A High-Precision Detection Method, System and Storage Medium for Transmission Line Defects

Abstract

This application relates to the technical field of equipment maintenance. This application provides a high-precision detection method, system, and storage medium for transmission line defects. The method includes obtaining original image data; adaptively denoising the obtained original image data to obtain denoised image data; for the denoised image data, training a convolutional neural network for identifying transmission line defects using a two-stage object detection neural network algorithm, and after inputting the denoised image data into the trained convolutional neural network, outputting a power object detection result. After obtaining the original inspection image data, adaptively denoise the inspection image in the transform domain and the spatial domain to obtain denoised image data with good performance, and then use a two-stage object detection neural network algorithm to train the identification of transmission line defects, providing high-quality inspection image data for the high-precision judgment of the transmission line object detection algorithm to improve the defect detection accuracy.

Description

Technical Field

[0001] This application relates to the technical field of equipment maintenance, and particularly to a high-precision detection method, system and storage medium for transmission line defects based on adaptive inspection image denoising. Background Art

[0002] At present, the inspection methods for power grid transmission lines mainly include manual methods, drone inspections and helicopter inspections. The inspection images of transmission equipment such as transmission towers, key transmission equipment, and tower identification signs taken under harsh weather conditions such as heavy fog, heavy rain, and strong winds are not clear and have high noise, making it difficult to directly apply them to intelligent transmission line inspection technologies. In addition, the noise in the inspection images is prone to safety hazards such as pin shedding and grading ring inclination that cannot be detected in time. Front-line line operation and maintenance personnel need to repeatedly inspect the transmission lines, increasing the work burden and line operation and maintenance costs. Therefore, it is very necessary to perform denoising on the inspection images.

[0003] The purpose of image denoising is to remove the noisy parts in the image through technical means. The inspection images of transmission lines in complex shooting environments are often contaminated by noise during the processes of acquisition, capture, and transmission. Noise is the main factor affecting image quality. According to the impact on the signal, noise is divided into additive and multiplicative. Assuming that the noise is related to the image, an additive noise model is adopted. According to the characteristics of the noise signal, noise can be divided into salt-and-pepper noise, Gaussian noise, Rayleigh noise, etc. Existing denoising methods mainly utilize the pixel information in the image and adopt methods such as local averaging, energy transformation, and weighted averaging, which have problems such as poor denoising effects and inability to achieve adaptive image denoising. Summary of the Invention

[0004] Aiming at the problems of noise, unclearness, and jitter in transmission line inspection images, the main purpose of this application is to provide a high-precision detection method, device, equipment and computer-readable storage medium for transmission line defects based on adaptive inspection image denoising. By adaptively denoising the inspection image data in the transform domain and the spatial domain, and then performing training, high-quality inspection image data is provided for the high-precision judgment of the transmission line target detection algorithm, so as to improve the defect detection accuracy.

[0005] To achieve the above purpose, this application provides a high-precision detection method for transmission line defects, which adaptively denoises the inspection image data in the transform domain and the spatial domain, and then performs training. The high-precision detection method for transmission line defects includes:

[0006] Obtain the original image data;

[0007] Adaptively denoise the obtained original image data to obtain denoised image data;

[0008] For the denoised image data, a two-stage object detection neural network algorithm is used to train a convolutional neural network for identifying transmission line defects. After training, when the denoised image data is input into the convolutional neural network, the power object detection result is output.

[0009] Optionally, the original image data is video image data obtained by helicopter inspection.

[0010] Optionally, adaptive noise removal is performed on the acquired original image data based on a combination of transform domain and spatial domain methods. According to the image quality evaluation rules and the denoising algorithm, the picture noise parameter information is obtained, and the inspection videos and image data in the original image data are preprocessed to obtain denoised image video data.

[0011] Optionally, the adaptive noise removal method based on a combination of transform domain and spatial domain methods includes:

[0012] Noise parameter evaluation based on blocks and filtering, adaptively estimating the parameters of the mixture Gaussian noise; using the estimated noise parameters for image denoising, and fusing the data of multiple denoised images to obtain a denoised image.

