CN113112489B - Insulator string-dropping fault detection method based on cascade detection model - Google Patents
Insulator string-dropping fault detection method based on cascade detection model Download PDFInfo
- Publication number
- CN113112489B CN113112489B CN202110437796.7A CN202110437796A CN113112489B CN 113112489 B CN113112489 B CN 113112489B CN 202110437796 A CN202110437796 A CN 202110437796A CN 113112489 B CN113112489 B CN 113112489B
- Authority
- CN
- China
- Prior art keywords
- layer
- insulator
- convolutional layer
- scale
- convolution
- 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.)
- Active
Links
- 239000012212 insulator Substances 0.000 title claims abstract description 174
- 238000001514 detection method Methods 0.000 title claims abstract description 154
- 238000012360 testing method Methods 0.000 claims abstract description 20
- 238000012549 training Methods 0.000 claims abstract description 18
- 230000005540 biological transmission Effects 0.000 claims abstract description 16
- 238000007689 inspection Methods 0.000 claims abstract description 11
- 238000000034 method Methods 0.000 claims abstract description 11
- 238000011176 pooling Methods 0.000 claims description 47
- 230000007704 transition Effects 0.000 claims description 42
- 230000004927 fusion Effects 0.000 claims description 15
- 238000005070 sampling Methods 0.000 claims description 14
- 238000010586 diagram Methods 0.000 claims description 13
- 239000000284 extract Substances 0.000 claims description 8
- 238000013135 deep learning Methods 0.000 claims description 7
- 101100076840 Arabidopsis thaliana MFL1 gene Proteins 0.000 description 4
- 101100533322 Candida albicans (strain SC5314 / ATCC MYA-2876) SFL2 gene Proteins 0.000 description 4
- 101100428781 Human herpesvirus 6A (strain Uganda-1102) U8 gene Proteins 0.000 description 4
- 101100428786 Human herpesvirus 6A (strain Uganda-1102) U9 gene Proteins 0.000 description 4
- 101100126953 Oryza sativa subsp. japonica KCS20 gene Proteins 0.000 description 4
- 101100511042 Oryza sativa subsp. japonica LFL1 gene Proteins 0.000 description 4
- 101150043825 SFL1 gene Proteins 0.000 description 4
- 101100538885 Saccharomyces cerevisiae (strain ATCC 204508 / S288c) TUP1 gene Proteins 0.000 description 4
- 230000009466 transformation Effects 0.000 description 4
- 101100203828 Arabidopsis thaliana SPP3A gene Proteins 0.000 description 3
- 101100534124 Oryza sativa subsp. japonica SPP3 gene Proteins 0.000 description 3
- 102100030684 Sphingosine-1-phosphate phosphatase 1 Human genes 0.000 description 3
- 101710168942 Sphingosine-1-phosphate phosphatase 1 Proteins 0.000 description 3
- 102100030677 Sphingosine-1-phosphate phosphatase 2 Human genes 0.000 description 3
- 101710168938 Sphingosine-1-phosphate phosphatase 2 Proteins 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 238000005728 strengthening 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
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000010292 electrical insulation Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- 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
- G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
- G06V10/464—Salient features, e.g. scale invariant feature transforms [SIFT] using a plurality of salient features, e.g. bag-of-words [BoW] representations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30164—Workpiece; Machine component
Abstract
The invention provides an insulator string drop fault detection method based on a cascade detection model, which relates to the technical field of insulator fault detection of a power transmission line.A power transmission line insulator aerial image is collected, and an insulator data set and an insulator string drop data set are manufactured; establishing and training an insulator positioning detection model and an insulator fault detection model; the method comprises the steps of firstly positioning the position of an insulator through a trained insulator positioning detection model, then carrying out fault detection on the positioned insulator by using an insulator fault detection model, and finally outputting positioning and fault detection results on a test image. The method can effectively improve the accuracy of insulator string-dropping fault detection; the method has good robustness for detecting insulators with different scales, complex background interference and shielding in aerial images, and can effectively realize the detection of the insulator string failure in the power transmission line inspection image.
Description
Technical Field
The invention relates to the technical field of insulator fault detection of a power transmission line, in particular to an insulator string drop fault detection method based on a cascade detection model.
Background
The insulators are numerous in the power transmission and transformation lines and play an important role in electrical insulation and mechanical connection. Because the power transmission and transformation line spans various complex natural geographic environments, the power transmission and transformation line is easy to have a string-dropping fault after being exposed to an outdoor environment for a long time. The insulator string falling fault can affect the safe and stable operation of the whole power transmission line, and even cause huge economic loss to a power grid. In order to ensure the normal operation of the transmission line, the insulator string-dropping fault detection becomes a primary task of the intelligent detection of the transmission line. In order to realize automation and real-time inspection of the power transmission and transformation line, an image processing technology gradually replaces manual screening and aerial inspection of inspection pictures. However, the traditional target detection algorithm aims at massive inspection pictures, and both the detection speed and the detection effect cannot meet the actual application requirements.
With the development of deep neural networks, particularly the successful application of deep convolutional networks in image recognition, deep learning detection models have been applied to power routing inspection. The existing insulator fault detection is realized by adopting a deep learning model, and because the insulator string falling faults are not easy to collect in real life, the number of the insulator string falling faults is more than that of insulators. The insulator fault detection is carried out by adopting a single detection model, so that the detection accuracy and the recall rate are not high, and the class imbalance is easy to occur in the training process.
Disclosure of Invention
Technical problem to be solved
Aiming at the defect problems in the prior art, the invention provides a cascading detection model-based insulator string drop fault detection method.
