CN115546664A

CN115546664A - Cascaded network-based insulator self-explosion detection method and system

Info

Publication number: CN115546664A
Application number: CN202211207805.4A
Authority: CN
Inventors: 魏鑫; 丁亚洲; 吏军平; 戴明松; 王小丽; 王新安; 田其; 黄先绪; 冯发杰; 王汉广
Original assignee: PowerChina Hubei Electric Engineering Co Ltd
Current assignee: PowerChina Hubei Electric Engineering Co Ltd
Priority date: 2022-09-30
Filing date: 2022-09-30
Publication date: 2022-12-30

Abstract

The invention discloses a cascade network-based insulator self-explosion detection method and a system, wherein the method comprises the following steps: a cascade network comprising a target detection model and an instance segmentation model is constructed and trained in advance; then, acquiring aerial insulating subimages of the area to be monitored, and preprocessing the aerial insulating subimages; then, detecting the preprocessed aerial insulator image by using a target detection model of the cascade network to obtain a detection result image of the insulator; cutting the detection result image according to the positioning boundary frame, and taking the cut image as an interested area; and finally, segmenting the insulator missing fault of the region of interest by using a Mask R-CNN example segmentation network, and obtaining a detection result. The invention can solve the problems of complex background, image distortion and low signal-to-noise ratio, and improve the detection precision.

Description

Cascading network-based insulator self-explosion detection method and system

Technical Field

The invention relates to the technical field of power line detection, in particular to a cascading network-based insulator self-explosion detection method and system.

Background

The electric power infrastructure, the traffic infrastructure, the water conservancy infrastructure and the like form a material foundation for the life and industrial production of residents in our current society together. In electric power infrastructure, an overhead transmission line is a main long-distance power transmission mode in China. In order to ensure the normal and safe operation of the overhead transmission line, various protection devices need to be configured on the electric power tower of the overhead transmission line, and the insulator is an important protection device on the electric power tower and is arranged on the electric power tower of the overhead transmission line to play the roles of electrical insulation and mechanical fixation. The insulator can be divided into three main types of insulators, namely a glass insulator, a ceramic insulator and a composite insulator according to different manufacturing materials, the insulators made of different materials have different physicochemical properties, and the corresponding insulators are selected according to specific application scenes and tasks in actual use. Under the influence of outdoor biochemical environment changes such as rainwater, thunder and lightning, the glass insulator can crack and be electrically broken in actual use, and then is self-exploded into small fragments, so that a certain ring of the insulator string is lost, the protection function of the power line is lost, and the safety of the power transmission line is threatened. In the electric power inspection work, the self-explosion condition of the insulator needs to be counted by workers, so that the safety accident of the electric power circuit is timely found and maintained, and the larger harm is generated.

Traditional insulator spontaneous explosion detection work mainly relies on the manual work observation in the electric power patrols and examines, and the staff need climb the observation equipment such as electricity tower or use the telescope on the spot and find the spontaneous explosion phenomenon that exists in the insulator cluster, and the danger coefficient is great consumes more manpower and time cost simultaneously.

Along with the maturity of unmanned aerial vehicle technique and the development of computer technology, electric power is patrolled and examined work and is utilized unmanned aerial vehicle to carry on optical camera, plans unmanned aerial vehicle cruise route in advance and shoots position and angle, acquires insulator equipment optical image data on the power line automatically. The insulator image data acquired by the unmanned aerial vehicle is analyzed and processed, and the insulator spontaneous explosion phenomenon is rapidly, accurately and automatically identified. However, the method is influenced by factors such as field natural environment change, seasonal climate and unmanned aerial vehicle shooting angle, the background difference of the insulator images shot by the unmanned aerial vehicle is large, the shapes, sizes and hues of the insulator strings are different, and when the traditional image processing method is used for detecting the self-explosion of the insulator, an independent filter needs to be arranged for separating the image insulator strings from the background according to various conditions, so that the self-explosion detection work is carried out, and the complicated image change of the insulator strings of the unmanned aerial vehicle is difficult to deal with.

Disclosure of Invention

The invention provides a cascade network-based insulator self-explosion detection method and system, which are used for solving or at least partially solving the technical problem of poor detection effect in the prior art.

