CN113095253A - Insulator detection method for unmanned aerial vehicle to inspect power transmission line - Google Patents
Insulator detection method for unmanned aerial vehicle to inspect power transmission line Download PDFInfo
- Publication number
- CN113095253A CN113095253A CN202110422013.8A CN202110422013A CN113095253A CN 113095253 A CN113095253 A CN 113095253A CN 202110422013 A CN202110422013 A CN 202110422013A CN 113095253 A CN113095253 A CN 113095253A
- Authority
- CN
- China
- Prior art keywords
- layer
- convolutional layer
- insulator
- output
- convolutional
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 95
- 239000012212 insulator Substances 0.000 title claims abstract description 91
- 230000005540 biological transmission Effects 0.000 title claims abstract description 19
- 238000007689 inspection Methods 0.000 claims abstract description 25
- 238000012360 testing method Methods 0.000 claims abstract description 20
- 238000012549 training Methods 0.000 claims abstract description 17
- 238000000034 method Methods 0.000 claims abstract description 14
- 238000013135 deep learning Methods 0.000 claims abstract description 6
- 238000011176 pooling Methods 0.000 claims description 66
- 230000004927 fusion Effects 0.000 claims description 23
- 238000013136 deep learning model Methods 0.000 claims description 7
- 238000007499 fusion processing Methods 0.000 claims description 3
- 238000009413 insulation Methods 0.000 claims description 3
- 239000000284 extract Substances 0.000 abstract description 3
- 238000012216 screening Methods 0.000 abstract description 3
- 230000007547 defect Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 230000008034 disappearance Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000004880 explosion Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000003672 processing method Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- HPTJABJPZMULFH-UHFFFAOYSA-N 12-[(Cyclohexylcarbamoyl)amino]dodecanoic acid Chemical compound OC(=O)CCCCCCCCCCCNC(=O)NC1CCCCC1 HPTJABJPZMULFH-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 238000010292 electrical insulation Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000005055 memory storage Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Multimedia (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Image Analysis (AREA)
Abstract
The invention provides an insulator detection method for an unmanned aerial vehicle to patrol and examine a power transmission line, which relates to the technical field of target detection and comprises the following steps: s1, acquiring aerial insulating subimages of the unmanned aerial vehicle routing inspection power transmission line, and establishing a training set and a test set; s2, establishing an insulator detection model based on deep learning; s3, inputting the insulator images of the training set into the detection model, carrying out network setting, and repeating iterative training to obtain an insulator detection model; and S4, inputting the insulator image of the test set into the trained insulator detection model, and outputting the detection result of the insulator in the test set image. The method effectively extracts the target characteristics in the complex environment, ensures the detection precision and occupies less memory resources; the positioning of the insulator sub-targets in the inspection image is effectively realized, and meanwhile, the inspection image can be screened, so that the burden of manual screening is reduced, and the method has a wide application prospect.
Description
Technical Field
The invention relates to the technical field of digital image processing and target detection in power transmission line equipment detection, in particular to an insulator detection method for unmanned aerial vehicle inspection of a power transmission line.
Background
With the overhead transmission and distribution network being more and more widely distributed, the regular inspection of the power line is taken as an important work for guaranteeing the continuous power supply and the safe operation of the power facility. Insulators serve as indispensable parts of power lines, and have dual functions of electrical insulation and mechanical support. Faults such as insulator cracks, surface pollution and damage are very likely to damage the safe operation of a power system, and large-area power failure or huge economic loss occurs in a power grid. The traditional insulator detection generally adopts a manual inspection method, which is simple, but has extremely low efficiency and certain danger. In the existing power line inspection process, inspection personnel hold a shooting device to shoot photos or identify insulators by naked eyes. Usually, the insulator is firstly photographed or visually identified by a photographing device held by an unmanned aerial vehicle or a patrol person. After the inspection is finished, the time is spent on browsing, identifying and inspecting the shot picture in the inspection process, and the insulator fault is identified manually, so that the inspection is time-consuming, and the false inspection and the missing inspection are easy to occur.