(II) technical scheme
In order to achieve the purpose, the invention is realized by the following technical scheme:
a cascading detection model-based insulator string drop fault detection method comprises the following steps:
s1, collecting a power transmission line inspection image, and establishing an insulator image set and an insulator fault image set;
s2, traversing the insulator image set and the insulator fault image set, and marking the insulator position and the insulator fault position in the image to obtain an insulator data set and an insulator fault data set;
s3, dividing the insulator data set into an insulator training set and an insulator testing set, and dividing the insulator fault data set into an insulator fault training set and an insulator fault testing set;
s4, establishing an insulator positioning detection model based on deep learning, wherein the insulator positioning detection model specifically comprises a backbone network, a pyramid pooling network, a feature fusion network and a target detection network which are connected in sequence; establishing a three-scale predicted YOLOv4-tiny insulator fault detection model;
s5, training an insulator positioning detection model and an insulator fault detection model by using an insulator training set and an insulator fault training set respectively to obtain a trained insulator positioning detection model and a trained insulator fault detection model;
and S6, inputting the insulator test set image and the insulator fault test set image into the insulator positioning detection model in the step S5, detecting the position of the insulator by using the insulator positioning detection model, taking the position of the insulator as an interested area, performing fault detection on the insulator in the interested area by using the insulator fault detection model, and finally outputting a detection result on the insulator test set image.
According to an embodiment of the invention, the deep learning-based insulator positioning detection model comprises a backbone network, a pyramid pooling network, a feature fusion network and a detection network which are connected in sequence; the backbone network comprises a first transition convolutional layer, a first CSP module, a second CSP module, a third CSP module, a fourth CSP module and a fifth CSP module which are sequentially connected, the first transition convolutional layer is a convolutional layer of 3 multiplied by 32, image characteristics are extracted, a characteristic diagram of 416 multiplied by 32 is obtained, the first CSP module extracts the image characteristics, obtaining a 208 × 208 × 64 feature map, extracting image features by the second CSP module to obtain a 104 × 104 × 128 feature map, extracting image features by the third CSP module to obtain a 52 × 52 × 256 feature map, extracting image features by the fourth CSP module to obtain a 26 × 26 × 512 feature map, extracting image features by the fifth CSP module to obtain a 13 × 13 × 1024 feature map; and the output of the third CSP module, the output of the fourth CSP module and the output of the fifth CSP module are subjected to pooling operation of a three-scale pyramid pooling network, three-scale feature fusion is performed by using a feature fusion network, and 52 multiplied by 52, 26 multiplied by 26 and 13 multiplied by 13 features are output to predict insulators in the image after passing through a residual error network.
According to an embodiment of the present invention, the first CSP module includes a first downsampling layer, a first bypass convolutional layer, and 1 residual error unit, where the first downsampling layer is a convolutional layer with a 3 × 3 × 64 step size of 2, the first bypass convolutional layer is a convolutional layer with a 3 × 3 × 64, the residual error unit is formed by connecting a 1 × 1 × 32 convolutional layer, a 3 × 3 × 64 convolutional layer, and a short cut, and an output of the residual error unit is fused with an output of the first bypass convolutional layer to obtain a feature 208 × 208 × 64.
According to an embodiment of the present invention, the second CSP module includes a second downsampling layer, a second bypass convolution layer, and 2 residual error units, the second downsampling layer is a convolution layer with a 3 × 3 × 128 step size of 2, the second bypass convolution layer is a convolution layer with a 3 × 3 × 128 step size, the residual error unit is formed by connecting 1 × 1 × 64 convolution layers, 3 × 3 × 128 convolution layers, and a short cut, and an output of the residual error unit and an output of the second bypass convolution layer are fused to obtain a characteristic 104 × 104 × 128.
According to an embodiment of the present invention, the third CSP module includes a third downsampling layer, a second transition convolutional layer, a third bypass convolutional layer, 4 sense units, and 4 residual error units, the third downsampling layer is a convolutional layer with 3 × 3 × 256 step size of 2, the third bypass convolutional layer is a convolutional layer with 3 × 3 × 256, the second transition convolutional layer is a 1 × 1 × 128 convolutional layer, outputs of the third downsampling layer are respectively connected to the third bypass convolutional layer and the second transition convolutional layer, outputs of the second transition convolutional layer are connected to 4 sequentially connected sense units, the sense unit is composed of a 1 × 1 × 32 convolutional layer, a 3 × 3 × 32 convolutional layer, and a Concat connection, outputs of the sense unit are connected to 4 sequentially connected residual error units, the residual error unit is composed of a 1 × 1 × 128 convolutional layer, a 3 × 3 × 256 convolutional layer, and a short cut connection, and outputs of the residual error units and the third bypass convolutional layer are fused to 256 × 52 characteristics.
According to an embodiment of the present invention, the fourth CSP module includes a fourth down-sampling layer, a third transition convolution layer, a fourth bypass convolution layer, 4 sense units, and 4 residual units, the fourth down-sampling layer is a convolution layer with 3 × 3 × 512 step size of 2, the fourth bypass convolution layer is a convolution layer with 3 × 3 × 512, the third transition convolution layer is a 1 × 1 × 256 convolution layer, outputs of the fourth down-sampling layer are respectively connected to the fourth bypass convolution layer and the third transition convolution layer, outputs of the third transition convolution layer are connected to 4 sequentially connected sense units, the sense unit is formed by connecting 1 × 1 × 64 convolution layer, 3 × 3 × 64 convolution layer, and Concat, outputs of the sense unit are connected to 4 sequentially connected residual units, the residual unit is formed by connecting 1 × 1 × 256 convolution layer, 3 × 3 × 512 convolution layer, and short cut, outputs of the residual unit and the fourth bypass convolution layer are 26 × 26 outputs, and features are fused with outputs of the fourth bypass convolution layer.
