CN107274451A - Isolator detecting method and device based on shared convolutional neural networks - Google Patents
Isolator detecting method and device based on shared convolutional neural networks Download PDFInfo
- Publication number
- CN107274451A CN107274451A CN201710347616.XA CN201710347616A CN107274451A CN 107274451 A CN107274451 A CN 107274451A CN 201710347616 A CN201710347616 A CN 201710347616A CN 107274451 A CN107274451 A CN 107274451A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- layer
- msubsup
- network
- 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.)
- Pending
Links
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 143
- 238000000034 method Methods 0.000 title claims abstract description 38
- 239000012212 insulator Substances 0.000 claims abstract description 99
- 238000012549 training Methods 0.000 claims abstract description 57
- 238000001514 detection method Methods 0.000 claims abstract description 40
- 230000005540 biological transmission Effects 0.000 claims abstract description 32
- 238000007689 inspection Methods 0.000 claims abstract description 16
- 238000011176 pooling Methods 0.000 claims description 52
- 238000013507 mapping Methods 0.000 claims description 23
- 238000011478 gradient descent method Methods 0.000 claims description 20
- 238000010586 diagram Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 7
- 230000007246 mechanism Effects 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000011897 real-time detection Methods 0.000 abstract description 7
- 230000005611 electricity Effects 0.000 abstract 3
- 230000006870 function Effects 0.000 description 26
- 238000004364 calculation method Methods 0.000 description 8
- 238000005315 distribution function Methods 0.000 description 7
- 238000012360 testing method Methods 0.000 description 6
- 238000013135 deep learning Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000010292 electrical insulation Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- 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
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of isolator detecting method and device based on shared convolutional neural networks.Methods described includes：The transmission line of electricity image of transformer station is shot using crusing robot；The optimum position of insulator in the transmission line of electricity image is obtained using the shared convolutional neural networks of RPN networks and Fast R CNN network trainings.The present invention is trained the insulator in obtained shared convolutional neural networks, detection transmission line of electricity image using the RPN networks and Fast R CNN networks of shared part convolutional layer and pond layer；It can reduce computation complexity relative to prior art, reach the real-time detection to insulator under complex background, realize the accurate identification and positioning of insulator in robot inspection image.
Description
Technical Field
The invention relates to the field of power equipment key power equipment target detection, in particular to an insulator detection method and device based on a shared convolutional neural network.
Background
In recent years, ensuring the reliability and the operation condition of a power transmission line becomes an important content for building a smart grid. The safe operation of the power transformation equipment is the premise for ensuring the stability and safety of the power system. The insulator is used as an indispensable insulating element of a power transmission line, and the operation condition of the insulator directly influences the reliability and safety of a power grid. Meanwhile, the insulator plays the roles of electrical insulation and support in the power transmission line; and the problems of dirt, cracks, breakage and the like on the surface seriously threaten the safe operation of the power transmission line. According to statistics, the accident with the highest proportion of the current power system faults is caused by insulator defects. Therefore, it is important to monitor the condition of the insulator and complete the fault diagnosis in time.
At present, an inspection robot and an unmanned aerial vehicle become important modes of power inspection, a camera loaded on a platform is used for acquiring a large amount of insulator image information, and if the large amount of images are judged and read by naked eyes of workers, the workload is high, phenomena of missing judgment and misjudgment are easy to occur, and potential safety hazards existing in insulators are difficult to accurately find. The important premise for realizing automatic fault detection is to identify and locate the insulator in the image, so that it is very necessary to research an automatic insulator detection method.
Because the background of the insulator pictures shot by the inspection robot and the unmanned aerial vehicle is complex, most of the traditional insulator detection methods mostly need manual detection one by one, so that a large amount of manpower and material resources are consumed, casualties are easily caused, and phenomena such as missing judgment and misjudgment easily occur.
At present, although some automatic insulator identification methods appear, the methods are prone to generate false identification results because the background of an image of a substation is complex and other electric equipment similar to the insulator in shape, such as a current transformer, a lightning arrester and the like, exists in the image; or when the data volume is increased, the calculation complexity and the calculation time are greatly increased, and the requirement of real-time detection cannot be met.
Disclosure of Invention
The present invention provides a shared convolutional neural network based insulator detection method and apparatus that overcomes, or at least partially solves, the above-mentioned problems.
According to one aspect of the invention, an insulator detection method based on a shared convolutional neural network is provided, which comprises the following steps:
shooting an image of a power transmission line of the transformer substation by using the inspection robot;
and acquiring the optimal position of the insulator in the image of the power transmission line by utilizing the shared convolutional neural network trained by the RPN network and the Fast R-CNN network.
The invention provides an insulator detection method based on a shared convolutional neural network, which is characterized in that the insulator in an image of a power transmission line is detected by using the shared convolutional neural network obtained by training an RPN network and a Fast R-CNN network of a convolutional layer and a pooling layer which are shared partially; compared with the prior art, the method can reduce the calculation complexity, achieve the real-time detection of the insulator under the complex background, and realize the accurate identification and positioning of the insulator in the inspection image of the robot.
Further, the shared convolutional neural network is obtained by:
s1, performing iterative training on the RPN by using the power transmission line image training sample set, and acquiring a first coarse candidate region and an initial RPN by using mapping mechanisms with different proportions and different sizes;
s2, performing iterative training on the Fast R-CNN network by using the first coarse candidate region to obtain an initial Fast R-CNN network;
s3, keeping the convolution layer parameters of the initial Fast R-CNN network unchanged, and performing first parameter fine adjustment on the initial RPN network by using the training sample set to obtain a second coarse candidate region and an optimized RPN network;
s4, keeping the convolution layer parameters of the initial Fast R-CNN network unchanged, and performing second parameter fine adjustment on the initial Fast R-CNN network by using the training sample set and the second coarse candidate region to obtain an optimized Fast R-CNN network;
wherein the optimized RPN network and the optimized Fast R-CNN share a partial convolutional layer and a pooling layer, thereby obtaining a shared convolutional neural network.
Further, the S1 further includes:
s1.1, constructing an RPN network comprising 6 convolutional layers, 2 pooling layers, 1 classification layer and 1 frame regression layer;
s1.2, inputting the training sample set into the RPN, extracting features through a convolutional layer, obtaining a feature map of a seventh layer as a first feature map through pooling layer mapping features, and generating anchor boxes with different sizes and different proportions by adopting mapping mechanisms with different proportions and different sizes;
s1.3, based on the anchor boxes, selecting a positive sample and a negative sample according to a first preset rule, unifying the corresponding features of the positive sample and the negative sample into the same size, inputting the unified features into a classification layer and a frame regression layer to obtain the first coarse candidate region, and obtaining an initial RPN.
Further, the S2 further includes:
s2.1, constructing a Fast R-CNN network comprising 5 convolutional layers, 2 pooling layers, 1 ROI pooling layer, 1 classification layer and 1 frame regression layer;
s2.2, inputting the training sample set into the Fast R-CNN network, extracting features through convolutional layers, mapping the features through pooling layers, and acquiring the feature of the last convolutional layer as a second feature map;
s2.3, mapping the coarse candidate region onto the second feature map, and adjusting each feature on the second feature map to be a fixed size through the ROI pooling layer;
and S2.4, inputting each characteristic into a classification layer and a frame regression layer, selecting a positive sample and a negative sample according to a second preset rule, and updating the weight of each layer of the Fast R-CNN network by using a random gradient descent method and a back propagation algorithm to obtain an initial Fast R-CNN network.
Further, iteratively training the PRN network in S1 and fine-tuning the first parameter in S3 includes:
inputting a training sample set into the initial RPN network, and updating the weights of the last convolutional layer, the classification layer and the frame regression layer of the initial RPN network by using a back propagation method and a gradient descent method;
the iterative training of the Fast R-CNN network in S2 and the second parameter fine-tuning in S4 include:
inputting the coarse candidate region into a Fast R-CNN network, and updating the weights of the ROI pooling layer, the classification layer and the frame regression layer in the initial Fast R-CNN network by using a gradient descent method and a back propagation algorithm.
Further, the convolutional layer of the RPN network in S1.2 and the convolutional layer of the Fast R-CNN network in S2.2 are respectively characterized by the following formulas:
the pooling layer maps features by:
wherein,an input feature map representing the l-th layer;
a characteristic diagram representing the output of each layer, l ∈ {1,2,3,4,5,6} represents the number of layers, k is the convolution kernel, b is the bias value, down () is the sampling function, β represents the weight of the pooling layer, and the activation function is
The step of generating anchor boxes with different sizes and proportions in S1.2 comprises the following steps:
sequentially sliding the 3 x 3 sliding windows on the first feature map output by the seventh layer, and mapping the center points of the sliding windows to the original image;
selecting the area of the pixels around the central point as 1282、2562And 5122And 9 anchor boxes with the length-width ratios of 1:1, 1:2 and 2:1 respectively, and mapping each anchor box into a 256-dimensional vector.
