CN110335270A - Transmission line of electricity defect inspection method based on the study of hierarchical regions Fusion Features - Google Patents

Transmission line of electricity defect inspection method based on the study of hierarchical regions Fusion Features Download PDF

Info

Publication number
CN110335270A
CN110335270A CN201910614536.5A CN201910614536A CN110335270A CN 110335270 A CN110335270 A CN 110335270A CN 201910614536 A CN201910614536 A CN 201910614536A CN 110335270 A CN110335270 A CN 110335270A
Authority
CN
China
Prior art keywords
feature
network
layer
transmission line
fusion features
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
CN201910614536.5A
Other languages
Chinese (zh)
Other versions
CN110335270B (en
Inventor
赵振兵
李延旭
戚银城
赵文清
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.)
Shandong University
North China Electric Power University
NARI Group Corp
Zhejiang Dahua Technology Co Ltd
Zhiyang Innovation Technology Co Ltd
Original Assignee
North China Electric Power University
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 North China Electric Power University filed Critical North China Electric Power University
Priority to CN201910614536.5A priority Critical patent/CN110335270B/en
Publication of CN110335270A publication Critical patent/CN110335270A/en
Application granted granted Critical
Publication of CN110335270B publication Critical patent/CN110335270B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/08—Locating faults in cables, transmission lines, or networks
    • G01R31/081—Locating faults in cables, transmission lines, or networks according to type of conductors
    • G01R31/085—Locating faults in cables, transmission lines, or networks according to type of conductors in power transmission or distribution lines, e.g. overhead
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/08—Locating faults in cables, transmission lines, or networks
    • G01R31/088—Aspects of digital computing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G01N2021/8883—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges involving the calculation of gauges, generating models
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G01N2021/8887—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges based on image processing techniques
    • 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
    • G06T2207/20—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
    • G06T2207/20092—Interactive image processing based on input by user
    • G06T2207/20104—Interactive definition of region of interest [ROI]
    • 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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Multimedia (AREA)
  • Evolutionary Biology (AREA)
  • Biomedical Technology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Signal Processing (AREA)
  • Analytical Chemistry (AREA)
  • Immunology (AREA)
  • Chemical & Material Sciences (AREA)
  • Biochemistry (AREA)
  • Pathology (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses the transmission line of electricity defect inspection methods learnt based on hierarchical regions Fusion Features, comprising: constructs and transfer Faster R-CNN model；The target signature that core network extracts is obtained into target area by RPN net regression；Local level provincial characteristics is generated by carrying out RoI pooling operation to input picture, selects e-learning to generate weight required for Fusion Features by depth and merges further feature region and shallow-layer characteristic area；And last prediction result is generated by sorter network and Recurrent networks.The present invention merges weight using the provincial characteristics that depth selection network generates self study, save the time of adjusting parameter, and the fusion feature for enabling model learning to obtain preferably adapts to the defects detection task under different complex situations, depth model using area feature is predicted, strengthen model to the learning ability for extracting target local feature, reduces model in the actual environment because of the erroneous detection problem of complex background and the class inherited generation of transmission line of electricity defect image.

