An online defect identification method and system for overhead distribution lines

Abstract

The present invention relates to an online defect identification method and system for overhead distribution lines. The method includes obtaining a dataset of inspection images of the distribution network; performing exposure correction processing on the inspection images of the distribution network in the dataset of inspection images of the distribution network; performing noise reduction processing on the inspection images of the distribution network in the dataset of inspection images of the distribution network after correction by using a wavelet packet function; performing data enhancement processing on the inspection images of the distribution network in the noise-reduced dataset of inspection images of the distribution network after noise reduction to determine an updated dataset of inspection images of the distribution network; using the updated dataset of inspection images of the distribution network and the Faster R-CNN algorithm to determine an overhead distribution line defect identification model; and using the overhead distribution line defect identification model to perform defect identification on the inspection images of the distribution network obtained in real time. The present invention can improve the identification efficiency and accuracy.

Description

Technical Field

[0001] The present invention relates to the field of overhead distribution line defect identification, and particularly to an online defect identification method and system for overhead distribution lines. Background Art

[0002] In view of the existing disadvantages in distribution inspection, such as slow inspection efficiency and high work intensity, by effectively applying computer artificial intelligence identification technology, a typical defect feature data set of overhead distribution lines is constructed, and a detection model for typical defects of overhead distribution lines is established based on image recognition technology to achieve the classification of typical defects, object detection, and the application of image recognition at the device end.

[0003] However, during the collection of the data set, the occurrence probability of some defect categories is low and the number of images is small, resulting in an unbalanced phenomenon between classes in the data set, and further leading to low detection efficiency of the detection model.

[0004] Therefore, based on the above problems, there is an urgent need to provide a new identification method to improve the identification efficiency and accuracy. Summary of the Invention

[0005] The purpose of the present invention is to provide an online defect identification method and system for overhead distribution lines, which can improve the identification efficiency and accuracy.

[0006] To achieve the above purpose, the present invention provides the following solutions:

[0007] An online defect identification method for overhead distribution lines, comprising:

[0008] Obtaining a distribution network inspection image data set; the distribution network inspection image data set includes distribution network inspection images of different defect types; the defect types include: channel type defects, conductor type defects, insulator type defects, metal oxide arrester type defects, and electronic component type defects;

[0009] Performing exposure correction processing on the distribution network inspection images in the distribution network inspection image data set;

[0010] For the distribution network inspection images in the corrected distribution network inspection image data set, performing noise reduction processing using a wavelet packet function; wherein, using the formula To determine the wavelet decomposition coefficient; Is the wavelet decomposition coefficient obtained through the thresholding function, ω i,k Is the decomposition coefficient of the wavelet, and λ is the size of the threshold;

[0011] Performing data enhancement processing on the denoised distribution network inspection images in the denoised distribution network inspection image data set to determine the updated distribution network inspection image data set;

[0012] Using the updated inspection image dataset of the distribution network and the Faster R-CNN algorithm, determine the defect recognition model for overhead distribution lines;

[0013] Use the defect recognition model for overhead distribution lines to identify defects in the inspection images of the distribution network obtained in real time.

[0014] Optionally, the exposure correction processing includes: adopting a deep learning method for single-model correction of overexposure and underexposure.

[0015] Optionally, before performing noise reduction processing on the inspection images of the distribution network in the corrected inspection image dataset of the distribution network using wavelet packet functions, it further includes:

[0016] Use Fourier transform to convert the inspection images of the distribution network in the corrected inspection image dataset of the distribution network from the spatial domain to the frequency domain.

[0017] Optionally, the data augmentation processing includes: spatial domain enhancement algorithm and frequency domain enhancement algorithm.

[0018] Optionally, the data augmentation processing includes: histogram equalization method.

[0019] Optionally, the histogram equalization method includes:

[0020] Determine the histogram corresponding to the inspection images of the distribution network in the denoised inspection image dataset of the distribution network after noise reduction processing;

[0021] Perform gray level transformation on the histogram using the cumulative distribution function;

[0022] Merge according to the gray level values of the histogram after gray level transformation.