According to an embodiment of the present invention, the fifth CSP module includes a fifth downsampling layer, a fourth transition convolutional layer, a fifth bypass convolutional layer, 4 sense units, and 4 residual units, the fifth downsampling layer is a convolutional layer with 3 × 3 × 1024 step size of 2, the fifth bypass convolutional layer is a convolutional layer with 3 × 3 × 1024, the fourth transition convolutional layer is a 1 × 1 × 512 convolutional layer, outputs of the fifth downsampling layer are respectively connected to the fifth bypass convolutional layer and the fourth transition convolutional layer, outputs of the fourth transition convolutional layer are connected to 4 sequentially connected sense units, the sense unit is composed of a 1 × 1 × 128 convolutional layer, a 3 × 3 × 128 convolutional layer, and a Concat connection, outputs of the sense unit are connected to 4 sequentially connected residual units, the residual unit is composed of a 1 × 1 × 512 convolutional layer, a 3 × 3 × 1024 convolutional layer, and a shcutt connection, and outputs of the residual unit and the fourth bypass convolutional layer output obtain a fused bypass 13 × 13 characteristic.
According to an embodiment of the present invention, the pyramid pooling network is a three-scale pyramid pooling structure, the output feature layer 52 × 52 × 256 is connected to the large-scale pyramid pooling structure, the output feature layer 26 × 26 × 512 is connected to the medium-scale pyramid pooling structure, the output feature layer 13 × 13 × 1024 is connected to the small-scale pyramid pooling structure, each scale pyramid pooling structure includes three largest pooling layers, and the sizes of filters corresponding to the three largest pooling layers are 13 × 13, 9 × 9, and 5 × 5, respectively.
According to an embodiment of the invention, three effective feature layers extracted by a backbone network are used for a feature fusion network, wherein 52 × 52 × 256 of the feature layers correspond to a first large-scale feature layer LFL0, 26 × 26 × 512 of the feature layers correspond to a first medium-scale feature layer MFL0, and 13 × 13 × 1024 of the feature layers correspond to a first small-scale feature layer SFL0; the first small-scale feature layer SFL0 obtains a second small-scale feature layer SFL1 through pyramid pooling structure SPP1 and convolution operation, the first medium-scale feature layer MFL0 obtains a second medium-scale feature layer MFL1 through pyramid pooling structure SPP2 and convolution operation, and the first large-scale feature layer LFL0 obtains a second large-scale feature layer LFL1 through pyramid pooling structure SPP3 and convolution operation; the second small-scale feature layer SFL is subjected to upsampling operation and then fused with the second medium-scale feature layer MFL1 to obtain a third medium-scale feature layer MFL2, and the third medium-scale feature layer MFL2 is subjected to upsampling operation and then fused with the second large-scale feature layer LFL1 to obtain a third large-scale feature layer LFL2; the third large-scale feature layer LFL2 is fused with the third medium-scale feature layer MFL2 after downsampling operation to obtain a fourth medium-scale feature layer MFL3, and the fourth medium-scale feature layer MFL3 is fused with the second small-scale feature layer SFL1 after downsampling operation to obtain a third small-scale feature layer SFL2; and the output features 13 × 13, 26 × 26 and 52 × 52 of the third small-scale feature layer SFL2, the fourth medium-scale feature layer MFL3 and the third large-scale feature layer LFL2 are respectively subjected to three residual modules and then are sent to a three-scale target detection layer, and the three-scale target detection layer respectively predicts the insulator images with the feature scales of 13 × 13, 26 × 26 and 52 × 52.
(III) advantageous effects
The invention has the beneficial effects that: a cascade detection model-based insulator string drop fault detection method comprises the steps of firstly, positioning the position of an insulator in an image by using an insulator positioning detection model, and then, carrying out fault detection on the positioned insulator by using an insulator fault detection model, so that the accuracy and the recall rate of insulator fault detection can be effectively improved; the method has good robustness for detecting insulators with different scales, complex background interference and shielding in aerial images, and can effectively realize insulator fault detection in power transmission line inspection; according to the insulator positioning detection model, cross-stage local network strengthening feature multiplexing and propagation are introduced into a backbone network, so that the target feature under a complex environment can be effectively extracted, the detection precision is guaranteed, and the detection accuracy and recall rate of the target are improved; the insulator fault detection model adopts the three-scale predicted YOLOv4-tiny, YOLOv4-tiny not only has small memory occupation amount, but also has fast detection speed and higher detection accuracy.
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 the detection method of the present invention;
FIG. 2 is a diagram of a model structure for insulator positioning and detection according to the present invention;
FIG. 3 (a) is a first CSP module structure diagram;
FIG. 3 (b) is a diagram of a second CSP module structure;
FIG. 3 (c) is a diagram of a third CSP module structure;
FIG. 3 (d) is a diagram of a fourth CSP module structure;
FIG. 3 (e) is a diagram of a fifth CSP module structure;
FIG. 4 is a diagram of a pyramid pooling network architecture;
FIG. 5 is a diagram of a residual module structure;
FIG. 6 (a) is a detection result of the positioning detection model of the present invention with a field as a background;
FIG. 6 (b) is a field-based detection result of the fault detection model of the present invention;
FIG. 7 (a) is the detection result of the positioning detection model of the present invention with river as the background;
FIG. 7 (b) is a detection result of the fault detection model of the present invention with river as background;
FIG. 8 (a) is a diagram of the detection result of the positioning detection model of the present invention with the power frame as the background;
FIG. 8 (b) is a diagram of the detection results of the fault detection model of the present invention with the power rack as the background;
FIG. 9 (a) shows the detection results of the localization detection model of the present invention for complex background interference;
fig. 9 (b) shows the detection result of the fault detection model of the present invention for complex background interference.
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 a string-dropping fault of an insulator based on a cascade detection model includes the following steps:
s1, collecting power transmission line inspection aerial images, and establishing an insulator image set and an insulator fault image set.
And S2, traversing the image set, and marking the insulator position and the insulator fault position in the image by using a Label-image marking tool to obtain an insulator data set and an insulator fault data set.
