts the adaptive interpolation strength of each pixel point of the infrared grayscale image based on the visible light image, obtains a high-resolution infrared image more efficiently, and makes subsequent fault detection more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 A flow chart of a method for detecting a fault in an electric power device provided by an embodiment of the present invention;

[0044] Figure 2 A schematic diagram of an infrared grayscale image of an electric power device to be detected provided by an embodiment of the present invention;

[0045] Figure 3 A structural block diagram of a power equipment fault detection device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0046] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of a method, device and system for detecting faults of electric power equipment proposed by the present invention, its specific implementation, structure, features and effects in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0047] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0048] The specific scheme of a method, device and system for detecting faults of electric power equipment provided by the present invention is described in detail below with reference to the accompanying drawings.

[0049] Embodiment 1:

[0050] See also Figure 1 , which shows a flow chart of a method for detecting a fault of an electric power device provided by an embodiment of the present invention, the method comprising the following steps:

[0051] S1: Obtain an infrared grayscale image and a visible light grayscale image of the power equipment to be tested; and screen out suspected defective pixels based on the deviation of the pixels in the infrared grayscale image.

[0052] In an embodiment of the present invention, an infrared imager and an industrial camera are set up to shoot the power equipment to be tested, and an infrared image and a visible light image of the power equipment to be tested are obtained. Since the shooting may be affected by environmental factors such as sunlight, wind, and dust scattering, and the detector, A / D converter and other devices of the instrument itself will generate noise, the infrared image and the visible light image are preprocessed to obtain an infrared grayscale image and a visible light grayscale image. Among them, the image preprocessing process can specifically include image grayscale processing, filtering and denoising processing, image enhancement processing and background removal processing. It should be noted that the image preprocessing process is a technical means well known to those skilled in the art. Specifically, the grayscale weighting method can be selected for grayscale, the bilateral filtering method can be used for filtering and denoising, the adaptive histogram equalization can be used for image enhancement, and the background difference method can be used for background removal, etc., which will not be repeated here.

[0053] Due to the difference in shooting equipment, the resolution of infrared images is often lower than that of visible light images, which greatly limits the ability of infrared images to express details. Therefore, it is necessary to upsample and interpolate infrared images to improve the resolution. For power equipment, its heating condition can well reflect its working status. The heating characteristics of normal power equipment will remain within the normal range, while faulty power equipment will have abnormal heating conditions, and significant pixel value changes will appear in the infrared grayscale image. When improving the resolution in the future, more attention should be paid to the parts with higher deviations. Please refer to Figure 2 , which shows a schematic diagram of an infrared grayscale image of an electric power device to be detected provided by an embodiment of the present invention.

[0054] Therefore, the suspected defective pixels are first screened out through the deviation of the pixels in the infrared grayscale image. Preferably, in the embodiment of the present invention, all the pixel values ​​in the infrared grayscale image are arranged in ascending order to obtain the pixel sequence of the infrared grayscale image, and the pixel difference between each two adjacent pixel values ​​in the pixel sequence is calculated. The minimum pixel value of the two pixel values ​​corresponding to the maximum pixel difference is used as the dividing pixel value, and the pixels in the infrared grayscale image that are greater than the dividing pixel value are used as suspected defective pixels. Through the degree of continuous pixel value mutation, the pixels with higher pixel values ​​are used as suspected defective pixels that need more attention, and the interpolation quality is focused on when the resolution is subsequently improved.

[0055] In other embodiments of the present invention, the infrared grayscale image can also be directly divided using the Otsu threshold algorithm, and the pixels with higher pixel values ​​can be regarded as suspected defective pixels. The Otsu threshold algorithm is a technical means well known to those skilled in the art and will not be elaborated here.

[0056] S2: In the upsampling interpolation process of the infrared grayscale image, a weight model is constructed based on the weight sensitivity coefficient and the distance; for each interpolation pixel, the pixel value of each interpolation pixel is obtained by combining the distribution of local non-interpolation pixels with the weight model; the initial interpolation quality is obtained based on the distribution of local suspected defect pixels of the interpolation pixel and the deviation of the pixel value between the interpolation pixel and the visible light grayscale image.