[0013] Optionally, the noise parameter evaluation includes:

[0014] Obtain the input noise image, and perform high-pass filtering on the original image of the noise image to obtain a high-frequency image;

[0015] Separate the high-frequency image into several non-overlapping blocks, and statistically calculate the variance peak value of the segmented image;

[0016] Sum the absolute values of the pixel values of the high-frequency image in the blocks near the peak and take the average to obtain two regional noise variance estimation values.

[0017] Optionally, the data fusion of multiple denoised images includes:

[0018] Use the two regional noise variance estimation values obtained by noise parameter evaluation to denoise the input sound image to obtain a smoothed image and a noise residue image;

[0019] Perform high-pass filtering on the noise residue image to obtain the noise level;

[0020] Compare the noise level with a preset threshold. If the pixel point noise level value is greater than the preset threshold, output the pixel value to select the smoothed image, otherwise output the pixel value to select the noise residue image;

[0021] Complete the data fusion to obtain the output denoised image.

[0022] Optionally, the high-precision detection method for transmission line defects further includes using a Faster-RCNN deep learning network to track and detect the inspection targets of transmission lines. The input end of the Faster-RCNN deep learning network is the original inspection video image of the transmission line, and the output end is the power target detection result.

[0023] Optionally, in the Faster-RCNN deep learning network, the target region extraction layer, the region screening layer, and the target analysis layer are developed based on the deep learning network layer and are connected to the existing convolutional layer and fully connected layer. The target region extraction layer and the region screening layer train the network layer parameters according to the marked result region, and the convolutional layer and the target analysis layer are adjusted according to the marked result loss function.

[0024] In addition, to achieve the above object, the present application further provides a high-precision detection system for transmission line defects. The high-precision detection system for transmission line defects includes: a data acquisition module for acquiring original image data; an image denoising module for adaptively denoising the acquired original image data to obtain denoised image data; and a defect recognition module for, for the denoised image data, training a convolutional neural network for identifying transmission line defects by using a two-stage object detection neural network algorithm, and outputting a power target detection result after inputting the denoised image data into the trained convolutional neural network.

[0025] In addition, to achieve the above object, the present application further provides a high-precision detection device for transmission line defects. The high-precision detection device for transmission line defects includes a processor, a memory, and a high-precision detection program for transmission line defects stored on the memory and executable by the processor. When the high-precision detection program for transmission line defects is executed by the processor, the steps of the high-precision detection method for transmission line defects as described above are implemented.

[0026] In addition, to achieve the above object, the present application further provides a computer-readable storage medium. A high-precision detection program for transmission line defects is stored on the computer-readable storage medium. When the high-precision detection program for transmission line defects is executed by a processor, the steps of the high-precision detection method for transmission line defects as described above are implemented.

[0027] The present application provides a high-precision detection method for transmission line defects. The method acquires original image data; performs image adaptive denoising to obtain well-performing denoised image data; and for the denoised image data, uses a two-stage object detection neural network algorithm to train the identification of transmission line defects. After acquiring the original inspection image data, the inspection image is adaptively denoised in the transform domain and the spatial domain to obtain well-performing denoised image data, and then a two-stage object detection neural network algorithm is used to train the identification of transmission line defects.

[0028] These aspects or other aspects of the present application will be more clearly understood in the following description of the embodiments. It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. Brief Description of the Drawings

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. In the drawings:

[0030] Figure 1 It is a flowchart of the high-precision detection method for transmission line defects of the present application;

[0031] Figure 2 It is a flowchart of the noise parameter evaluation in the high-precision detection method for transmission line defects of the present application;

[0032] Figure 3 It is a flowchart of the noise parameter estimation in the high-precision detection method for transmission line defects of the present application;

[0033] Figure 4 It is a flowchart of the data fusion in the high-precision detection method for transmission line defects of the present application;