Further, the classification function of the classification layer of the RPN network in S1.3 and the classification function of the classification layer of the Fast R-CNN network in S2.4 are softmax functions of the following formula:
wherein P (i) is the probability of the class to which it belongs,for model parameters, x is the input and k is the number of classification categories.
The bounding box regression layer adjusts each anchor box region using the following equation:
wherein x and y represent coordinates of a center point of each insulating subframe, w and h represent a length and a width of each insulating subframe, and t represents a prediction box;
wherein x, y, w and h are coordinates of the center point, length and width of the prediction frame, and xa、ya、waAnd haCenter point coordinates, length and width, x, representing candidate region box*、y*、w*、h*Representing the coordinates of the center point, the length and the width of the real box.
Further, the loss function in the iterative training of the PRN network in S1 and the first parameter fine tuning in S3 is:
wherein q isiThe probability of being the target is predicted for the anchor,is 0 or 1, if the positive sample is 1, if the negative sample is 0; t is ti＝{tx,ty,tw,thDenotes the 4 parameterized coordinates of the predicted bounding box, LclsLogarithmic loss in two categories (target and non-target), LregIs the regression loss.
Further, the loss function in the iterative training of the Fast R-CNN network in S2 and the second parameter fine tuning in S4 is:
L(p,u,tu,v)＝Lcls(p,u)+λ[u≥1]Lloc(tu,v)
wherein L isclsAs a function of classification level loss, Lcls＝-logpu；
LlocAs a function of the loss of positioning of the bezel,wherein,
v＝(vx,vy,vw,vh) Representing the coordinates of the predicted insulating subframe,representing the coordinates of a real insulating subframe.
According to another aspect of the present invention, there is also provided an insulator detection apparatus based on a shared convolutional neural network, including:
the image acquisition module is used for shooting the image of the power transmission line of the transformer substation by using the inspection robot; and
and the insulator detection module is used for acquiring the optimal position of the insulator in the power transmission line image by utilizing the shared convolutional neural network trained by the RPN network and the Fast R-CNN network.
The invention provides an insulator detection method and device based on a shared convolutional neural network, which comprises the steps of firstly shooting a power transmission line of a transformer substation through a polling robot to obtain an image of the power transmission line; then, detecting insulators in the images of the power transmission line by using a shared convolutional neural network obtained by training an RPN network and a Fast R-CNN network of a convolutional layer and a pooling layer of the shared part; compared with the prior art, the method can reduce the calculation complexity, achieve the real-time detection of the insulator under the complex background, and realize the accurate identification and positioning of the insulator in the inspection image of the robot.
Drawings
Fig. 1 is a schematic diagram of an insulator detection method based on a shared convolutional neural network according to an embodiment of the present invention;
FIG. 2 is a schematic diagram illustrating a step of obtaining a shared convolutional neural network according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a shared convolutional neural network according to an embodiment of the present invention;
FIG. 4 is a schematic diagram illustrating the result of insulator detection according to an embodiment of the present invention;
fig. 5 is a schematic diagram illustrating the recall rate and the accuracy of the insulator detection result according to the embodiment of the present invention.
Detailed Description
The following detailed description of embodiments of the present invention is provided in connection with the accompanying drawings and examples. The following examples are intended to illustrate the invention but are not intended to limit the scope of the invention.
In recent years, with the development of artificial intelligence, deep learning is increasingly applied to the fields of object classification, voice recognition, target detection and the like, and breakthrough progress is achieved. Convolutional neural networks, which are one type of deep learning networks, are widely used in image processing. The method can automatically extract the characteristic information of the image, is beneficial to classification and target detection, enables the detection result to be more accurate, can reach the standard of real-time detection, and lays a foundation for judging the fault of the insulator in the future.
As shown in fig. 1, a shared convolutional neural network-based insulator detection method includes:
shooting an image of a power transmission line of the transformer substation by using the inspection robot;
and acquiring the optimal position of the insulator in the image of the power transmission line by utilizing the shared convolutional neural network trained by the RPN network and the Fast R-CNN network.
The embodiment provides an insulator detection method based on a shared convolutional neural network, which comprises the steps of firstly shooting a power transmission line of a transformer substation through an inspection robot to obtain an image of the power transmission line; then, detecting insulators in the images of the power transmission line by using a shared convolutional neural network obtained by training an RPN network and a Fast R-CNN network of a convolutional layer and a pooling layer of the shared part; compared with the prior art, the method can reduce the calculation complexity, achieve the real-time detection of the insulator under the complex background, and realize the accurate identification and positioning of the insulator in the inspection image of the robot.
In this embodiment, the RPN network and the Fast R-CNN network are both convolutional neural networks, and the shared convolutional neural network is a convolutional neural network obtained by sharing part of network layers between the RPN network and the Fast R-CNN network.
As shown in fig. 2, in one embodiment, the shared convolutional neural network is obtained by:
s1, performing iterative training on the RPN by using the power transmission line image training sample set, and acquiring a first coarse candidate region and an initial RPN by using mapping mechanisms with different proportions and different sizes;
s2, performing iterative training on the Fast R-CNN network by using the first coarse candidate region to obtain an initial Fast R-CNN network;
s3, keeping the convolution layer parameters of the initial Fast R-CNN network unchanged, and performing first parameter fine adjustment on the initial RPN network by using the training sample set to obtain a second coarse candidate region and an optimized RPN network;
s4, keeping the convolution layer parameters of the initial Fast R-CNN network unchanged, and performing second parameter fine adjustment on the initial Fast R-CNN network by using the training sample set and the second coarse candidate region to obtain an optimized Fast R-CNN network;
wherein the optimized RPN network and the optimized Fast R-CNN share a partial convolutional layer and a pooling layer, thereby obtaining a shared convolutional neural network.
In the embodiment, firstly, a training sample set is utilized to respectively and independently train an RPN network and a Fast R-CNN network to obtain an initial RPN network and an initial Fast R-CNN network; and mutually performing parameter fine adjustment by using the initial Fast R-CNN network and the initial RPN network again to obtain the final shared convolutional neural network.
The training sample set is a certain number of electric transmission line images. In this embodiment, before training the RPN network and Fast R-CNN network by using a training sample, firstly, the position of an insulator in an image is labeled by using label software, coordinates of the insulator at the upper left corner and the lower right corner in the image are recorded, and a label 'insulator' of the insulator is given.
The training network of the present embodiment is divided into two parts: inputting a sample training set with labels into an RPN network to obtain a series of insulator candidate regions with high quality and small quantity; and a second part, inputting the obtained insulator candidate region into a Fast R-CNN network to obtain the optimal position of the final insulator.
In this embodiment, the RPN network and the Fast R-CNN network share the parameters of the convolutional layer, so that the amount of calculation is small, the accuracy is high, and end-to-end detection is realized. Under the condition that the background of the transformer substation is complex, the method is good in detection performance, and the problem of instantaneity of current insulator detection is solved.
In one embodiment, the S1 further includes:
s1.1, constructing an RPN network comprising 6 convolutional layers, 2 pooling layers, 1 classification layer and 1 frame regression layer;
s1.2, inputting the training sample set into the RPN, extracting features through a convolutional layer, obtaining a feature map of a seventh layer as a first feature map through pooling layer mapping features, and generating anchor boxes with different sizes and different proportions by adopting mapping mechanisms with different proportions and different sizes;
s1.3, based on the anchor boxes, selecting a positive sample and a negative sample according to a first preset rule, unifying the corresponding features of the positive sample and the negative sample into the same size, inputting the unified features into a classification layer and a frame regression layer to obtain the first coarse candidate region, and obtaining an initial RPN.
In this embodiment, the RPN network constructed by S1.1 is composed of 6 convolutional layers, 2 pooling layers, 1 classification layer, and 1 frame regression layer. The first layer is a convolution layer, the size of a convolution kernel is 7 x 7, and 96 characteristic graphs are output; the second layer is a pooling layer, and the size of the core window is 3 x 3; the third layer is a convolution layer, the size of a convolution kernel is 5 x 5, and 256 characteristic graphs are output; the fourth layer is a pooling layer, and the size of the nuclear window is 3 x 3; the fifth layer, the sixth layer, the seventh layer and the eighth layer are convolution layers, the sizes of convolution kernels are all 3 x 3, and the number of output characteristic graphs is 384, 256 and 256 in sequence; the ninth layer and the tenth layer are connected in parallel and are respectively a classification layer and a frame regression layer. The above layers are connected together to obtain the RPN network for outputting the insulator candidate region.