Along with the continuous expansion of each country's electric power demand, the development of intelligent equipment such as unmanned aerial vehicle and high definition digtal camera, the continuous maturity of technologies such as image processing, machine learning and degree of depth learning, unmanned aerial vehicle carries on image acquisition equipment and has become the main mode that transmission line patrolled and examined. The insulator detection in the aerial image is realized by utilizing the traditional image processing method and the deep learning method, and remarkable results are obtained. The traditional image processing method depends on various feature extraction algorithms and is very sensitive to background interference, different feature extraction methods need to be designed for different types of insulator faults, and the simultaneous design of a detection model for realizing multi-insulator fault detection is impossible. Insulator detection based on the deep learning model can achieve good performance and has the potential to meet the requirements of real-time application. YOLO-v2 and YOLO-v3 are widely applied to the field of target detection as typical end-to-end target detection models, and achieve good detection effects. However, because the number of the layers of the YOLO-v2 and YOLO-v3 networks is large, the weight file after model training is large, and the method is actually applied to occupy large memory resources and is not suitable for being applied to the real-time detection of insulators in a power transmission line by an unmanned aerial vehicle. YOLO-tiny has good performance in the aspects of running time and memory storage, but due to different shooting angles and shooting distances of aerial images, insulators are difficult to accurately detect in a complex background.
Disclosure of Invention
Technical problem to be solved
The invention provides an insulator detection method for unmanned aerial vehicle inspection of a power transmission line, aiming at the defect that YOLO-v2 and YOLO-v3 detection models occupy larger memory resources and are not suitable for unmanned aerial vehicles to detect insulators in the power transmission line in real time.
(II) technical scheme
In order to achieve the purpose, the invention is realized by the following technical scheme:
an insulator detection method for unmanned aerial vehicle inspection transmission line comprises the following steps:
s1, acquiring aerial insulation sub-images of the unmanned aerial vehicle inspection power transmission line, and establishing a training set and a test set for detecting insulators by a deep learning model;
s2, establishing an insulator detection model based on deep learning, wherein the detection model specifically comprises a backbone network layer, a feature fusion network layer, a spatial pyramid pooling layer and a target detection layer which are sequentially connected;
s3, inputting the insulator image of the training set in the step S1 into the detection model in the step S2, carrying out network setting, and repeating iterative training to obtain an insulator detection model;
and S4, inputting the insulator detection model trained in the step S3 into the insulator image of the test set in the step S1, and outputting the detection result of the insulator in the test set image.
According to an embodiment of the present invention, in step S1, 4500 insulator images for aerial photography by the unmanned aerial vehicle are selected, and the image resolution is adjusted to 416 × 416; marking the position of the insulator in the Image by using a Label-Image marking tool, and establishing an insulator detection data set; 3000 marked insulator images are selected as a training set, and the rest 1500 insulator images are used as a testing set.
According to an embodiment of the present invention, the deep learning model for insulator detection in step S2 includes a backbone network layer, a feature fusion network layer, a pyramid pooling layer, and a target detection layer; the input features of the detection model are 416 × 416 × 3, the backbone network layer is used for extracting insulator image features, and the sizes of the extracted image features are 208 × 208 × 16, 104 × 104 × 32, 52 × 52 × 64, 26 × 26 × 128, 13 × 13 × 256, 13 × 13 × 512 and 13 × 13 × 1024 respectively; the feature fusion network layer performs fusion processing on features of three scales of 52 × 52, 26 × 26 and 13 × 13, three inputs of the feature fusion network layer are 52 × 52 × 64, 26 × 26 × 128 and 13 × 13 × 1024 respectively, and three outputs of the feature fusion network layer are 52 × 52 × 128, 26 × 26 × 256 and 13 × 13 × 512 respectively; the input of the pyramid pooling layer is connected with the output of the feature fusion network layer, and the sizes of the output image features of the pyramid pooling layer are respectively 52 multiplied by 512, 26 multiplied by 1024 and 13 multiplied by 2048; and the input of the target detection layer is connected with the output of the pyramid pooling layer, and the target detection layer respectively predicts insulator images with three scales of 52 × 52, 26 × 26 and 13 × 13.
According to an embodiment of the present invention, the backbone network layer includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a first residual module, a second residual module, a third residual module, a fourteenth convolutional layer, a third pooling layer, and a fifteenth convolutional layer, which are connected in sequence, the first residual module is composed of a third convolutional layer, a fourth convolutional layer, and a fifth convolutional layer, the second residual module is composed of a sixth convolutional layer, a seventh convolutional layer, an eighth convolutional layer, a ninth convolutional layer, and a tenth convolutional layer, and the third residual module is composed of an eleventh convolutional layer, a twelfth convolutional layer, and a thirteenth convolutional layer.