[0034] Figure 5 It is a flowchart of the image denoising and fusion process in the high-precision detection method for transmission line defects of the present application;

[0035] Figure 6 It is a schematic block diagram of the structure of the Faster R-CNN model in the high-precision detection method for transmission line defects of the present application;

[0036] Figure 7 It is a schematic diagram of the structure of the RPN in the high-precision detection method for transmission line defects of the present application;

[0037] Figure 8 It is a schematic diagram of the structure of the Classification part network in the high-precision detection method for transmission line defects of the present application;

[0038] Figure 9 It is a system block diagram of the high-precision detection system for transmission line defects of the present application;

[0039] Figure 10 It is a schematic block diagram of the structure of the high-precision detection device for transmission line defects of the present application.

[0040] The realization of the purpose, functional features and advantages of the present application will be further described with reference to the embodiments and the drawings. Detailed Embodiments

[0041] Next, in combination with the accompanying drawings and specific embodiments, the present application will be further described. It should be noted that, on the premise of no conflict, the following-described embodiments or technical features can be arbitrarily combined to form new embodiments.

[0042] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0043] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0044] The flowchart shown in the accompanying drawings is only an example, and does not necessarily include all contents and operations / steps, nor does it necessarily need to be executed in the described order. For example, some operations / steps can also be decomposed, combined or partially merged, so the actual execution order may change according to the actual situation.

[0045] It should be understood that the terms used in the specification of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the specification of the present application and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include the plural forms.

[0046] Next, in combination with the accompanying drawings, some embodiments of the present application will be described in detail. Without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0047] The embodiments of the present application provide a high-precision detection method, system and storage medium for transmission line defects. Different from the previous methods of using pixel information in images and adopting methods such as local averaging, energy transformation, and weighted averaging for denoising, aiming at problems such as noise, blurriness, and jitter in transmission line inspection images. The purpose of the present application is to study an adaptive video and image denoising technology based on the combination of the transform domain and the spatial domain method, obtain the picture noise parameter information according to the image quality evaluation rules and the denoising algorithm, preprocess the inspection video and image data, remove the influence of coding noise on the data, and obtain high-quality denoised image and video data. The Faster-RCNN deep learning network is used to realize the tracking and detection of transmission line inspection targets, and the RPN is directly used to generate detection frames, improving the generation speed of the detection frames for identifying transmission line defects.

[0048] In some embodiments, the high-precision detection method for transmission line defects can be applied to a high-precision detection device for transmission line defects. The high-precision detection device for transmission line defects can be a device with display and processing functions such as a PC, a portable computer, a mobile terminal, etc., but is not limited thereto.

[0049] Please refer to Figure 1 , Figure 1 which is a schematic flowchart of the first embodiment of the high-precision detection method for transmission line defects of the present application. In the embodiments of the present application, the high-precision detection method for transmission line defects includes the following steps:

[0050] Step S10, obtain original image data.

[0051] In some embodiments, the original image data is video image data obtained by helicopter inspection.

[0052] Step S20, adaptively denoise the obtained original image data to obtain denoised image data.

[0053] In some embodiments, the image is adaptively denoised to obtain denoised image data with good performance.

[0054] Step S30, for the denoised image data, use a two-stage object detection neural network algorithm to train a convolutional neural network for identifying transmission line defects. After the trained convolutional neural network inputs the denoised image data, it outputs a power object detection result.

[0055] In some embodiments, for the denoised image data, use a two-stage object detection neural network algorithm to train for identifying transmission line defects.

[0056] A high-precision detection method for transmission line defects provided by the embodiments of the present application, after obtaining the original inspection image data, adaptively denoises the inspection image through the transform domain and the spatial domain to obtain denoised image data with good performance, and then uses a two-stage object detection neural network algorithm to train for identifying transmission line defects.