[0057] In the method of upsampling and interpolating infrared grayscale images, bicubic interpolation can usually be used to perform interpolation with higher precision. However, the weight analysis of local pixels by bicubic interpolation is relatively complicated, which makes the interpolation process inefficient. Therefore, in an embodiment of the present invention, the weight model of the interpolation pixels is reconstructed to reduce the complexity of the weight analysis process. At the same time, in order to ensure the accuracy of the weight analysis, the weight sensitivity coefficient is added to facilitate subsequent adaptive adjustment and improve the accuracy requirement.

[0058] The embodiment of the present invention is based on the bicubic interpolation method. The upsampling interpolation process is mainly to calculate the pixel value of the interpolation point after enlarging the image. The length and width of the image are two-dimensional, that is, the X axis and the Y axis. Let X=M, Y=N of the visible light grayscale image, and x=m, y=n of the infrared grayscale image. According to the high resolution of the visible light image, this embodiment enlarges the resolution of the infrared grayscale image to the same as that of the visible light grayscale image, that is, K×(m×n)=M×N. The pixel value of each interpolation pixel point is obtained by analyzing the pixel value of the local non-interpolation pixel point in the original infrared grayscale image. The local non-interpolation pixel points are 16 pixel points around the interpolation pixel point. It should be noted that the specific enlargement process and the local pixel point acquisition process are well-known technical means in the bicubic interpolation method that are well known to those skilled in the art, and will not be described in detail here.

[0059] Preferably, in an embodiment of the present invention, the weight model constructed based on the weight sensitivity coefficient and the distance includes:

[0060] Taking the i-th local non-interpolation pixel of the a-th interpolation pixel as an example, the expression of the weight model is:

[0061] Wherein, H(x) represents the weight model of the i-th local non-interpolation pixel in the horizontal direction, d represents the distance between the a-th interpolation pixel and the i-th local non-interpolation pixel in the horizontal direction, H(y) represents the weight model of the i-th local non-interpolation pixel in the vertical direction, d′ represents the distance between the a-th interpolation pixel and the i-th local non-interpolation pixel in the vertical direction, and k represents the weight sensitivity coefficient.

[0062] The computational complexity of weight analysis is reduced by quadratic functions. The weight model is a quadratic function that opens downward. The smaller the distance, the larger the weight, and the greater the influence of the pixel value of the pixel point. The weight sensitivity coefficient k represents the sensitivity of the weight to the change in distance. The larger the k, the more sensitive the weight is to the change in distance. In the interpolated image, the boundaries between different infrared feature areas will be clearer, but excessive sensitivity may cause jaggedness. On the contrary, the smaller k is, the less sensitive the weight is to the change in distance. In the interpolated image, the gradient change at the edge will be smoother, but excessive insensitivity may cause artifacts of varying degrees in the image.

[0063] Therefore, an initial value is first set for the weight sensitivity coefficient, and then adaptively adjusted according to the specific imaging conditions of different images to improve the accurate distribution of interpolation strength while reducing complexity. In an embodiment of the present invention, the weight sensitivity coefficient can be set to 0.5. The specific initial setting implementer can adjust it according to the specific implementation scenario to reduce the number of subsequent iterations.

[0064] First, upsampling interpolation is performed through the initial weight model to obtain the pixel value of each interpolation pixel. In the embodiment of the present invention, the pixel value is obtained based on the pixel value and weight of the local non-interpolation pixel of each interpolation pixel. For any interpolation pixel, the horizontal axis weight and vertical axis weight of each local non-interpolation pixel are obtained through the distance between each local non-interpolation pixel and the interpolation pixel based on the weight model. The horizontal axis weight is obtained through the weight model of the local non-interpolation pixel in the horizontal direction, and the vertical axis weight is obtained through the weight model of the local non-interpolation pixel in the vertical direction. In the embodiment of the present invention, the distance can be calculated using the Euclidean distance. The Euclidean distance is a technical means well known to those skilled in the art and will not be described in detail here.