S3, dividing the data set into a training set and a testing set, wherein the training data set and the testing data set are divided according to the ratio of 2:1, selecting the proportion.
4500 aerial insulator images are selected, wherein the insulator fault images comprise 1331 aerial insulator fault images, 3000 insulator images are selected to serve as a training set, wherein 809 marked insulator fault images are included, and the rest 1500 images are used as a test set.
S4, establishing an insulator positioning detection model based on deep learning, wherein the detection model specifically comprises a backbone network, a pyramid pooling network, a feature fusion network and a target detection network which are connected in sequence; and establishing a three-scale prediction YOLOv4-tiny insulator fault detection model.
And S5, training the insulator positioning detection model and the insulator fault detection model respectively by using the insulator image training set and the insulator fault image training set to obtain the trained insulator positioning detection model and insulator fault detection model.
And S6, inputting the test set image into the insulator positioning detection model in the step S5, detecting the position of the insulator by using the insulator positioning detection model, taking the position of the insulator as an interested area, performing fault detection on the insulator in the interested area by using the insulator fault detection model, and outputting a detection result on the insulator image of the test set.
With reference to fig. 2, the deep learning-based insulator positioning detection model includes a backbone network, a pyramid pooling network, a feature fusion network, and a detection network, which are connected in sequence; the backbone network comprises a first transition convolutional layer, a first CSP module (CSP cross-stage local network), a second CSP module, a third CSP module, a fourth CSP module and a fifth CSP module which are sequentially connected, wherein the first transition convolutional layer is a 3 × 3 × 32 convolutional layer and extracts image features to obtain a 416 × 416 × 32 feature map, the first CSP module extracts image features to obtain a 208 × 208 × 64 feature map, the second CSP module extracts image features to obtain a 104 × 104 × 128 feature map, the third CSP module extracts image features to obtain a 52 × 52 × 256 feature map, the fourth CSP module extracts image features to obtain a 26 × 26 × 512 feature map, and the fifth CSP module extracts image features to obtain a 13 × 13 × 1024 feature map; and the output of the third CSP module, the output of the fourth CSP module and the output of the fifth CSP module are subjected to pooling operation of a three-scale pyramid pooling network, three-scale feature fusion is performed by using a feature fusion network, and 52 multiplied by 52, 26 multiplied by 26 and 13 multiplied by 13 features are output to predict insulators in the image after passing through a residual error network.
With reference to fig. 3 (a), the first CSP module includes a first downsampling layer, a first bypass convolution layer, and 1 residual error unit, where the first downsampling layer is a convolution layer with a 3 × 3 × 64 step size of 2, the first bypass convolution layer is a convolution layer with a 3 × 3 × 64 step size, the residual error unit is formed by connecting 1 × 1 × 32 convolution layers, 3 × 3 × 64 convolution layers, and a shortcut, and an output of the residual error unit is fused with an output of the first bypass convolution layer to obtain a feature 208 × 208 × 64.
With reference to fig. 3 (b), the second CSP module includes a second down-sampling layer, a second bypass convolution layer, and 2 residual error units, where the second down-sampling layer is a convolution layer with a 3 × 3 × 128 step size of 2, the second bypass convolution layer is a convolution layer with a 3 × 3 × 128 step size, the residual error unit is formed by connecting 1 × 1 × 64 convolution layers, 3 × 3 × 128 convolution layers, and a shortcut, and an output of the residual error unit and an output of the second bypass convolution layer are fused to obtain a characteristic 104 × 104 × 128.
With reference to fig. 3 (c), the third CSP module includes a third downsampling layer, a second transition convolutional layer, a third bypass convolutional layer, 4 density connection units, and 4 residual error units, the third downsampling layer is a convolutional layer with a 3 × 3 × 256 step size of 2, the third bypass convolutional layer is a convolutional layer with a 3 × 3 × 256 step size, the second transition convolutional layer is a 1 × 1 × 128 convolutional layer, outputs of the third downsampling layer are respectively connected to the third bypass convolutional layer and the second transition convolutional layer, outputs of the second transition convolutional layer are connected to 4 sequentially connected residual error units, the residual error unit is formed by connecting 1 × 1 × 32 convolutional layers, 3 × 3 × 32 convolutional layers, and Concat, outputs of the density units are connected to 4 sequentially connected residual error units, the residual error unit is formed by connecting 1 × 1 × 128 convolutional layers, 3 × 3 × 256 convolutional layers, and a short residual error unit, and outputs of the bypass unit are fused to 52 × 52 × 256 features of the third output.
With reference to fig. 3 (d), the fourth CSP module includes a fourth downsampling layer, a third transition convolutional layer, a fourth bypass convolutional layer, 4 sense units, and 4 residual error units, where the fourth downsampling layer is a convolutional layer with a 3 × 3 × 512 step size of 2, the fourth bypass convolutional layer is a convolutional layer with a 3 × 3 × 512 step size, the third transition convolutional layer is a 1 × 1 × 256 convolutional layer, outputs of the fourth downsampling layer are respectively connected to the fourth bypass convolutional layer and the third transition convolutional layer, outputs of the third transition convolutional layer are connected to 4 sequentially connected sense units, the sense unit is formed by connecting 1 × 1 × 64 convolutional layers, 3 × 3 × 64 convolutional layers, and Concat, outputs of the sense unit are connected to 4 sequentially connected residual error units, the residual error unit is formed by connecting 1 × 1 × 256 convolutional layers, 3 × 512 convolutional layers, and a short cut, and outputs of the residual error units and the fourth bypass convolutional layers obtain 26 × 26 residual error fusion characteristics.