[0057] Based on the above Figure 1 shown embodiments, in an embodiment of the present application, in step S20, the high-precision detection method for transmission line defects further includes:

[0058] Perform adaptive noise removal on the obtained original image data based on the combination of the transform domain and the spatial domain method, obtain picture noise parameter information according to the image quality evaluation rule and the denoising algorithm, preprocess the inspection video and image data in the original image data, and obtain denoised image video data.

[0059] Due to the influence of imaging equipment and external environmental noise interference during the digitization and transmission of helicopter inspection video image data, noisy images often occur. At the same time, different regions of the image are usually contaminated by different intensities of noise. This project uses a comprehensive denoising method combining the transform domain and the spatial domain to process the helicopter inspection video image data and eliminate the influence of background noise on the inspection image defect recognition task.

[0060] In the embodiment of this application, an adaptive noise removal method combining the transform domain and the spatial domain methods is adopted. This method first uses block-based and filtering-based noise parameter estimation to adaptively estimate the parameters of the mixture Gaussian noise, and then uses the estimated noise parameters for image denoising. Multiple denoised images are simply data-fused to finally obtain a denoised image with good performance.

[0061] Among them, the adaptive noise removal method combining the transform domain and the spatial domain methods includes:

[0062] Evaluating the noise parameters based on blocks and filtering to adaptively estimate the parameters of the mixture Gaussian noise; using the estimated noise parameters for image denoising, and data-fusing multiple denoised images to obtain a denoised image.

[0063] Among them. Please refer to Figure 2 As shown, the evaluation of the noise parameters includes:

[0064] S201. Obtain the input noise image, and perform high-pass filtering on the original image of the noise image to obtain a high-frequency image;

[0065] S202. Divide the high-frequency image into several non-overlapping blocks, and statistically calculate the variance peak value of the image after block division;

[0066] S203. Sum the absolute values of the pixel values of the high-frequency image in the blocks near the peak and take the average to obtain two regional noise variance estimation values.

[0067] Refer to Figure 3 As shown, in the first step of noise parameter estimation, first input the noise image F to obtain the high-frequency image G. Then divide the high-frequency image into several non-overlapping blocks. In the third step, statistically calculate the variances of the image after block division to obtain two variance peak values, sum the absolute values of the pixel values of the high-frequency image in the area near the peak and take the average to obtain the variance estimation value of the regional noise.

[0068] Among them, please refer to Figure 4 As shown, the data fusion of multiple denoised images includes:

[0069] S211. Use the two regional noise variance estimation values obtained from the noise parameter evaluation to denoise the input sound image to obtain a smoothed image and a noise residue image;

[0070] S212. Perform high-pass filtering on the noise residual image to obtain the noise level.

[0071] S213. Compare the noise level with a preset threshold. If the noise level value of a pixel is greater than the preset threshold, output the pixel value to select the smoothed image; otherwise, output the pixel value to select the noise residual image. Complete data fusion to obtain the output denoised image.

[0072] As Figure 5 shown, first, use the two variance estimation values obtained by the noise estimation part to perform the first denoising operation on the noise image F, obtaining an over-smoothed image with better denoising effect for large-intensity noise and a noise residual image with better denoising effect for small-intensity noise. In the second step, perform high-pass filtering on the noise residual image to obtain the noise level. Compare the noise level with the threshold. If the noise level value of a pixel is greater than the threshold, output the pixel value to select the smoothed image; otherwise, output the pixel value to select the noise residual image. Complete data fusion to obtain the output denoised image.

[0073] In some embodiments, the high-precision detection method for transmission line defects further includes using the Faster-RCNN deep learning network to track and detect the inspection targets of transmission lines. The input end of the Faster-RCNN deep learning network is the original transmission line inspection video image, and the output end is the power target detection result.

[0074] In this embodiment, in the deep learning architecture of the Faster-RCNN deep learning network, the input end is the original transmission line inspection video image, and the output end is the power target detection result. Among them, the input image does not go through empirical model analysis and directly extracts high-dimensional feature maps through the deep convolutional layer. The output feature maps are used to support three functions:

[0075] 1) The target region extraction layer extracts the potential target regions in the feature map according to the feature map vector and outputs the target regions.