According to an embodiment of the present invention, the size of the input insulator image is 416 × 416 × 3, the input image is connected to a first convolution layer, the first convolution layer is a 3 × 3 × 16 convolution layer, the output of the first convolution layer is connected to a first pooling layer, the first pooling layer is a maximum pooling layer with 2 × 2 step size of 2, the output is 208 × 208 × 16, and the output of the first pooling layer is connected to a second convolution layer; the second convolution layer is a 3 multiplied by 32 convolution layer, the output of the second convolution layer is connected with the second pooling layer, the second pooling layer is a maximum pooling layer with 2 multiplied by 2 step length being 2, the output is 104 multiplied by 32, and the output of the second pooling layer is connected with the first residual module; the third convolution layer, the fourth convolution layer and the fifth convolution layer are connected in sequence, the output of the third convolution layer is directly connected with the output of the fifth convolution layer to form a first residual error module, the third convolution layer is a convolution layer with a 3 x 64 step length of 2, the fourth convolution layer is a 1 x 32 convolution layer, the fifth convolution layer is a 3 x 64 convolution layer, the first residual error module is used for extracting an image feature with a size of 52 x 64, and the output of the first residual error module is connected with the second residual error module; the sixth convolutional layer, the seventh convolutional layer, the eighth convolutional layer, the ninth convolutional layer and the tenth convolutional layer are sequentially connected, the sixth convolutional layer output, the eighth convolutional layer output and the tenth convolutional layer output are directly connected to form a second residual error module, the sixth convolutional layer is a convolutional layer with the 3 x 128 step length of 2, the seventh convolutional layer is a convolutional layer with the 1 x 64, the eighth convolutional layer is a convolutional layer with the 3 x 128, the ninth convolutional layer is a convolutional layer with the 1 x 64, and the tenth convolutional layer is a convolutional layer with the 3 x 128, the second residual error module is used for extracting the image feature size of 26 x 128, and the second residual error module output is connected with the third residual error module; the eleventh convolutional layer, the twelfth convolutional layer and the thirteenth convolutional layer are sequentially connected, the output of the eleventh convolutional layer is directly connected with the output of the thirteenth convolutional layer to form a third residual error module, the eleventh convolutional layer is a convolutional layer with the step length of 3 multiplied by 256 of 2, the twelfth convolutional layer is a 1 multiplied by 128 convolutional layer, the thirteenth convolutional layer is a 3 multiplied by 256 convolutional layer, the third residual error module is used for extracting the image feature with the size of 13 multiplied by 256, and the output of the third residual error module is connected with the fourteenth convolutional layer; the fourteenth convolutional layer is a 3 × 3 × 512 convolutional layer, the output is 13 × 13 × 512, the fourteenth convolutional layer is connected with the third pooling layer, the third pooling layer is a maximum pooling layer with a step size of 1 of 2 × 2, the output of the third pooling layer is connected with the fifteenth convolutional layer, the fifteenth convolutional layer is a 3 × 3 × 1024 convolutional layer, and the output is 13 × 13 × 1024.
(III) advantageous effects
The invention has the beneficial effects that: a depth learning model for insulator detection consists of a backbone network layer, a feature fusion network layer, a pyramid pooling layer and a target detection layer, has good robustness to interference caused by a complex background to an insulator image, can effectively realize positioning of an insulator sub-target in an inspection image, provides a basis for subsequent defect detection, can screen the image obtained by inspection, reduces the burden of manual screening and has wide application prospect; compared with a YOLO-v2 and a YOLO-v3 detection model, the detection model reduces the number of network layers for extracting the target features, and can effectively extract the target features in a complex environment and occupy less memory resources while ensuring the detection precision by extracting the insulator features in the image through the backbone network layer, the feature fusion network layer, the pyramid pooling layer and the target detection layer.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a flow chart of an insulator testing method of the present invention;
fig. 2 is a view showing a structure of an insulator inspection model according to the present invention.
Detailed Description
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 derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
With reference to fig. 1, a method for detecting an insulator of an unmanned aerial vehicle routing inspection power transmission line includes the following steps:
s1, acquiring aerial insulation sub-images of the unmanned aerial vehicle inspection power transmission line, and establishing a training set and a test set for detecting the insulator by the deep learning model.