In order to solve the above technical problem, a first aspect of the present invention provides a method for detecting self-explosion of an insulator based on a cascade network, including:

a cascade network comprising a target detection model and an instance segmentation model is constructed and trained in advance, wherein the target detection model adopts a network structure of a YOLOV5 algorithm, and the instance segmentation model adopts a Mask R-CNN instance segmentation network;

acquiring aerial insulator images of a region to be monitored, and preprocessing the aerial insulator images;

detecting the preprocessed aerial insulator image by using a target detection model of the cascade network to obtain a detection result image of the insulator, wherein the detection result image comprises position information of the insulator string, and the position information is a positioning boundary frame;

cutting the detection result image according to the positioning boundary frame, and taking the cut image as an interested area;

and segmenting the insulator missing fault of the region of interest by using a Mask R-CNN example segmentation network, and obtaining a detection result.

In one embodiment, the Mask R-CNN instance segmentation network comprises a backbone network, a feature pyramid module, a candidate frame generation network and a Mask generation module.

In one embodiment, the feature pyramid module is a FPN, and includes a bottom-up pyramid and a top-down pyramid, and then the two pyramids are connected in the horizontal direction to fuse features with shallow high resolution and deep rich semantic information.

In one implementation, the segmentation of the insulator missing fault of the region of interest by using a Mask R-CNN instance segmentation network comprises two stages, wherein the first stage is to scan an input region of interest image and generate a region proposal, the region proposal represents a region which possibly contains a target, and the second stage branches the category and the boundary box offset of a predicted image through a convolution network and outputs an object pixel level Mask for each region of interest through another Mask prediction branch.

Based on the same inventive concept, the second aspect of the present invention provides a cascade network based insulator self-explosion detection system, comprising:

the network construction module is used for constructing and training a cascade network comprising a target detection model and an example segmentation model in advance, wherein the target detection model adopts a network structure of a YOLOV5 algorithm, and the example segmentation model adopts a Mask R-CNN example segmentation network;

the image preprocessing module is used for acquiring aerial insulating subimages of the area to be monitored and preprocessing the aerial insulating subimages;

the target detection module is used for detecting the preprocessed aerial insulator image by using a target detection model of the cascade network to obtain a detection result image of the insulator, wherein the detection result image comprises position information of the insulator string, and the position information is a positioning boundary frame;

the image cutting module is used for cutting the detection result image according to the positioning boundary box and taking the cut image as an interested area;

and the example segmentation module is used for segmenting the insulator missing fault of the region of interest by utilizing a Mask R-CNN example segmentation network and obtaining a detection result.

Based on the same inventive concept, a third aspect of the present invention provides a computer-readable storage medium, on which a computer program is stored, which when executed, performs the method of the first aspect.

Based on the same inventive concept, a fourth aspect of the present invention provides a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method of the first aspect when executing the program.

Compared with the prior art, the invention has the advantages and beneficial technical effects as follows:

the application provides an insulator spontaneous explosion detection method based on a cascade network, which converts the problem of positioning of a spontaneous explosion glass insulator into the problem of positioning and segmentation in computer vision. The method based on deep learning is used for deeply researching aerial insulator image processing, recognition and defect positioning algorithms, improves image recognition speed and recognition accuracy, and helps to improve the digitization and intelligence degree of power transmission line detection. And a YOLOV5 network with good performance and high speed in the current stage target detection network is selected, so that the insulator string is positioned at high precision. A Mask R-CNN example segmentation network is adopted in the self-explosion insulator detection process, example segmentation is used as the combination of target detection and semantic segmentation, a target can be detected in an image, each pixel is labeled, two tasks of self-explosion point positioning and shape segmentation are directly completed through only one network, and the defect detection process is simplified. And a cascade network method is skillfully adopted, the two detection networks are combined, the position information of the insulator string target detected by the YOLOV5 network is utilized to assist a second Mask R-CNN example segmentation network, the insulator string is cut out by utilizing the target detection result, the complex picture background is reduced, the defect position of the insulator is highlighted, the example segmentation network is enabled to focus on learning the smaller categories, the identification and segmentation precision of the defective insulator is further improved, and the purpose of accurately segmenting and positioning the self-explosion insulator is achieved. The method of the cascade network can solve the problems of complex background, image distortion and low signal-to-noise ratio by processing the pictures.

Drawings

In order to more clearly illustrate the embodiments or technical solutions of the present invention, the drawings used in the embodiments or technical solutions in the prior art are briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without creative efforts.

Fig. 1 is a flowchart of a cascading network-based insulator spontaneous explosion detection method according to an embodiment of the present invention;

fig. 2 is a schematic diagram of a YOLOV5 network involved in a target detection model in the method provided by the embodiment of the present invention;

fig. 3 is a schematic diagram of a Mask R-CNN example division network related to an example division module of an insulator breakage portion in the method according to the embodiment of the present invention;

FIG. 4 is a schematic structural diagram of a computer-readable storage medium according to an embodiment of the present invention;

fig. 5 is a schematic structural diagram of a computer device according to an embodiment of the present invention.