[0076] 2) The region screening layer extracts and reads the target region extraction layer output and the high-dimensional feature map, sorts them according to the probability of the target object appearing in the extracted regions, and outputs the high-probability feature map regions.

[0077] 3) The target analysis layer analyzes and classifies the regional targets according to the output of the region screening layer, and can classify and mark pixels and regions according to the classification requirements.

[0078] In the Faster-RCNN deep learning network, the target region extraction layer, region screening layer, and target analysis layer are developed based on the deep learning network layer and are connected to the existing convolutional layer and fully connected layer. The target region extraction layer and region screening layer train the network layer parameters according to the marked result region, and the convolutional layer and target analysis layer are adjusted according to the marked result loss function.

[0079] In the embodiment of the present application, the target region extraction layer, region screening layer, and target analysis layer in the deep learning network are developed based on the deep learning network layer and are connected to the existing convolutional layer and fully connected layer. The end-to-end deep learning network architecture is trained through a large number of images, where the target region extraction layer and screening layer train the network layer parameters according to the marked result region, and the convolutional layer and target analysis layer are adjusted according to the marked result loss function.

[0080] In the convolutional domain, see Figure 6 As shown, for all convolutional layers: kernel_size = 3, pad = 1. All convolutions are padded (pad = 1, that is, a circle of 0s is filled), resulting in the original image (M x N) becoming (M + 2) x (N + 2) in size. After performing a 3x3 convolution, the output is M x N. This setting causes the conv layer in the Conv layers not to change the size of the input and output matrices. Similarly, for all pooling layers, kernel_size = 2, stride = 2. In this way, each M x N matrix passing through the pooling layer will become (M / 2) * (N / 2) in size.

[0081] In summary, in the entire Conv layers, the conv and relu layers do not change the input and output sizes, and only the pooling layer makes the output length and width become 1 / 2 of the input. In conclusion, an M x N-sized matrix is fixed to become (M / 16) x (N / 16) after passing through the Conv layers. In this way, the feature maps generated in the convolutional domain can all correspond to the original image, reducing the possibility of target loss or omission.

[0082] Since the classic object detection methods for generating detection boxes are very time-consuming. For example, OpenCV adaboost uses a sliding window + image pyramid to generate detection boxes; or RCNN uses the SS (Selective Search) method to generate detection boxes. However, the Faster R-CNN of the present application abandons the traditional sliding window and SS methods and directly uses the RPN to generate detection boxes. This is also a huge advantage of the Faster R-CNN, which can greatly improve the generation speed of detection boxes.

[0083] In the embodiment of the present application, the specific structure of the RPN network is as Figure 7As shown in the figure. It can be seen that the RPN network is actually divided into two lines. The upper line classifies the anchors through softmax to obtain foreground and background (the detection target is the foreground), and the lower line is used to calculate the bounding box regression offsets for the anchors to obtain accurate proposals. The final Proposal layer is responsible for integrating the foreground anchors and the bounding box regression offsets to obtain proposals, and at the same time removing the proposals that are too small or exceed the boundaries. When the whole network reaches the Proposal Layer, the dynamic target localization is basically completed.

[0084] In the embodiments of the present application, in the Classification part of Faster R-CNN, using the obtained proposal feature maps, each proposal is calculated through the full connect layer and softmax to determine which specific category it belongs to (such as person, car, TV, etc.), and the cls_prob probability vector is output; at the same time, the bounding box regression is used again to obtain the position offset bbox_pred of each proposal, which is used to regress a more accurate target detection box, so as to achieve accurate recognition and detection of multiple targets in a moving state. The network structure of the Classification part is as Figure 8 shown.

[0085] In some embodiments of the present application, due to factors such as meteorological conditions or air pollution during aerial inspection, there are many haze days. After the images obtained outdoors are absorbed and scattered by particles such as water droplets and dust in the air, degraded images are formed. Due to the interference, the contrast and resolution of the images are both poor, which affects subsequent work such as image analysis and understanding.