[0023] An on-line defect identification system for overhead distribution lines, comprising:

[0024] A dataset acquisition module for acquiring an inspection image dataset of the distribution network; the inspection image dataset of the distribution network includes inspection images of different defect types; the defect types include: channel type defects, conductor type defects, insulator type defects, metal oxide arrester type defects, and electronic component type defects;

[0025] An exposure correction processing module for performing exposure correction processing on the inspection images of the distribution network in the inspection image dataset of the distribution network;

[0026] A noise reduction processing module for performing noise reduction processing on the inspection images of the distribution network in the corrected inspection image dataset of the distribution network using wavelet packet functions; wherein, using the formula Determine the wavelet decomposition coefficient; is the wavelet decomposition coefficient obtained through the thresholding function, ω i,kis the decomposition coefficient of the wavelet, and λ is the size of the threshold;

[0027] A data enhancement processing module for performing data enhancement processing on the power distribution network inspection images in the denoised power distribution network inspection image dataset after denoising processing to determine an updated power distribution network inspection image dataset;

[0028] An identification model determination module for using the updated power distribution network inspection image dataset and the Faster R-CNN algorithm to determine an overhead power distribution line defect identification model;

[0029] A defect identification module for using the overhead power distribution line defect identification model to perform defect identification on the power distribution network inspection images obtained in real time.

[0030] An on-line identification system for overhead power distribution line defects, comprising: at least one processor, at least one memory, and computer program instructions stored in the memory, and when the computer program instructions are executed by the processor, the on-line identification method for overhead power distribution line defects is implemented.

[0031] According to the specific embodiments provided by the present invention, the following technical effects are disclosed by the present invention:

[0032] The on-line identification method and system for overhead power distribution line defects provided by the present invention increase the number of samples by performing exposure correction, denoising, and data enhancement processing on the power distribution network inspection image dataset with a small number of samples. And the present invention adopts a new method for determining the wavelet packet threshold, which improves the denoising performance, and further ensures the stability of the samples in subsequent data enhancement. The quality of the samples is greatly enhanced, the utilization efficiency of the samples is improved, and guarantee is provided for model training. Furthermore, the identification efficiency and accuracy can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention, and for those of ordinary skill in the art, other drawings can be obtained according to these drawings without creative efforts.

[0034] Figure 1 is a schematic flow chart of an on-line identification method for overhead power distribution line defects provided by the present invention;

[0035] Figure 2 are schematic diagrams of three threshold functions;

[0036] Figure 3 is a comparison chart of hard threshold, soft threshold, and new threshold;

[0037] Figure 4 For the hard threshold, soft threshold, and new threshold noise reduction graphs;

[0038] Figure 5 Schematic diagram of the structure of an on-line defect identification system for overhead distribution lines provided by the present invention. Specific embodiments

[0039] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0040] The purpose of the present invention is to provide an on-line defect identification method and system for overhead distribution lines, which can improve the identification efficiency and accuracy.

[0041] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] Figure 1 Schematic diagram of the flow of an on-line defect identification method for overhead distribution lines provided by the present invention, as Figure 1 shown, an on-line defect identification method for overhead distribution lines provided by the present invention includes:

[0043] S101, obtaining a dataset of inspection images of the distribution network; the dataset of inspection images of the distribution network includes inspection images of the distribution network with different defect types; the defect types include: channel defects, conductor defects, insulator defects, metal oxide arrester defects, and electronic component defects;

[0044] According to the standard Q / GDW745-2012 "Classification Standard for Defects of Distribution Network Equipment", a list of artificial intelligence image recognition models for overhead distribution lines is formed, which is generally divided into 8 categories: overhead lines, pole-mounted vacuum switches, pole-mounted SF6 switches, pole-mounted disconnect switches, drop fuses, capacitors, metal oxide arresters, and distribution transformers.