Detailed Description

The inventor of the application discovers through a large amount of research and practice that in the prior art, the method mainly comprises two methods, wherein the first method is to directly position the defect position by taking the defect insulator as a target, the other method is to adopt a three-stage processing mode, firstly, the insulator is identified by using a Yolov3 target detection algorithm based on a GIoU strategy, then, an improved U-Net segmentation network is adopted to obtain a binaryzation insulator image, and finally, the insulator defects are detected by using mathematical modeling to judge the position and the number of the insulator chain defects.

However, the first method combines with the insulator picture defect detection method of Faster R-CNN + ResNet101+ FPN to perform denoising and anti-shake processing on the collected sample picture, and then directly locates the defect position by taking the defect insulator as a target. However, this method has a drawback that the picture samples are not distorted due to the jitter. And the insulator background of the aerial image is accompanied by complex backgrounds of forests, rivers, farmlands and the like due to long distance, large scale and complex structure of the power transmission line. And the insulator breakage positions occupy only a small portion of the pixels compared to the entire image. In some extreme cases, it is also difficult to find the fracture location by human eye recognition. Therefore, after simple pretreatment, the target detection model is directly used for positioning the defect position, and the precision is difficult to improve.

In the second method, firstly, a YOLO series target detection network has been developed for multiple generations after v3, and the performance and the speed are both improved higher than those of the v 3; then, the method can only divide the target insulator image by using the U-Net dividing network during the dividing task, and can not directly find the defects and position the defects. And the process of judging the defects and positioning can be completed only by pixel statistical mathematical modeling. Therefore, the detection process in the method is too complex and can be simplified, and the detection of the segmented networks has a space for improving the improvement.

Based on the method, the cascade network combining the target detection model and the example segmentation network is innovatively provided, and the YOLOV5 network with good performance and high speed in the target detection network at the current stage is selected at the target detection stage, so that the high-precision positioning of the insulator string is realized. A Mask R-CNN example segmentation network is adopted in an example segmentation stage, example segmentation is used as combination of target detection and semantic segmentation, a target can be detected in an image, each pixel is labeled, two tasks of self-explosion point positioning and shape segmentation can be completed only through one network, and a defect detection process is simplified. And a cascade network method is skillfully adopted, the two detection networks are combined, the position information of the insulator string target detected by the YOLOV5 network is utilized to assist a second Mask R-CNN example segmentation network, the insulator string is cut out by utilizing the target detection result, the complex picture background is reduced, the defect position of the insulator is highlighted, the example segmentation network is enabled to focus on learning the smaller categories, the identification and segmentation precision of the defective insulator is further improved, and the purpose of accurately segmenting and positioning the self-explosion insulator is achieved. The method of the cascade network can solve the problems of complex background, image distortion and low signal-to-noise ratio by processing the pictures.

In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.

Example one

The embodiment of the invention provides a cascade network-based insulator self-explosion detection method, which comprises the following steps:

a cascade network comprising a target detection model and an example segmentation model is constructed and trained in advance, wherein the target detection model adopts a network structure of a YOLOV5 algorithm, and the example segmentation model adopts a Mask R-CNN example segmentation network;

acquiring aerial insulating sub-images of an area to be monitored, and preprocessing the aerial insulating sub-images;

As shown in fig. 1, an implementation flow chart of the method for detecting the self-explosion of the insulator based on the cascade network provided by the invention mainly includes two parts, namely a YOLOV5 target detection network and a Mask R-CNN example segmentation network.

Referring to fig. 2, it is a model structure of YOLOv5 network, and the YOLOv5 network is composed of four main parts: the device comprises an input end, a Backbone network (Backbone), a feature fusion module (hack) and an output end.

Firstly, an Input end (Input) randomly cuts, scales and splices an Input image, finds the most appropriate self-adaptive anchor frame for calculation, and improves the positioning accuracy of the small damaged target defect of the insulator; then, a Backbone network (Backbone) is aggregated on the fine granularities of different images, and characteristic diagrams of three scales are extracted; secondly, a feature fusion module (hack) enlarges the receptive field of the feature map by adding a Spatial Pyramid Pooling Structure (SPP), uses 3 groups of multi-scale maximum Pooling layers, realizes feature fusion by splicing high and low feature layers obtained by upsampling (upsampling) to obtain a new feature map so as to improve the propagation of low-layer features, and then transmits the features from weak to strong through a Path Aggregation Network (PAN) from bottom to top, so that the feature layers realize more feature fusion, the two are combined to operate, and the capability of Network feature fusion is enhanced; the final Output (Output) is the predicted part of the network, and the target frame is screened by Non-Maximum suppression (NMS), the image features are predicted, a bounding box is generated, and the class is predicted. In one embodiment, the Mask R-CNN instance segmentation network comprises a backbone network, a feature pyramid module, a candidate frame generation network and a Mask generation module.