[0086] In the embodiments of the present application, it also includes image dehazing. The main task of image dehazing technology is to remove the influence of weather factors on image quality, which is mainly divided into two categories: foggy image restoration and foggy image enhancement. The enhancement method of foggy images starts from improving the image contrast, which is just the application of ordinary enhancement algorithms to foggy images. The purpose is that after the image contrast is enhanced, it is more suitable for the visual habits of human eyes and the input habits of machine vision, rather than real dehazing, and there are often problems such as edge information loss or over-saturation.

[0087] In the embodiments of the present application, the foggy image restoration method is based on the physical process of foggy image degradation, establishes a model describing image degradation, and through relevant algorithms, combines the degradation model, uses the dark channel prior algorithm to obtain the relevant parameters of the foggy image imaging model, and then inversely deduces the real scene information and inversely calculates the image degradation process, so as to restore the foggy blur caused by fog. Such a processing method is truly defogging from a physical sense, restoring the true nature of the image, and generally there will be no information loss. The general process of defogging foggy images is to clear the foggy degraded images based on the image restoration method of the physical model.

[0088] In some embodiments, video images are defogged based on the dark channel priority algorithm. Among them, the dark channel priority defogging method is a defogging method based on the dark channel priority rule of outdoor natural scenes. The dark channel priority rule holds that in an outdoor natural scene image without fog, in the non-sky area, some pixels have at least one color channel with a very low value.

[0089] When using the dark channel priority algorithm to defog video images, the following steps are included:

[0090] In the first step, input the foggy image. According to the definition of the dark channel, calculate the dark channel feature map of the whole image, that is, take the minimum value within the window Ω in the RGB three channels centered on the pixel point x, and then take the minimum value of the three channels as the value of the dark channel of the pixel point x. The dark channel priority rule believes that after the dark channel priority processing, the dark channel value of the image J will approach zero. If there are a large number of pixels with relatively high brightness in the dark channel image, then these brightness values should come from the fog in the air or the sky.

[0091] In the second step, sample one-thousandth of the total number of pixels in the dark channel that are the brightest, and record the coordinates of these pixel points (x, y). According to the coordinates of these points, find the corresponding pixel points in the foggy image I, and calculate the average value of the RGB channels of each point, that is, the global atmospheric light value.

[0092] In the third step, according to the atmospheric light model of the foggy image, solve the transmittance map from the foggy image and the atmospheric light value. The higher the brightness in the transmittance map indicates that the passing property of the scene color here is better, or it can be understood that it is closer to the viewpoint.

[0093] In the fourth step, after refining the transmittance, solve the defogged image J from the dark channel priority degradation formula.

[0094] Therefore, the high-precision detection method for transmission line defects in this application is based on an adaptive video and image denoising technology that combines transform domain and spatial domain methods. According to the image quality evaluation rules and denoising algorithms, the picture noise parameter information is obtained, and the inspection video and image data are preprocessed to remove the influence of coding noise on the data, and high-quality denoised image and video data are obtained. The Faster-RCNN deep learning network is used to implement the tracking and detection of transmission line inspection targets, and the RPN is directly used to generate detection frames, which improves the generation speed of detection frames for identifying transmission line defects.

[0095] In addition, the embodiment of this application also provides a high-precision detection system for transmission line defects.

[0096] Referring to Figure 9 , Figure 9 is a schematic diagram of the functional modules of the first embodiment of the high-precision detection system for transmission line defects in this application.

[0097] In the embodiment of this application, the high-precision detection system for transmission line defects includes:

[0098] A data acquisition module 10, which is used to acquire original image data;

[0099] An image denoising module 20, which is used to calculate the driving factors of the implied carbon emissions in the target area according to the input-output data, where the driving factors include the final demand of the industry, the complete consumption coefficient, and the direct carbon emission intensity; and

[0100] A defect identification module 30, which is used to calculate the implied carbon emissions caused by the final demand data according to the driving factors.