[0045] According to the subordinate relationship of the equipment type, components, component types, and parts of the overhead distribution line, the visible light image pixel size is not less than 1920×1080, the image is clear and good, the brightness is uniform, the contrast is moderate, and various sample libraries are established.

[0046] The samples of the dataset of inspection images of the distribution network are as follows:

[0047] Channel defects (insufficient distance from trees) 8671 pieces, accounting for 43%;

[0048] There are 4,867 wire - related defects (including damaged wire insulation, broken and loose strands, and foreign objects), accounting for 24%;

[0049] There are 3,896 insulator - related defects (including breakage, fouling, and loose fixation), accounting for 20%;

[0050] There are 690 metal - oxide arrester - related defects, accounting for 3%;

[0051] There are 2,034 electronic - component - related defects (including pole - mounted vacuum switches, pole - mounted SF6 switches, pole - mounted disconnect switches, drop - out fuses, capacitors, and distribution transformers), accounting for 10%.

[0052] S102. Perform exposure correction processing on the distribution network inspection image in the distribution network inspection image dataset; the exposure correction processing includes: adopting a deep - learning method of single - model correction for over - exposure and under - exposure.

[0053] As a specific embodiment, given an 8 - bit sRGB input image I with exposure problems, generate an output image Y with correct exposure, and propose to correct the color first and then the details in sequence. And apply the Laplacian pyramid to the input image to obtain a multi - resolution representation as the network input.

[0054] S103. For the distribution network inspection images in the corrected distribution network inspection image dataset, perform noise reduction processing using wavelet packet functions;

[0055] Wavelet packet signal denoising provides a more complex and flexible analysis method because wavelet packet analysis decomposes both the low - frequency part and the high - frequency part of the upper layer simultaneously, having a more accurate local analysis ability.

[0056] When performing wavelet packet decomposition on a signal, multiple wavelet packet bases can be used. Usually, the best wavelet packet base, that is, the optimal base, is selected according to the requirements of the analyzed signal. The selection criterion for the optimal base is the entropy criterion.

[0057] The wavelet packet threshold denoising process is mainly divided into 4 steps as follows.

[0058] (1) Wavelet packet decomposition of the signal. Select a wavelet packet base and determine the required decomposition level, and then perform wavelet packet decomposition on the signal.

[0059] (2) Selection of the optimal wavelet packet base.

[0060] (3) Thresholding of wavelet packet decomposition coefficients. For each wavelet packet decomposition coefficient, select an appropriate threshold to perform threshold quantization processing on the coefficients after wavelet packet decomposition. That is, use the formula to determine the wavelet decomposition coefficients; is the wavelet decomposition coefficient obtained through the thresholding function, ω i,k is the decomposition coefficient of the wavelet, λ is the size of the threshold. When then at this time, this new threshold function will tend to the soft threshold function; and when θ → 0, then at this time, this new threshold function will tend to the hard threshold.

[0061] As Figure 2 shown, in the same coordinate system, the schematic diagrams of the three functions of hard threshold, soft threshold and new threshold are given respectively. In practical applications, different effects can be achieved by adjusting the value of the adjustable factor θ. This compromise method can not only make up for the constant deviation existing between i,k and ω, but also overcome the discontinuity problem existing in the hard thresholding function, and can better improve the signal-to-noise ratio of the signal and improve the problem of signal curve distortion.

[0062] (4) Wavelet packet reconstruction of the signal. Perform wavelet packet reconstruction on the low-frequency coefficients and the processed high-frequency coefficients. To use wavelet packets for denoising, the wavelet packet basis and the number of decomposition levels need to be selected first. Wavelets with good symmetry do not generate phase distortion, and wavelets with good regularity are easy to obtain smooth reconstructed signals.

[0063] As Figure 3 shown, the graphs of simulating these three functions in LabVIEW are presented. The middle graph is the graph of the new threshold function, and obviously its value range is between the hard thresholding and the soft thresholding.