[0065] The product of the pixel value, horizontal axis weight and vertical axis weight of each local non-interpolation pixel of the interpolation pixel is calculated to obtain the pixel influence value of each local non-interpolation pixel of the interpolation pixel, reflecting the pixel value distribution of each local non-interpolation pixel to the interpolation pixel. Then the sum of the pixel influence values ​​of all local non-interpolation pixels of the interpolation pixel is used as the pixel value of the interpolation pixel. As an example, the pixel value of the interpolation pixel is obtained as follows:

[0066] In the formula, f a Represented as the pixel value of the a-th interpolated pixel, F i is represented by the pixel value of the i-th local non-interpolation pixel of the a-th interpolation pixel, N is represented by the total number of local non-interpolation pixels of the a-th interpolation pixel, H i (x) represents the horizontal axis weight of the i-th local non-interpolation pixel of the a-th interpolation pixel, H i (y) is the vertical axis weight of the i-th local non-interpolation pixel of the a-th interpolation pixel, F i ×H i (x)×H i (y) is represented as the pixel influence value of the i-th local non-interpolation pixel of the a-th interpolation pixel.

[0067] After completing the preliminary interpolation of each interpolation pixel, the interpolation quality after interpolation is obtained to evaluate whether the interpolation effect meets expectations. Considering the possible faults of the power equipment, the closer the interpolation pixel is to the distribution of the suspected pixel, the more attention the interpolation pixel needs to be paid when conducting quality analysis. The quality of the pixel value distribution is judged by the visible light grayscale image with better resolution.

[0068] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the initial interpolation quality includes:

[0069] First, for any interpolated pixel point, if there is no suspected defective pixel point within the preset neighborhood of the interpolated pixel point, it means that the location of the interpolated pixel point requires less attention, and the attention weight of the interpolated pixel point is set to the preset weight. Otherwise, it means that a suspected defective pixel point has appeared locally, and the number of suspected defective pixels within the preset neighborhood of the interpolated pixel point is counted and normalized to obtain the attention weight of the interpolated pixel point. The more suspected defective pixels there are locally, the more important the situation that needs attention is.

[0070] In a specific implementation of an embodiment of the present invention, the preset neighborhood range is set to a range size of 7 with a side length of 7 centered on the interpolation pixel point, and the preset weight is set to 0.1. The specific numerical value implementer can adjust it according to the specific implementation situation, and no limitation is made here. It should be noted that normalization is a technical means well known to those skilled in the art, and the choice of normalization can be linear normalization or standard normalization, etc. The specific normalization method is not limited here.

[0071] Furthermore, the difference between the pixel value of the interpolated pixel and the pixel value of the pixel at the corresponding position in the visible light grayscale image is used as the interpolation deviation value of the interpolated pixel. Taking the visible light grayscale image as a reference, if the quality of the interpolation is better, the pixel values ​​at the corresponding positions should be more similar. Therefore, the quality judgment is made by the grayscale difference.

[0072] Finally, the interpolation deviation values ​​are weighted and normalized by the attention weights of the interpolation pixels to obtain the initial interpolation quality. By weighting the attention weights, the interpolation quality is mainly focused on the more important parts, improving the interpolation of the important parts and making the interpolation strength of the subsequent important parts better adjusted. As an example, the expression of the initial interpolation quality is:

[0073] Q=norm(∑ A a＝1 (M a ×|f a -f′ a |)); where Q represents the initial interpolation quality, A represents the total number of interpolated pixels, and M a It is expressed as the attention weight of the a-th interpolated pixel, f a Represented as the pixel value of the ath interpolated pixel, f′ a It is represented as the pixel value of the ath interpolated pixel at the corresponding position in the visible light grayscale image, |f a -f′ a | represents the interpolation deviation value of the ath interpolation pixel point, and norm() represents the normalization processing function.

[0074] At this point, the interpolation under the initial weight sensitivity coefficient is completed, and the initial interpolation quality is obtained.

[0075] S3: When the initial interpolation quality does not meet the expected quality, the weight sensitivity coefficient of each interpolation pixel is iteratively adjusted; in each iteration, based on the difference between the initial interpolation quality and the expected quality, according to the difference in local blur between each interpolation pixel and the corresponding position in the visible light grayscale image, and the weight sensitivity coefficient, a new weight sensitivity coefficient of each interpolation pixel is obtained; a new weight model is obtained through the new weight sensitivity coefficient, and a new pixel value of each interpolation pixel is obtained through the new weight model, and a new initial interpolation quality is obtained based on the new pixel value.