4500 aerial insulator images of the unmanned aerial vehicle are selected, and the image resolution is adjusted to 416 multiplied by 416; marking the position of the insulator in the Image by using a Label-Image marking tool, and establishing an insulator detection data set; 3000 marked insulator images are selected as a training set, and the rest 1500 insulator images are used as a testing set.
S2, establishing an insulator detection model based on YOLO deep learning, wherein the detection model specifically comprises a backbone network layer, a feature fusion network layer, a spatial pyramid pooling layer (SPP) and a target detection layer which are sequentially connected.
S3, inputting the insulator image of the training set in the step S1 into the detection model in the step S2, carrying out network setting, and repeating iterative training to obtain an insulator detection model;
and S4, inputting the insulator detection model trained in the step S3 into the insulator image of the test set in the step S1, and outputting the detection result of the insulator in the test set image.
With reference to fig. 2, the deep learning model for insulator detection in step S2 includes a backbone network layer, a feature fusion network layer, a three-scale pyramid pooling layer, and a three-scale target detection layer; the input features of the detection model are 416 × 416 × 3, the backbone network layer is used for extracting insulator image features, and the sizes of the extracted image features are 208 × 208 × 16, 104 × 104 × 32, 52 × 52 × 64, 26 × 26 × 128, 13 × 13 × 256, 13 × 13 × 512 and 13 × 13 × 1024 respectively; the feature fusion network layer performs fusion processing on features of three scales of 52 × 52, 26 × 26 and 13 × 13, three inputs of the feature fusion network layer are 52 × 52 × 64, 26 × 26 × 128 and 13 × 13 × 1024 respectively, and three outputs of the feature fusion network layer are 52 × 52 × 128, 26 × 26 × 256 and 13 × 13 × 512 respectively; the input of the three-scale pyramid pooling layer is connected with the output of the feature fusion network layer, and the sizes of the output image features of the three-scale pyramid pooling layer are respectively 52 multiplied by 512, 26 multiplied by 1024 and 13 multiplied by 2048; and the input of the three-scale target detection layer is connected with the output of the three-scale pyramid pooling layer, and the three-scale target detection layer respectively predicts insulator images of three scales of 52 × 52, 26 × 26 and 13 × 13.
Table 1 shows a backbone network layer structure of the insulator detection model of the present invention, and with reference to fig. 2 and table 1, the backbone network layer includes a first convolution layer, a first pooling layer, a second convolution layer, a second pooling layer, a first residual module, a second residual module, a third residual module, a fourteenth convolution layer, a third pooling layer, and a fifteenth convolution layer, which are sequentially connected, the first residual module includes a third convolution layer, a fourth convolution layer, and a fifth convolution layer, the second residual module includes a sixth convolution layer, a seventh convolution layer, an eighth convolution layer, a ninth convolution layer, and a tenth convolution layer, and the third residual module includes an eleventh convolution layer, a twelfth convolution layer, and a thirteenth convolution layer. The convolution layer performs convolution operation on input information and convolution kernels with set pixel sizes to obtain feature vectors. The pooling layer performs down-sampling operation according to the definition made by the parameter template, takes the maximum value in the image block with the set pixel size as output, and reduces the data processing amount on the basis of saving useful information through the down-sampling process.