And then, cutting the image by using the acquired positioning bounding box to obtain an amplified insulator string serving as an ROI (region of interest) so as to improve the signal-to-noise ratio. Therefore, the position information of the insulator string target detected by the YOLOV5 network can be used for assisting the second Mask R-CNN example segmentation network, the insulator string can be cut out by using the target detection result, the complex picture background is reduced, the defect position of the insulator is highlighted, the example segmentation network is enabled to focus on learning the smaller types, and the detection effect is improved.

And segmenting the insulator missing fault of the region of interest by using a Mask R-CNN network, thereby solving the defect diagnosis problem of the insulator string disc unit.

In one embodiment, the feature pyramid module is an FPN, and includes a bottom-up pyramid and a top-down pyramid, and then the two pyramids are connected in the horizontal direction to fuse features with a shallow high resolution layer and a deep rich semantic information layer.

In one embodiment, the segmentation of the insulator missing fault of the region of interest by using a Mask R-CNN example segmentation network comprises two stages, wherein the first stage is to scan an input region of interest image and generate a region proposal, the region proposal represents a region which possibly contains a target, and the second stage branches the type and the boundary box offset of a prediction image through a convolution network and outputs an object pixel level Mask for each region of interest through another Mask prediction branch.

In particular, mask R-CNN adds improvements in several respects. Firstly, a Feature Pyramid (Feature Pyramid Network) is adopted, when target detection is carried out by using Fast R-CNN in the prior art, whether RPN or Fast R-CNN, roI acts on the last layer of Feature map, which is not problematic in detecting a large target, but causes problems in detecting a small target. For small objects, when performing convolution pooling to the last layer of feature map, semantic information may not be actually available because for a method of mapping the RoI to the feature map, coordinates are directly divided by the step size, and obviously the further the convolution, the smaller the mapping is in the past, and may even be unavailable. If the characteristics of multiple levels and multiple scales can be combined, the detection accuracy can be greatly improved.

FPN is a well-designed multi-scale detection method. It comprises a pyramid from bottom to top, a pyramid from top to bottom and a transverse connection. The bottom-up path may be any convolution network that extracts features from the original image, specifically, resNet is used as a skeleton network and is divided into 5 stages according to the size of the feature map. Wherein, the conv2, conv3, conv4 and conv5 output by the last layer of each of stage2, stage3, stage4 and stage5 are respectively defined as { C2, C3, C4, C5}, and their step sizes relative to the original picture are {4,8, 16, 32}, respectively. The upsampling is performed from the top and the bottom, and the upsampling directly uses nearest neighbor upsampling instead of deconvolution operation, so that the upsampling is simple on one hand and can reduce training parameters on the other hand. The top-down channel generates a signature pyramid similar in size to the bottom-up channel. The horizontal connection is to fuse the up-sampling result and the feature map with the same size generated from bottom to top, that is, the convolution and addition operation between two corresponding levels of the two pyramids. Specifically, each layer in { C2, C3, C4, C5} is subjected to a conv 1x1 operation (1 x1 convolution is used for reducing the number of channels), no active function operation is performed, output channels are all set to be the same 256 channels, and then the sum operation is performed with the up-sampled feature map. After the fusion, the fused features are further processed by using a convolution kernel of 3 × 3, so as to eliminate aliasing effect of upsampling.

The FPN improves the standard feature extraction pyramid by adding a second pyramid that takes the high-level features from the first pyramid and passes them to lower layers. By doing so, it allows each level of features to access both lower level and higher level features simultaneously. The main reason that FPN is superior to other single convolution neural networks is that it maintains strong semantic features at different resolution scales, because FPN naturally utilizes the pyramid form of CNN-level features, and designs top-down structures and transverse connections at the same time, so as to fuse features with shallow high resolution and deep rich semantic information, thereby generating a feature pyramid with strong semantic information at all scales. Therefore, the characteristic pyramid of multi-scale strong semantic information is quickly constructed from a single input image with a single scale, and obvious cost is not generated. The FPN is a flexible component, can be matched with different basic networks, and can be used in different algorithm frameworks to extract multi-scale features.