[0076] Adaptive iterative adjustment is performed on each interpolation pixel through analysis of the interpolation quality. In the embodiment of the present invention, the judgment is made by setting the expected quality. When the initial interpolation quality is smaller, the overall interpolation situation is better. The expected quality is set to 0.2, and the specific value can be adjusted by the implementer.

[0077] When the initial interpolation quality does not meet the expected quality, the position of each interpolation pixel is compared with the visible light grayscale image, and the weight sensitivity coefficient of each interpolation pixel is iteratively adjusted. In each iteration, the difference between the initial interpolation quality and the expected quality reflects the maximum adjustable degree of the weight sensitivity coefficient. The weight sensitivity coefficient of each interpolation pixel is adjusted in different directions and degrees according to the deviation of the local blur degree to obtain a better new weight sensitivity coefficient.

[0078] Preferably, in an embodiment of the present invention, a method for obtaining a new weight sensitivity coefficient of each interpolation pixel point includes:

[0079] First, the difference between the initial interpolation quality and the expected quality is used as the adjustment range indicator of the current iteration. The higher the quality difference, the greater the range of adjustment required.

[0080] Further, for any interpolated pixel point, the mean value of the gradient value of the interpolated pixel point in a preset local range is calculated to obtain the local blurriness of the interpolated pixel point. The higher the gradient value, the clearer the local pixel point boundary is, and vice versa, the more serious the blurriness of the local edge is. At the same time, the mean value of the gradient value of the interpolated pixel point in the preset local range of the corresponding position in the visible light grayscale image is calculated to obtain the reference blurriness of the interpolated pixel point. The clarity of the local pixel value after interpolation is judged based on the degree of the local edge of the visible light.

[0081] In a specific implementation of the embodiment of the present invention, the preset local range is set to a range size with a side length of 9 centered on the interpolation pixel point, and the specific value is not limited here.

[0082] The difference between the reference blur and the local blur of the interpolated pixel is used as the adjustment coefficient of the interpolated pixel. When the reference blur is large, it means that the local clarity of the pixel after interpolation is insufficient and the weight sensitivity coefficient is small. Therefore, the adjustment coefficient is positive, and the weight sensitivity coefficient is increased and adjusted. When the deviation is larger, the degree of adjustment is greater. On the contrary, when the reference blur is small, it means that the local pixel after the difference is too clear and the weight sensitivity coefficient is large. Therefore, the adjustment coefficient is negative, and the weight sensitivity coefficient is reduced and adjusted. When the deviation is larger, the degree of adjustment is greater.

[0083] The product of the adjustment coefficient of the interpolation pixel and the adjustment range index is used as the adjustment amount of the interpolation pixel, reflecting the value of each interpolation pixel that needs to be adjusted adaptively. Finally, the sum of the adjustment amount of the interpolation pixel and the weight sensitivity coefficient is calculated as the new weight sensitivity coefficient of the interpolation pixel.

[0084] In the current iteration, the new weight sensitivity coefficient is used to replace the original weight sensitivity coefficient of the weight model to obtain a new weight model. The new weight model is used again to adjust only the weights of the local non-interpolation pixels of the interpolation pixels, and the new pixel values ​​are obtained by the above-mentioned pixel value acquisition method of the interpolation pixels. Based on the new pixel values, the new initial interpolation quality is obtained again by the above-mentioned initial interpolation quality acquisition method.

[0085] At this point, the adaptive adjustment analysis of the weight sensitivity coefficients in the iterative process is completed, and the initial interpolation quality after each iteration is obtained.

[0086] S4: until the initial interpolation quality does not meet the expected quality or reaches the termination condition, a high-resolution infrared image after interpolation is obtained; and fault detection is performed based on the high-resolution infrared image.

[0087] The new initial interpolation quality at each iteration is re-evaluated against the expected quality. If it still does not meet expectations, the next iterative adjustment is performed until the expected quality is met or the termination condition is reached. In the embodiment of the present invention, the termination condition is that the number of iterations meets the preset number of times. When the number of iterations is large, it means that the adjustment has approached the optimal situation and can be stopped in advance to improve the interpolation efficiency. The preset number of times can be set to 10. The specific value can be adjusted by the implementer according to the specific implementation situation and is not limited here.