[0064] As Figure 4 shown, these three threshold functions can all achieve the denoising effect as a whole, and the original signals are all restored to varying degrees. Especially for the first 400 sampling points, the denoising effects of these three threshold functions are not very obvious in comparison. However, in the part of the last 400 data, the comparison of the denoising effects is very obvious. Figure 4 In the left part of , there are many sawteeth and spikes on the signal curve after the sampling point 400, and the signal is not smooth enough. This is caused by the inherent disadvantage of the hard threshold method, that is, discontinuity; Figure 4 In the middle part of , between the sampling points 500 and 700, the signal curve obtained by the soft threshold method is significantly smoother than that of the hard threshold method, but there are still residual burr parts around the curve; Figure 4The right part in the middle shows the effect after denoising by the new threshold function. Obviously, the smoothness of its signal curve is the best. The most obvious comparison is between sampling points 500 and 800, which is sufficient to prove that the new threshold function designed in this paper can better make up for the deficiencies of traditional soft and hard thresholding methods, preserve the useful signals submerged in noise, remove the interference caused by useless noise, make the denoised signal smoother, and thus can better reconstruct the original signal.

[0065] The signal-to-noise ratio (Signal / Noise) and the root mean square error in engineering measurement, i.e., the standard error (RMSerror), are used for further verification. Through specific calculations, the magnitude of the values can be compared to further evaluate the noise reduction effects of these three algorithms.

[0066] The signal-to-noise ratio is usually denoted by SNR. Generally, the higher the signal-to-noise ratio, the greater the useful signal and the smaller the noise. Its calculation formula is:

[0067]

[0068] In the formula, nx(i) is the original signal, and xd(i) is the denoised signal.

[0069] The standard error is usually denoted by RMSE and can better reflect the extremely large and extremely small errors in signal measurement. Its calculation formula is:

[0070]

[0071] Through calculation, the noise reduction effects of the above three threshold algorithms are calculated respectively, and Table 1 is obtained.

[0072] Table 1

[0073]

[0074] It can be clearly seen from the data in Table 1 that the signal-to-noise ratios from small to large are the soft threshold algorithm, the hard threshold algorithm, and the new threshold algorithm, while the standard errors from small to large are the new threshold algorithm, the hard threshold algorithm, and the soft threshold algorithm. Thus, it can be known that whether in terms of the signal-to-noise ratio index or the standard error index, the new wavelet packet threshold algorithm designed in this paper is superior to the traditional soft and hard thresholding methods in terms of the signal noise reduction effect. Therefore, it is proved that this method is an improvement on the classical wavelet packet threshold method, can better improve the signal-to-noise ratio of the signal, and has practical significance.

[0075] Before S103, it also includes:

[0076] The distribution network inspection images in the corrected distribution network inspection image dataset are transformed from the spatial domain to the frequency domain by using Fourier transform.

[0077] S104. Perform data augmentation on the power grid inspection images in the denoised power grid inspection image dataset after noise reduction processing to determine the updated power grid inspection image dataset;

[0078] As a specific embodiment, the data augmentation processing includes: spatial domain enhancement algorithms and frequency domain enhancement algorithms. Spatial domain enhancement algorithms can be divided into two major categories. One is point operation, and this type of algorithm processes each pixel of the image point by point during enhancement, regardless of its surrounding pixels. The other is neighborhood operation, also known as template operation, and the template operation algorithm is related to adjacent pixels. The main spatial domain enhancement algorithms include: direct gray mapping, histogram transformation, linear filtering, non-linear filtering, and local enhancement, etc. According to the different spatial domain transformation functions, the processing can be divided into direct gray transformation and histogram-based transformation.