And the classification layer of the RPN is used for judging whether the region on the sample image is an insulator candidate region.
Before S1.2 in this embodiment, initializing parameters of each layer of an initially constructed RPN network further includes: and initializing the parameters to be trained in the RPN network by using a Gaussian distribution function random number with the average value of 0 and the standard deviation of 0.01. The learning rate of the RPN network is set to 0.01, the learning rate is divided by 10 for 5000 iterations, and the maximum number of iterations is 15000.
Wherein the Gaussian distribution function is as follows:
where μ is the mean and σ is the standard deviation.
In this embodiment, a 10-layer RPN network is constructed through S1.1, and the constructed RPN network is initialized; then S1.2 is carried out, wherein S1.2 is completed by a convolution layer and a pooling layer and comprises a first layer to an eighth layer; and then S1.3 is implemented, wherein the S1.3 is completed by a classification layer and a frame regression layer, and anchor boxes output by the eighth layer are screened by the classification layer and the frame regression layer to obtain an insulator coarse candidate region and obtain an initial RPN network.
The step of S1.3, wherein the selection of the positive sample and the negative sample according to the first preset rule specifically comprises: and selecting the area with the intersection union (IOU) ratio to the real insulator subframe more than 0.7 as a positive sample and selecting the area with the intersection union ratio to the real insulator subframe less than 0.3 as a negative sample.
IOU＝SI/(SA+SB-SI)
Wherein S isARepresenting the area of the real insulating subframe, SBDenotes the area, S, of the anchor box mapped to the originalIIndicating the area of intersection of the two.
In one embodiment, the S2 further includes:
s2.1, constructing a Fast R-CNN network comprising 5 convolutional layers, 2 pooling layers, 1 ROI pooling layer, 1 classification layer and 1 frame regression layer;
s2.2, inputting the training sample set into the Fast R-CNN network, extracting features through convolutional layers, mapping the features through pooling layers, and acquiring the feature of the last convolutional layer as a second feature map;
s2.3, mapping the coarse candidate region onto the second feature map, and adjusting each feature on the second feature map to be a fixed size through the ROI pooling layer;
and S2.4, inputting each characteristic into a classification layer and a frame regression layer, selecting a positive sample and a negative sample according to a second preset rule, and updating the weight of each layer of the Fast R-CNN network by using a random gradient descent method and a back propagation algorithm to obtain an initial Fast R-CNN network.
In this embodiment, the Fast R-CNN network constructed in S2.1 is composed of 5 convolutional layers, 2 pooling layers, 1 ROI pooling layer, 1 classification layer, and 1 frame regression layer. The first seven layers have the same structure as the RPN network in the S1.1, the eighth layer is an ROI pooling layer, and then the output of the ROI pooling layer is respectively sent to a frame regression layer and a classification layer to obtain the final Fast R-CNN network.
Similarly, in this embodiment, before S2.2, the initialization of the Fast R-CNN network includes: the parameters to be trained in the network are initialized by Gaussian distribution function random numbers with the average value of 0 and the standard deviation of 0.01. The learning rate of the network is set to 0.01, the learning rate is divided by 10 per 10000 iterations, and the maximum number of iterations is 30000. The gaussian distribution function described in this embodiment is the same as the gaussian distribution function initialized by the RPN network in S1.1.
In this example, Fast R-CNN network is constructed by S2.1, and after initialization, S2.2, S2.3 and S2.4 are implemented. The method specifically comprises the following steps: inputting a training sample set into the Fast R-CNN network, and performing layer-by-layer operation through a convolutional layer and a pooling layer to obtain characteristics; and then mapping the insulator candidate region obtained by the RPN network in the S1 to a feature map obtained by the seventh convolutional layer, and fixing each feature to be a uniform size through the ROI pooling layer. All features are then fed into the classification and bounding box regression layers of the Fast R-CNN network.
In this embodiment S2.4, selecting the positive sample and the negative sample according to the second preset rule specifically includes: and selecting positive samples with the ratio of intersection union set of the real insulator frames being more than or equal to 0.5, and the rest negative samples.
In one embodiment, iteratively training the PRN network in S1 and fine-tuning the first parameter in S3 includes:
inputting a training sample set into the initial RPN network, and updating the weights of the last convolutional layer, the classification layer and the frame regression layer of the initial RPN network by using a back propagation method and a gradient descent method;
in this embodiment, the parameters of the trained initial Fast R-CNN network model convolutional layer are kept unchanged for fine tuning the initial RPN network. And inputting the training sample set into the initial RPN, and updating the weights of the last convolution layer, the classification layer and the frame regression layer of the initial RPN only by using a back propagation method and a gradient descent method at the moment so as to obtain the trained optimized RPN.
The iterative training of the Fast R-CNN network in S2 and the second parameter fine-tuning in S4 include:
inputting the coarse candidate region into a Fast R-CNN network, and updating the weights of the ROI pooling layer, the classification layer and the frame regression layer in the initial Fast R-CNN network by using a gradient descent method and a back propagation algorithm.
In this embodiment, the candidate region output by the initial RPN network is sent to the initial Fast R-CNN network, the convolutional layer parameters of the initial Fast R-CNN network model are kept unchanged, the weights of the ROI pooling layer, the classification layer and the frame regression layer in the Fast R-CNN network are updated by using a gradient descent method and a back propagation algorithm, and the trained optimized Fast R-CNN network is obtained for accurate detection of the insulator.
In one embodiment, the convolutional layer of the RPN network in S1.2 and the convolutional layer of the Fast R-CNN network in S2.2, respectively, are characterized by the following equations:
the pooling layer maps features by:
wherein,an input feature map representing the l-th layer;
l ∈ {1,2,3,4,5,6} represents the number of layers, k is the convolution kernel, b is the bias value, down () is the sampling function, β represents the weight of the pooling layer, the activation function is
The step of generating anchor boxes with different sizes and proportions in S1.2 comprises the following steps:
sequentially sliding the 3 x 3 sliding windows on the first feature map output by the seventh layer, and mapping the center points of the sliding windows to the original image;
selecting the area of the pixels around the central point as 1282、2562And 5122And 9 anchor boxes with the length-width ratios of 1:1, 1:2 and 2:1 respectively, and mapping each anchor box into a 256-dimensional vector.
In one embodiment, the classification function of the classification layer of the RPN network in S1.3 and the classification function of the classification layer of the Fast R-CNN network in S2.4 is a softmax function of the following formula:
wherein P (i) is the probability of the class to which it belongs,for model parameters, x is the input and k is the number of classification categories.
The bounding box regression layer adjusts each anchor box region using the following equation:
wherein x and y represent coordinates of a center point of each insulating subframe, w and h represent a length and a width of each insulating subframe, and t represents a prediction box;
wherein x, y, w and h are coordinates of the center point, length and width of the prediction frame, and xa、ya、waAnd haCenter point coordinates, length and width, x, representing candidate region box*、y*、w*、h*Representing the coordinates of the center point, the length and the width of the real box.
In this embodiment, S1.3 sends the feature vector of the eighth layer to the classification layer, determines whether the layer is an insulator by the classification layer, and adjusts the position of the frame of the insulator by the frame regression layer.
In this embodiment, S1.3 further includes: and updating the weight value of each layer by using a back propagation algorithm and a random gradient descent method, wherein the loss function in the updating process is the same as the loss function of the first parameter fine adjustment in the S3.
In one embodiment, the loss function in the iterative training of the PRN network in S1 and the first parameter fine tuning process in S3 is:
wherein q isiThe probability of being the target is predicted for the anchor,is 0 or 1, if the positive sample is 1, if the negative sample is 0; t is ti＝{tx,ty,tw,thDenotes the 4 parameterized coordinates of the predicted bounding box, LclsLogarithmic loss in two categories (target and non-target), LregIs the regression loss.
The loss function in this embodiment is a loss function in the process of adjusting the weight of the RPN network by using a back propagation algorithm and a random gradient descent method, and is the same as the loss function of the initial RPN network obtained by using the back propagation algorithm and the random gradient descent method in S1.3.
In the back propagation process, the residual E can be reduced by continuously adjusting the parameters of each layer.
The derivative of the residual E to the bias value b is:
the derivative of the residual to the weight W is:
the parameters for each layer are then updated using the following formula:
where η is a learning rate and indicates a gradient descent speed.