The size of an input insulator image is 416 multiplied by 3, the input image is connected with a first convolution layer, the first convolution layer is a 3 multiplied by 16 convolution layer, the output of the first convolution layer is connected with a first pooling layer, the first pooling layer is a maximum pooling layer with 2 multiplied by 2 step length being 2, the output is 208 multiplied by 16, and the output of the first pooling layer is connected with a second convolution layer; the second convolution layer is a 3 multiplied by 32 convolution layer, the output of the second convolution layer is connected with the second pooling layer, the second pooling layer is a maximum pooling layer with 2 multiplied by 2 step length being 2, the output is 104 multiplied by 32, and the output of the second pooling layer is connected with the first residual module; the third convolution layer, the fourth convolution layer and the fifth convolution layer are connected in sequence, the output of the third convolution layer is directly connected with the output of the fifth convolution layer to form a first residual error module, the third convolution layer is a convolution layer with a 3 x 64 step length of 2, the fourth convolution layer is a 1 x 32 convolution layer, the fifth convolution layer is a 3 x 64 convolution layer, the first residual error module is used for extracting an image feature with a size of 52 x 64, and the output of the first residual error module is connected with the second residual error module; the sixth convolutional layer, the seventh convolutional layer, the eighth convolutional layer, the ninth convolutional layer and the tenth convolutional layer are sequentially connected, the sixth convolutional layer output, the eighth convolutional layer output and the tenth convolutional layer output are directly connected to form a second residual error module, the sixth convolutional layer is a convolutional layer with the 3 x 128 step length of 2, the seventh convolutional layer is a convolutional layer with the 1 x 64, the eighth convolutional layer is a convolutional layer with the 3 x 128, the ninth convolutional layer is a convolutional layer with the 1 x 64, and the tenth convolutional layer is a convolutional layer with the 3 x 128, the second residual error module is used for extracting the image feature size of 26 x 128, and the second residual error module output is connected with the third residual error module; the eleventh convolutional layer, the twelfth convolutional layer and the thirteenth convolutional layer are sequentially connected, the output of the eleventh convolutional layer is directly connected with the output of the thirteenth convolutional layer to form a third residual error module, the eleventh convolutional layer is a convolutional layer with the step length of 3 multiplied by 256 of 2, the twelfth convolutional layer is a 1 multiplied by 128 convolutional layer, the thirteenth convolutional layer is a 3 multiplied by 256 convolutional layer, the third residual error module is used for extracting the image feature with the size of 13 multiplied by 256, and the output of the third residual error module is connected with the fourteenth convolutional layer; the fourteenth convolutional layer is a 3 × 3 × 512 convolutional layer, the output is 13 × 13 × 512, the fourteenth convolutional layer is connected with the third pooling layer, the third pooling layer is a maximum pooling layer with 2 × 2 and the step length of 1, the output of the third pooling layer is connected with the fifteenth convolutional layer, the fifteenth convolutional layer is a 3 × 3 × 1024 convolutional layer, and the output is 13 × 13 × 1024; and the fifth convolutional layer output (52 multiplied by 64), the tenth convolutional layer output (26 multiplied by 128) and the fifteenth convolutional layer output (13 multiplied by 1024) are respectively connected with the feature fusion network layer. Three residual modules are introduced into the backbone network layer, so that gradient disappearance or gradient explosion can be effectively avoided.
Table 1: the backbone network layer structure of the insulator detection model of the invention
In order to verify the effectiveness of the detection model, the detection model is compared with three existing classical YOLO detection models (YOLO-tiny, YOLO-v2 and YOLO-v3) in a test set. The experimental conditions were as follows: in terms of hardware, the CPU is of 3.60GHzCoreTMi9-9900K, and the total memory is 32 GB; the GPU is NVIDIA GeForce GTX 3080 with 10G memory. In terms of software, CUDA 11.1 and cuDNN 8.0.5 accelerators are provided, and Open CV 3.4.0, Visual Studio 2017, Windows 10 operating system and Dark-net deep learning framework are provided.
As shown in Table 2, the test indexes (average accuracy and memory occupation) of the four detection models are that the number of backbone network layers of the YOLO-v3 is the largest, so that the weight file of the YOLO-v3 after training is larger, and the memory occupation is larger and reaches 240 MB; the backbone network layer of the YOLO-tiny is composed of 7 convolutional layers and 6 pooling layers, the number of network layers is small, and the memory occupation amount is only 33 MB; the memory occupation amount of a YOLO-v2 detection model reaches 197MB, and the memory occupation amount of the detection model reaches 146 MB. Compared with the YOLO-v2 and YOLO-v3 detection models, the memory occupancy of the detection model is reduced by 51MB and 94MB respectively. The average accuracy is used for evaluating the performance of the detection model for detecting the insulator target on the test set, and the average accuracy of the four detection networks (YOLO-tiny, YOLO-v2, YOLO-v3 and the detection model of the invention) is 72.8%, 80.9%, 90.3% and 89.7% respectively. Therefore, the average precision and the memory occupation amount are comprehensively considered, and the detection model has better detection performance.