[0079] Then, important detail information of the image is enhanced by means of Fourier transform or other orthogonal transforms. The main functions of frequency domain image enhancement are to remove noise, enhance edges, improve contrast, improve the image display quality, and enrich hierarchical information, etc. The algorithm uses a frequency domain transformation function to transform the image processing space to the frequency domain for processing during image enhancement. After the enhancement processing, the enhanced result is then inverse-transformed to the original image space, which is the final enhanced result. The frequency domain transformation function can be selected from transformation functions such as Fourier transform and wavelet transform. The image pixels are analyzed after the image undergoes frequency domain transformation. The details and noise in the image correspond to the high-frequency components in the frequency domain, and the background and slowly changing parts of the image correspond to the low-frequency components in the frequency domain. During the transformation process, digital filtering methods can be used to change different frequency components according to different enhancement purposes to achieve image enhancement. According to the different selected filters in frequency domain enhancement, the frequency domain enhancement methods can be divided into: low-pass filter enhancement, high-pass filter enhancement, band-stop filter enhancement, and homomorphic filter enhancement.

[0080] Image enhancement technology uses a series of techniques to improve the visual effect of an image, or to convert the image into a form that is more suitable for human or machine analysis and processing. It does not follow the criterion of image fidelity, but selectively highlights certain information that is meaningful for human or machine analysis, suppresses useless information, and improves the use value of the image.

[0081] As another specific embodiment, the data augmentation processing includes: histogram equalization method. Histogram equalization is to correct the histogram of the original image through a transformation function into a uniform histogram, and then correct the original image according to the equalized histogram.

[0082] The histogram equalization method includes:

[0083] Using the formula P r (r k ) = n k / N Determine the histogram corresponding to the power distribution network inspection image in the denoised power distribution network inspection image dataset after noise reduction processing;

[0084] Use the formula Perform gray-scale transformation on the histogram using the cumulative distribution function;

[0085] Merge according to the gray-scale values of the histogram after gray-scale transformation. That is, merge the gray-scale values that are equal or approximately equal together.

[0086] S105, Use the updated power distribution network inspection image dataset and the Faster R-CNN algorithm to determine the overhead power distribution line defect recognition model;

[0087] S106, Use the overhead power distribution line defect recognition model to perform defect recognition on the power distribution network inspection images obtained in real time.

[0088] Figure 5 The structural schematic diagram of an on-line defect recognition system for overhead power distribution lines provided by the present invention is as Figure 5 shown. An on-line defect recognition system for overhead power distribution lines provided by the present invention includes:

[0089] A dataset acquisition module 201 for acquiring a power distribution network inspection image dataset; the power distribution network inspection image dataset includes power distribution network inspection images of different defect types; the defect types include: channel type defects, conductor type defects, insulator type defects, metal oxide arrester type defects, and electronic component type defects;

[0090] An exposure correction processing module 202 for performing exposure correction processing on the power distribution network inspection images in the power distribution network inspection image dataset;

[0091] A noise reduction processing module 203 for performing noise reduction processing on the power distribution network inspection images in the corrected power distribution network inspection image dataset using a wavelet packet function; wherein, use the formula to determine the wavelet decomposition coefficient; is the wavelet decomposition coefficient obtained through the thresholding function, ω i,k is the decomposition coefficient of the wavelet, and λ is the size of the threshold;

[0092] A data enhancement processing module 204 for performing data enhancement processing on the power distribution network inspection images in the denoised power distribution network inspection image dataset after noise reduction processing to determine an updated power distribution network inspection image dataset;

[0093] An identification model determination module 205 for using the updated power distribution network inspection image dataset and the Faster R-CNN algorithm to determine an overhead power distribution line defect recognition model;

[0094] A defect recognition module, which is used to recognize defects in the inspection images of the distribution network obtained in real time by using the overhead distribution line defect recognition model.

[0095] In order to execute the method corresponding to the first embodiment above to achieve the corresponding functions and technical effects, the present invention also provides an online defect recognition system for overhead distribution lines, including: at least one processor, at least one memory, and computer program instructions stored in the memory. When the computer program instructions are executed by the processor, the online defect recognition method for overhead distribution lines is implemented.

[0096] In this specification, each embodiment is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0097] In this article, specific examples are used to elaborate on the principle and implementation mode of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation mode and application scope. In summary, the content of this specification should not be construed as a limitation on the present invention.

