CN106504233A - Image electric power widget recognition methodss and system are patrolled and examined based on the unmanned plane of Faster R CNN - Google Patents

Image electric power widget recognition methodss and system are patrolled and examined based on the unmanned plane of Faster R CNN Download PDF

Info

Publication number
CN106504233A
CN106504233A CN201610906708.2A CN201610906708A CN106504233A CN 106504233 A CN106504233 A CN 106504233A CN 201610906708 A CN201610906708 A CN 201610906708A CN 106504233 A CN106504233 A CN 106504233A
Authority
CN
China
Prior art keywords
network
electric power
cnn
training
image
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201610906708.2A
Other languages
Chinese (zh)
Other versions
CN106504233B (en
Inventor
蒋斌
王万国
刘越
刘俍
苏建军
慕世友
任志刚
杨波
李超英
傅孟潮
孙晓斌
李宗谕
李建祥
赵金龙
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
State Grid Intelligent Technology Co Ltd
Original Assignee
State Grid Corp of China SGCC
Electric Power Research Institute of State Grid Shandong Electric Power Co Ltd
Shandong Luneng Intelligence Technology Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by State Grid Corp of China SGCC, Electric Power Research Institute of State Grid Shandong Electric Power Co Ltd, Shandong Luneng Intelligence Technology Co Ltd filed Critical State Grid Corp of China SGCC
Priority to CN201610906708.2A priority Critical patent/CN106504233B/en
Publication of CN106504233A publication Critical patent/CN106504233A/en
Application granted granted Critical
Publication of CN106504233B publication Critical patent/CN106504233B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133—Distances to prototypes
    • G06F18/24137—Distances to cluster centroïds
    • G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/13—Satellite images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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
    • G06T2207/10—Image acquisition modality
    • G06T2207/10016—Video; Image sequence
    • 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]
    • 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

Abstract

The invention discloses patrolling and examining image electric power widget recognition methodss and system based on the unmanned plane of Faster R CNN；It includes that step is as follows：Pre-training is carried out to ZFnet models, extracts the characteristic pattern that unmanned plane patrols and examines image；Network model's training is proposed to the RPN regions that initialization is obtained, obtain extracted region network, candidate region frame is generated using extracted region network on the characteristic pattern of image, the feature in the frame of candidate region is extracted, extract position feature and the further feature of target；Using the position feature of target, further feature and characteristic pattern, the Faster R CNN detection networks obtained by initialization are trained, and obtain electric power widget detection model；Step (4)：Actual electric power widget recognition detection is carried out using electric power widget detection model.Beneficial effects of the present invention：The electric power widget identification positioning that plurality of classes is carried out using Faster R CNN can reach the recognition speed per a nearly 80ms and 92.7% accuracy rate.