[0088] When the iteration is stopped, the interpolated infrared grayscale image is used as a high-resolution infrared image. The resolution of the high-resolution infrared image at this time is relatively high and can be used for subsequent fault detection. In an embodiment of the present invention, the high-resolution infrared image of the power equipment to be detected is input into the trained network model to output the fault detection result.

[0089] In a specific implementation of an embodiment of the present invention, R-FCN is selected to obtain a trained network model to realize fault recognition of infrared images of power equipment to be detected. The main idea of ​​the regional fully convolutional network (R-FCN) is "position sensitive score map". The network model consists of three parts: feature extraction network, region generation network and region of interest subnet. The infrared images with existing fault annotations are used as training sets, and the training sets are used as input to train R-FCN. The PR curve is used to evaluate R-FCN. The training parameters are adjusted according to the evaluation results to improve the accuracy of detection and obtain a trained network model. In the trained network model, the infrared image is used as input and the fault result is used as output.

[0090] In summary, in the upsampling interpolation process, the present invention obtains pixel values ​​of interpolated pixels through a weight model constructed by weight sensitivity coefficients and distances, and obtains initial interpolation quality based on the distribution of local suspected defective pixels of the interpolated pixels and the deviation from the visible light grayscale image. The strength of the weight model is adjusted through the weight sensitivity coefficient, and the interpolation evaluation is performed through the difference with the visible light grayscale image, so that the coefficient can be adaptively adjusted for each interpolated pixel subsequently. Further, when the initial interpolation quality is not satisfactory to the expected situation, each interpolation pixel point is subjected to adaptive coefficient iteration. In each iteration, the difference between the initial interpolation quality and the expected situation reflects the total strength of the adjustable coefficient. According to the local fuzzy difference and weight sensitivity coefficient of each interpolation pixel point at the corresponding position of the visible light, a new weight sensitivity coefficient is obtained. The weight sensitivity coefficient is adaptively adjusted for each interpolation pixel point, and a new pixel point is obtained based on a new weight model constructed based on the new weight sensitivity coefficient. The new initial interpolation quality is re-obtained to iterate and satisfy the evaluation conditions, so that each interpolation pixel point is subjected to adaptive interpolation strength acquisition, and finally a better high-resolution infrared image is obtained for fault detection. The present invention adjusts the adaptive interpolation strength of each pixel point of the infrared grayscale image based on the visible light image, obtains a high-resolution infrared image more efficiently, and makes subsequent fault detection more accurate.

[0091] Embodiment 2:

[0092] The present invention provides a device for detecting a fault in an electric power device. Figure 3 , which shows a structural block diagram of a power equipment fault detection device provided by an embodiment of the present invention, the device includes:

[0093] The data acquisition unit is used to obtain the infrared grayscale image and the visible light grayscale image of the power equipment to be tested; and to screen out the suspected defective pixels according to the deviation of the pixels in the infrared grayscale image;

[0094] The interpolation quality analysis unit is used to construct a weight model based on the weight sensitivity coefficient and the distance during the upsampling interpolation process of the infrared grayscale image; the pixel value of each interpolation pixel is obtained by combining the distribution of local non-interpolation pixels with the weight model; the initial interpolation quality is obtained according to the distribution of local suspected defective pixels of the interpolation pixel and the deviation of the pixel value between the interpolation pixel and the visible light grayscale image;

[0095] An iterative adjustment unit is used to iteratively adjust the weight sensitivity coefficient of each interpolation pixel when the initial interpolation quality does not meet the expected quality; in each iteration, based on the difference between the initial interpolation quality and the expected quality, according to the difference in the local blur between each interpolation pixel and the corresponding position in the visible light grayscale image, and the weight sensitivity coefficient, a new weight sensitivity coefficient of each interpolation pixel is obtained; a new weight model is obtained through the new weight sensitivity coefficient, and a new pixel value of each interpolation pixel is obtained through the new weight model, and a new initial interpolation quality is obtained according to the new pixel value;

[0096] The detection unit is used to obtain a high-resolution infrared image after interpolation until the initial interpolation quality does not meet the expected quality or reaches a termination condition; and perform fault detection based on the high-resolution infrared image.