[0101] Among them, each module in the above high-precision detection system for transmission line defects corresponds to each step in the embodiment of the above high-precision detection method for transmission line defects, and its functions and implementation processes will not be elaborated here one by one.

[0102] The high-precision detection method and device for transmission line defects in this application can be implemented in the form of a computer program, and this computer program can run on a high-precision detection device for transmission line defects as shown in Figure 10 .

[0103] Please refer to Figure 10 , Figure 10 is a schematic block diagram of the structure of the high-precision detection device for transmission line defects in this application.

[0104] Please refer to Figure 10 , this high-precision detection device for transmission line defects includes a processor and a memory connected through a system bus, where the memory can include a non-volatile storage medium and an internal memory.

[0105] The processor is used to provide computing and control capabilities to support the operation of the entire high-precision detection device for transmission line defects.

[0106] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can be made to execute any high-precision detection method for transmission line defects.

[0107] It should be understood that the processor can be a central processing unit (CPU), and this processor can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor can be a microprocessor or this processor can also be any conventional processor, etc.

[0108] Among them, the processor is used to run the computer program stored in the memory to implement the various embodiments of the high-precision detection method for transmission line defects in this application, which will not be elaborated here.

[0109] In addition, the embodiments of this application also provide a computer-readable storage medium.

[0110] The high-precision detection program for transmission line defects is stored on the computer-readable storage medium of this application. When the high-precision detection program for transmission line defects is executed by the processor, the steps of the high-precision detection method for transmission line defects as described above are implemented.

[0111] Among them, the method implemented when the high-precision detection program for transmission line defects is executed can refer to the various embodiments of the high-precision detection method for transmission line defects in this application, which will not be elaborated here.

[0112] It should be noted that in this article, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or system. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, article or system including that element.

[0113] The serial numbers of the embodiments of the present application above are only for description and do not represent the superiority or inferiority of the embodiments.

[0114] The present application can be used in numerous general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multi-processor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and so on. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in a distributed computing environment where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0115] Through the description of the above embodiments, those skilled in the art can clearly understand that the above embodiment methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above and includes several instructions to enable a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in various embodiments of the present application.

[0116] The present application provides a high-precision detection method, system, and storage medium for transmission line defects. The method obtains original image data; performs image adaptive denoising to obtain well-performing denoised image data; and for the denoised image data, uses a two-stage object detection neural network algorithm to train for the recognition of transmission line defects. After obtaining the original inspection image data, the inspection image is adaptively denoised in the transform domain and the spatial domain to obtain well-performing denoised image data. Then, a two-stage object detection neural network algorithm is used to train for the recognition of transmission line defects. It can obtain picture noise parameter information according to the image quality evaluation rule and the denoising algorithm, preprocess the inspection video and image data, remove the influence of coding noise on the data, and obtain high-quality denoised image and video data. The Faster-RCNN deep learning network is used to implement the tracking and detection of transmission line inspection targets, and the RPN is directly used to generate detection frames, which improves the generation speed of the detection frames for the recognition of transmission line defects.

[0117] The above are only the preferred embodiments of the present application, and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be similarly included in the patent protection scope of the present application.