Example two

Based on the same inventive concept, the embodiment provides a cascade network-based insulator spontaneous explosion detection system, which comprises:

the network construction module is used for constructing and training a cascade network comprising a target detection model and an instance segmentation model in advance, wherein the target detection model adopts a network structure of a YOLOV5 algorithm, and the instance segmentation model adopts a Mask R-CNN instance segmentation network;

the image preprocessing module is used for acquiring aerial insulating sub-images of the area to be monitored and preprocessing the aerial insulating sub-images;

and the instance segmentation module is used for segmenting the insulator missing faults of the region of interest by utilizing a Mask R-CNN instance segmentation network and obtaining a detection result.

Since the system described in the second embodiment of the present invention is a system used for implementing the method for detecting self-explosion of an insulator based on a cascade network in the first embodiment of the present invention, a person skilled in the art can understand the specific structure and deformation of the system based on the method described in the first embodiment of the present invention, and thus, details are not described herein. All systems adopted by the method in the first embodiment of the invention belong to the protection scope of the invention.

EXAMPLE III

Based on the same inventive concept, please refer to fig. 4, the present invention further provides a computer readable storage medium 300, on which a computer program 311 is stored, which when executed implements the method as described in the first embodiment.

Since the computer-readable storage medium introduced in the third embodiment of the present invention is a computer-readable storage medium used for implementing the method for detecting self-explosion of an insulator based on a cascade network in the first embodiment of the present invention, based on the method introduced in the first embodiment of the present invention, persons skilled in the art can understand the specific structure and deformation of the computer-readable storage medium, and therefore, no further description is given here. Any computer readable storage medium used in the method of the first embodiment of the present invention falls within the intended scope of the present invention.

Example four

Based on the same inventive concept, the present application further provides a computer device, as shown in fig. 5, including a memory 401, a processor 402, and a computer program 403 stored in the memory and capable of running on the processor, where the processor executes the above program to implement the method in the first embodiment.

Since the computer device introduced in the fourth embodiment of the present invention is a computer device used for implementing the method for detecting self-explosion of an insulator based on a cascade network in the first embodiment of the present invention, based on the method introduced in the first embodiment of the present invention, a person skilled in the art can know the specific structure and deformation of the computer device, and thus, no further description is provided herein. All the computer devices used in the method in the first embodiment of the present invention are within the scope of the present invention.

As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present invention has been described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including the preferred embodiment and all changes and modifications that fall within the scope of the invention.

It will be apparent to those skilled in the art that various modifications and variations can be made in the embodiments of the present invention without departing from the spirit or scope of the embodiments of the invention. Thus, if such modifications and variations of the embodiments of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to encompass such modifications and variations.

Claims

1. An insulator self-explosion detection method based on a cascade network is characterized by comprising the following steps:

and segmenting the insulator missing fault of the region of interest by using a Mask R-CNN instance segmentation network, and obtaining a detection result.

2. The cascade network-based insulator spontaneous explosion detection method as recited in claim 1, wherein the Mask R-CNN instance segmentation network comprises a backbone network, a feature pyramid module, a candidate box generation network and a Mask generation module.

3. The cascade network based insulator spontaneous explosion detection method as claimed in claim 2, wherein the characteristic pyramid module is FPN, and comprises a bottom-up pyramid and a top-down pyramid, and then the two pyramids are connected in a transverse manner to fuse the characteristics with a shallow layer with high resolution and a deep layer with rich semantic information.

4. The cascade network-based insulator spontaneous explosion detection method as claimed in claim 2, wherein the segmentation of the missing insulator fault of the region of interest by using the Mask R-CNN instance segmentation network comprises two stages, the first stage is to scan an input image of the region of interest and generate a region proposal, the region proposal represents a region which may contain a target, and the second stage branches the category and the bounding box offset of a predicted image through a convolution network and outputs an object pixel level Mask for each region of interest through another Mask prediction branch.

5. The utility model provides an insulator spontaneous explosion detecting system based on cascade network which characterized in that includes:

the target detection module is used for detecting the preprocessed aerial insulator images by using a target detection model of the cascade network to obtain a detection result image of the insulator, wherein the detection result image comprises position information of the insulator string, and the position information is a positioning boundary frame;

6. A computer-readable storage medium, on which a computer program is stored, which program, when executed, carries out the method of any one of claims 1 to 4.

7. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the method according to any of claims 1 to 4 when executing the program.