Description

Based on the unmanned plane of Faster R-CNN patrol and examine image electric power widget recognition methodss and System
Technical field
The present invention relates to a kind of unmanned plane based on Faster R-CNN is patrolled and examined image electric power widget recognition methodss and is System.
Background technology
In recent years with unmanned plane (Unmanned Aerial Vehicle, UAV)) the gradually popularization applied, power-line patrolling Unmanned plane is subject to the extensive concern of each bulk power grid company and is demonstrated and popularization and application.Unmanned plane line walking has field work low The characteristics of risk, low cost and flexible operation；Meanwhile, also bring the interior industry workload of walking operation to increase so that magnanimity Data need just obtain final patrolling and examining report through substantial amounts of artificial interpretation.
At present, power components identification remains in traditional identification aspect based on shallow-layer feature, by Fine design Shallow-layer feature, such as SIFT (Scale-invariant feature transform), rim detection symbol, HOG (Histogram Of Oriented Gridients), or image segmentation is carried out based on part circumference skeleton, adaptive threshold etc., so as to reach Purpose to identification.But these methods are often based on particular category to realize in design principle, to insulator, wire etc. Single part can fully excavate its mode of appearance feature, but its accuracy rate is low, and not there is extensibility；And method structure pine Dissipate, lack carries out comprehensively utilizing and then reach the purpose of global optimum's identification to low-level feature.
By comparison, the contour detecting of Malik team and hierarchical image segmentation method and Multiscale combination polymerization (Multiscale Combinatorial Grouping, MCG) method, and J.Uijlings and K.van de Sande etc. What people proposed is given multiple low level features based on the target identification method of selective search (Selective Search) Carry out global optimization and build the normal form of hierarchy Model, improve accuracy rate, but these methods still do not possess with sample number Amount increases the ability for lifting recognition accuracy.
Chinese invention patent (application number：201510907472.X, patent name：A kind of transmission line of electricity small parts identification Method), although the relational implementation that this method can be according to wire with small parts to conductor spacer, the identification of stockbridge damper and is determined Position, but for the recognition efficiency under complex background is with recognition effect and bad, it is not met by the needs at scene.
Content of the invention
The purpose of the present invention is exactly to solve the above problems, there is provided a kind of unmanned plane based on Faster R-CNN is patrolled and examined DPM, SPPNet and Faster R-CNN recognition methodss are contrasted by image electric power widget recognition methodss and system respectively Analysis, the data set for patrolling and examining data structure using the electric power widget of actual acquisition carry out test checking to three kinds of methods.Experiment As a result show, the recognition methodss of image electric power widget are patrolled and examined to realize the identification of electric power widget be feasible based on deep learning , and the electric power widget identification positioning for carrying out plurality of classes using Faster R-CNN can reach the knowledge per a nearly 80ms Other speed and 92.7% accuracy rate.RCNN (Region based Convolutional Neural Network), based on area The convolutional neural networks that domain is proposed.
To achieve these goals, the present invention is adopted the following technical scheme that：
Image electric power widget recognition methodss are patrolled and examined based on the unmanned plane of FasterR-CNN, as follows including step：
Step (1)：Pre-training is carried out to ZFnet models, extracts the characteristic pattern that unmanned plane patrols and examines image；RPN regions are carried View network model and Faster R-CNN detection networks are initialized；
Step (2)：Network model's training is proposed to the RPN regions that initialization is obtained, obtains extracted region network, using area Domain is extracted network and generates candidate region frame on the characteristic pattern of image, the feature in the frame of candidate region is extracted, is extracted The position feature of target and further feature；
Step (3)：Using the position feature of target, further feature and characteristic pattern, the Faster R- obtained by initialization CNN detection networks are trained, and obtain electric power widget detection model；
Step (4)：Actual electric power widget recognition detection is carried out using electric power widget detection model.
Pre-training is carried out using ImageNet categorical data set pair ZFnet models and obtains the ZFnet models after pre-training.
The step of pre-training is carried out using ImageNet categorical data set pair ZFnet models is as follows：
ZFnet models are 8 layer network structures, comprising 5 convolutional layers and 3 full articulamentums, from top to bottom number ground floor, Add the operation of Max Pooling pondizations behind the second layer and layer 5 convolutional layer.
The step of pre-training is carried out using ImageNet categorical data set pair ZFnet models is as follows：
ZFNet models are counted from top to bottom,
First convolutional layer carries out convolution to the ImageNet categorized data sets being input into, and after convolution, carries out first Max Pooling pondizations are operated；
Second convolutional layer carries out convolution to the result that first Max Pooling pondization is operated, and after convolution, carries out second Individual Max Pooling pondizations operation；
3rd convolutional layer carries out convolution to the result that second Max Pooling pondization is operated；After convolution,
4th convolutional layer carries out convolution to the convolution results of the 3rd convolutional layer；After convolution,
5th convolutional layer carries out convolution to the convolution results of the 4th convolutional layer, after convolution, carries out the 3rd Max Pooling pondizations are operated；
256 output units are obtained after 3rd Max Pooling pondizations operation, characteristic pattern (Feature is formed Map).
Propose that to RPN regions network model and Faster R-CNN detection networks are initialized：
Set the initialization informations such as sliding step, the sliding window size of RPN network extraction modules.
According to the output unit number that training dataset classification number sets detection network, and initialize the weight of detection network Parameter.
Step (2)：With electric power widget picture construction electric power widget image set, trained with electric power widget image set When network is proposed in RPN regions after initialization, using back-propagation algorithm to initialization after region propose that network carries out tuning.
Network is proposed with the image of arbitrary size as input in RPN regions, and RPN regions propose that network exports several comprising time Select the candidate region frame of target.
Propose that network convolutional layer, convolution operation are adopted in the 5th convolutional layer of a ZFnet models RPN region added behind Operated with sliding type, for a window is opened on the target location of each hand labeled on ZFNet aspect of model figures, sharp Propose that network carries out convolution algorithm with RPN regions, obtain corresponding 256 dimensional feature vector in each position, 256 dimensional feature Vector is used for the further feature reflected in the window of each position；
Propose 9 kinds of convolution kernel letters that network convolutional layer is combined into using 3 kinds of different sizes and 3 kinds of different proportions in RPN regions Several positions to the window comprising candidate target carry out Objective extraction, obtain the position feature with regard to target, by the position of target Feature is input to step (3) as the training data of input.
Measurable by 256 dimensional feature vector：
1) window of the corresponding position of 256 dimensional feature vectors belongs to the probit of foreground/background；
2) include the window of candidate target near the corresponding position of 256 dimensional feature vectors relative to 256 dimensional feature vectors pair The deviation of the window of the position that answers.
3 kinds of different proportions are referred to：1:1、1:2 and 2:1.
Step (3) step is as follows：
Step (3-1)：The further feature that 5th convolutional layer is obtained, full connection form high dimensional feature vector and realize image Global feature description, as the input of the full articulamentum FC6 of layer 6.
The input that the position feature that RPN regions proposal network convolutional layer is obtained is also served as the full articulamentum FC6 of layer 6；
It is that the full mode for connecting carries out data exchange between layer 6 full articulamentum FC6 and layer 7 prediction interval FC7；
Layer 7 prediction interval FC7 includes sort module and regression block；
Sort module is used for the type of judging characteristic, and regression block is used for the target location of precise positioning feature；
Step (3-2)：Calculating network whole loss function simultaneously carries out each layer parameter optimization of network according to loss function.
Loss function L (p, k*,t,t*)：
Training dataset is divided into K+1 classes, k*Represent correct tag along sort, p=(p0,...,pK) presentation class is for k Probability,Represent the calculated indicia framing information of regression block：The abscissa of indicia framing, the vertical seat of indicia framing Mark, the height of the wide and indicia framing of indicia framing.
Computing formula is：
Label according to demarcating in training set carries out the small parameter perturbations training for detecting network, by stochastic gradient descent side (Stochastic Gradient Descent, SGD), alternative optimization network parameter.
Regression block parameter constant, the fixed cluster in optimized regression module parameter is fixed in Optimum Classification module parameter Module parameter is constant.
The high dimensional feature vector refers to 4096 dimensional feature vectors.
Measurable by layer 7 prediction interval FC7：
1) regional frame comprising candidate target belongs to the probability of any classification；
2) the precise information set of destination object external surrounding frame, the position including target in feature description is correspondingly originally inputted 2 translational movements of the transverse and longitudinal coordinate on image location information, and zoom in or out two on transverse and longitudinal coordinate axle put Contracting coefficient.By the precise information set for possessing this 4 parameters, Accurate Calibration of the target in original image is realized.
Propose that network finally have shared convolutional layer and define an electric power with Faster R-CNN detection networks in RPN regions Widget detection model.
Step (4) step is as follows：
Step (4-1)：According to the electric power widget detection model that training is obtained, initialization electric power widget detection model ginseng Number；
Step (4-2)：Input picture, obtains to each class included in image after electric power widget detection model Other probit and positional information；
Step (4-3)：According to positional information, target particular location is gazed in artwork subscript.
The electric power widget includes conductor spacer, grading ring and stockbridge damper.
RPN, Region Proposal Networks
Image electric power widget identifying system is patrolled and examined based on the unmanned plane of FasterR-CNN, including：
Pre-training module：Pre-training is carried out to ZFnet models, extracts the characteristic pattern that unmanned plane patrols and examines image；To RPN regions Propose that network model and Faster R-CNN detection networks are initialized；
Characteristic extracting module：Network model's training is proposed to the RPN regions that initialization is obtained, obtains extracted region network, Candidate region frame is generated using extracted region network on the characteristic pattern of image, the feature in the frame of candidate region is extracted, Extract position feature and the further feature of target；
Model training module：Using the position feature of target, further feature and characteristic pattern, the Faster obtained by initialization R-CNN detection networks are trained, and obtain electric power widget detection model；
Detection module：Actual electric power widget recognition detection is carried out using electric power widget detection model.
Beneficial effects of the present invention：
1 is all compared to the accuracy and efficiency that electric power widget is recognized using Faster R-CNN even depth learning algorithm Height, experiment show the deep learning method based on statistics can be made to realize to patrolling and examining video or figure using specific GPU computing units The real-time target detection of picture and identification, can patrol and examine the accurate bat of the intelligent processing method and patrol unmanned machine of image for later stage unmanned plane Take the photograph and lay a good foundation.
2 compared with Sppnet and Fast R-CNN, and Faster R-CNN methods had both breached the time of zoning proposal Bottleneck, can guarantee that preferable discrimination again.
The region that 3 the inventive method are adopted proposes that network and detection network have good generalization ability, can recognize that Part is blocked and the middle conductor spacer through iron, and the part to various different directions all correctly can be recognized.
Description of the drawings
Fig. 1 is the schematic network structure that the present invention is used；
Fig. 2 is the joint network training process of part identification；
Fig. 3 is network training process schematic diagram；
Fig. 4 is detection identification process；
Fig. 5 (a) is raw-data map；
Fig. 5 (b) is training sample data figure；
Fig. 6 (a) is conductor spacer, stockbridge damper Detection results figure；
Fig. 6 (b) is grading ring, conductor spacer Detection results schematic diagram.
Specific embodiment
The invention will be further described with embodiment below in conjunction with the accompanying drawings.
From 2012, along with the development of high-performance GPU parallel computation, deep learning Remarkable Progress On Electric Artificial, base is surmounted In shallow-layer feature and the traditional method of linear classifier, become the leader of field of target recognition, PASCAL (pattern Analysis, statistical modelling and computational learning) and ILSVRC (Imagenet Large Scale Vision Recognition Challenge) contest become evaluation universal identification algorithm Sample Storehouse base Standard, has witnessed the breakthrough of deep learning method and has stepped up.The present invention is by the research to deep learning in terms of identification, pin Problem and data characteristicses are recognized to power components, identification and the test sample storehouse of three class power components is constructed, be have studied DPM (Deformable Part Models), it is based on RCNN (Region based Convolutional Neural Network) Sppnet (Spatial pyramid pooling networks) and tri- kinds of algorithms of Faster R-CNN, and to the little portion of electric power Part is identified assignment test and checking, analyzes the quality of each algorithm, provides the algorithm suitable for power components identification, is Later stage unmanned plane patrols and examines the intelligent processing method of image and the accurate shooting of unmanned plane image is laid a good foundation.
1 classics DPM methods and RCNN
In terms of identification, key issue has two, one be target location determination；Another is that target class is other sentences Not.The two problems are to be mutually related.Different according to the originally determined mode in target location, recognition methodss can be classified as two Class：One class is by the way of sliding window, judges whether destination object by window；Another kind of proposed using region Whether mode, first concentrates and generates the regional frame that may include destination object, then judge each candidate frame comprising destination object one by one. The typical algorithm of sliding window mode recognition methodss is deformable part model DPM；Propose the typical algorithm that mode is recognized in region It is the convolutional neural networks RCNN based on region.
1.1 deformable part models DPM
Deformable part model DPM method is the classical Target Recognition Algorithms proposed by P.Felzenszwalb.In detection In the stage, DPM as a sliding window operation on characteristics of image pyramid, generally built by HOG features by characteristics of image pyramid Vertical.DPM is by optimizing the scoring function of a comprehensive part distortion cost function and images match score to each sliding window Give a score.The optimal result of whole PASCAL identifications matches in 2007 to 2011 is all by DPM methods and its mutation Obtain.However, the defect that DPM methods also have which intrinsic：1) sliding window mode is actually a kind of traversal search The mode of (exhaustive search), this cause DPM that not there is the feasibility of scale expansion, and with image resolution ratio Raising, the amount of calculation of DPM methods increases by geometric progression；2) test result indicate that, as training samples number increases, DPM The recognition accuracy of method can reach saturation, it is impossible to make full use of unmanned plane obtain image easily advantage improving the standard of identification True rate.
The 1.2 convolutional neural networks RCNN proposed based on region
Convolutional neural networks method RCNN based on region that Ross etc. was proposed in 2014, greatly improves PASCAL knowledges Recognition accuracy, by taking PASCAL VOC2012 contests as an example, is brought up to 53.3% from 41% (DPM++) by the accuracy rate of other contest (RCNN), it becomes the typical scenario being identified based on region proposal mode.In detection-phase, it is divided into following 4 steps：
1) a large amount of candidate regions are generated with some visible sensation methods (such as Selective Search) first.
2) character representation is carried out with convolutional neural networks CNN to each candidate region secondly, ultimately form high dimensional feature to Amount.