Table 2: test indexes of four detection models
In summary, in the embodiment of the present invention, the depth learning model for insulator detection is composed of a backbone network layer, a feature fusion network layer, a three-dimensional pyramid pooling layer, and a three-dimensional target detection layer, so that the method has a good robustness to the interference of a complex background on an insulator image, can effectively realize the positioning of the insulator sub-targets in the inspection image, provides a basis for the subsequent defect detection, and can simultaneously screen the images obtained by inspection, reduce the burden of manual screening, and has a wide application prospect; compared with a YOLO-v2 and a YOLO-v3 detection model, the detection model reduces the number of network layers for extracting the target features, and can effectively extract the target features in a complex environment and occupy less memory resources while ensuring the detection precision by extracting the insulator features in the image through a backbone network layer, a feature fusion network layer, a pyramid pooling layer and a target detection layer; according to the invention, three residual modules are introduced into the backbone network layer, so that gradient disappearance or gradient explosion can be effectively avoided.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (5)
1. The utility model provides a method for unmanned aerial vehicle patrols and examines insulator detection of transmission line which characterized in that includes following step:
s1, acquiring aerial insulation sub-images of the unmanned aerial vehicle inspection power transmission line, and establishing a training set and a test set for detecting insulators by a deep learning model;
s2, establishing an insulator detection model based on deep learning, wherein the detection model specifically comprises a backbone network layer, a feature fusion network layer, a spatial pyramid pooling layer and a target detection layer which are sequentially connected;
s3, inputting the insulator image of the training set in the step S1 into the detection model in the step S2, carrying out network setting, and repeating iterative training to obtain an insulator detection model;
and S4, inputting the insulator detection model trained in the step S3 into the insulator image of the test set in the step S1, and outputting the detection result of the insulator in the test set image.
2. The method for detecting the insulator of the unmanned aerial vehicle inspection transmission line according to claim 1, characterized in that: 4500 unmanned aerial vehicle aerial insulator images are selected in the step S1, and the image resolution is adjusted to 416 x 416; marking the position of the insulator in the Image by using a Label-Image marking tool, and establishing an insulator detection data set; 3000 marked insulator images are selected as a training set, and the rest 1500 insulator images are used as a testing set.
3. The method according to claim 1, wherein the deep learning model for insulator detection in step S2 includes a backbone network layer, a feature fusion network layer, a pyramid pooling layer, and a target detection layer; the input features of the detection model are 416 × 416 × 3, the backbone network layer is used for extracting insulator image features, and the sizes of the extracted image features are 208 × 208 × 16, 104 × 104 × 32, 52 × 52 × 64, 26 × 26 × 128, 13 × 13 × 256, 13 × 13 × 512 and 13 × 13 × 1024 respectively; the feature fusion network layer performs fusion processing on features of three scales of 52 × 52, 26 × 26 and 13 × 13, three inputs of the feature fusion network layer are 52 × 52 × 64, 26 × 26 × 128 and 13 × 13 × 1024 respectively, and three outputs of the feature fusion network layer are 52 × 52 × 128, 26 × 26 × 256 and 13 × 13 × 512 respectively; the input of the pyramid pooling layer is connected with the output of the feature fusion network layer, and the sizes of the output image features of the pyramid pooling layer are respectively 52 multiplied by 512, 26 multiplied by 1024 and 13 multiplied by 2048; and the input of the target detection layer is connected with the output of the pyramid pooling layer, and the target detection layer respectively predicts insulator images with three scales of 52 × 52, 26 × 26 and 13 × 13.
4. The method as claimed in claim 3, wherein the backbone network layer includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a first residual module, a second residual module, a third residual module, a fourteenth convolutional layer, a third pooling layer, and a fifteenth convolutional layer, which are connected in sequence, the first residual module includes a third convolutional layer, a fourth convolutional layer, and a fifth convolutional layer, the second residual module includes a sixth convolutional layer, a seventh convolutional layer, an eighth convolutional layer, a ninth convolutional layer, and a tenth convolutional layer, and the third residual module includes an eleventh convolutional layer, a twelfth convolutional layer, and a thirteenth convolutional layer.