Claims

1. A high-precision detection method for transmission line defects, characterized in that The high-precision detection method for transmission line defects includes the following steps: Obtain the original image data; Adaptive denoising of the obtained original image data to obtain denoised image data; For the denoised image data, use a two-stage object detection neural network algorithm to train a convolutional neural network for identifying transmission line defects. After training, input the denoised image data into the convolutional neural network and output the power object detection result; Adaptive denoising of the obtained original image data to obtain denoised image data, including: adaptive noise removal of the obtained original image data based on the combination of the transform domain and the spatial domain method, obtaining the picture noise parameter information according to the image quality evaluation rule and the denoising algorithm, preprocessing the inspection video and image data in the original image data, and obtaining the denoised image video data; The adaptive noise removal method based on the combination of the transform domain and the spatial domain method includes: noise parameter evaluation based on blocks and filtering, adaptively estimating the parameters of the mixture Gaussian noise; using the estimated noise parameters for image denoising, and performing data fusion on multiple denoised images to obtain a denoised image; The noise parameter evaluation includes: obtaining the input noise image, performing high-pass filtering on the original image data to obtain a high-frequency image; dividing the high-frequency image into several non-overlapping blocks, and statistically obtaining two variance peaks of the variance of the segmented image; summing the absolute values of the pixel values of the high-frequency image in the blocks near the peaks and taking the average to obtain two regional noise variance estimation values; where the pixel values of the high-frequency image in the blocks near the peaks are the pixel values of the high-frequency image in the peak neighborhood.

2. The high-precision detection method for transmission line defects according to claim 1, wherein The original image data is video image data obtained by helicopter inspection.

3. The high-precision detection method for transmission line defects according to claim 1, characterized in that The data fusion of the multiple denoised images includes: Using the two regional noise variance estimation values obtained by noise parameter evaluation to denoise the input noise image to obtain a smoothed image and a noise residue image; Performing high-pass filtering on the noise residue image to obtain the noise degree; Comparing the noise degree with a preset threshold. If the pixel point noise degree value is greater than the preset threshold, output the pixel value to select the smoothed image, otherwise output the pixel value to select the noise residue image; Complete the data fusion to obtain the output denoised image.

4. The high-precision detection method for transmission line defects according to claim 1, wherein, The high-precision detection method for transmission line defects also includes using the Faster-RCNN deep learning network to track and detect the inspection targets of the transmission line. The input end of the Faster-RCNN deep learning network is the original transmission line inspection video image, and the output end is the power object detection result.

5. The high-precision detection method for transmission line defects according to claim 4, characterized in that In the Faster-RCNN deep learning network, the target region extraction layer, the region screening layer, and the target analysis layer are developed based on the deep learning network layer, and are connected to the existing convolutional layer and fully connected layer. The target region extraction layer and the region screening layer train the network layer parameters according to the marked result region, and the convolutional layer and the target analysis layer are adjusted according to the marked result loss function.

6. A high-precision detection system for transmission line defects, characterized in that, The high-precision detection system for transmission line defects includes: A data acquisition module for obtaining the original image data; An image denoising module for adaptively denoising the obtained original image data to obtain denoised image data; A defect recognition module, which is used to train a convolutional neural network for power transmission line defect recognition by using a two-stage object detection neural network algorithm for the denoised image data. After the training, when the denoised image data is input into the convolutional neural network, a power object detection result is output; An image denoising module, specifically used for: Performing adaptive noise removal on the acquired original image data based on a combination of transform domain and spatial domain methods, obtaining picture noise parameter information according to the image quality evaluation rule and the denoising algorithm, preprocessing the inspection video and image data in the original image data, and obtaining denoised image video data; The adaptive noise removal method based on a combination of transform domain and spatial domain methods includes: noise parameter evaluation based on blocks and filtering, adaptively estimating the parameters of mixed Gaussian noise; using the estimated noise parameters for image denoising, and performing data fusion on multiple denoised images to obtain a denoised image; The noise parameter evaluation includes: acquiring an input noise image, and performing high-pass filtering on the original image data to obtain a high-frequency image; dividing the high-frequency image into several non-overlapping blocks, and statistically obtaining two variance peaks of the variance of the image after block division; performing absolute value summation and averaging on the pixel values of the high-frequency image of the blocks near the peaks to obtain two regional noise variance estimation values; wherein, the pixel values of the high-frequency image of the blocks near the peaks are the pixel values of the high-frequency image within the peak neighborhood.

7. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a high-precision detection program for power transmission line defects is stored on the computer-readable storage medium. When the high-precision detection program for power transmission line defects is executed by a processor, the steps of the high-precision detection method for power transmission line defects according to any one of claims 1 to 5 are implemented.

Images