In one embodiment, the loss function during iterative training of the Fast R-CNN network in S2 and the second parameter fine tuning in S4 is:
L(p,u,tu,v)＝Lcls(p,u)+λ[u≥1]Lloc(tu,v)
wherein L isclsAs a function of classification level loss, Lcls＝-logpu；
LlocAs a function of the loss of positioning of the bezel,wherein,
v＝(vx,vy,vw,vh) Representing the coordinates of the predicted insulating subframe,representing the coordinates of a real insulating subframe.
The loss function in this embodiment is a loss function in the process of adjusting the Fast R-CNN network weight by using a back propagation algorithm and a random gradient descent method, and is the same as the loss function of the initial FastR-CNN network obtained by using the back propagation algorithm and the random gradient descent method in S2.4.
The invention also provides an insulator detection device based on the shared convolutional neural network, which comprises the following components:
the image acquisition module is used for shooting the image of the power transmission line of the transformer substation by using the inspection robot; and
and the insulator detection module is used for acquiring the optimal position of the insulator in the power transmission line image by utilizing the shared convolutional neural network trained by the RPN network and the Fast R-CNN network.
The invention is further analyzed with reference to the following figures.
As shown in FIG. 3, the present invention is mainly composed of two networks, including an RPN network and a Fast R-CNN network. The method is divided into two parts: inputting a sample training set with labels into an RPN network to obtain a series of insulator candidate regions with high quality and small quantity; and a second part, inputting the obtained insulator candidate region into a Fast R-CNN network to obtain the optimal position of the final insulator.
The first part comprises the steps of:
step 11: marking a training sample, collecting an image shot by a transformer substation inspection robot, marking the position of an insulator in the image by using Labelimg software, recording the coordinates of the insulator at the upper left corner and the lower right corner in the image, and giving a label 'insulator' of the insulator.
Step 12: and constructing the RPN network. The network consists of five convolutional layers, two pooling layers, a classification layer (judging whether the insulator candidate region exists) and a frame regression layer. The first layer is a convolution layer, the convolution kernel size is 7 x 7, 96 feature maps are output, the second layer is a pooling layer, the kernel window size is 3 x 3, the third layer is a convolution layer, the convolution kernel size is 5 x 5, 256 feature maps are output, the fourth layer is a pooling layer, the kernel window size is 3 x 3, the fifth layer and the sixth layer areAnd the seven layers are convolution layers, the sizes of convolution kernels are all 3 x 3, and the number of output feature maps is 384, 384 and 256 in sequence. The eighth layer is a convolution layer, the convolution kernel is 3 x 3, 3 x 3 sliding windows are used for sliding on the characteristic diagram output by the seventh layer in sequence, the center point of each sliding window is mapped to the original drawing, and the area of pixels around the center point is 1282、2562And 5122And mapping each anchor box into a 256-dimensional vector, and inputting the 256-dimensional vector into an eighth-layer classification layer and a ninth-layer frame regression layer respectively, wherein the length-width ratios of the 9 anchor boxes are 1:1, 1:2 and 2: 1. And connecting the layers together to obtain the RPN network for outputting the insulator candidate region.
Step 13: initializing parameters of each layer of the RPN convolutional neural network, and initializing parameters to be trained in the network by using Gaussian distribution function random numbers with the mean value of 0 and the standard deviation of 0.01.
Step 14: and inputting the training sample set into an RPN network, and optimizing the network by using a back propagation algorithm and a gradient descent method. And selecting a region with the ratio of the intersection union of the real insulator frame and the insulator frame larger than 0.7 as a positive sample, and selecting a region with the ratio of the intersection union of the real insulator frame and the insulator frame smaller than 0.3 as a negative sample. The learning rate of the network was set to 0.01, and the learning rate was divided by 10 for 5000 iterations, with a maximum number of iterations of 15000. And obtaining the trained RPN network model.
The second part comprises the following steps:
step 21: constructing Fast R-CNN network. Because the RPN network and the Fast R-CNN network share a part of the convolutional layer and the pooling layer, the Fast R-CNN network adds the ROI pooling layer behind the seventh layer of the RPN network, and then the output of the ROI pooling layer is respectively sent to the frame regression layer and the classification layer to obtain the final Fast R-CNN network.
Step 22: initializing parameters of each layer of the Fast R-CNN convolutional neural network, and initializing parameters to be trained in the network by using Gaussian distribution function random numbers with the mean value of 0 and standard deviation of 0.01.
Step 23: and sending the insulator candidate region obtained by the RPN network into a Fast R-CNN network for training. And optimizing the network by utilizing a back propagation algorithm and a gradient descent method. The learning rate of the network is set to 0.01, the learning rate is divided by 10 for 5000 iterations, and the maximum number of iterations is 20000. And selecting positive samples with the ratio of intersection union set of the real insulator frames being more than or equal to 0.5, and the rest negative samples. And continuously training and optimizing to obtain the trained Fast R-CNN network model.
The RPN network and the Fast R-CNN network are trained independently and do not show the parameters of the shared convolutional layer. The following steps are presented to reduce the detection time and to share convolutional layer parameters.
And step 3: and keeping the parameters of the trained Fast R-CNN network model convolution layer unchanged for fine tuning the RPN network. And inputting the training sample set into the RPN, and updating the weights of the last convolution layer, the classification layer and the frame regression layer of the RPN only by using a back propagation method and a gradient descent method to obtain the trained RPN.
And 4, step 4: and sending the candidate areas output by the sample training set and the RPN network into a Fast R-CNN network, keeping the parameter of the convolutional layer of the FastR-CNN network model unchanged, and updating the weight values of the ROI pooling layer, the classification layer and the frame regression layer in the Fast R-CNN network by using a gradient descent method and a back propagation algorithm to obtain a trained Fast R-CNN network for accurately detecting the insulator.
Therefore, the insulator detection method based on the shared convolutional neural network provided by the invention is completed in a training stage. The specific insulator testing process is further described below.
The insulator detection method based on the shared convolutional neural network utilizes a test set of 300 insulator images to carry out testing, wherein the test set comprises the objects of insulators, transmission lines, trees and the like, and the purpose of the experiment is to identify and position the insulators in the images.
Computer environment for the experiments of the invention: the operating system is Linux 14.04 version, the display card is GTX980ti, and the software platform comprises: MatlabR2014a, caffe.
Inputting the test set into the trained network model to obtain a test result; then, the detection effect of the method is measured by utilizing the recall rate and the accuracy.
Recall is the number of correctly identified insulators/number of all insulators.
The accuracy rate is the number of correctly identified insulators/the number of all detected targets.
As shown in fig. 5, it can be seen from the recall rate and the accuracy that the method of the present invention has higher recall rate and accuracy, which indicates that the method has better robustness.
In summary, the invention provides an insulator detection method based on a shared convolutional neural network, which detects insulators in an image of a power transmission line by using the shared convolutional neural network obtained by training an RPN network and a Fast R-CNN network of a convolutional layer and a pooling layer which are shared partially; compared with the prior art, the method has the advantages that the calculation complexity is reduced, the real-time detection of the insulator under the complex background is achieved, and the accurate identification and positioning of the insulator in the inspection image of the robot are realized. The RPN network and the Fast R-CNN network share the parameters of the convolutional layer, so that the calculation amount is small during detection, the accuracy rate is high, and end-to-end detection is realized. Under the condition that the background of the transformer substation is complex, the method is good in detection performance, and the problem of instantaneity of current insulator detection is solved.
Finally, the method of the present invention is only a preferred embodiment and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. An insulator detection method based on a shared convolutional neural network is characterized by comprising the following steps:
shooting an image of a power transmission line of the transformer substation by using the inspection robot;
and acquiring the optimal position of the insulator in the image of the power transmission line by utilizing the shared convolutional neural network trained by the RPN network and the Fast R-CNN network.
2. The method of claim 1, wherein the shared convolutional neural network is obtained by:
s1, performing iterative training on the RPN by using the power transmission line image training sample set, and acquiring a first coarse candidate region and an initial RPN by using mapping mechanisms with different proportions and different sizes;
s2, performing iterative training on the Fast R-CNN network by using the first coarse candidate region to obtain an initial Fast R-CNN network;
s3, keeping the convolution layer parameters of the initial Fast R-CNN network unchanged, and performing first parameter fine adjustment on the initial RPN network by using the training sample set to obtain a second coarse candidate region and an optimized RPN network;
s4, keeping the convolution layer parameters of the initial Fast R-CNN network unchanged, and performing second parameter fine adjustment on the initial Fast R-CNN network by using the training sample set and the second coarse candidate region to obtain an optimized Fast R-CNN network;
wherein the optimized RPN network and the optimized Fast R-CNN share a partial convolutional layer and a pooling layer, thereby obtaining a shared convolutional neural network.