3) and then, these characteristic quantities are sent into linear classifiers and calculate category score, for judging included object.
4) last, a fine recurrence is carried out to the position and size of targeted peripheral frame.
Compared with the traversal search mode of the sliding window of DPM, the region of the first step proposes it is selective search, uses The amount of calculation of feature extraction below can be effectively reduced in front 2000 regions of highest scoring, can tackle scale problem well； Convolutional neural networks CNN carries out parallel computation using graphics calculations unit GPU in realization, and computational efficiency is substantially better than DPM side Method (is calculated using list CPU in realization)；External surrounding frame returns and the accuracy positioned by target is further lifted.RCNN is in training Also there are following 4 steps in stage.
1) concentrate the candidate region generated per pictures first with selective search, and each candidate region is carried with CNN Feature is taken, CNN uses ImageNet networks (the up to a million figures of 1000 classes for match of being classified by ILSVRC for training here As training is obtained).
2) secondly, tuning, tuning establishing criteria are carried out to ImageNet networks using candidate region and the feature for extracting Back-propagation algorithm carry out, start to adjust each layer weight backward from characteristic layer.
3) and then, with characteristic layer output high dimensional feature vector sum target class label as input, Training Support Vector Machines.
4) last, training carries out the recurrence device of fine recurrence to targeted peripheral frame position and size.
RCNN methods considerably beyond DPM methods, become the allusion quotation being identified based on deep learning in accuracy rate and efficiency Type scheme.2014 and 2015, the researcher of Ross and Microsoft Research, Asia proposed improved RCNN methods successively, wrapped Include be firstly introduced spatial pyramid pond layer so as to relax to be input into dimension of picture limit and improve the SPPnet of accuracy rate；Adopt Tuning can be carried out to whole network with adaptive scale pondization so as to improving the Fast R- of the accuracy rate of deep layer Network Recognition CNN；It is finally Faster R-CNN, he is searched by building the selectivity that network is proposed to replace time overhead big in exquisite region Suo Fangfa, proposes the big bottleneck problem of time overhead so as to break zoning, make the Real time identification of view-based access control model feature into For possibility.The present invention is mainly told about and power components is identified using Faster R-CNN methods.
The 2 power components identification positioning based on Faster R-CNN methods
Compared with Sppnet and Fast R-CNN, Faster R-CNN methods had both breached the time bottle of zoning proposal Neck, can guarantee that preferable discrimination again.Therefore, the present invention extracts electric power widget based on Faster R-CNN recognition methodss Identification feature and carry out target recognition checking.
The present invention is carried out to the Caffe CNN storehouses that the training of network and the detection of test sample are based on increasing income.Caffe is One clear and efficient deep learning framework, its readability, terseness and performance are all very outstanding, and have been directly integrated convolution Neutral net nervous layer.Due to the depth convolutional network characteristic of itself, when accelerating computing greatly shorten Algorithm for Training with GPU Between, Caffe also provides corresponding interface.
As shown in figure 1, ZFnet models are 8 layer network structures, comprising 5 convolutional layers and 3 full articulamentums, from top to bottom Number adds the operation of Max Pooling pondizations behind ground floor, the second layer and layer 5 convolutional layer.
The step of pre-training is carried out using ImageNet categorical data set pair ZFnet models is as follows：
ZFNet models are counted from top to bottom,
First convolutional layer carries out convolution to the ImageNet categorized data sets being input into, and after convolution, carries out first Max Pooling pondizations are operated；
Second convolutional layer carries out convolution to the result that first Max Pooling pondization is operated, and after convolution, carries out second Individual Max Pooling pondizations operation；
3rd convolutional layer carries out convolution to the result that second Max Pooling pondization is operated；After convolution,
4th convolutional layer carries out convolution to the convolution results of the 3rd convolutional layer；After convolution,
5th convolutional layer carries out convolution to the convolution results of the 4th convolutional layer, after convolution, carries out the 3rd Max Pooling pondizations are operated；
256 output units are obtained after 3rd Max Pooling pondizations operation, characteristic pattern (Feature is formed Map).
The further feature that 5th convolutional layer is obtained, full connection form the global feature that high dimensional feature vector realizes image Description, used as the input of the full articulamentum FC6 of layer 6.
The input that the position feature that RPN regions proposal network convolutional layer is obtained is also served as the full articulamentum FC6 of layer 6；
It is that the full mode for connecting carries out data exchange between layer 6 full articulamentum FC6 and layer 7 prediction interval FC7；
Layer 7 prediction interval FC7 includes sort module and regression block；
Sort module is used for the type of judging characteristic, and regression block is used for the target location of precise positioning feature；
The network of 2.1 pairs of power components identifications is trained
Faster R-CNN methods include two CNN networks：Propose network RPN (Regional Proposal in region Network) and Fast R-CNN detection network.The key step of training stage is as shown in Fig. 2 Fig. 3 is to RPN networks and detection Network carries out joint training figure.
(1) pre-training CNN model
RPN networks and detection network are required for the ImageNet networks of pre-training to be initialized, and tool can be selected to be of five storeys The ZFnet networks (Zeiler and Fergus) of network structure, it is also possible to select the VGG16 networks with 16 layer network structures (Simonyan and Zisserman modal).Because the present invention build data set scale less, therefore select ZFnet networks.
Using the training data (1,200,000 images, 1000 classes) in ILSVRC2012 image classification tasks to ZFnet models Carry out pre-training.Propose that network and detection network all add specific layer on the basis of ZFnet and obtain in region.ZFnet was once Higher classification accuracy has been reached on ILSVRC competition data collection.ZFnet includes 5 convolutional layers, behind some convolutional layers Add Max Pooling pond layers and 3 characteristic layers being fully connected.
The convolutional layer of last convolutional layer of ZFNet, i.e., the 5th includes 256 passages, is referred to as characteristic pattern (Feature Map).Characteristic pattern can intuitively be interpreted as that the deep layer convolution feature of original image, the further feature of similar object are sufficiently close to； And the further feature of different type objects is widely different, i.e., on characteristic pattern, object has good separability, and this exactly depth is refreshing Through remarkable feature learning that the ability of network is located and expression ability.
The present invention is using the ZFnet networks pair obtained by ILSVRC categorized data sets (1,200,000 pictures, 1000 classes) training Propose that network and detection network are initialized in region.It is demonstrated experimentally that using by big data quantity and more multi-class Sample Storehouse (phase Compared with small data quantity and the Sample Storehouse of less classification) sorter network trained initialization identification network, accuracy rate is higher.
(2) RPN network trainings
Training set of images is built with electric power image of component, but power components image set and pre-training image set either classification All there is very big difference in quantity or image style.When with electric power image of component collection training RPN networks, previous step is directly used To region, the ZFnet model initialization RPN of pre-training, propose that network carries out tuning using back-propagation algorithm.
RPN networks can export a series of regional frames that may include target afterwards with the image of arbitrary size as input. As shown in figure 3, in a CONV5 little convolutional layer added behind of ZFnet, this little convolutional layer is transported using sliding type Make, for each position (position on corresponding original image) on characteristic pattern, convolution algorithm is carried out by little convolutional layer, i.e., A wicket is opened in this position carries out convolution algorithm, obtains corresponding 256 dimensional vector in same position (due to there is 256 to lead to Road), the vector reflects the further feature in the position wicket (a certain window on corresponding original image).256 tieed up by this Characteristic vector can predict：1) the position wicket belongs to the probit of foreground/background, i.e. score；2) wrap near the position Window containing target represents with 4 parameters that relative to the deviation of the position wicket 2 translate, 2 scalings.
By experimental analysiss, using 3 kinds of different sizes and 3 kinds of different proportions (1:1,1:2,2:1) 9 kinds of benchmark being combined into Wicket is predicted to the position of the window comprising target, proposes can region more accurate.
(3) Fast R-CNN detections network training
Independent detection network, detection are trained based on Fast R-CNN methods according to the region proposed issue that previous step is generated Network is also using ZFnet pre-training model initializations.
The feature extraction that 5 layers of convolutional network are carried out to input picture, the 5th layer of characteristic pattern (CONV5) is one 256 × 256 Characteristic pattern, take out CONV5 on corresponding depth characteristic, the whole features in 256 passages are connected into a higher-dimension (4096 Dimension) characteristic vector, referred to as FC6 characteristic layers, the characteristic layer of another 4096 dimension added behind, formation FC7 are adopted between FC6 and FC7 With being fully connected.Measurable by FC7 characteristic layers：1) candidate region frame belongs to the probability of any classification, i.e. score；2) destination object The more suitably position of external surrounding frame, with it, relative to 2 of candidate region frame translations and 2 scalings, totally 4 parameters are represented.Pass through The Information Pull back-propagation algorithm of labelling is finely adjusted to the detection network in advance.When ZFnet is trained, used is not electricity Power parts data, is trained the model for obtaining to be misfitted with power components, carries out arameter optimization with electric power parts data.Reversely pass Broadcast the process that arameter optimization is exactly completed using stochastic gradient descent method, adjust the weight parameter of network.
The CNN of (4) two networks is shared and combines tuning
Two networks are individually trained and the parameter of unrealized convolutional network is shared.
The detection network that is trained using the 3rd step is initializing RPN networks, and fixed shared depth convolutional layer.I.e. excellent Using same input data when change RPN networks and sort module.As shown in figure 3, the special part to RPN networks is adjusted Excellent, in order to corresponding with detection network, this part is called the FC layers of RPN networks, and two such network just have shared depth convolutional layer；
Finally, fixed shared convolutional layer, carries out tuning to the FC layers of Fast R-CNN.Two such network just have shared Convolutional layer simultaneously defines a united network.
2.2 detection identification process
From training above, two networks can finally share same 5 layers of convolutional neural networks, and this allows for whole Individual detection process completes to detect identification process by need to only completing serial convolution algorithm, thoroughly solves original region and propose step The big bottleneck problem of time overhead.
The process of detection identification is as shown in figure 4, implementation step is：
1) serial convolution algorithm is carried out to whole image first, obtains characteristic pattern CONV5；
2) propose that network generates a large amount of candidate region frames on characteristic pattern by region；
3) non-maximum suppression is carried out, keep score higher front 300 frames to candidate region frame；
The feature for taking out candidate region inframe on characteristic pattern forms high dimensional feature vector, is obtained by detection network calculations classification Point, and predict more suitably targeted peripheral frame position.
3 experimental results are contrasted
Unmanned plane filmed image have resolution higher, comprising the less feature of target, the angle of filmed image has many Sample and certain randomness, are suitably for the Sample Storehouse that deep learning method is provided with enough tolerance amounts.In an experiment, we will know Other 3 class small electrical parts space rod, stockbridge damper and grading ring.
The process of 3.1 training samples
Data set covers spring, summer, autumn and winter four from multi-rotor unmanned aerial vehicle and helicopter routing inspection image from seasonal Individual season.Raw video pixel size is 5184 × 3456 (as shown in Fig. 5 (a)), intercepts square little based on target Block image, unified scaling to 500 × 500 (such as Fig. 5 (b)), as training sample, so process the size for causing target in sample Ratio is close to the ratio of sample in PASCAL identification contests.
3.2 training sets and test set build
This test, for each base part of conductor spacer, grading ring and stockbridge damper, using 1500 training samples, Totally 4500 sample composing training collection；Per 500 testing images of class, totally 1500 images constitute test set.To every in training set Complete small electrical part labelling its external surrounding frame (imperfect in the training set picture or quilt not being blocked for occurring in pictures The power components that blocks not labelling)；And to test set, all power components occurred in every pictures to be marked, including not complete Whole and being blocked.
During test, when the external surrounding frame overlapping area of the external surrounding frame that identifies and labelling reaches more than the 90% of labelling external surrounding frame It is treated as once successfully recognizing.In this test, the accuracy of identification is passed judgment on accuracy and recall rate, wherein accuracy is The correct external surrounding frame number of target category label is divided by all external surrounding frame numbers for marking；Recall rate be target category label just External surrounding frame number of the true external surrounding frame number divided by all standards.As the classification of this identification only has three types, because This accuracy and recall rate that respectively each class power components are recognized are counted.
3.3 experimental result
The present invention realizes convolutional neural networks model using the Caffe frameworks that Berkeley vision is developed with learning center.Make With 3.2 section build training sets and test set, by Faster R-CNN methods with carried out based on Selective Search methods The SPPnet methods and DPM methods that region is proposed is contrasted, and test result is as shown in table 1.
The contrast of 1 the inventive method of table and SPPnet recognition accuracy on test set
The accuracy rate that Faster R-CNN methods are recognized as can be seen from Table 1 is apparently higher than SPPnet and DPM, and DPM side Method accuracy rate is minimum.Mainly due to region, this proposes that network can be with producing ratio SPPnet more accurately candidate frame, and DPM methods Detected using sliding window, it is characterized by HOG features, rather than depth training characteristics.Additionally, the inventive method is in network 2nd step of training carries out tuning to the weight of whole characteristic layers and convolutional layer, and SPPnet only tuning characteristic layers, so as to limit Recognition accuracy.It should be noted that the region that the inventive method is adopted proposes network and detection network with extensive well Ability, can recognize that part is blocked and the middle conductor spacer through iron, and the part to various different directions all can be just Really recognize, and other two methods are then slightly more inferior.Fig. 6 (a)-Fig. 6 (b) is to three kinds of electric power using Faster R-CNN methods The result of widget identification.
The present invention all test be based on same server and tested, test set picture size be 5184 × 3456, DPM methods realize that based on CPU Faster R-CNN methods and SPPnet are using Nivdia Titan Black GPU (6G video memorys) carries out convolutional calculation, and identification process consumes 3G video memorys.In addition, the non-maximum of the inventive method suppresses also to adopt Realized with GPU.From table 2 it can be seen that the operation time of DPM is in minute level, it is impossible to carry out time efficiency with other two methods Contrast；For typical RCNN methods as SPPnet, region proposes to occupy the main calculating time；And present invention side Method, as sharing for convolution feature (proposes that network and the special layers of detection network are all added on shared characteristic pattern CONV5 in region Behind) so that region proposes that the time is almost negligible, and detection time be able to can be completed in nearly 80ms.
Result of the test shows, the reality that can be realized patrolling and examining image based on specific graphics acceleration card using deep learning method When detect.
2 the inventive method of table and SPPNet, DPM method calculate time overhead contrast
Although the above-mentioned accompanying drawing that combines is described to the specific embodiment of the present invention, not to present invention protection model The restriction that encloses, one of ordinary skill in the art should be understood that on the basis of technical scheme those skilled in the art are not The various modifications that makes by needing to pay creative work or deformation are still within protection scope of the present invention.