5. The insulator detection method for the unmanned aerial vehicle to inspect the transmission line according to the claim 4, characterized in that the size of the input insulator image is 416 x 3, the input image is connected with the first convolution layer, the first convolution layer is 3 x 16 convolution layer, the output of the first convolution layer is connected with the first pooling layer, the first pooling layer is the largest pooling layer with 2 x 2 step length being 2, the output is 208 x 16, the output of the first pooling layer is connected with the second convolution layer; the second convolution layer is a 3 multiplied by 32 convolution layer, the output of the second convolution layer is connected with the second pooling layer, the second pooling layer is a maximum pooling layer with 2 multiplied by 2 step length being 2, the output is 104 multiplied by 32, and the output of the second pooling layer is connected with the first residual module; the third convolution layer, the fourth convolution layer and the fifth convolution layer are connected in sequence, the output of the third convolution layer is directly connected with the output of the fifth convolution layer to form a first residual error module, the third convolution layer is a convolution layer with a 3 x 64 step length of 2, the fourth convolution layer is a 1 x 32 convolution layer, the fifth convolution layer is a 3 x 64 convolution layer, the first residual error module is used for extracting an image feature with a size of 52 x 64, and the output of the first residual error module is connected with the second residual error module; the sixth convolutional layer, the seventh convolutional layer, the eighth convolutional layer, the ninth convolutional layer and the tenth convolutional layer are sequentially connected, the sixth convolutional layer output, the eighth convolutional layer output and the tenth convolutional layer output are directly connected to form a second residual error module, the sixth convolutional layer is a convolutional layer with the 3 x 128 step length of 2, the seventh convolutional layer is a convolutional layer with the 1 x 64, the eighth convolutional layer is a convolutional layer with the 3 x 128, the ninth convolutional layer is a convolutional layer with the 1 x 64, and the tenth convolutional layer is a convolutional layer with the 3 x 128, the second residual error module is used for extracting the image feature size of 26 x 128, and the second residual error module output is connected with the third residual error module; the eleventh convolutional layer, the twelfth convolutional layer and the thirteenth convolutional layer are sequentially connected, the output of the eleventh convolutional layer is directly connected with the output of the thirteenth convolutional layer to form a third residual error module, the eleventh convolutional layer is a convolutional layer with the step length of 3 multiplied by 256 of 2, the twelfth convolutional layer is a 1 multiplied by 128 convolutional layer, the thirteenth convolutional layer is a 3 multiplied by 256 convolutional layer, the third residual error module is used for extracting the image feature with the size of 13 multiplied by 256, and the output of the third residual error module is connected with the fourteenth convolutional layer; the fourteenth convolutional layer is a 3 × 3 × 512 convolutional layer, the output is 13 × 13 × 512, the fourteenth convolutional layer is connected with the third pooling layer, the third pooling layer is a maximum pooling layer with a step size of 1 of 2 × 2, the output of the third pooling layer is connected with the fifteenth convolutional layer, the fifteenth convolutional layer is a 3 × 3 × 1024 convolutional layer, and the output is 13 × 13 × 1024.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110422013.8A CN113095253B (en) | 2021-04-20 | 2021-04-20 | Insulator detection method for unmanned aerial vehicle inspection transmission line |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110422013.8A CN113095253B (en) | 2021-04-20 | 2021-04-20 | Insulator detection method for unmanned aerial vehicle inspection transmission line |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113095253A true CN113095253A (en) | 2021-07-09 |
CN113095253B CN113095253B (en) | 2023-05-23 |
Family
ID=76678713
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110422013.8A Active CN113095253B (en) | 2021-04-20 | 2021-04-20 | Insulator detection method for unmanned aerial vehicle inspection transmission line |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113095253B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116580285A (en) * | 2023-07-14 | 2023-08-11 | 西安益迪惟科智能科技有限责任公司 | Railway insulator night target identification and detection method |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160117845A1 (en) * | 2014-10-27 | 2016-04-28 | King Fahd University Petroleum and Minerals | Contamination level estimation method for high voltage insulators |
CN106504233A (en) * | 2016-10-18 | 2017-03-15 | 国网山东省电力公司电力科学研究院 | Image electric power widget recognition methodss and system are patrolled and examined based on the unmanned plane of Faster R CNN |