3. The method of claim 2, wherein the S1 further comprises:
s1.1, constructing an RPN network comprising 6 convolutional layers, 2 pooling layers, 1 classification layer and 1 frame regression layer;
s1.2, inputting the training sample set into the RPN, extracting features through a convolutional layer, obtaining a feature map of a seventh layer as a first feature map through pooling layer mapping features, and generating anchor boxes with different sizes and different proportions by adopting mapping mechanisms with different proportions and different sizes;
s1.3, based on the anchor boxes, selecting a positive sample and a negative sample according to a first preset rule, unifying the corresponding features of the positive sample and the negative sample into the same size, inputting the unified features into a classification layer and a frame regression layer to obtain the first coarse candidate region, and obtaining an initial RPN.
4. The method of claim 2, wherein the S2 further comprises:
s2.1, constructing a Fast R-CNN network comprising 5 convolutional layers, 2 pooling layers, 1 ROI pooling layer, 1 classification layer and 1 frame regression layer;
s2.2, inputting the training sample set into the Fast R-CNN network, extracting features through convolutional layers, mapping the features through pooling layers, and acquiring the feature of the last convolutional layer as a second feature map;
s2.3, mapping the coarse candidate region onto the second feature map, and adjusting each feature on the second feature map to be a fixed size through the ROI pooling layer;
and S2.4, inputting each characteristic into a classification layer and a frame regression layer, selecting a positive sample and a negative sample according to a second preset rule, and updating the weight of each layer of the Fast R-CNN network by using a random gradient descent method and a back propagation algorithm to obtain an initial Fast R-CNN network.
5. The method of claim 2,
iteratively training the PRN network in S1 and fine-tuning the first parameter in S3 includes:
inputting a training sample set into the initial RPN network, and updating the weights of the last convolutional layer, the classification layer and the frame regression layer of the initial RPN network by using a back propagation method and a gradient descent method;
the iterative training of the Fast R-CNN network in S2 and the second parameter fine-tuning in S4 include:
inputting the coarse candidate region into a Fast R-CNN network, and updating the weights of the ROI pooling layer, the classification layer and the frame regression layer in the initial Fast R-CNN network by using a gradient descent method and a back propagation algorithm.
6. The method according to claim 3, wherein the convolutional layer of the RPN network in S1.2 and the convolutional layer of the Fast R-CNN network in S2.2 are characterized by the following equations, respectively:
<mrow> <msubsup> <mi>Y</mi> <mi>j</mi> <mi>l</mi> </msubsup> <mo>=</mo> <mi>f</mi> <mrow> <mo>(</mo> <msubsup> <mi>&Sigma;X</mi> <mi>i</mi> <mrow> <mi>l</mi> <mo>-</mo> <mn>1</mn> </mrow> </msubsup> <mo>*</mo> <msubsup> <mi>k</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> <mi>l</mi> </msubsup> <mo>+</mo> <msubsup> <mi>b</mi> <mi>j</mi> <mi>l</mi> </msubsup> <mo>)</mo> </mrow> <mo>;</mo> </mrow>
the pooling layer maps features by:
<mrow> <msubsup> <mi>Y</mi> <mi>j</mi> <mi>l</mi> </msubsup> <mo>=</mo> <mi>f</mi> <mrow> <mo>(</mo> <msubsup> <mi>&beta;</mi> <mi>j</mi> <mi>l</mi> </msubsup> <mi>d</mi> <mi>o</mi> <mi>w</mi> <mi>n</mi> <mo>(</mo> <msubsup> <mi>X</mi> <mi>j</mi> <mrow> <mi>l</mi> <mo>-</mo> <mn>1</mn> </mrow> </msubsup> <mo>)</mo> <mo>+</mo> <msubsup> <mi>b</mi> <mi>j</mi> <mi>l</mi> </msubsup> <mo>)</mo> </mrow> <mo>;</mo> </mrow>
wherein,input feature diagram representing the l-th layerL ∈ {1,2,3,4,5,6} represents the number of layers, k is the convolution kernel, b is the bias value, and down () is the sampling functionNumber, β, represents the weight of the pooling layer, with the activation function being
The step of generating anchor boxes with different sizes and proportions in S1.2 comprises the following steps:
sequentially sliding the 3 × 3 sliding windows on the first feature map output by the seventh layer, and mapping the center points of the sliding windows to the original map of the first feature map;
selecting the area of the pixels around the central point as 1282、2562And 5122And 9 anchor boxes with the length-width ratios of 1:1, 1:2 and 2:1 respectively, and mapping each anchor box into a 256-dimensional vector.
7. The method of claim 3, wherein the classification function of the classification layer of the RPN network in S1.3 and the classification function of the classification layer of the Fast R-CNN network in S2.4 is a softmax function of the formula:
<mrow> <mi>P</mi> <mrow> <mo>(</mo> <mi>i</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <mi>exp</mi> <mrow> <mo>(</mo> <msubsup> <mi>&theta;</mi> <mi>i</mi> <mi>T</mi> </msubsup> <mi>x</mi> <mo>)</mo> </mrow> </mrow> <mrow> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>k</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>K</mi> </msubsup> <mi>exp</mi> <mrow> <mo>(</mo> <msubsup> <mi>&theta;</mi> <mi>k</mi> <mi>T</mi> </msubsup> <mi>x</mi> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>,</mo> </mrow>
whereinP (i) is the probability of belonging to the category,for model parameters, x is the input and k is the number of classification categories.
The bounding box regression layer adjusts each anchor box region using the following equation:
<mrow> <msub> <mi>t</mi> <mi>x</mi> </msub> <mo>=</mo> <mfrac> <mrow> <mi>x</mi> <mo>-</mo> <msub> <mi>x</mi> <mi>a</mi> </msub> </mrow> <msub> <mi>w</mi> <mi>a</mi> </msub> </mfrac> <mo>,</mo> <msub> <mi>t</mi> <mi>y</mi> </msub> <mo>=</mo> <mfrac> <mrow> <mi>y</mi> <mo>-</mo> <msub> <mi>y</mi> <mi>a</mi> </msub> </mrow> <msub> <mi>h</mi> <mi>a</mi> </msub> </mfrac> <mo>,</mo> </mrow>
<mrow> <msub> <mi>t</mi> <mi>w</mi> </msub> <mo>=</mo> <mi>l</mi> <mi>o</mi> <mi>g</mi> <mrow> <mo>(</mo> <mfrac> <mi>w</mi> <msub> <mi>w</mi> <mi>a</mi> </msub> </mfrac> <mo>)</mo> </mrow> <mo>,</mo> <msub> <mi>t</mi> <mi>h</mi> </msub> <mo>=</mo> <mi>l</mi> <mi>o</mi> <mi>g</mi> <mrow> <mo>(</mo> <mfrac> <mi>h</mi> <msub> <mi>h</mi> <mi>a</mi> </msub> </mfrac> <mo>)</mo> </mrow> <mo>,</mo> </mrow>
<mrow> <msubsup> <mi>t</mi> <mi>x</mi> <mo>*</mo> </msubsup> <mo>=</mo> <mfrac> <mrow> <msup> <mi>x</mi> <mo>*</mo> </msup> <mo>-</mo> <msub> <mi>x</mi> <mi>a</mi> </msub> </mrow> <msub> <mi>w</mi> <mi>a</mi> </msub> </mfrac> <mo>,</mo> <msubsup> <mi>t</mi> <mi>y</mi> <mo>*</mo> </msubsup> <mo>=</mo> <mfrac> <mrow> <msup> <mi>y</mi> <mo>*</mo> </msup> <mo>-</mo> <msub> <mi>y</mi> <mi>a</mi> </msub> </mrow> <msub> <mi>h</mi> <mi>a</mi> </msub> </mfrac> <mo>,</mo> </mrow>
<mrow> <msubsup> <mi>t</mi> <mi>w</mi> <mo>*</mo> </msubsup> <mo>=</mo> <mi>l</mi> <mi>o</mi> <mi>g</mi> <mrow> <mo>(</mo> <mfrac> <msup> <mi>w</mi> <mo>*</mo> </msup> <msub> <mi>w</mi> <mi>a</mi> </msub> </mfrac> <mo>)</mo> </mrow> <mo>,</mo> <msubsup> <mi>t</mi> <mi>h</mi> <mo>*</mo> </msubsup> <mo>=</mo> <mi>l</mi> <mi>o</mi> <mi>g</mi> <mrow> <mo>(</mo> <mfrac> <msup> <mi>h</mi> <mo>*</mo> </msup> <msub> <mi>h</mi> <mi>a</mi> </msub> </mfrac> <mo>)</mo> </mrow> <mo>,</mo> </mrow>
wherein x and y represent coordinates of a center point of each insulating subframe, w and h represent a length and a width of each insulating subframe, and t represents a prediction box;
wherein x, y, w and h are coordinates of the center point, length and width of the prediction frame, and xa、ya、waAnd haRepresenting candidate regionsCenter point coordinate, length and width of frame, x*、y*、w*、h*Representing the coordinates of the center point, the length and the width of the real box.