Claims

1. An on-line defect identification method for overhead distribution lines, characterized in that including: obtaining a dataset of inspection images of a distribution network; the dataset of inspection images of the distribution network includes inspection images of the distribution network with different defect types; the defect types include: channel - type defects, conductor - type defects, insulator - type defects, metal - oxide arrester - type defects, and electronic - component - type defects; performing exposure correction processing on the inspection images of the distribution network in the dataset of inspection images of the distribution network; For the inspected images of the distribution network in the corrected dataset of inspected images of the distribution network, wavelet packet functions are used for noise reduction processing; among them, the formula is used to determine the wavelet decomposition coefficients; is the wavelet decomposition coefficient obtained through the thresholding function, ω i,k is the decomposition coefficient of the wavelet, and λ is the size of the threshold; performing data enhancement processing on the inspection images of the distribution network in the denoised dataset of inspection images of the distribution network after denoising processing to determine an updated dataset of inspection images of the distribution network; using the updated dataset of inspection images of the distribution network and the Faster R - CNN algorithm to determine a defect recognition model for overhead distribution lines; using the defect recognition model for overhead distribution lines to perform defect recognition on the inspection images of the distribution network obtained in real time.

2. The on-line identification method for overhead distribution line defects according to claim 1, characterized in that The exposure correction processing includes: adopting a deep - learning method of single - model correction for overexposure and underexposure.

3. The on-line defect identification method for overhead distribution lines according to claim 1, characterized in that Before performing denoising processing on the inspection images of the distribution network in the corrected dataset of inspection images of the distribution network using a wavelet packet function, it also includes: using Fourier transform to transform the inspection images of the distribution network in the corrected dataset of inspection images of the distribution network from the spatial domain to the frequency domain.

4. The on-line identification method for overhead distribution line defects according to claim 1, characterized in that, The data enhancement processing includes: a spatial - domain enhancement algorithm and a frequency - domain enhancement algorithm.

5. The on-line identification method for defects of overhead distribution lines according to claim 1, characterized in that, The data enhancement processing includes: a histogram equalization method.

6. The on-line defect identification method for overhead distribution lines according to claim 5, characterized in that, The histogram equalization method includes: determining the histogram corresponding to the inspection images of the distribution network in the denoised dataset of inspection images of the distribution network after denoising processing; performing gray - level transformation on the histogram using a cumulative distribution function; merging according to the gray - level values of the histogram after gray - level transformation.

7. An on-line defect identification system for overhead distribution lines, characterized in that, including: a dataset acquisition module for obtaining a dataset of inspection images of the distribution network; the dataset of inspection images of the distribution network includes inspection images of the distribution network with different defect types; the defect types include: channel - type defects, conductor - type defects, insulator - type defects, metal - oxide arrester - type defects, and electronic - component - type defects; an exposure correction processing module for performing exposure correction processing on the inspection images of the distribution network in the dataset of inspection images of the distribution network; A noise reduction processing module is used to perform noise reduction processing on the power distribution network inspection images in the corrected power distribution network inspection image dataset by using a wavelet packet function. Among them, the wavelet decomposition coefficients are determined by using the formula to determine the wavelet decomposition coefficients; is the wavelet decomposition coefficient obtained through the thresholding function, ω i,k is the decomposition coefficient of the wavelet, and λ is the size of the threshold; a data enhancement processing module for performing data enhancement processing on the inspection images of the distribution network in the denoised dataset of inspection images of the distribution network after denoising processing to determine an updated dataset of inspection images of the distribution network; a recognition model determination module for using the updated dataset of inspection images of the distribution network and the Faster R - CNN algorithm to determine a defect recognition model for overhead distribution lines; a defect recognition module for using the defect recognition model for overhead distribution lines to perform defect recognition on the inspection images of the distribution network obtained in real time.

8. An on-line defect identification system for overhead distribution lines, characterized in that including: at least one processor, at least one memory, and computer program instructions stored in the memory, which implement an online defect recognition method for overhead distribution lines according to any one of claims 1 - 6 when the computer program instructions are executed by the processor.

Images