Claims (10)

1. image electric power widget recognition methodss are patrolled and examined based on the unmanned plane of FasterR-CNN, be it is characterized in that, including step such as Under：
Step (1)：Pre-training is carried out to ZFnet models, extracts the characteristic pattern that unmanned plane patrols and examines image；Net is proposed to RPN regions Network model and Faster R-CNN detection networks are initialized；
Step (2)：Network model's training is proposed to the RPN regions that initialization is obtained, is obtained extracted region network, is carried using region Taking network and candidate region frame being generated on the characteristic pattern of image, the feature in the frame of candidate region is extracted, target is extracted Position feature and further feature；
Step (3)：Using the position feature of target, further feature and characteristic pattern, the Faster R-CNN inspections obtained by initialization Survey grid network is trained, and obtains electric power widget detection model；
Step (4)：Actual electric power widget recognition detection is carried out using electric power widget detection model.
2. image electric power widget recognition methodss are patrolled and examined based on the unmanned plane of FasterR-CNN as claimed in claim 1, which is special Levying is, carries out pre-training using ImageNet categorical data set pair ZFnet models and obtains the ZFnet models after pre-training.
3. image electric power widget recognition methodss are patrolled and examined based on the unmanned plane of FasterR-CNN as claimed in claim 2, which is special Levying is, as follows the step of carry out pre-training using ImageNet categorical data set pair ZFnet models：
ZFnet models are 8 layer network structures, comprising 5 convolutional layers and 3 full articulamentums, from top to bottom number ground floor, second Add the operation of Max Pooling pondizations behind layer and layer 5 convolutional layer.
4. image electric power widget recognition methodss are patrolled and examined based on the unmanned plane of FasterR-CNN as claimed in claim 2, which is special Levying is, as follows the step of carry out pre-training using ImageNet categorical data set pair ZFnet models：
ZFNet models are counted from top to bottom,
First convolutional layer carries out convolution to the ImageNet categorized data sets being input into, and after convolution, carries out first Max Pooling pondizations are operated；
Second convolutional layer carries out convolution to the result that first Max Pooling pondization is operated, and after convolution, carries out second MaxPooling pondizations are operated；
3rd convolutional layer carries out convolution to the result that second Max Pooling pondization is operated；After convolution,
4th convolutional layer carries out convolution to the convolution results of the 3rd convolutional layer；After convolution,
5th convolutional layer carries out convolution to the convolution results of the 4th convolutional layer, after convolution, carries out the 3rd Max Pooling Pondization is operated；
256 output units are obtained after 3rd Max Pooling pondizations operation, characteristic pattern is formed.
5. image electric power widget recognition methodss are patrolled and examined based on the unmanned plane of FasterR-CNN as claimed in claim 1, which is special Levying is, step (2)：With electric power widget picture construction electric power widget image set, first with the training of electric power widget image set When network is proposed in RPN regions after beginningization, using back-propagation algorithm to initialization after region propose that network carries out tuning.
6. image electric power widget recognition methodss are patrolled and examined based on the unmanned plane of FasterR-CNN as claimed in claim 5, which is special Levying is, network is proposed with the image of arbitrary size as input in RPN regions, and RPN regions propose that network exports several comprising candidate The candidate region frame of target；
Propose network convolutional layer in the 5th convolutional layer of a ZFnet models RPN region added behind, convolution operation is using cunning Flowing mode is operated, and for a window is opened on the target location of each hand labeled on ZFNet aspect of model figures, is utilized RPN regions propose that network carries out convolution algorithm, obtain corresponding 256 dimensional feature vector in each position, 256 dimensional feature to Measure for reflecting the further feature in the window of each position；
Propose 9 kinds of convolution kernel function pairs that network convolutional layer is combined into using 3 kinds of different sizes and 3 kinds of different proportions in RPN regions The position of the window comprising candidate target carries out Objective extraction, obtains the position feature with regard to target, by the position feature of target As the training data of input, step (3) is input to.
7. image electric power widget recognition methodss are patrolled and examined based on the unmanned plane of FasterR-CNN as claimed in claim 6, which is special Levying is, measurable by 256 dimensional feature vector：
1) window of the corresponding position of 256 dimensional feature vectors belongs to the probit of foreground/background；
2) near the corresponding position of 256 dimensional feature vectors, the window comprising candidate target is corresponding relative to 256 dimensional feature vectors The deviation of the window of position.
8. image electric power widget recognition methodss are patrolled and examined based on the unmanned plane of FasterR-CNN as claimed in claim 1, which is special Levying is, step (3) step is as follows：
Step (3-1)：The further feature that 5th convolutional layer is obtained, full connection form high dimensional feature vector and realize the whole of image Body characteristicses are described, used as the input of the full articulamentum FC6 of layer 6；
The input that the position feature that RPN regions proposal network convolutional layer is obtained is also served as the full articulamentum FC6 of layer 6；
It is that the full mode for connecting carries out data exchange between layer 6 full articulamentum FC6 and layer 7 prediction interval FC7；
Layer 7 prediction interval FC7 includes sort module and regression block；
Sort module is used for the type of judging characteristic, and regression block is used for the target location of precise positioning feature；
Step (3-2)：Calculating network whole loss function simultaneously carries out each layer parameter optimization of network according to loss function；
Loss function L (p, k*,t,t*)：
L ( p , k * , t , t * ) = - log p k * + λ [ k * ≥ 1 ] Σ i ∈ { x , y , w , h } smooth L 1 ( t , t i * )
Training dataset is divided into K+1 classes, k*Represent correct tag along sort, p=(p0,...,pK) presentation class is general for k Rate,Represent the calculated indicia framing information of regression block：The abscissa of indicia framing, the vertical seat of indicia framing Mark, the height of the wide and indicia framing of indicia framing；
Computing formula is：
smooth L 1 ( t , t i * ) = 0.5 x 2 i f | x | < 1 | x | - 0.5 o t h e r w i s e ;
Label according to demarcating in training set carries out the small parameter perturbations training for detecting network, by stochastic gradient descent side, replaces Optimize network parameter；
In Optimum Classification module parameter, fixed regression block parameter constant, the fixed cluster mould in optimized regression module parameter Block parameter constant；
Propose that network and Faster R-CNN detection network finally have shared convolutional layer and define the little portion of electric power in RPN regions Part detection model.
9. image electric power widget recognition methodss are patrolled and examined based on the unmanned plane of FasterR-CNN as claimed in claim 1, which is special Levying is, step (4) step is as follows：
Step (4-1)：According to the electric power widget detection model that training is obtained, electric power widget detection model parameter is initialized；
Step (4-2)：Input picture, obtains to each classification included in image after electric power widget detection model Probit and positional information；
Step (4-3)：According to positional information, target particular location is gazed in artwork subscript.
10. image electric power widget identifying system is patrolled and examined based on the unmanned plane of FasterR-CNN, be it is characterized in that, including：
Pre-training module：Pre-training is carried out to ZFnet models, extracts the characteristic pattern that unmanned plane patrols and examines image；RPN regions are proposed Network model and Faster R-CNN detection networks are initialized；
Characteristic extracting module：Network model's training is proposed to the RPN regions that initialization is obtained, obtains extracted region network, utilize Extracted region network generates candidate region frame on the characteristic pattern of image, and the feature in the frame of candidate region is extracted, and extracts Position feature and further feature to target；
Model training module：Using the position feature of target, further feature and characteristic pattern, the Faster R- obtained by initialization CNN detection networks are trained, and obtain electric power widget detection model；
Detection module：Actual electric power widget recognition detection is carried out using electric power widget detection model.
CN201610906708.2A 2016-10-18 2016-10-18 Unmanned plane inspection image electric power widget recognition methods and system based on Faster R-CNN Active CN106504233B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610906708.2A CN106504233B (en) 2016-10-18 2016-10-18 Unmanned plane inspection image electric power widget recognition methods and system based on Faster R-CNN

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610906708.2A CN106504233B (en) 2016-10-18 2016-10-18 Unmanned plane inspection image electric power widget recognition methods and system based on Faster R-CNN

Publications (2)

Publication Number Publication Date
CN106504233A true CN106504233A (en) 2017-03-15
CN106504233B CN106504233B (en) 2019-04-09

Family

ID=58294243

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610906708.2A Active CN106504233B (en) 2016-10-18 2016-10-18 Unmanned plane inspection image electric power widget recognition methods and system based on Faster R-CNN

Country Status (1)

Country Link
CN (1) CN106504233B (en)