With reference to fig. 3 (e), the fifth CSP module includes a fifth downsampling layer, a fourth transition convolutional layer, a fifth bypass convolutional layer, 4 transmit units, and 4 residual units, where the fifth downsampling layer is a convolutional layer with a 3 × 3 × 1024 step size of 2, the fifth bypass convolutional layer is a convolutional layer with a 3 × 3 × 1024, the fourth transition convolutional layer is a 1 × 1 × 512 convolutional layer, outputs of the fifth downsampling layer are respectively connected to the fifth bypass convolutional layer and the fourth transition convolutional layer, outputs of the fourth transition convolutional layer are connected to 4 sequentially connected transmit units, the transmit unit is formed by connecting 1 × 1 × 128 convolutional layers, 3 × 3 × 128 convolutional layers, and Concat, outputs of the transmit units are connected to 4 sequentially connected residual units, the residual units are formed by connecting 1 × 1 × 512 convolutional layers, 3 × 3 × 1024 convolutional layers, and outputs of the fourth bypass convolutional layers, and output of the 4 × 13 × 1024 residual units are fused to obtain a fused feature.
The pyramid pooling network used for the insulator positioning detection model is a three-scale pyramid pooling structure, an output feature layer 52 x 256 is connected with a large-scale pyramid pooling structure SPP3, an output feature layer 26 x 512 is connected with a medium-scale pyramid pooling structure SPP2, an output feature layer 13 x 1024 is connected with a small-scale pyramid pooling structure SPP1, and in combination with fig. 4, each scale pyramid pooling structure comprises three maximum pooling layers, and the sizes of filters corresponding to the three maximum pooling layers are respectively 13 x 13, 9 x 9 and 5 x 5. And obtaining three local feature graphs by inputting the features through three maximal pooling operations of different scales, and then fusing the input features and the three local feature graphs to obtain the feature graph. The use of the three-scale pyramid pooling structure can greatly increase the receiving range of the local area characteristic diagram, obtain richer local characteristic information and improve the accuracy of prediction.
With reference to fig. 1, three effective feature layers extracted by the backbone network are used in the feature fusion network, where 52 × 52 × 256 of the feature layers correspond to the first large-scale feature layer LFL0, 26 × 26 × 512 of the feature layers correspond to the first medium-scale feature layer MFL0, and 13 × 13 × 1024 of the feature layers correspond to the first small-scale feature layer SFL0; the first small-scale feature layer SFL0 obtains a second small-scale feature layer SFL1 through pyramid pooling structure SPP1 and convolution operation, the first medium-scale feature layer MFL0 obtains a second medium-scale feature layer MFL1 through pyramid pooling structure SPP2 and convolution operation, and the first large-scale feature layer LFL0 obtains a second large-scale feature layer LFL1 through pyramid pooling structure SPP3 and convolution operation; the second small-scale feature layer SFL is subjected to upsampling operation and then fused with the second medium-scale feature layer MFL1 to obtain a third medium-scale feature layer MFL2, and the third medium-scale feature layer MFL2 is subjected to upsampling operation and then fused with the second large-scale feature layer LFL1 to obtain a third large-scale feature layer LFL2; the third large-scale feature layer LFL2 is fused with the third medium-scale feature layer MFL2 after downsampling operation to obtain a fourth medium-scale feature layer MFL3, and the fourth medium-scale feature layer MFL3 is fused with the second small-scale feature layer SFL1 after downsampling operation to obtain a third small-scale feature layer SFL2; and the output features 13 × 13, 26 × 26 and 52 × 52 of the third small-scale feature layer SFL2, the fourth medium-scale feature layer MFL3 and the third large-scale feature layer LFL2 are respectively subjected to three residual modules and then are sent to a three-scale target detection layer, and the three-scale target detection layer respectively predicts the insulator images with the feature scales of 13 × 13, 26 × 26 and 52 × 52.
With reference to fig. 5, the third residual module is composed of 3 residual units, each residual unit is composed of a 1 × 1 × 128 convolutional layer, a 3 × 3 × 256 convolutional layer, and a shortcut, and an output characteristic of the third residual module is 52 × 52 × 256; the second residual error module consists of 3 residual error units, each residual error unit consists of a 1 × 1 × 256 convolutional layer, a 3 × 3 × 512 convolutional layer and a short connection, and the output characteristic of the second residual error module is 26 × 26 × 512; the first residual module consists of 3 residual units, the residual units consist of 1 multiplied by 512 convolutional layers, 3 multiplied by 1024 convolutional layers and shortcut connection, and the output characteristic of the first residual module is 13 multiplied by 1024. The target detection network adopts the residual error module to replace 5 convolutional layers, so that the disappearance of characteristic gradients or gradient explosion can be avoided.
In order to verify the effectiveness of the insulator positioning detection model, a comparison experiment is carried out on the detection model and YOLOv3 in a test set. The experimental conditions were as follows: in terms of hardware, the CPU is of 3.60GHz
Core i9-9900K, and the total memory is 32GB; 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. The test indexes (average accuracy, accuracy and recall) of the two test models are shown in table 1. The average accuracy rates of the two detection networks are respectively 90.3% and 94.4%; the accuracy rates of the two detection networks are respectively 90% and 94%; the recall rates of the two detection networks were 91% and 95%, respectively. Therefore, the average accuracy rate, the accuracy rate and the recall rate are comprehensively considered, and the detection model has better detection performance compared with the YOLOv 3. The insulator fault detection model adopts three-scale prediction YOLOv4-tiny, and the memory occupancy of YOLOv4-tiny is only 25MB, which is smaller than 33MB of YOLOv3-tiny, and is far smaller than the memory occupancy of YOLOv3 and YOLOv 4. And the YOLOv4-tiny detection speed is high, the detection of insulator fault images of more than 200 pieces per second can be realized, and the detection accuracy rate reaches more than 90%.