CN112200178A (en) * | 2020-09-01 | 2021-01-08 | 广西大学 | Transformer substation insulator infrared image detection method based on artificial intelligence |
CN112229845A (en) * | 2020-10-12 | 2021-01-15 | 国网河南省电力公司濮阳供电公司 | Unmanned aerial vehicle high-precision winding tower intelligent inspection method based on visual navigation technology |
-
2021
- 2021-04-20 CN CN202110422013.8A patent/CN113095253B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160117845A1 (en) * | 2014-10-27 | 2016-04-28 | King Fahd University Petroleum and Minerals | Contamination level estimation method for high voltage insulators |
CN106504233A (en) * | 2016-10-18 | 2017-03-15 | 国网山东省电力公司电力科学研究院 | Image electric power widget recognition methodss and system are patrolled and examined based on the unmanned plane of Faster R CNN |
CN112200178A (en) * | 2020-09-01 | 2021-01-08 | 广西大学 | Transformer substation insulator infrared image detection method based on artificial intelligence |
CN112229845A (en) * | 2020-10-12 | 2021-01-15 | 国网河南省电力公司濮阳供电公司 | Unmanned aerial vehicle high-precision winding tower intelligent inspection method based on visual navigation technology |
Non-Patent Citations (2)
Title |
---|
邱志斌等: "基于目标检测的电网涉鸟故障相关鸟种智能识别", 电网技术 * |
陈志豪等: "基于多尺度密集网络的配网架空输电线路绝缘子识别", 激光与光电子学进展 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116580285A (en) * | 2023-07-14 | 2023-08-11 | 西安益迪惟科智能科技有限责任公司 | Railway insulator night target identification and detection method |
CN116580285B (en) * | 2023-07-14 | 2023-09-22 | 西安益迪惟科智能科技有限责任公司 | Railway insulator night target identification and detection method |
Also Published As
Publication number | Publication date |
---|---|
CN113095253B (en) | 2023-05-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108846418B (en) | Cable equipment temperature abnormity positioning and identifying method | |
CN107742093B (en) | Real-time detection method, server and system for infrared image power equipment components | |
CN112200178B (en) | Transformer substation insulator infrared image detection method based on artificial intelligence | |
CN109544501B (en) | Transmission equipment defect detection method based on unmanned aerial vehicle multi-source image feature matching | |
CN110288586A (en) | A kind of multiple dimensioned transmission line of electricity defect inspection method based on visible images data | |
CN110335260B (en) | Power cable damage detection method based on light convolutional neural network | |
CN108665468A (en) | A kind of device and method extracting tangent tower insulator chain based on dimensionality reduction gray scale texture-form matrix | |
CN116503318A (en) | Aerial insulator multi-defect detection method, system and equipment integrating CAT-BiFPN and attention mechanism | |
CN113095253B (en) | Insulator detection method for unmanned aerial vehicle inspection transmission line | |
Wanguo et al. | Typical defect detection technology of transmission line based on deep learning | |
CN113112489A (en) | Insulator string-dropping fault detection method based on cascade detection model | |
CN113159334A (en) | Electrical equipment infrared image real-time detection and diagnosis method based on light-weight deep learning | |
CN114627044A (en) | Solar photovoltaic module hot spot detection method based on deep learning | |
CN113095420A (en) | Insulator fault detection method based on improved YOLOv3 | |
CN113486873B (en) | Transformer substation equipment inspection method and system based on big data and artificial intelligence | |
CN113095252A (en) | Insulator fault image identification method for power inspection | |
CN105043275A (en) | Image processing-based icing thickness measurement method for power transmission line | |
CN114494186A (en) | Fault detection method for high-voltage power transmission and transformation line electrical equipment | |
CN113610837A (en) | External insulation equipment discharge fault diagnosis method based on deep learning and edge calculation | |
CN113033556A (en) | Insulator rapid distinguishing and positioning method and system based on machine vision | |
CN116192049B (en) | Safety detection method and system applied to distributed photovoltaic project | |
Yan et al. | Research on fault diagnosis of transmission line based on SIFT feature | |
Gu et al. | Object detection of overhead transmission lines based on improved YOLOv5s | |
CN113111802B (en) | Insulator detection model based on deep neural network | |
Hao et al. | Research on distribution infrared image defect system based on improved Mask-RCNN |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right | ||
TR01 | Transfer of patent right |
Effective date of registration: 20240202 Address after: Room 043, F2002, 20th Floor, Building 4-A, Xixian Financial Port, Fengdong New City Energy Jinmao District, Xi'an City, Shaanxi Province, 710018 Patentee after: Shaanxi Zitan Optical Measurement Technology Co.,Ltd. Country or region after: China Address before: 169 Jianshe West Road, Chizhou City, Anhui Province Patentee before: CHIZHOU University Country or region before: China |