Cited By (104)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106877237A (en) * 2017-03-16 2017-06-20 天津大学 A kind of method of insulator missing in detection transmission line of electricity based on Aerial Images
CN107016357A (en) * 2017-03-23 2017-08-04 北京工业大学 A kind of video pedestrian detection method based on time-domain convolutional neural networks
CN107133943A (en) * 2017-04-26 2017-09-05 贵州电网有限责任公司输电运行检修分公司 A kind of visible detection method of stockbridge damper defects detection
CN107139179A (en) * 2017-05-26 2017-09-08 西安电子科技大学 A kind of intellect service robot and method of work
CN107194318A (en) * 2017-04-24 2017-09-22 北京航空航天大学 The scene recognition method of target detection auxiliary
CN107229904A (en) * 2017-04-24 2017-10-03 东北大学 A kind of object detection and recognition method based on deep learning
CN107230205A (en) * 2017-05-27 2017-10-03 国网上海市电力公司 A kind of transmission line of electricity bolt detection method based on convolutional neural networks
CN107239731A (en) * 2017-04-17 2017-10-10 浙江工业大学 A kind of gestures detection and recognition methods based on Faster R CNN
CN107256544A (en) * 2017-04-21 2017-10-17 南京天数信息科技有限公司 A kind of prostate cancer image diagnosing method and system based on VCG16
CN107274451A (en) * 2017-05-17 2017-10-20 北京工业大学 Isolator detecting method and device based on shared convolutional neural networks
CN107273836A (en) * 2017-06-07 2017-10-20 深圳市深网视界科技有限公司 A kind of pedestrian detection recognition methods, device, model and medium
CN107368845A (en) * 2017-06-15 2017-11-21 华南理工大学 A kind of Faster R CNN object detection methods based on optimization candidate region
CN107392901A (en) * 2017-07-24 2017-11-24 国网山东省电力公司信息通信公司 A kind of method for transmission line part intelligence automatic identification
CN107451997A (en) * 2017-07-31 2017-12-08 南昌航空大学 A kind of automatic identifying method of the welding line ultrasonic TOFD D scanning defect types based on deep learning
CN107507172A (en) * 2017-08-08 2017-12-22 国网上海市电力公司 Merge the extra high voltage line insulator chain deep learning recognition methods of infrared visible ray
CN107545263A (en) * 2017-08-02 2018-01-05 清华大学 A kind of object detecting method and device
CN107680090A (en) * 2017-10-11 2018-02-09 电子科技大学 Based on the electric transmission line isolator state identification method for improving full convolutional neural networks
CN107742093A (en) * 2017-09-01 2018-02-27 国网山东省电力公司电力科学研究院 A kind of infrared image power equipment component real-time detection method, server and system
CN107808116A (en) * 2017-09-28 2018-03-16 中国科学院合肥物质科学研究院 A kind of wheat spider detection method based on the fusion study of depth multilayer feature
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
CN107808425A (en) * 2017-11-28 2018-03-16 刘松林 Oil-gas pipeline cruising inspection system and its method for inspecting based on unmanned plane image
CN107844790A (en) * 2017-11-15 2018-03-27 上海捷售智能科技有限公司 A kind of vegetable identification and POS and method based on image recognition
CN107895367A (en) * 2017-11-14 2018-04-10 中国科学院深圳先进技术研究院 A kind of stone age recognition methods, system and electronic equipment
CN107944396A (en) * 2017-11-27 2018-04-20 国网安徽省电力有限公司经济技术研究院 A kind of disconnecting link state identification method based on improvement deep learning
CN107944860A (en) * 2017-11-15 2018-04-20 上海捷售智能科技有限公司 A kind of bakery identification and cash register system and method based on neutral net
CN108010025A (en) * 2017-12-14 2018-05-08 浙江大学 Switch and indicator lamp positioning and state identification method of screen cabinet based on RCNN
CN108022235A (en) * 2017-11-23 2018-05-11 中国科学院自动化研究所 High-voltage power transmission tower critical component defect identification method
CN108037133A (en) * 2017-12-27 2018-05-15 武汉市智勤创亿信息技术股份有限公司 A kind of power equipments defect intelligent identification Method and its system based on unmanned plane inspection image
CN108122003A (en) * 2017-12-19 2018-06-05 西北工业大学 A kind of Weak target recognition methods based on deep neural network
CN108171117A (en) * 2017-12-05 2018-06-15 南京南瑞信息通信科技有限公司 Electric power artificial intelligence visual analysis system based on multinuclear heterogeneous Computing
CN108171136A (en) * 2017-12-21 2018-06-15 浙江银江研究院有限公司 A kind of multitask bayonet vehicle is to scheme to search the system and method for figure
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
CN108229307A (en) * 2017-11-22 2018-06-29 北京市商汤科技开发有限公司 For the method, apparatus and equipment of object detection
CN108280855A (en) * 2018-01-13 2018-07-13 福州大学 A kind of insulator breakdown detection method based on Fast R-CNN
CN108334815A (en) * 2018-01-11 2018-07-27 深圳供电局有限公司 Method for inspecting, on off state recognition methods and the system of second power equipment
CN108537222A (en) * 2018-04-13 2018-09-14 湖南阳光电力科技有限公司 A kind of image-recognizing method and system for electric instrument
CN108537170A (en) * 2018-04-09 2018-09-14 电子科技大学 A kind of power equipment firmware unmanned plane inspection pin missing detection method
CN108564563A (en) * 2018-03-07 2018-09-21 浙江大学 A kind of tire X-ray defect detection method based on Faster R-CNN
CN108597053A (en) * 2018-04-25 2018-09-28 北京御航智能科技有限公司 Shaft tower and channel targets identification based on image data and neural network and defect diagnostic method
CN108615057A (en) * 2018-04-28 2018-10-02 广东电网有限责任公司 A kind of abnormality recognition method of the cable tunnel lighting apparatus based on CNN
CN108647682A (en) * 2018-05-17 2018-10-12 电子科技大学 A kind of brand Logo detections and recognition methods based on region convolutional neural networks model
CN108681752A (en) * 2018-05-28 2018-10-19 电子科技大学 A kind of image scene mask method based on deep learning
CN108734694A (en) * 2018-04-09 2018-11-02 华南农业大学 Thyroid tumors ultrasonoscopy automatic identifying method based on faster r-cnn
CN108764072A (en) * 2018-05-14 2018-11-06 浙江工业大学 A kind of blood cell subsets image classification method based on Multiscale Fusion
CN108776777A (en) * 2018-05-25 2018-11-09 武汉理工大学 The recognition methods of spatial relationship between a kind of remote sensing image object based on Faster RCNN
CN108830225A (en) * 2018-06-13 2018-11-16 广东工业大学 The detection method of target object, device, equipment and medium in terahertz image
CN108846325A (en) * 2018-05-28 2018-11-20 广州极飞科技有限公司 Planing method, device, storage medium and the processor of target area operation
CN108846418A (en) * 2018-05-24 2018-11-20 广东电网有限责任公司 A kind of positioning of cable machinery temperature anomaly and recognition methods
CN108875732A (en) * 2018-01-11 2018-11-23 北京旷视科技有限公司 Model training and example dividing method, device and system and storage medium
CN108921057A (en) * 2018-06-19 2018-11-30 厦门大学 Prawn method for measuring shape of palaemon, medium, terminal device and device based on convolutional neural networks
CN109117789A (en) * 2018-08-10 2019-01-01 国网上海市电力公司 A kind of switchgear inspection image segmentation configuration processing method
CN109190502A (en) * 2018-08-10 2019-01-11 北京百度网讯科技有限公司 Method and apparatus for generating location information
CN109376768A (en) * 2018-09-21 2019-02-22 福州大学 A kind of Aerial Images shaft tower Sign Board method for diagnosing faults based on deep learning
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
CN109472767A (en) * 2018-09-07 2019-03-15 浙江大丰实业股份有限公司 Stage lamp miss status analysis system
CN109522930A (en) * 2018-10-17 2019-03-26 天津大学 A kind of object detecting method based on type of barrier prediction
CN109598290A (en) * 2018-11-22 2019-04-09 上海交通大学 A kind of image small target detecting method combined based on hierarchical detection
CN109685930A (en) * 2018-12-19 2019-04-26 深圳供电局有限公司 Power components method for inspecting, device, computer equipment and storage medium
CN109766884A (en) * 2018-12-26 2019-05-17 哈尔滨工程大学 A kind of airfield runway foreign matter detecting method based on Faster-RCNN
CN109767427A (en) * 2018-12-25 2019-05-17 北京交通大学 The detection method of train rail fastener defect
CN109767423A (en) * 2018-12-11 2019-05-17 西南交通大学 A kind of crack detection method of bituminous pavement image
CN109784385A (en) * 2018-12-29 2019-05-21 广州海昇计算机科技有限公司 A kind of commodity automatic identifying method, system, device and storage medium
CN109784375A (en) * 2018-12-22 2019-05-21 国网内蒙古东部电力有限公司 Adaptive transformer part detection recognition method based on Faster RCNN
CN109829456A (en) * 2017-11-23 2019-05-31 腾讯科技（深圳）有限公司 Image-recognizing method, device and terminal
CN109886232A (en) * 2019-02-28 2019-06-14 燊赛(上海)智能科技有限公司 A kind of power grid image identification system neural network based
CN109886286A (en) * 2019-01-03 2019-06-14 武汉精测电子集团股份有限公司 Object detection method, target detection model and system based on cascade detectors
CN109993168A (en) * 2019-04-09 2019-07-09 成都鹏业软件股份有限公司 Intelligent polling method
CN110009702A (en) * 2019-04-16 2019-07-12 聊城大学 Fall webworms larva net curtain image position method for intelligence spray robot
CN110008877A (en) * 2019-03-27 2019-07-12 国网内蒙古东部电力有限公司 A kind of substation isolating-switch detection recognition method based on Faster RCNN
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
CN110298287A (en) * 2019-06-24 2019-10-01 国网上海市电力公司 A kind of power equipment automatic identifying method based on deep learning
CN110321867A (en) * 2019-07-09 2019-10-11 西安电子科技大学 Shelter target detection method based on part constraint network
CN110379036A (en) * 2019-06-26 2019-10-25 广东康云科技有限公司 Intelligent substation patrol recognition methods, system, device and storage medium
CN110443130A (en) * 2019-07-01 2019-11-12 国网湖南省电力有限公司 A kind of electric distribution network overhead wire abnormal state detection method
CN110490342A (en) * 2019-09-19 2019-11-22 江苏新绿能科技有限公司 A kind of contact net static geometric parameter detection method based on Faster R-CNN
CN110516551A (en) * 2019-07-29 2019-11-29 上海交通大学烟台信息技术研究院 A kind of line walking positional shift identifying system, method and the unmanned plane of view-based access control model
CN110766127A (en) * 2018-07-25 2020-02-07 北京深鉴智能科技有限公司 Neural network computing special circuit and related computing platform and implementation method thereof
CN110826505A (en) * 2019-11-11 2020-02-21 卓迎 Toy image detection and identification method and device based on convolutional neural network
CN111103629A (en) * 2018-10-25 2020-05-05 杭州海康威视数字技术股份有限公司 Target detection method and device, NVR (network video recorder) equipment and security check system
CN111144208A (en) * 2019-11-22 2020-05-12 北京航天控制仪器研究所 Automatic detection and identification method for marine vessel target and target detector
CN111179262A (en) * 2020-01-02 2020-05-19 国家电网有限公司 Electric power inspection image hardware fitting detection method combined with shape attribute
CN111209915A (en) * 2019-12-25 2020-05-29 上海航天控制技术研究所 Three-dimensional image synchronous identification and segmentation method based on deep learning
CN111222372A (en) * 2018-11-26 2020-06-02 珠海格力电器股份有限公司 Person searching method, device, storage medium and processor
CN111310899A (en) * 2020-02-19 2020-06-19 山东大学 Electric power defect identification method based on symbiotic relationship and small sample learning
CN111353451A (en) * 2020-03-06 2020-06-30 深圳市赛为智能股份有限公司 Battery car detection method and device, computer equipment and storage medium
CN111401215A (en) * 2020-03-12 2020-07-10 杭州涂鸦信息技术有限公司 Method and system for detecting multi-class targets
CN111598013A (en) * 2020-05-19 2020-08-28 广东电网有限责任公司 Nut-pin state identification method and related device
CN111724130A (en) * 2020-06-17 2020-09-29 广东电网有限责任公司 Distribution network machine patrol operation management system and management method
CN111784633A (en) * 2020-05-26 2020-10-16 西安理工大学 Insulator defect automatic detection algorithm for power inspection video
KR20200124280A (en) * 2018-09-27 2020-11-02 베이징 센스타임 테크놀로지 디벨롭먼트 컴퍼니 리미티드 Motion recognition, driving motion analysis method and device, electronic device
CN112016594A (en) * 2020-08-05 2020-12-01 中山大学 Collaborative training method based on domain self-adaptation
CN112132826A (en) * 2020-10-12 2020-12-25 国网河南省电力公司濮阳供电公司 Pole tower accessory defect inspection image troubleshooting method and system based on artificial intelligence
CN112180984A (en) * 2020-10-22 2021-01-05 江汉大学 Unmanned aerial vehicle auxiliary flight device based on artificial intelligence and flight control method
CN112257621A (en) * 2020-10-28 2021-01-22 贵州电网有限责任公司 Equipment image identification method for unmanned aerial vehicle inspection
CN112256906A (en) * 2020-10-23 2021-01-22 安徽启新明智科技有限公司 Method, device and storage medium for marking annotation on display screen
CN112364754A (en) * 2020-11-09 2021-02-12 云南电网有限责任公司迪庆供电局 Bolt defect detection method and system
CN112455676A (en) * 2019-09-09 2021-03-09 中国电力科学研究院有限公司 Intelligent monitoring and analyzing system and method for health state of photovoltaic panel
CN112947519A (en) * 2021-02-05 2021-06-11 北京御航智能科技有限公司 Unmanned aerial vehicle inspection method and device and edge calculation module
CN113095253A (en) * 2021-04-20 2021-07-09 池州学院 Insulator detection method for unmanned aerial vehicle to inspect power transmission line
CN113159228A (en) * 2021-05-19 2021-07-23 江苏奥易克斯汽车电子科技股份有限公司 Garbage classification identification method and device based on deep learning and intelligent garbage can
WO2021232670A1 (en) * 2020-05-22 2021-11-25 深圳技术大学 Pcb component identification method and device
US11682190B2 (en) * 2019-04-10 2023-06-20 Axis Ab Method, system, and device for detecting an object in a distorted image
CN116432089A (en) * 2023-05-15 2023-07-14 厦门星拉科技有限公司 Electric power internet of things inspection system and method
CN117292283A (en) * 2023-11-24 2023-12-26 成都庆龙航空科技有限公司 Target identification method based on unmanned aerial vehicle