[0097] It should be noted that the device provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device is divided into different functional modules to complete all or part of the functions described above. Since the specific implementation process of an electric power equipment fault detection device in this embodiment is the same as the specific implementation process of an electric power equipment fault detection method described above, it will not be described in detail here.

[0098] Embodiment 3:

[0099] The present invention provides an electric power equipment fault detection system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any one of the above-mentioned electric power equipment fault detection methods when executing the computer program.

[0100] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0101] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A method for detecting a fault in an electric power device, characterized in that: The method comprises: Obtain infrared grayscale images and visible light grayscale images of the power equipment to be tested; filter out suspected defective pixels based on the deviation of pixels in the infrared grayscale image; In the upsampling interpolation process of the infrared grayscale image, a weight model is constructed based on the weight sensitivity coefficient and the distance; the pixel value of each interpolated pixel is obtained by combining the distribution of local non-interpolated pixels and the weight model; the initial interpolation quality is obtained based on the distribution of local suspected defect pixels of the interpolated pixels and the deviation of the pixel values ​​between the interpolated pixels and the visible light grayscale image; When the initial interpolation quality does not meet the expected quality, the weight sensitivity coefficient of each interpolation pixel is iteratively adjusted; in each iteration, based on the difference between the initial interpolation quality and the expected quality, according to the difference in the local blur between each interpolation pixel and the corresponding position in the visible light grayscale image, and the weight sensitivity coefficient, a new weight sensitivity coefficient of each interpolation pixel is obtained; a new weight model is obtained through the new weight sensitivity coefficient, and a new pixel value of each interpolation pixel is obtained through the new weight model, and a new initial interpolation quality is obtained based on the new pixel value; Until the initial interpolation quality does not meet the expected quality or reaches the termination condition, a high-resolution infrared image after interpolation is obtained; fault detection is performed based on the high-resolution infrared image; The weight model constructed based on the weight sensitivity coefficient and the distance includes: First The interpolated pixel Taking a local non-interpolation pixel as an example, the expression of the weight model is: ; ; In the formula, Expressed as The weight model of local non-interpolation pixels in the horizontal direction, Expressed as The interpolated pixel and the The distance between local non-interpolated pixels in the horizontal direction, Expressed as The weight model of local non-interpolation pixels in the vertical direction, Expressed as The interpolated pixel and the The distance between local non-interpolated pixels in the vertical direction, Expressed as weight sensitivity coefficient; The method for obtaining the pixel value of each interpolation pixel point comprises: For any interpolation pixel point, based on the weight model, the horizontal axis weight and the vertical axis weight of each local non-interpolation pixel point are obtained by calculating the distance between each local non-interpolation pixel point and the interpolation pixel point; Calculate the product of the pixel value of each local non-interpolation pixel point of the interpolation pixel point, the horizontal axis weight, and the vertical axis weight to obtain the pixel influence value of each local non-interpolation pixel point of the interpolation pixel point; The sum of the pixel influence values ​​of all local non-interpolation pixel points of the interpolation pixel point is used as the pixel value of the interpolation pixel point.

2. A method for detecting a fault in an electric power device according to claim 1, characterized in that: The method for obtaining the initial interpolation quality includes: For any interpolation pixel point, if there is no suspected defective pixel point within the preset neighborhood range of the interpolation pixel point, the attention weight of the interpolation pixel point is set to the preset weight; otherwise, the number of suspected defective pixels within the preset neighborhood range of the interpolation pixel point is counted and normalized to obtain the attention weight of the interpolation pixel point; The difference between the pixel value of the interpolated pixel point and the pixel value of the pixel point at the corresponding position in the visible light grayscale image is used as the interpolation deviation value of the interpolated pixel point; The interpolation deviation values ​​are weighted summed and normalized by the attention weights of the interpolation pixels to obtain the initial interpolation quality.