8. The method of claim 5 wherein the loss function in iteratively training a PRN network in S1 and in the first parameter fine tuning process in S3 is:
<mrow> <mi>L</mi> <mrow> <mo>(</mo> <msub> <mi>q</mi> <mi>i</mi> </msub> <mo>,</mo> <msub> <mi>t</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <msub> <mi>N</mi> <mrow> <mi>c</mi> <mi>l</mi> <mi>s</mi> </mrow> </msub> </mfrac> <munder> <mo>&Sigma;</mo> <mi>i</mi> </munder> <msub> <mi>L</mi> <mrow> <mi>c</mi> <mi>l</mi> <mi>s</mi> </mrow> </msub> <mrow> <mo>(</mo> <msub> <mi>q</mi> <mi>i</mi> </msub> <mo>,</mo> <msubsup> <mi>q</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>)</mo> </mrow> <mo>+</mo> <mi>&alpha;</mi> <mfrac> <mn>1</mn> <msub> <mi>N</mi> <mrow> <mi>r</mi> <mi>e</mi> <mi>g</mi> </mrow> </msub> </mfrac> <munder> <mo>&Sigma;</mo> <mi>i</mi> </munder> <msubsup> <mi>q</mi> <mi>i</mi> <mo>*</mo> </msubsup> <msub> <mi>L</mi> <mrow> <mi>r</mi> <mi>e</mi> <mi>g</mi> </mrow> </msub> <mrow> <mo>(</mo> <msub> <mi>t</mi> <mi>i</mi> </msub> <mo>,</mo> <msubsup> <mi>t</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>)</mo> </mrow> </mrow>
wherein q isiThe probability of being the target is predicted for the anchor,is 0 or 1, if the positive sample is 1, if the negative sample is 0; t is ti＝{tx,ty,tw,thDenotes the 4 parameterized coordinates of the predicted bounding box, LclsLogarithmic loss in two categories (target and non-target), LregIs the regression loss.
9. The method of claim 5, wherein the loss function during iterative training of Fast R-CNN networks in S2 and the second parameter fine tuning in S4 is:
L(p,u,tu,v)＝Lcls(p,u)+λ[u≥1]Lloc(tu,v)
wherein L isclsAs a function of classification level loss, Lcls＝-logpu；
LlocAs a function of the loss of positioning of the bezel,wherein,
v＝(vx,vy,vw,vh) Representing the coordinates of the predicted insulating subframe,representing the coordinates of a real insulating subframe.
10. An insulator detection device based on a shared convolutional neural network, comprising:
the image acquisition module is used for shooting the image of the power transmission line of the transformer substation by using the inspection robot; and
and the insulator detection module is used for acquiring the optimal position of the insulator in the power transmission line image by utilizing the shared convolutional neural network trained by the RPN network and the Fast R-CNN network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710347616.XA CN107274451A (en) | 2017-05-17 | 2017-05-17 | Isolator detecting method and device based on shared convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710347616.XA CN107274451A (en) | 2017-05-17 | 2017-05-17 | Isolator detecting method and device based on shared convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107274451A true CN107274451A (en) | 2017-10-20 |
Family
ID=60064110
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710347616.XA Pending CN107274451A (en) | 2017-05-17 | 2017-05-17 | Isolator detecting method and device based on shared convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107274451A (en) |
Cited By (45)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107808141A (en) * | 2017-11-08 | 2018-03-16 | 国家电网公司 | A kind of electric transmission line isolator explosion recognition methods based on deep learning |
| CN107862383A (en) * | 2017-11-09 | 2018-03-30 | 睿魔智能科技（东莞）有限公司 | A kind of multitask deep learning method and system perceived for human visual |
| CN108171112A (en) * | 2017-12-01 | 2018-06-15 | 西安电子科技大学 | Vehicle identification and tracking based on convolutional neural networks |
| CN108197637A (en) * | 2017-12-11 | 2018-06-22 | 国网上海市电力公司 | A kind of high pressure cabinet switch detection method based on deep learning |
| CN108205687A (en) * | 2018-02-01 | 2018-06-26 | 通号通信信息集团有限公司 | Based on focus mechanism positioning loss calculation method and system in object detection system |
| CN108280855A (en) * | 2018-01-13 | 2018-07-13 | 福州大学 | A kind of insulator breakdown detection method based on Fast R-CNN |
| CN108345911A (en) * | 2018-04-16 | 2018-07-31 | 东北大学 | Surface Defects in Steel Plate detection method based on convolutional neural networks multi-stage characteristics |
| CN108416299A (en) * | 2018-03-09 | 2018-08-17 | 重庆览辉信息技术有限公司 | A kind of power circuit intelligent inspection system |
| CN108492297A (en) * | 2017-12-25 | 2018-09-04 | 重庆理工大学 | The MRI brain tumors positioning for cascading convolutional network based on depth and dividing method in tumor |
| CN108564125A (en) * | 2018-04-16 | 2018-09-21 | 华北电力大学（保定） | A kind of insulator image classification method and system |
| CN108876144A (en) * | 2018-06-13 | 2018-11-23 | 国网山东省电力公司经济技术研究院 | A kind of pre- site selecting method of substation based on deep learning algorithm |
| CN108875593A (en) * | 2018-05-28 | 2018-11-23 | 上海交通大学 | Visible images weather recognition methods based on convolutional neural networks |
| CN108921840A (en) * | 2018-07-02 | 2018-11-30 | 北京百度网讯科技有限公司 | Display screen peripheral circuit detection method, device, electronic equipment and storage medium |
| CN109086765A (en) * | 2018-08-01 | 2018-12-25 | 北京嘀嘀无限科技发展有限公司 | Licence plate recognition method, device, medium, server and automobile data recorder |
| CN109242844A (en) * | 2018-09-04 | 2019-01-18 | 青岛大学附属医院 | Pancreatic tumour automatic recognition system based on deep learning, computer equipment, storage medium |
| CN109255776A (en) * | 2018-07-23 | 2019-01-22 | 中国电力科学研究院有限公司 | A kind of transmission line of electricity split pin defect automatic identifying method |
| CN109410190A (en) * | 2018-10-15 | 2019-03-01 | 广东电网有限责任公司 | Shaft tower based on High Resolution Remote Sensing Satellites image falls disconnected detection model training method |
| CN109446925A (en) * | 2018-10-08 | 2019-03-08 | 中山大学 | A kind of electric device maintenance algorithm based on convolutional neural networks |
| CN109472220A (en) * | 2018-10-23 | 2019-03-15 | 广东电网有限责任公司 | A kind of substation's worker safety helmet detection method and its system based on Faster R-CNN |
| CN109544501A (en) * | 2018-03-22 | 2019-03-29 | 广东电网有限责任公司清远供电局 | A kind of transmission facility defect inspection method based on unmanned plane multi-source image characteristic matching |
| CN109784205A (en) * | 2018-12-25 | 2019-05-21 | 国网河北省电力有限公司电力科学研究院 | A kind of weeds intelligent identification Method based on multispectral inspection image |
| CN109782771A (en) * | 2019-02-26 | 2019-05-21 | 西安交通大学 | A kind of orchard mobile robot and edge of a field forward method |
| CN109859207A (en) * | 2019-03-06 | 2019-06-07 | 华南理工大学 | A kind of defect inspection method of high density flexible substrate |
| CN109886947A (en) * | 2019-02-19 | 2019-06-14 | 浙江理工大学 | The high-tension bus-bar defect inspection method of convolutional neural networks based on region |
| CN109961423A (en) * | 2019-02-15 | 2019-07-02 | 平安科技（深圳）有限公司 | A kind of pulmonary nodule detection method based on disaggregated model, server and storage medium |
| CN110120074A (en) * | 2019-05-10 | 2019-08-13 | 清研同创机器人（天津）有限公司 | A kind of hot line robot cable localization method under complex environment |
| CN110136131A (en) * | 2019-05-24 | 2019-08-16 | 贵州电网有限责任公司 | A kind of zero resistance insulator detection method based on deep learning |
| CN110175985A (en) * | 2019-04-22 | 2019-08-27 | 国网江苏省电力有限公司电力科学研究院 | Carbon fiber composite core wire damage detecting method, device and computer storage medium |
| CN110189292A (en) * | 2019-04-15 | 2019-08-30 | 浙江工业大学 | A kind of cancer cell detection method based on Faster R-CNN and density estimation |
| CN110287786A (en) * | 2019-05-20 | 2019-09-27 | 特斯联（北京）科技有限公司 | Based on artificial intelligence anti-tampering vehicle information recognition method and device |