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104217225A (en) * 2014-09-02 2014-12-17 中国科学院自动化研究所 A visual target detection and labeling method
CN104573731A (en) * 2015-02-06 2015-04-29 厦门大学 Rapid target detection method based on convolutional neural network
CN104573669A (en) * 2015-01-27 2015-04-29 中国科学院自动化研究所 Image object detection method
CN105825511A (en) * 2016-03-18 2016-08-03 南京邮电大学 Image background definition detection method based on deep learning

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104217225A (en) * 2014-09-02 2014-12-17 中国科学院自动化研究所 A visual target detection and labeling method
CN104573669A (en) * 2015-01-27 2015-04-29 中国科学院自动化研究所 Image object detection method
CN104573731A (en) * 2015-02-06 2015-04-29 厦门大学 Rapid target detection method based on convolutional neural network
CN105825511A (en) * 2016-03-18 2016-08-03 南京邮电大学 Image background definition detection method based on deep learning

Cited By (144)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106877237A (en) * 2017-03-16 2017-06-20 天津大学 A kind of method of insulator missing in detection transmission line of electricity based on Aerial Images
CN107016357A (en) * 2017-03-23 2017-08-04 北京工业大学 A kind of video pedestrian detection method based on time-domain convolutional neural networks
CN107016357B (en) * 2017-03-23 2020-06-16 北京工业大学 Video pedestrian detection method based on time domain convolutional neural network
CN107239731A (en) * 2017-04-17 2017-10-10 浙江工业大学 A kind of gestures detection and recognition methods based on Faster R CNN
CN107256544A (en) * 2017-04-21 2017-10-17 南京天数信息科技有限公司 A kind of prostate cancer image diagnosing method and system based on VCG16
CN107194318B (en) * 2017-04-24 2020-06-12 北京航空航天大学 Target detection assisted scene identification method
CN107194318A (en) * 2017-04-24 2017-09-22 北京航空航天大学 The scene recognition method of target detection auxiliary
CN107229904A (en) * 2017-04-24 2017-10-03 东北大学 A kind of object detection and recognition method based on deep learning
CN107133943A (en) * 2017-04-26 2017-09-05 贵州电网有限责任公司输电运行检修分公司 A kind of visible detection method of stockbridge damper defects detection
CN107133943B (en) * 2017-04-26 2018-07-06 贵州电网有限责任公司输电运行检修分公司 A kind of visible detection method of stockbridge damper defects detection
CN107274451A (en) * 2017-05-17 2017-10-20 北京工业大学 Isolator detecting method and device based on shared convolutional neural networks
CN107139179A (en) * 2017-05-26 2017-09-08 西安电子科技大学 A kind of intellect service robot and method of work
CN107139179B (en) * 2017-05-26 2020-05-29 西安电子科技大学 Intelligent service robot and working method
CN107230205A (en) * 2017-05-27 2017-10-03 国网上海市电力公司 A kind of transmission line of electricity bolt detection method based on convolutional neural networks
CN107273836A (en) * 2017-06-07 2017-10-20 深圳市深网视界科技有限公司 A kind of pedestrian detection recognition methods, device, model and medium
CN107368845A (en) * 2017-06-15 2017-11-21 华南理工大学 A kind of Faster R CNN object detection methods based on optimization candidate region
CN107392901A (en) * 2017-07-24 2017-11-24 国网山东省电力公司信息通信公司 A kind of method for transmission line part intelligence automatic identification
CN107451997A (en) * 2017-07-31 2017-12-08 南昌航空大学 A kind of automatic identifying method of the welding line ultrasonic TOFD D scanning defect types based on deep learning
CN107545263A (en) * 2017-08-02 2018-01-05 清华大学 A kind of object detecting method and device
CN107507172A (en) * 2017-08-08 2017-12-22 国网上海市电力公司 Merge the extra high voltage line insulator chain deep learning recognition methods of infrared visible ray
CN107742093B (en) * 2017-09-01 2020-05-05 国网山东省电力公司电力科学研究院 Real-time detection method, server and system for infrared image power equipment components
CN107742093A (en) * 2017-09-01 2018-02-27 国网山东省电力公司电力科学研究院 A kind of infrared image power equipment component real-time detection method, server and system
CN107808116B (en) * 2017-09-28 2020-05-05 中国科学院合肥物质科学研究院 Wheat and wheat spider detection method based on deep multilayer feature fusion learning
CN107808116A (en) * 2017-09-28 2018-03-16 中国科学院合肥物质科学研究院 A kind of wheat spider detection method based on the fusion study of depth multilayer feature
CN107680090A (en) * 2017-10-11 2018-02-09 电子科技大学 Based on the electric transmission line isolator state identification method for improving full convolutional neural networks
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
CN107895367B (en) * 2017-11-14 2021-11-30 中国科学院深圳先进技术研究院 Bone age identification method and system and electronic equipment
CN107895367A (en) * 2017-11-14 2018-04-10 中国科学院深圳先进技术研究院 A kind of stone age recognition methods, system and electronic equipment
CN107944860A (en) * 2017-11-15 2018-04-20 上海捷售智能科技有限公司 A kind of bakery identification and cash register system and method based on neutral net
CN107844790A (en) * 2017-11-15 2018-03-27 上海捷售智能科技有限公司 A kind of vegetable identification and POS and method based on image recognition
KR102407664B1 (en) * 2017-11-22 2022-06-15 베이징 센스타임 테크놀로지 디벨롭먼트 컴퍼니 리미티드 Method, apparatus and apparatus for object detection
CN108229307A (en) * 2017-11-22 2018-06-29 北京市商汤科技开发有限公司 For the method, apparatus and equipment of object detection
KR20200018612A (en) * 2017-11-22 2020-02-19 베이징 센스타임 테크놀로지 디벨롭먼트 컴퍼니 리미티드 Method, apparatus and apparatus for object detection
US11222441B2 (en) 2017-11-22 2022-01-11 Beijing Sensetime Technology Development Co., Ltd. Methods and apparatuses for object detection, and devices
CN108229307B (en) * 2017-11-22 2022-01-04 北京市商汤科技开发有限公司 Method, device and equipment for object detection
CN108022235A (en) * 2017-11-23 2018-05-11 中国科学院自动化研究所 High-voltage power transmission tower critical component defect identification method
CN109829456A (en) * 2017-11-23 2019-05-31 腾讯科技（深圳）有限公司 Image-recognizing method, device and terminal
CN108022235B (en) * 2017-11-23 2020-07-28 中国科学院自动化研究所 Method for identifying defects of key components of high-voltage transmission iron tower
CN107944396B (en) * 2017-11-27 2021-12-28 国网安徽省电力有限公司经济技术研究院 Knife switch state identification method based on improved deep learning
CN107944396A (en) * 2017-11-27 2018-04-20 国网安徽省电力有限公司经济技术研究院 A kind of disconnecting link state identification method based on improvement deep learning
CN107808425A (en) * 2017-11-28 2018-03-16 刘松林 Oil-gas pipeline cruising inspection system and its method for inspecting based on unmanned plane image
WO2019109771A1 (en) * 2017-12-05 2019-06-13 南京南瑞信息通信科技有限公司 Power artificial-intelligence visual-analysis system on basis of multi-core heterogeneous parallel computing
CN108171117B (en) * 2017-12-05 2019-05-21 南京南瑞信息通信科技有限公司 Electric power artificial intelligence visual analysis system based on multicore heterogeneous Computing
CN108171117A (en) * 2017-12-05 2018-06-15 南京南瑞信息通信科技有限公司 Electric power artificial intelligence visual analysis system based on multinuclear heterogeneous Computing
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
CN108010025A (en) * 2017-12-14 2018-05-08 浙江大学 Switch and indicator lamp positioning and state identification method of screen cabinet based on RCNN
CN108010025B (en) * 2017-12-14 2022-05-13 浙江大学 Switch and indicator lamp positioning and state identification method of screen cabinet based on RCNN
CN108122003A (en) * 2017-12-19 2018-06-05 西北工业大学 A kind of Weak target recognition methods based on deep neural network
CN108171136A (en) * 2017-12-21 2018-06-15 浙江银江研究院有限公司 A kind of multitask bayonet vehicle is to scheme to search the system and method for figure
CN108171136B (en) * 2017-12-21 2020-12-15 浙江银江研究院有限公司 System and method for searching images by images for vehicles at multi-task gate
CN108037133B (en) * 2017-12-27 2021-01-26 武汉市智勤创亿信息技术股份有限公司 Intelligent electric power equipment defect identification method and system based on unmanned aerial vehicle inspection image
CN108037133A (en) * 2017-12-27 2018-05-15 武汉市智勤创亿信息技术股份有限公司 A kind of power equipments defect intelligent identification Method and its system based on unmanned plane inspection image
CN108875732A (en) * 2018-01-11 2018-11-23 北京旷视科技有限公司 Model training and example dividing method, device and system and storage medium
CN108875732B (en) * 2018-01-11 2022-07-12 北京旷视科技有限公司 Model training and instance segmentation method, device and system and storage medium
CN108334815A (en) * 2018-01-11 2018-07-27 深圳供电局有限公司 Method for inspecting, on off state recognition methods and the system of second power equipment
CN108280855A (en) * 2018-01-13 2018-07-13 福州大学 A kind of insulator breakdown detection method based on Fast R-CNN
CN108564563A (en) * 2018-03-07 2018-09-21 浙江大学 A kind of tire X-ray defect detection method based on Faster R-CNN
CN108734694A (en) * 2018-04-09 2018-11-02 华南农业大学 Thyroid tumors ultrasonoscopy automatic identifying method based on faster r-cnn
CN108537170A (en) * 2018-04-09 2018-09-14 电子科技大学 A kind of power equipment firmware unmanned plane inspection pin missing detection method
CN108537222A (en) * 2018-04-13 2018-09-14 湖南阳光电力科技有限公司 A kind of image-recognizing method and system for electric instrument
CN108597053A (en) * 2018-04-25 2018-09-28 北京御航智能科技有限公司 Shaft tower and channel targets identification based on image data and neural network and defect diagnostic method
CN108615057B (en) * 2018-04-28 2020-07-14 广东电网有限责任公司 CNN-based abnormity identification method for cable tunnel lighting equipment
CN108615057A (en) * 2018-04-28 2018-10-02 广东电网有限责任公司 A kind of abnormality recognition method of the cable tunnel lighting apparatus based on CNN
CN108764072B (en) * 2018-05-14 2021-04-06 浙江工业大学 Blood cell subtype image classification method based on multi-scale fusion
CN108764072A (en) * 2018-05-14 2018-11-06 浙江工业大学 A kind of blood cell subsets image classification method based on Multiscale Fusion
CN108647682A (en) * 2018-05-17 2018-10-12 电子科技大学 A kind of brand Logo detections and recognition methods based on region convolutional neural networks model
CN108846418A (en) * 2018-05-24 2018-11-20 广东电网有限责任公司 A kind of positioning of cable machinery temperature anomaly and recognition methods