Table 1: test indexes of two detection models
In order to verify the accuracy of the insulator positioning and fault detection of the detection model of the invention under different background interferences and different scales, fig. 6-9 show the detection results of the positioning detection model and the fault detection model of the invention. Fig. 6 shows the experimental results of the field-based interference, where 3 insulators are located by the insulator location model in fig. 6 (a), and 1 insulator fault is detected by the insulator fault detection model in fig. 6 (b) based on fig. 6 (a). Fig. 7 shows the experimental results of the river as the background interference, where 3 insulators are located by the insulator location model in fig. 7 (a), and 3 insulator faults are detected by the insulator fault detection model in fig. 7 (b) based on fig. 7 (a). Fig. 8 is an experimental result of interference using a power rack as a background, in fig. 8 (a), the insulator positioning model positions 3 insulators including insulators of a shielding portion, and in fig. 8 (b), the insulator fault detection model detects 2 insulator faults on the basis of fig. 8 (a). Fig. 9 shows the experimental result under the complex background interference, where 4 insulators are located by the insulator location model in fig. 9 (a), and fig. 9 (b) shows that 2 insulator faults are detected by the insulator fault detection model based on fig. 9 (a).
In summary, in the embodiment of the present invention, the insulator string drop fault detection method based on the cascade detection model first locates the insulator in the image by using the insulator location detection model, and then performs fault detection on the located insulator by using the insulator fault detection model, so that the accuracy of insulator fault detection can be effectively improved; the method has good robustness for insulator detection with different scales, complex background interference and shielding in aerial images, and can effectively realize insulator fault detection in power transmission line inspection; according to the insulator positioning detection model, cross-stage local network strengthening feature multiplexing and propagation are introduced into a backbone network, so that the target feature under a complex environment can be effectively extracted, the detection precision is guaranteed, and the detection accuracy and recall rate of the target are improved; the insulator fault detection model adopts the three-scale predicted YOLOv4-tiny, YOLOv4-tiny not only has small memory occupation amount, but also has fast detection speed and higher detection accuracy.
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. A cascading detection model-based insulator string drop fault detection method is characterized by comprising the following steps:
s1, collecting a power transmission line inspection image, and establishing an insulator image set and an insulator fault image set;
s2, traversing the insulator image set and the insulator fault image set, and marking the insulator position and the insulator fault position in the image to obtain an insulator data set and an insulator fault data set;
s3, dividing the insulator data set into an insulator training set and an insulator testing set, and dividing the insulator fault data set into an insulator fault training set and an insulator fault testing set;
s4, establishing an insulator positioning detection model based on deep learning, wherein the insulator positioning detection model specifically comprises a backbone network, a pyramid pooling network, a feature fusion network and a target detection network which are connected in sequence; establishing a three-scale prediction YOLOv4-tiny insulator fault detection model;
s5, training an insulator positioning detection model and an insulator fault detection model by utilizing an insulator training set and an insulator fault training set respectively to obtain a trained insulator positioning detection model and a trained insulator fault detection model;
s6, inputting the insulator test set image and the insulator fault test set image into the insulator positioning detection model in the step S5, detecting the position of the insulator by using the insulator positioning detection model, taking the position of the insulator as an interested area, performing fault detection on the insulator in the interested area by using the insulator fault detection model, and finally outputting a detection result on the insulator test set image;
the backbone network comprises a first transition convolutional layer, a first CSP module, a second CSP module, a third CSP module, a fourth CSP module and a fifth CSP module which are sequentially connected, the first transition convolutional layer is a convolutional layer of 3 multiplied by 32, image characteristics are extracted, a characteristic diagram of 416 multiplied by 32 is obtained, the first CSP module extracts the image characteristics, obtaining a 208 × 208 × 64 feature map, extracting image features by the second CSP module to obtain a 104 × 104 × 128 feature map, extracting image features by the third CSP module to obtain a 52 × 52 × 256 feature map, extracting image features by the fourth CSP module to obtain a 26 × 26 × 512 feature map, extracting image features by the fifth CSP module to obtain a 13 × 13 × 1024 feature map; the outputs of the third CSP module, the fourth CSP module and the fifth CSP module are respectively subjected to pyramid pooling network pooling operation, three-scale feature fusion is carried out by using a feature fusion network, and 52 × 52, 26 × 26 and 13 × 13 features are output to predict the insulator position and the insulator fault position in the image after passing through a residual error network;
the first CSP module comprises a first down-sampling layer, a first bypass convolution layer and a residual error unit, wherein the first down-sampling layer is a convolution layer with a 3 x 64 step length of 2, the first bypass convolution layer is a convolution layer with a 3 x 64 step length, the residual error unit is formed by connecting a 1 x 32 convolution layer, a 3 x 64 convolution layer and a short cut, and the output of the residual error unit and the output of the first bypass convolution layer are fused to obtain a characteristic of 208 x 64;
the second CSP module comprises a second down-sampling layer, a second bypass convolution layer and 2 residual error units, wherein the second down-sampling layer is a convolution layer with 3 x 128 step length of 2, the second bypass convolution layer is a convolution layer with 3 x 128, the residual error units are formed by connecting 1 x 64 convolution layer, 3 x 128 convolution layer and short cut, and the output of the residual error units and the output of the second bypass convolution layer are fused to obtain a characteristic of 104 x 128;
the third CSP module comprises a third downsampling layer, a second transition convolutional layer, a third bypass convolutional layer, 4 sense units and 4 residual error units, wherein the third downsampling layer is a convolutional layer with a 3 × 3 × 256 step length of 2, the third bypass convolutional layer is a convolutional layer with a 3 × 3 × 256 step length, the second transition convolutional layer is a 1 × 1 × 128 convolutional layer, the output of the third downsampling layer is respectively connected with the third bypass convolutional layer and the second transition convolutional layer, the output of the second transition convolutional layer is connected with the 4 sense units which are connected in sequence, the sense unit is formed by connecting the 1 × 1 × 32 convolutional layer, the 3 × 3 × 32 convolutional layer and a Concat, the output of the sense unit is connected with the 4 residual error units which are connected in sequence, the residual error unit is formed by connecting the 1 × 1 × 128 convolutional layer, the 3 × 3 × 256 convolutional layer and a short cut, and the output of the residual error unit and the third bypass convolutional layer are fused to obtain a characteristic 52 × 52 × 256.