| CN110443280A (en) * | 2019-07-05 | 2019-11-12 | 北京达佳互联信息技术有限公司 | Training method, device and the storage medium of image detection model |
| CN110610122A (en) * | 2019-07-09 | 2019-12-24 | 国网江苏省电力有限公司扬州供电分公司 | Visible light imaging big data artificial intelligence detection diagnosis method for power transmission equipment |
| CN110633717A (en) * | 2018-06-21 | 2019-12-31 | 北京京东尚科信息技术有限公司 | Training method and device for target detection model |
| CN110895692A (en) * | 2018-09-13 | 2020-03-20 | 浙江宇视科技有限公司 | Vehicle brand identification method and device and readable storage medium |
| CN110991230A (en) * | 2019-10-25 | 2020-04-10 | 湖北富瑞尔科技有限公司 | Method and system for detecting ships by remote sensing images in any direction based on rotating candidate frame |
| CN111126424A (en) * | 2018-11-01 | 2020-05-08 | 南京大学 | Ultrasonic image classification method based on convolutional neural network |
| CN111507975A (en) * | 2020-04-23 | 2020-08-07 | 西南交通大学 | Method for detecting abnormity of outdoor insulator of traction substation |
| CN111539919A (en) * | 2020-04-16 | 2020-08-14 | 北京天下宏图空间信息技术有限公司 | Method and device for judging position and routing inspection of tower part |
| CN111860439A (en) * | 2020-07-31 | 2020-10-30 | 广东电网有限责任公司 | Unmanned aerial vehicle inspection image defect detection method, system and equipment |
| CN111914720A (en) * | 2020-07-27 | 2020-11-10 | 长江大学 | Method and device for identifying insulator burst of power transmission line |
| CN112115986A (en) * | 2020-08-31 | 2020-12-22 | 南京航空航天大学 | Power transmission line scene classification method based on lightweight neural network |
| CN112329550A (en) * | 2020-10-16 | 2021-02-05 | 中国科学院空间应用工程与技术中心 | Weak supervision learning-based disaster-stricken building rapid positioning evaluation method and device |
| CN113469950A (en) * | 2021-06-08 | 2021-10-01 | 海南电网有限责任公司电力科学研究院 | Method for diagnosing abnormal heating defect of composite insulator based on deep learning |
| CN114460086A (en) * | 2021-12-15 | 2022-05-10 | 国网河北省电力有限公司电力科学研究院 | Detection method for inclined strain insulator string |
| CN115620211A (en) * | 2022-12-06 | 2023-01-17 | 广东祥利科技有限公司 | Performance data processing method and system of flame-retardant low-smoke halogen-free sheath |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106156744A (en) * | 2016-07-11 | 2016-11-23 | 西安电子科技大学 | SAR target detection method based on CFAR detection with degree of depth study |
| CN106250812A (en) * | 2016-07-15 | 2016-12-21 | 汤平 | A kind of model recognizing method based on quick R CNN deep neural network |
| 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 |
-
2017
- 2017-05-17 CN CN201710347616.XA patent/CN107274451A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106156744A (en) * | 2016-07-11 | 2016-11-23 | 西安电子科技大学 | SAR target detection method based on CFAR detection with degree of depth study |
| CN106250812A (en) * | 2016-07-15 | 2016-12-21 | 汤平 | A kind of model recognizing method based on quick R CNN deep neural network |
| 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 |
Non-Patent Citations (3)
| Title |
|---|
| ROSS GIRSHICK: "Fast R-CNN", 《COMPUTER VISION》 * |
| SHAOQING REN 等: "Faster R-CNN:Towards Real-Time Object Detection with Region ProposalNetworks", 《ARXIV》 * |
| 王琴芳，应娜: "基于区域建议网络的行人检测", 《通信技术》 * |
Cited By (59)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107808141A (en) * | 2017-11-08 | 2018-03-16 | 国家电网公司 | A kind of electric transmission line isolator explosion recognition methods based on deep learning |
| CN107862383A (en) * | 2017-11-09 | 2018-03-30 | 睿魔智能科技（东莞）有限公司 | A kind of multitask deep learning method and system perceived for human visual |
| CN108171112A (en) * | 2017-12-01 | 2018-06-15 | 西安电子科技大学 | Vehicle identification and tracking based on convolutional neural networks |
| CN108171112B (en) * | 2017-12-01 | 2021-06-01 | 西安电子科技大学 | Vehicle identification and tracking method based on convolutional neural network |
| CN108197637A (en) * | 2017-12-11 | 2018-06-22 | 国网上海市电力公司 | A kind of high pressure cabinet switch detection method based on deep learning |
| CN108492297A (en) * | 2017-12-25 | 2018-09-04 | 重庆理工大学 | The MRI brain tumors positioning for cascading convolutional network based on depth and dividing method in tumor |
| CN108280855A (en) * | 2018-01-13 | 2018-07-13 | 福州大学 | A kind of insulator breakdown detection method based on Fast R-CNN |
| CN108205687A (en) * | 2018-02-01 | 2018-06-26 | 通号通信信息集团有限公司 | Based on focus mechanism positioning loss calculation method and system in object detection system |
| CN108205687B (en) * | 2018-02-01 | 2022-04-01 | 通号通信信息集团有限公司 | Attention mechanism-based positioning loss calculation method and system in target detection system |
| CN108416299A (en) * | 2018-03-09 | 2018-08-17 | 重庆览辉信息技术有限公司 | A kind of power circuit intelligent inspection system |
| CN109544501A (en) * | 2018-03-22 | 2019-03-29 | 广东电网有限责任公司清远供电局 | A kind of transmission facility defect inspection method based on unmanned plane multi-source image characteristic matching |
| CN108564125A (en) * | 2018-04-16 | 2018-09-21 | 华北电力大学（保定） | A kind of insulator image classification method and system |
| CN108345911A (en) * | 2018-04-16 | 2018-07-31 | 东北大学 | Surface Defects in Steel Plate detection method based on convolutional neural networks multi-stage characteristics |
| CN108345911B (en) * | 2018-04-16 | 2021-06-29 | 东北大学 | Steel plate surface defect detection method based on convolutional neural network multi-stage characteristics |
| CN108875593A (en) * | 2018-05-28 | 2018-11-23 | 上海交通大学 | Visible images weather recognition methods based on convolutional neural networks |
| CN108876144A (en) * | 2018-06-13 | 2018-11-23 | 国网山东省电力公司经济技术研究院 | A kind of pre- site selecting method of substation based on deep learning algorithm |
| CN108876144B (en) * | 2018-06-13 | 2021-08-17 | 国网山东省电力公司经济技术研究院 | Transformer substation pre-addressing method based on deep learning algorithm |
| CN110633717A (en) * | 2018-06-21 | 2019-12-31 | 北京京东尚科信息技术有限公司 | Training method and device for target detection model |
| CN108921840A (en) * | 2018-07-02 | 2018-11-30 | 北京百度网讯科技有限公司 | Display screen peripheral circuit detection method, device, electronic equipment and storage medium |
| JP2020530125A (en) * | 2018-07-02 | 2020-10-15 | 北京百度網訊科技有限公司 | Display screen peripheral circuit detection method, display screen peripheral circuit detection device, electronic devices and storage media |
| CN109255776A (en) * | 2018-07-23 | 2019-01-22 | 中国电力科学研究院有限公司 | A kind of transmission line of electricity split pin defect automatic identifying method |
| CN109255776B (en) * | 2018-07-23 | 2023-12-05 | 中国电力科学研究院有限公司 | Automatic identification method for cotter pin defect of power transmission line |
| CN109086765A (en) * | 2018-08-01 | 2018-12-25 | 北京嘀嘀无限科技发展有限公司 | Licence plate recognition method, device, medium, server and automobile data recorder |
| CN109242844A (en) * | 2018-09-04 | 2019-01-18 | 青岛大学附属医院 | Pancreatic tumour automatic recognition system based on deep learning, computer equipment, storage medium |
| CN110895692A (en) * | 2018-09-13 | 2020-03-20 | 浙江宇视科技有限公司 | Vehicle brand identification method and device and readable storage medium |
| CN110895692B (en) * | 2018-09-13 | 2023-04-07 | 浙江宇视科技有限公司 | Vehicle brand identification method and device and readable storage medium |
| CN109446925A (en) * | 2018-10-08 | 2019-03-08 | 中山大学 | A kind of electric device maintenance algorithm based on convolutional neural networks |
| CN109410190A (en) * | 2018-10-15 | 2019-03-01 | 广东电网有限责任公司 | Shaft tower based on High Resolution Remote Sensing Satellites image falls disconnected detection model training method |