CN108776777A (en) * 2018-05-25 2018-11-09 武汉理工大学 The recognition methods of spatial relationship between a kind of remote sensing image object based on Faster RCNN
CN108681752A (en) * 2018-05-28 2018-10-19 电子科技大学 A kind of image scene mask method based on deep learning
CN108681752B (en) * 2018-05-28 2023-08-15 电子科技大学 Image scene labeling method based on deep learning
CN108846325A (en) * 2018-05-28 2018-11-20 广州极飞科技有限公司 Planing method, device, storage medium and the processor of target area operation
CN108830225A (en) * 2018-06-13 2018-11-16 广东工业大学 The detection method of target object, device, equipment and medium in terahertz image
CN108921057A (en) * 2018-06-19 2018-11-30 厦门大学 Prawn method for measuring shape of palaemon, medium, terminal device and device based on convolutional neural networks
CN108921057B (en) * 2018-06-19 2021-06-01 厦门大学 Convolutional neural network-based prawn form measuring method, medium, terminal equipment and device
CN110766127B (en) * 2018-07-25 2022-09-23 赛灵思电子科技(北京)有限公司 Neural network computing special circuit and related computing platform and implementation method thereof
CN110766127A (en) * 2018-07-25 2020-02-07 北京深鉴智能科技有限公司 Neural network computing special circuit and related computing platform and implementation method thereof
CN109190502A (en) * 2018-08-10 2019-01-11 北京百度网讯科技有限公司 Method and apparatus for generating location information
CN109117789A (en) * 2018-08-10 2019-01-01 国网上海市电力公司 A kind of switchgear inspection image segmentation configuration processing method
CN109472767A (en) * 2018-09-07 2019-03-15 浙江大丰实业股份有限公司 Stage lamp miss status analysis system
CN109472767B (en) * 2018-09-07 2022-02-08 浙江大丰实业股份有限公司 Stage lamp missing state analysis system
CN109376768A (en) * 2018-09-21 2019-02-22 福州大学 A kind of Aerial Images shaft tower Sign Board method for diagnosing faults based on deep learning
CN109376768B (en) * 2018-09-21 2021-12-17 福州大学 Aerial image tower signboard fault diagnosis method based on deep learning
KR20200124280A (en) * 2018-09-27 2020-11-02 베이징 센스타임 테크놀로지 디벨롭먼트 컴퍼니 리미티드 Motion recognition, driving motion analysis method and device, electronic device
KR102470680B1 (en) * 2018-09-27 2022-11-25 베이징 센스타임 테크놀로지 디벨롭먼트 컴퍼니 리미티드 Motion recognition, driving motion analysis method and device, electronic device
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
CN109522930A (en) * 2018-10-17 2019-03-26 天津大学 A kind of object detecting method based on type of barrier prediction
CN111103629A (en) * 2018-10-25 2020-05-05 杭州海康威视数字技术股份有限公司 Target detection method and device, NVR (network video recorder) equipment and security check system
CN109598290A (en) * 2018-11-22 2019-04-09 上海交通大学 A kind of image small target detecting method combined based on hierarchical detection
CN111222372A (en) * 2018-11-26 2020-06-02 珠海格力电器股份有限公司 Person searching method, device, storage medium and processor
CN109767423A (en) * 2018-12-11 2019-05-17 西南交通大学 A kind of crack detection method of bituminous pavement image
CN109767423B (en) * 2018-12-11 2019-12-10 西南交通大学 Crack detection method for asphalt pavement image
CN109685930A (en) * 2018-12-19 2019-04-26 深圳供电局有限公司 Power components method for inspecting, device, computer equipment and storage medium
CN109784375A (en) * 2018-12-22 2019-05-21 国网内蒙古东部电力有限公司 Adaptive transformer part detection recognition method based on Faster RCNN
CN109767427A (en) * 2018-12-25 2019-05-17 北京交通大学 The detection method of train rail fastener defect
CN109766884A (en) * 2018-12-26 2019-05-17 哈尔滨工程大学 A kind of airfield runway foreign matter detecting method based on Faster-RCNN
CN109784385A (en) * 2018-12-29 2019-05-21 广州海昇计算机科技有限公司 A kind of commodity automatic identifying method, system, device and storage medium
CN109886286A (en) * 2019-01-03 2019-06-14 武汉精测电子集团股份有限公司 Object detection method, target detection model and system based on cascade detectors
CN109886232A (en) * 2019-02-28 2019-06-14 燊赛(上海)智能科技有限公司 A kind of power grid image identification system neural network based
CN110008877A (en) * 2019-03-27 2019-07-12 国网内蒙古东部电力有限公司 A kind of substation isolating-switch detection recognition method based on Faster RCNN
CN110008877B (en) * 2019-03-27 2023-07-18 国网内蒙古东部电力有限公司 Substation disconnecting switch detection and identification method based on Faster RCNN
CN109993168A (en) * 2019-04-09 2019-07-09 成都鹏业软件股份有限公司 Intelligent polling method
US11682190B2 (en) * 2019-04-10 2023-06-20 Axis Ab Method, system, and device for detecting an object in a distorted image
CN110009702A (en) * 2019-04-16 2019-07-12 聊城大学 Fall webworms larva net curtain image position method for intelligence spray robot
CN110009702B (en) * 2019-04-16 2023-08-04 聊城大学 Fall webworm larva screen image positioning method for intelligent spraying robot
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
CN110298287A (en) * 2019-06-24 2019-10-01 国网上海市电力公司 A kind of power equipment automatic identifying method based on deep learning
CN110379036A (en) * 2019-06-26 2019-10-25 广东康云科技有限公司 Intelligent substation patrol recognition methods, system, device and storage medium
CN110443130A (en) * 2019-07-01 2019-11-12 国网湖南省电力有限公司 A kind of electric distribution network overhead wire abnormal state detection method
CN110321867B (en) * 2019-07-09 2022-03-04 西安电子科技大学 Shielded target detection method based on component constraint network
CN110321867A (en) * 2019-07-09 2019-10-11 西安电子科技大学 Shelter target detection method based on part constraint network
CN110516551B (en) * 2019-07-29 2023-04-07 上海交通大学烟台信息技术研究院 Vision-based line patrol position deviation identification system and method and unmanned aerial vehicle
CN110516551A (en) * 2019-07-29 2019-11-29 上海交通大学烟台信息技术研究院 A kind of line walking positional shift identifying system, method and the unmanned plane of view-based access control model
CN112455676A (en) * 2019-09-09 2021-03-09 中国电力科学研究院有限公司 Intelligent monitoring and analyzing system and method for health state of photovoltaic panel
CN110490342A (en) * 2019-09-19 2019-11-22 江苏新绿能科技有限公司 A kind of contact net static geometric parameter detection method based on Faster R-CNN
CN110490342B (en) * 2019-09-19 2023-05-02 江苏新绿能科技有限公司 Contact net static geometrical parameter detection method based on Faster R-CNN
CN110826505A (en) * 2019-11-11 2020-02-21 卓迎 Toy image detection and identification method and device based on convolutional neural network
CN111144208A (en) * 2019-11-22 2020-05-12 北京航天控制仪器研究所 Automatic detection and identification method for marine vessel target and target detector
CN111209915A (en) * 2019-12-25 2020-05-29 上海航天控制技术研究所 Three-dimensional image synchronous identification and segmentation method based on deep learning
CN111209915B (en) * 2019-12-25 2023-09-15 上海航天控制技术研究所 Three-dimensional image synchronous recognition and segmentation method based on deep learning
CN111179262A (en) * 2020-01-02 2020-05-19 国家电网有限公司 Electric power inspection image hardware fitting detection method combined with shape attribute
CN111310899B (en) * 2020-02-19 2023-07-11 山东大学 Power defect identification method based on symbiotic relation and small sample learning
CN111310899A (en) * 2020-02-19 2020-06-19 山东大学 Electric power defect identification method based on symbiotic relationship and small sample learning
CN111353451A (en) * 2020-03-06 2020-06-30 深圳市赛为智能股份有限公司 Battery car detection method and device, computer equipment and storage medium
CN111401215A (en) * 2020-03-12 2020-07-10 杭州涂鸦信息技术有限公司 Method and system for detecting multi-class targets
CN111401215B (en) * 2020-03-12 2023-10-31 杭州涂鸦信息技术有限公司 Multi-class target detection method and system
CN111598013A (en) * 2020-05-19 2020-08-28 广东电网有限责任公司 Nut-pin state identification method and related device
WO2021232670A1 (en) * 2020-05-22 2021-11-25 深圳技术大学 Pcb component identification method and device
CN111784633B (en) * 2020-05-26 2024-02-06 西安理工大学 Insulator defect automatic detection algorithm for electric power inspection video
CN111784633A (en) * 2020-05-26 2020-10-16 西安理工大学 Insulator defect automatic detection algorithm for power inspection video
CN111724130A (en) * 2020-06-17 2020-09-29 广东电网有限责任公司 Distribution network machine patrol operation management system and management method
CN112016594B (en) * 2020-08-05 2023-06-09 中山大学 Collaborative training method based on field self-adaption
CN112016594A (en) * 2020-08-05 2020-12-01 中山大学 Collaborative training method based on domain self-adaptation
CN112132826A (en) * 2020-10-12 2020-12-25 国网河南省电力公司濮阳供电公司 Pole tower accessory defect inspection image troubleshooting method and system based on artificial intelligence
CN112180984A (en) * 2020-10-22 2021-01-05 江汉大学 Unmanned aerial vehicle auxiliary flight device based on artificial intelligence and flight control method
CN112256906A (en) * 2020-10-23 2021-01-22 安徽启新明智科技有限公司 Method, device and storage medium for marking annotation on display screen
CN112257621A (en) * 2020-10-28 2021-01-22 贵州电网有限责任公司 Equipment image identification method for unmanned aerial vehicle inspection
CN112364754A (en) * 2020-11-09 2021-02-12 云南电网有限责任公司迪庆供电局 Bolt defect detection method and system
CN112947519A (en) * 2021-02-05 2021-06-11 北京御航智能科技有限公司 Unmanned aerial vehicle inspection method and device and edge calculation module
CN113095253A (en) * 2021-04-20 2021-07-09 池州学院 Insulator detection method for unmanned aerial vehicle to inspect power transmission line
CN113095253B (en) * 2021-04-20 2023-05-23 池州学院 Insulator detection method for unmanned aerial vehicle inspection transmission line
CN113159228A (en) * 2021-05-19 2021-07-23 江苏奥易克斯汽车电子科技股份有限公司 Garbage classification identification method and device based on deep learning and intelligent garbage can
CN116432089A (en) * 2023-05-15 2023-07-14 厦门星拉科技有限公司 Electric power internet of things inspection system and method
CN117292283A (en) * 2023-11-24 2023-12-26 成都庆龙航空科技有限公司 Target identification method based on unmanned aerial vehicle
CN117292283B (en) * 2023-11-24 2024-02-13 成都庆龙航空科技有限公司 Target identification method based on unmanned aerial vehicle