2. The method as claimed in claim 1, wherein the fourth CSP module comprises a fourth down-sampling layer, a third transition convolutional layer, a fourth bypass convolutional layer, 4 sense units, and 4 residual units, the fourth down-sampling layer is a convolutional layer with a step size of 3 x 512, the fourth bypass convolutional layer is a convolutional layer with a step size of 3 x 512, the third transition convolutional layer is a convolutional layer with a step size of 1 x 256, the output of the fourth down-sampling layer is respectively connected with the fourth bypass convolution layer and the third transition convolution layer, the output of the third transition convolution layer is connected with 4 sequentially connected Dense units, each Dense unit is formed by connecting 1 × 1 × 64 convolution layers, 3 × 3 × 64 convolution layers and Concat, the output of each Dense unit is connected with 4 sequentially connected residual error units, each residual error unit is formed by connecting 1 × 1 × 256 convolution layers, 3 × 3 × 512 convolution layers and short, and the output of each residual error unit and the output of the fourth bypass convolution layer are fused to obtain a characteristic 26 × 26 × 512.
3. The method as claimed in claim 2, wherein the fifth CSP module includes a fifth downsampling layer, a fourth transition convolutional layer, a fifth bypass convolutional layer, 4 sense units, and 4 residual error units, the fifth downsampling layer is a convolutional layer with 3 × 3 × 1024 step size of 2, the fifth bypass convolutional layer is a convolutional layer with 3 × 3 × 1024, the fourth transition convolutional layer is a convolutional layer with 1 × 1 × 512, the outputs of the fifth downsampling layer are respectively connected to the fifth bypass convolutional layer and the fourth transition convolutional layer, the output of the fourth transition convolutional layer is connected to 4 sequentially connected sense units, the sense unit is composed of a convolutional layer with 1 × 128, a convolutional layer with 3 × 3 × 128, and a Concat connection, the output of the sense unit is connected to 4 sequentially connected residual error units, the residual error unit is composed of a convolutional layer with 1 × 1 × 128, a convolutional layer with 3 × 3 × 1024, and a short × t connection, and the output of the fourth bypass convolutional unit is fused with 13 side output.
4. The method for detecting the string failure of the insulator based on the cascading detection model as claimed in claim 1, wherein the pyramid pooling network is a three-dimensional pyramid pooling structure, the output feature layer 52 × 52 × 256 is connected to the large-scale pyramid pooling structure, the output feature layer 26 × 26 × 512 is connected to the medium-scale pyramid pooling structure, the output feature layer 13 × 13 × 1024 is connected to the small-scale pyramid pooling structure, each scale pyramid pooling structure includes three maximum pooling layers, and the filters corresponding to the three maximum pooling layers have sizes of 13 × 13, 9 × 9, and 5 × 5, respectively.
5. The method for detecting the insulator string drop fault based on the cascade detection model as claimed in claim 4, wherein three effective feature layers extracted through a backbone network are used in a feature fusion network, a feature layer 52 x 256 corresponds to a first large-scale feature layer, a feature layer 26 x 512 corresponds to a first medium-scale feature layer, and a feature layer 13 x 1024 corresponds to a first small-scale feature layer; the first small-scale feature layer is subjected to pyramid pooling and convolution operation to obtain a second small-scale feature layer, the first medium-scale feature layer is subjected to pyramid pooling and convolution operation to obtain a second medium-scale feature layer, and the first large-scale feature layer is subjected to pyramid pooling and convolution operation to obtain a second large-scale feature layer; the second small-scale feature layer is fused with the second medium-scale feature layer after being subjected to up-sampling operation to obtain a third medium-scale feature layer, and the third medium-scale feature layer is fused with the second large-scale feature layer after being subjected to up-sampling operation to obtain a third large-scale feature layer; the third large-scale feature layer is fused with the third medium-scale feature layer after downsampling operation to obtain a fourth medium-scale feature layer, and the fourth medium-scale feature layer is fused with the second small-scale feature layer after downsampling operation to obtain a third small-scale feature layer; and the output characteristics of the third small-scale characteristic layer, the fourth medium-scale characteristic layer and the third large-scale characteristic layer, namely 13 multiplied by 13, 26 multiplied by 26 and 52 multiplied by 52, are respectively subjected to three residual modules and then are sent to a three-scale target detection network, and the three-scale target detection network respectively predicts the insulator images with the characteristic scales of 13 multiplied by 13, 26 multiplied by 26 and 52 multiplied by 52.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110437796.7A CN113112489B (en) | 2021-04-22 | 2021-04-22 | Insulator string-dropping fault detection method based on cascade detection model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110437796.7A CN113112489B (en) | 2021-04-22 | 2021-04-22 | Insulator string-dropping fault detection method based on cascade detection model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113112489A CN113112489A (en) | 2021-07-13 |
| CN113112489B true CN113112489B (en) | 2022-11-15 |
Family
ID=76719754
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110437796.7A Active CN113112489B (en) | 2021-04-22 | 2021-04-22 | Insulator string-dropping fault detection method based on cascade detection model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113112489B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113538411A (en) * | 2021-08-06 | 2021-10-22 | 广东电网有限责任公司 | Insulator defect detection method and device |