3. A method for detecting a fault in an electric power device according to claim 1, characterized in that: The method for obtaining the new weight sensitivity coefficient of each interpolation pixel point includes: The difference between the initial interpolation quality and the expected quality is used as the adjustment range indicator for the current iteration; For any interpolation pixel point, calculate the mean value of the gradient value of the interpolation pixel point within a preset local range to obtain the local blurriness of the interpolation pixel point; calculate the mean value of the gradient value of the interpolation pixel point within a preset local range at the corresponding position of the interpolation pixel point in the visible light grayscale image to obtain the reference blurriness of the interpolation pixel point; The difference between the reference blurriness and the local blurriness of the interpolation pixel point is used as the adjustment coefficient of the interpolation pixel point; The product of the adjustment coefficient of the interpolation pixel point and the adjustment range index is used as the adjustment amount of the interpolation pixel point; the sum of the adjustment amount of the interpolation pixel point and the weight sensitivity coefficient is calculated as the new weight sensitivity coefficient of the interpolation pixel point.

4. A method for detecting a fault in an electric power device according to claim 1, characterized in that: The fault detection based on high-resolution infrared images includes: The high-resolution infrared image of the power equipment to be detected is input into the trained network model to output the fault detection results.

5. A method for detecting a fault in an electric power device according to claim 1, characterized in that: The termination condition is that the number of iterations meets a preset number.

6. A method for detecting a fault in an electric power device according to claim 1, characterized in that: The method for obtaining suspected defective pixels includes: Arrange all pixel values ​​in the infrared grayscale image in ascending order to obtain a pixel sequence of the infrared grayscale image; Calculate the pixel difference between every two adjacent pixel values ​​in the pixel sequence, and use the minimum pixel value of the two pixel values ​​corresponding to the maximum pixel difference as the partition pixel value; The pixels in the infrared grayscale image whose pixel values ​​are greater than the divided pixel values ​​are regarded as suspected defective pixels.

7. A power equipment fault detection device, characterized in that: include: A data acquisition unit, used to obtain an infrared grayscale image and a visible light grayscale image of the power equipment to be detected; Filter out suspected defective pixels through the deviation of pixels in the infrared grayscale image; The interpolation quality analysis unit is used to construct a weight model based on the weight sensitivity coefficient and the distance during the upsampling interpolation process of the infrared grayscale image; the pixel value of each interpolation pixel is obtained by combining the distribution of local non-interpolation pixels with the weight model; the initial interpolation quality is obtained according to the distribution of local suspected defective pixels of the interpolation pixel and the deviation of the pixel value between the interpolation pixel and the visible light grayscale image; The weight model constructed based on the weight sensitivity coefficient and the distance includes: First The interpolated pixel Taking a local non-interpolation pixel as an example, the expression of the weight model is: ; ; In the formula, Expressed as The weight model of local non-interpolation pixels in the horizontal direction, Expressed as The interpolated pixel and the The distance between local non-interpolated pixels in the horizontal direction, Expressed as The weight model of local non-interpolation pixels in the vertical direction, Expressed as The interpolated pixel and the The distance between local non-interpolated pixels in the vertical direction, Expressed as weight sensitivity coefficient; The method for obtaining the pixel value of each interpolation pixel point comprises: For any interpolation pixel point, based on the weight model, the horizontal axis weight and the vertical axis weight of each local non-interpolation pixel point are obtained by calculating the distance between each local non-interpolation pixel point and the interpolation pixel point; Calculate the product of the pixel value of each local non-interpolation pixel point of the interpolation pixel point, the horizontal axis weight, and the vertical axis weight to obtain the pixel influence value of each local non-interpolation pixel point of the interpolation pixel point; The sum of the pixel influence values ​​of all local non-interpolation pixel points of the interpolation pixel point is used as the pixel value of the interpolation pixel point; An iterative adjustment unit is used to iteratively adjust the weight sensitivity coefficient of each interpolation pixel when the initial interpolation quality does not meet the expected quality; in each iteration, based on the difference between the initial interpolation quality and the expected quality, according to the difference in the local blur between each interpolation pixel and the corresponding position in the visible light grayscale image, and the weight sensitivity coefficient, a new weight sensitivity coefficient of each interpolation pixel is obtained; a new weight model is obtained through the new weight sensitivity coefficient, and a new pixel value of each interpolation pixel is obtained through the new weight model, and a new initial interpolation quality is obtained according to the new pixel value; The detection unit is used to obtain a high-resolution infrared image after interpolation until the initial interpolation quality does not meet the expected quality or reaches a termination condition; and perform fault detection based on the high-resolution infrared image.

8. A power equipment fault detection system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the power equipment fault detection method as described in any one of claims 1 to 6 are implemented.

Images