Also Published As

Publication number Publication date
CN106504233B (en) 2019-04-09

Similar Documents

Publication Publication Date Title
CN106504233A (en) Image electric power widget recognition methodss and system are patrolled and examined based on the unmanned plane of Faster R CNN
CN109118479B (en) Capsule network-based insulator defect identification and positioning device and method
Yuliang et al. Detecting curve text in the wild: New dataset and new solution
CN106599773B (en) Deep learning image identification method and system for intelligent driving and terminal equipment
CN104978580B (en) A kind of insulator recognition methods for unmanned plane inspection transmission line of electricity
CN111091105A (en) Remote sensing image target detection method based on new frame regression loss function
CN110136154A (en) Remote sensing images semantic segmentation method based on full convolutional network and Morphological scale-space
CN104462494B (en) A kind of remote sensing image retrieval method and system based on unsupervised feature learning
CN108288075A (en) A kind of lightweight small target detecting method improving SSD
CN112070729B (en) Anchor-free remote sensing image target detection method and system based on scene enhancement
CN109508715A (en) A kind of License Plate and recognition methods based on deep learning
CN106599789A (en) Video class identification method and device, data processing device and electronic device
CN107808129A (en) A kind of facial multi-characteristic points localization method based on single convolutional neural networks
CN105205449A (en) Sign language recognition method based on deep learning
CN108875076A (en) A kind of quick trademark image retrieval method based on Attention mechanism and convolutional neural networks
CN109034035A (en) Pedestrian&#39;s recognition methods again based on conspicuousness detection and Fusion Features
CN110097029B (en) Identity authentication method based on high way network multi-view gait recognition
CN110008899B (en) Method for extracting and classifying candidate targets of visible light remote sensing image
CN110222604A (en) Target identification method and device based on shared convolutional neural networks
CN111340034A (en) Text detection and identification method and system for natural scene
CN108764330A (en) SAR image sorting technique based on super-pixel segmentation and convolution deconvolution network
CN111881716A (en) Pedestrian re-identification method based on multi-view-angle generation countermeasure network
CN113159215A (en) Small target detection and identification method based on fast Rcnn
Wang et al. Railway insulator detection based on adaptive cascaded convolutional neural network
CN113283409A (en) Airplane detection method in aerial image based on EfficientDet and Transformer

Legal Events

Date Code Title Description
C06 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
CP01 Change in the name or title of a patent holder

Address after: Wang Yue Central Road Ji'nan City, Shandong province 250002 City No. 2000

Co-patentee after: National Network Intelligent Technology Co., Ltd.

Patentee after: Electric Power Research Institute of State Grid Shandong Electric Power Company

Co-patentee after: State Grid Corporation of China

Address before: Wang Yue Central Road Ji'nan City, Shandong province 250002 City No. 2000

Co-patentee before: Shandong Luneng Intelligent Technology Co., Ltd.

Patentee before: Electric Power Research Institute of State Grid Shandong Electric Power Company

Co-patentee before: State Grid Corporation of China

CP01 Change in the name or title of a patent holder
TR01 Transfer of patent right

Effective date of registration: 20201027

Address after: 250101 Electric Power Intelligent Robot Production Project 101 in Jinan City, Shandong Province, South of Feiyue Avenue and East of No. 26 Road (ICT Industrial Park)

Patentee after: National Network Intelligent Technology Co.,Ltd.

Address before: Wang Yue Central Road Ji'nan City, Shandong province 250002 City No. 2000

Patentee before: ELECTRIC POWER RESEARCH INSTITUTE OF STATE GRID SHANDONG ELECTRIC POWER Co.

Patentee before: National Network Intelligent Technology Co.,Ltd.

Patentee before: STATE GRID CORPORATION OF CHINA

TR01 Transfer of patent right