| CN115619763A (en) * | 2022-10-31 | 2023-01-17 | 国网湖北省电力有限公司超高压公司 | Power transmission line small part image target detection method based on multistage reasoning detection |
| CN116596908B (en) * | 2023-05-30 | 2024-02-06 | 南京亦鑫同电气有限责任公司 | Wire and cable safety state assessment method and system based on data processing |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AU2019101133A4 (en) * | 2019-09-30 | 2019-10-31 | Bo, Yaxin MISS | Fast vehicle detection using augmented dataset based on RetinaNet |
| CN112101433A (en) * | 2020-09-04 | 2020-12-18 | 东南大学 | Automatic lane-dividing vehicle counting method based on YOLO V4 and DeepsORT |
| CN112183578A (en) * | 2020-09-01 | 2021-01-05 | 国网宁夏电力有限公司检修公司 | Target detection method, medium and system |
| CN112184666A (en) * | 2020-09-28 | 2021-01-05 | 国网江西省电力有限公司电力科学研究院 | Target equipment detection method, device and system |
| CN112258446A (en) * | 2020-08-31 | 2021-01-22 | 洛阳中信成像智能科技有限公司 | Industrial part defect detection method based on improved YOLO algorithm |
| CN112561899A (en) * | 2020-12-23 | 2021-03-26 | 镇江市高等专科学校 | Electric power inspection image identification method |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11195057B2 (en) * | 2014-03-18 | 2021-12-07 | Z Advanced Computing, Inc. | System and method for extremely efficient image and pattern recognition and artificial intelligence platform |
| CN110033453B (en) * | 2019-04-18 | 2023-02-24 | 国网山西省电力公司电力科学研究院 | Power transmission and transformation line insulator aerial image fault detection method based on improved YOLOv3 |
-
2021
- 2021-04-22 CN CN202110437796.7A patent/CN113112489B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AU2019101133A4 (en) * | 2019-09-30 | 2019-10-31 | Bo, Yaxin MISS | Fast vehicle detection using augmented dataset based on RetinaNet |
| CN112258446A (en) * | 2020-08-31 | 2021-01-22 | 洛阳中信成像智能科技有限公司 | Industrial part defect detection method based on improved YOLO algorithm |
| CN112183578A (en) * | 2020-09-01 | 2021-01-05 | 国网宁夏电力有限公司检修公司 | Target detection method, medium and system |
| CN112101433A (en) * | 2020-09-04 | 2020-12-18 | 东南大学 | Automatic lane-dividing vehicle counting method based on YOLO V4 and DeepsORT |
| CN112184666A (en) * | 2020-09-28 | 2021-01-05 | 国网江西省电力有限公司电力科学研究院 | Target equipment detection method, device and system |
| CN112561899A (en) * | 2020-12-23 | 2021-03-26 | 镇江市高等专科学校 | Electric power inspection image identification method |
Non-Patent Citations (3)
| Title |
|---|
| detection of power line insulator defects using aerial images analyzed with convolutional neural networks;xian tao et al.;《IEEE transactions on systems，man，and cybernetics：systems》;20181014;第50卷(第4期);第1486-1498页 * |
| performance enhancement of yolov3 by adding prediction layers with spatial pyramid pooling for vehicle detection;kwang-ju kim et al.;《2018 15th IEEE international conference on advanced video and signal》;20190214;第1-6页 * |
| 基于深度学习框架的装配机器人零件实时检测方法;余永维等;《兵工学报》;20201231(第10期);第2122-2130页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113112489A (en) | 2021-07-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113112489B (en) | Insulator string-dropping fault detection method based on cascade detection model | |
| CN110188685B (en) | Target counting method and system based on double-attention multi-scale cascade network | |
| CN110084124A (en) | Feature based on feature pyramid network enhances object detection method | |
| CN114240878A (en) | Routing inspection scene-oriented insulator defect detection neural network construction and optimization method | |
| CN112861729B (en) | Real-time depth completion method based on pseudo-depth map guidance | |
| CN109934222A (en) | A kind of insulator chain self-destruction recognition methods based on transfer learning | |
| CN115294476B (en) | Edge computing intelligent detection method and device for unmanned aerial vehicle power inspection | |
| CN115100549A (en) | Transmission line hardware detection method based on improved YOLOv5 | |
| CN112288700A (en) | Rail defect detection method | |
| CN116862847A (en) | Interactive segmentation method and system for infrared image power equipment | |
| CN116543346A (en) | Deep learning-based transmission line video mountain fire detection method | |
| CN112837281B (en) | Pin defect identification method, device and equipment based on cascade convolution neural network | |
| CN113012107B (en) | Power grid defect detection method and system | |
| CN113095252A (en) | Insulator fault image identification method for power inspection | |
| CN113095420A (en) | Insulator fault detection method based on improved YOLOv3 | |
| CN113095253B (en) | Insulator detection method for unmanned aerial vehicle inspection transmission line | |
| CN110033105A (en) | Suspension system fault detection method for unbalanced data set under complex working condition | |
| CN113111802B (en) | Insulator detection model based on deep neural network | |
| CN117113066B (en) | Transmission line insulator defect detection method based on computer vision | |
| CN113033489A (en) | Power transmission line insulator identification and positioning method based on lightweight deep learning algorithm | |
| CN113033556A (en) | Insulator rapid distinguishing and positioning method and system based on machine vision | |
| CN111935762A (en) | EWT and CNN-based distribution network fault diagnosis method and system under 5G carrier network | |
| Yang | Method of insulator defect detection based on the improved YOLOv5s | |
| CN117593268A (en) | Multi-network fusion type transmission line insulator defect detection method | |
| Zhao et al. | Lightweight Neural Network for Component Recognition and Fault Detection of Large-Scale Transmission Line |
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 |
Effective date of registration: 20231101 Address after: No. 186, Cuiping Road West, Laibin City, Guangxi Zhuang Autonomous Region Patentee after: LAIBIN POWER SUPPLY BUREAU OF GUANGXI POWER GRID CO.,LTD. Patentee after: Laibin Binsheng Electric Power Engineering Co.,Ltd. Address before: No. 169, Jianshe West Road, Guichi District, Chizhou City, Anhui Province 247099 Patentee before: CHIZHOU University |
|
| TR01 | Transfer of patent right |