| CN109472220A (en) * | 2018-10-23 | 2019-03-15 | 广东电网有限责任公司 | A kind of substation's worker safety helmet detection method and its system based on Faster R-CNN |
| CN111126424B (en) * | 2018-11-01 | 2023-06-23 | 南京大学 | Ultrasonic image classification method based on convolutional neural network |
| CN111126424A (en) * | 2018-11-01 | 2020-05-08 | 南京大学 | Ultrasonic image classification method based on convolutional neural network |
| CN109784205A (en) * | 2018-12-25 | 2019-05-21 | 国网河北省电力有限公司电力科学研究院 | A kind of weeds intelligent identification Method based on multispectral inspection image |
| CN109961423B (en) * | 2019-02-15 | 2024-05-31 | 平安科技（深圳）有限公司 | Lung nodule detection method based on classification model, server and storage medium |
| CN109961423A (en) * | 2019-02-15 | 2019-07-02 | 平安科技（深圳）有限公司 | A kind of pulmonary nodule detection method based on disaggregated model, server and storage medium |
| CN109886947A (en) * | 2019-02-19 | 2019-06-14 | 浙江理工大学 | The high-tension bus-bar defect inspection method of convolutional neural networks based on region |
| CN109782771B (en) * | 2019-02-26 | 2021-01-19 | 西安交通大学 | Orchard mobile robot and ground steering method |
| CN109782771A (en) * | 2019-02-26 | 2019-05-21 | 西安交通大学 | A kind of orchard mobile robot and edge of a field forward method |
| CN109859207A (en) * | 2019-03-06 | 2019-06-07 | 华南理工大学 | A kind of defect inspection method of high density flexible substrate |
| CN110189292A (en) * | 2019-04-15 | 2019-08-30 | 浙江工业大学 | A kind of cancer cell detection method based on Faster R-CNN and density estimation |
| CN110175985A (en) * | 2019-04-22 | 2019-08-27 | 国网江苏省电力有限公司电力科学研究院 | Carbon fiber composite core wire damage detecting method, device and computer storage medium |
| CN110120074A (en) * | 2019-05-10 | 2019-08-13 | 清研同创机器人（天津）有限公司 | A kind of hot line robot cable localization method under complex environment |
| CN110287786A (en) * | 2019-05-20 | 2019-09-27 | 特斯联（北京）科技有限公司 | Based on artificial intelligence anti-tampering vehicle information recognition method and device |
| CN110136131A (en) * | 2019-05-24 | 2019-08-16 | 贵州电网有限责任公司 | A kind of zero resistance insulator detection method based on deep learning |
| CN110443280B (en) * | 2019-07-05 | 2022-06-03 | 北京达佳互联信息技术有限公司 | Training method and device of image detection model and storage medium |
| CN110443280A (en) * | 2019-07-05 | 2019-11-12 | 北京达佳互联信息技术有限公司 | Training method, device and the storage medium of image detection model |
| CN110610122A (en) * | 2019-07-09 | 2019-12-24 | 国网江苏省电力有限公司扬州供电分公司 | Visible light imaging big data artificial intelligence detection diagnosis method for power transmission equipment |
| CN110991230A (en) * | 2019-10-25 | 2020-04-10 | 湖北富瑞尔科技有限公司 | Method and system for detecting ships by remote sensing images in any direction based on rotating candidate frame |
| CN110991230B (en) * | 2019-10-25 | 2023-08-11 | 湖北富瑞尔科技有限公司 | Method and system for detecting ship by using remote sensing images in any direction based on rotation candidate frame |
| CN111539919A (en) * | 2020-04-16 | 2020-08-14 | 北京天下宏图空间信息技术有限公司 | Method and device for judging position and routing inspection of tower part |
| CN111507975A (en) * | 2020-04-23 | 2020-08-07 | 西南交通大学 | Method for detecting abnormity of outdoor insulator of traction substation |
| CN111914720A (en) * | 2020-07-27 | 2020-11-10 | 长江大学 | Method and device for identifying insulator burst of power transmission line |
| CN111860439A (en) * | 2020-07-31 | 2020-10-30 | 广东电网有限责任公司 | Unmanned aerial vehicle inspection image defect detection method, system and equipment |
| CN112115986A (en) * | 2020-08-31 | 2020-12-22 | 南京航空航天大学 | Power transmission line scene classification method based on lightweight neural network |
| CN112115986B (en) * | 2020-08-31 | 2024-04-16 | 南京航空航天大学 | Lightweight neural network-based power transmission line scene classification method |
| CN112329550A (en) * | 2020-10-16 | 2021-02-05 | 中国科学院空间应用工程与技术中心 | Weak supervision learning-based disaster-stricken building rapid positioning evaluation method and device |
| CN113469950A (en) * | 2021-06-08 | 2021-10-01 | 海南电网有限责任公司电力科学研究院 | Method for diagnosing abnormal heating defect of composite insulator based on deep learning |
| CN114460086A (en) * | 2021-12-15 | 2022-05-10 | 国网河北省电力有限公司电力科学研究院 | Detection method for inclined strain insulator string |
| CN114460086B (en) * | 2021-12-15 | 2023-08-15 | 国网河北省电力有限公司电力科学研究院 | Detection method of inclined tension insulator string |
| CN115620211A (en) * | 2022-12-06 | 2023-01-17 | 广东祥利科技有限公司 | Performance data processing method and system of flame-retardant low-smoke halogen-free sheath |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107274451A (en) | Isolator detecting method and device based on shared convolutional neural networks | |
| CN111553303B (en) | Remote sensing orthographic image dense building extraction method based on convolutional neural network | |
| US10269138B2 (en) | UAV inspection method for power line based on human visual system | |
| CN109544501B (en) | Transmission equipment defect detection method based on unmanned aerial vehicle multi-source image feature matching | |
| CN108648169B (en) | Method and device for automatically identifying defects of high-voltage power transmission tower insulator | |
| CN107742093A (en) | A kind of infrared image power equipment component real-time detection method, server and system | |
| CN110033453A (en) | Based on the power transmission and transformation line insulator Aerial Images fault detection method for improving YOLOv3 | |
| CN112070134A (en) | Power equipment image classification method and device, power equipment and storage medium | |
| CN110780164B (en) | Insulator infrared fault positioning diagnosis method and device based on YOLO | |
| CN110929723B (en) | Identification method of transformer substation pointer instrument based on convolutional neural network | |
| CN111008641B (en) | Power transmission line tower external force damage detection method based on convolutional neural network | |
| CN111242144A (en) | Method and device for detecting abnormality of power grid equipment | |
| CN116503318A (en) | Aerial insulator multi-defect detection method, system and equipment integrating CAT-BiFPN and attention mechanism | |
| CN108876788B (en) | Insulator significance detection method based on multi-scale reconstruction error fusion | |
| CN111815576B (en) | Method, device, equipment and storage medium for detecting corrosion condition of metal part | |
| CN116385958A (en) | Edge intelligent detection method for power grid inspection and monitoring | |
| CN111539302B (en) | Glass insulator self-explosion identification method based on multi-scale deep disturbance neural network | |
| CN112419268A (en) | Method, device, equipment and medium for detecting image defects of power transmission line | |
| CN115937079A (en) | YOLO v 3-based rapid detection method for defects of power transmission line | |
| CN115359295A (en) | Decoupling knowledge distillation hardware target detection method and system | |
| CN114897858A (en) | Rapid insulator defect detection method and system based on deep learning | |
| CN110136097A (en) | Insulator fault identification method and device based on characteristic pyramid | |
| CN117173100B (en) | Polymer lithium ion battery production control system and method thereof | |
| CN116485802B (en) | Insulator flashover defect detection method, device, equipment and storage medium | |
| CN117496223A (en) | Light insulator defect detection method and device based on deep learning |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171020 |
|
| RJ01 | Rejection of invention patent application after publication |