Description

Transmission line of electricity defect inspection method based on the study of hierarchical regions Fusion Features
Technical field
The present invention relates to image analysis technology field, the transmission line of electricity for being based particularly on the study of hierarchical regions Fusion Features is lacked Fall into detection method.
Background technique
The carrier that transmission line of electricity is transmitted as remote electric energy is in for a long time in severe field environment, is easy by wind The destruction of power, sleet, animal and other factors can cause to stop to power off on a large scale when serious, and economic loss is inestimable.Cause This, carries out fining inspection maintenance into the vital task of current electric system to transmission line of electricity.Aircraft inspection at present has become For conventional routine inspection mode, high-efficient, economic cost is lower.But the main pain spot of inspection is the Aerial Images quickly to increase substantially Defects detection demand and artificial detection precision, efficiency it is relatively low between contradiction.Defects detection side based on deep learning Method has become research hotspot.
However carrying out automatic identification to transmission line of electricity defect using deep learning detection method has following two problem:
(1) transmission line of electricity upper-part is more, and the type of security risk is also complicated various, therefore urgent need one can be mentioned effectively Take the detection method of the plurality of classes defect of defect target signature in image.
(2) transmission line of electricity is widely distributed, and geographical environment, the seasonal climate in each area differ greatly, limited in data In the case of, it is especially serious to the interference of testing result that different regions are distributed target background in lower image, therefore in a practical situation Model inspection performance, which can generate, sharply to be declined.
The object detection method of deep learning can be divided into two classes: first is that the detection method of Two-Stage, i.e., by test problems It is divided into two stages, first stage generates candidate region, and second stage carries out classification and position correction, main generation to target Table model has region convolutional neural networks (Region with CNN, R-CNN), Fast R-CNN, Faster R-CNN etc.；Two It is the detection method of One-Stage, without generating candidate region, the type probability and location information of direct estimation target, allusion quotation Type representative has SSD (Single Shot MultiBox Detector) model and YOLO (You Look Only Once) model. However due to above-mentioned two, cause simply to apply deep learning detection model in transmission line of electricity defects detection not Good effect can be obtained, depth detection model is generally only focused in global characteristics, and localized region feature is insensitive, and deep Already present multi-scale feature fusion method fusion process is only simply added together or cascades in degree Learning Studies, cannot be abundant The ability for playing depth model autonomous learning, anisotropic in face of complicated variety in transmission line of electricity defects detection and huge background subtraction Challenge when performance decline it is serious.
Under the limited actual conditions of data, effectively enhance model to the extractability of target signature, by part Provincial characteristics learns with carrying out different depth different levels, makes model to the understanding of target signature more focused on local region of interest On domain.The method for improving Fusion Features proposes that depth selection network makes model can not only learn how to extract feature, moreover it is possible to learn The feature for how effectively merging different levels is practised, carries on the back the fusion feature learnt in face of different defect kind different targets It can be dynamically adjusted when the case where scape, establish the more defects detection models of transmission line of electricity for reaching and adapting to actual environment.
Summary of the invention
It is an object of the invention to overcome the deficiencies of the prior art and provide the transmissions of electricity learnt based on hierarchical regions Fusion Features Line defct detection method solves the complicated variety and target background greatest differences of defect in existing transmission line of electricity defects detection Property.
The purpose of the present invention is achieved through the following technical solutions:
Transmission line of electricity defect inspection method based on the study of hierarchical regions Fusion Features, comprising:
Construct and transfer Faster R-CNN model；
The target signature that core network extracts is obtained into target area by RPN net regression；
Local depth layer region feature is generated by carrying out RoI pooling operation to input picture, is selected by depth Network generates weight required for Fusion Features and merges further feature region and shallow-layer characteristic area；And by sorter network and Recurrent networks generate last prediction result.
Further, method is specific further include:
S1, the network pre-training based on ImageNet:
Use vgg16 network to extract network as depth characteristic, pre-training is carried out on the data set of ImageNet, it is pre- to instruct Initial parameter value of the weighted value got as model in transmission line of electricity defective data collection；
S2, anchor point frame generate:
It pre-sets and respectively corresponds various sizes of different proportion anchor, the characteristic point in image is extracted, by multiple features Composition characteristic figure is put, there are the anchor of k different dimension scales, model anchor to generate total for each characteristic point in characteristic pattern Quantity is the length and widths that k*w*h, w and h are respectively convolution characteristic pattern；
S3, partial-depth feature area-of-interest is extracted:
The weight that pre-training is obtained extracts the initial parameter of network as the aspect of model, by before vgg16 network To communication process, characteristic pattern of the image after the 5th group of convolutional layer in vgg16 network is obtained；Obtained feature is inputted again In RPN, regression forecasting is carried out to the anchors generated in step S2, localized deep feature area-of-interest required for obtaining；
S4, the matching of feature channel:
Since the number of active lanes of further feature and shallow-layer feature is inconsistent, melt in further feature region and shallow-layer characteristic area Channel matching is carried out using 1 × 1 convolution kernel before closing；
S5, local shallow-layer feature area-of-interest is extracted:
Bounding box regressand value is generated according to RPN network in step S3, the local shallow-layer feature of original input picture is carried out It extracts, then size adjusting is carried out by the pond RoI layer, obtain local shallow-layer feature area-of-interest, such as:
ylThe area-of-interest shallow-layer feature for being, l refer to first of subgraph,It is that maximum pondization calculates；
S6, depth select network:
Local depth layer feature is passed through global poolization operation respectively to even up convolution characteristic pattern, then by multi-layer perception (MLP) point The weight of Fusion Features needs is not obtained, such as:
Wherein ucIt is input convolution feature, C represents the port number of input neuron, and C/r is second layer neuronal quantity, α It is perceptron weight, β is perceptron offset, and multi-layer perception (MLP) first layer neuron prevents gradient disperse using ReLU function, Second layer neuron uses Sigmoid function, by output valve specification (0,1] between region；
The weight for selecting network to obtain by depth in the training process dynamically adjust by autonomous learning, the weight learnt It is whole, to adapt to the test problems under different target different background；
S7, hierarchical regions Fusion Features:
By the weight of step S6 difference generating region further feature and region shallow-layer Fusion Features, being weighted and calculate will The fusion of two parts characteristic pattern；
S8, classification prediction network:
Fused local feature is inputted into sorter network, sorter network is composed of full articulamentum and softmax layers, Each anchors recurrence frame portion is mitogenetic at n softmax output valve, and the category distribution that softmax output valve is constituted is as last Prediction result.
Further, while network processes are predicted in classification, further include regression forecasting network: regression forecasting network is by one A fully-connected network is constituted, and is exported as 4 × n value, and belong to linear activation.
Further, further include costing bio disturbance, specifically include:
Use and intersect entropy function as the loss function of classification prediction network, formula is as follows:
In formula, piIt is predicted value, uiIt is true value；
In bounding box recurrence, Smooth L is used1As loss function, formula is as follows:
Wherein, tuIt is predicted value, v is true value.
Further, further include training dataset:
Be trained transmission line of electricity defective data collection according to neural network backpropagation principle, setting training set and Test set is 7:3 using the quantitative proportion of image, and initial learning rate is that the batchsize in 0.001 and training process is 128；Whole parameters needed for obtaining model extraction feature, obtain trained deep learning model, using model directly to defeated Electric line inspection image carries out defects detection.Further, in step s3, it also specifically includes:
Regression forecasting, the region suggested position obtained after prediction are carried out to the anchors that step S2 is generated are as follows:
X'=x+w × px
Y'=y+h × py
W'=w × epw
H'=h × eph
In formula, x ', y ', w ', h ' are respectively the centre coordinate and Length x Width after anchors is returned, x, y, w, h generation respectively The centre coordinate and Length x Width of table anchors, px、py、ph、pwRepresent the regressand value of RPN neural network forecast；
Obtained frame will be returned cut and size resetting by RoI Pooling layersFeature after cutting Figure is having a size of H × W, having a size of H ' × W ' after adjustment, then characteristic pattern after cutting is divided into H/H ' × W/W ' subgraph, every height Figure uses maximum value pond, localized deep feature area-of-interest required for obtaining.
Further, the matching of feature channel is carried out by convolution kernel, specifically further include:
Further feature port number is set as ch, obtained by fortran,
Wherein, xclIt is the shallow-layer feature that vgg16 network extracts, k={ 1,2 ... ..., ch,It is neural network volume Product calculates.
Further, hierarchical regions Fusion Features are specific further include:
ByWithPass through y=Ffuse(y, weight)=weightlyl+weighthyhMelted It closes；
Wherein, yhAnd ylIt is deep layer and shallow-layer region of interest characteristic of field respectively, weight is the power that step S7 learns Weight, local feature is passed to classification respectively after fusion and Recurrent networks are predicted.
The beneficial effects of the present invention are:
The present invention extracts effective office by convolutional neural networks and RPN network respectively from image local feature Portion's further feature and local shallow-layer feature, reinforce model to the complete perception ability of different levels feature；
Depth selection network is proposed and established, the Local Feature Fusion weight of self study is generated using multi-layer perception (MLP), both The time of adjusting parameter is saved, the fusion feature for improving efficiency, and model learning being enable to obtain preferably adapts to different complexity Defects detection task under background；
Deep learning model using area level fusion feature is predicted, study of the model to target regional area is strengthened Ability reduces in practical applications because the complex background and class inherited of transmission line of electricity defect image cause data distribution different Cause the erroneous detection problem generated.
Detailed description of the invention
Fig. 1 is flow chart of the invention；
Fig. 2 is region depth layer fusion feature visualization result figure；
Fig. 3 is the defect result figure of inspection image detection；
Wherein, (a) is that insulator falls to go here and there defect, is (b) conducting wire broken lot defect, is (c) Bird's Nest defect, is (d) that cover board lacks Lose defect.
Specific embodiment
Illustrate embodiments of the present invention below by way of specific specific example, invention is further explained.
Before illustrating concrete scheme of the present invention, the special noun being related to is explained, specific as follows:
RPN: region candidate network；Pooling layers of RoI in RoI pooling:Faster R-CNN；Anchor: anchor Point frame.
Embodiment 1:
Based on the transmission line of electricity defect inspection method of hierarchical regions Fusion Features study, please refer to shown in attached drawing 1, comprising: Construct and transfer Faster R-CNN model；The target signature that core network extracts is obtained into target area by RPN net regression Domain；Local depth layer region feature is generated by carrying out RoI pooling operation to input picture, selects network to produce by depth Weight required for raw Fusion Features merges further feature region and shallow-layer characteristic area；And by sorter network and return net Network generates last prediction result.
The specific steps of which are as follows:
S1, the network pre-training based on ImageNet:
Use vgg16 network to extract network as depth characteristic, pre-training is carried out on the data set of ImageNet, it is pre- to instruct Initial parameter value of the weighted value got as model in transmission line of electricity defective data collection；
S2, anchor point frame generate:
It pre-sets and respectively corresponds various sizes of different proportion anchors, the characteristic point in image is extracted, by multiple features Point composition characteristic figure, in characteristic pattern each characteristic point there are the anchors of k different dimension scales, model anchor generation Total quantity is the length and widths that k*w*h, w and h are respectively convolution characteristic pattern；
S3, partial-depth feature area-of-interest is extracted:
The weight that pre-training is obtained extracts the initial parameter of network as the aspect of model, passes through the forward direction of vgg16 network Communication process obtains characteristic pattern of the image after the 5th group of convolutional layer in vgg16 network；Obtained feature is inputted into RPN again In, regression forecasting is carried out to the anchors generated in step S2, localized deep feature area-of-interest required for obtaining；It is right The anchors that step S2 is generated carries out regression forecasting, the region suggested position obtained after prediction are as follows:
X'=x+w × px
Y'=y+h × py
In formula, x ', y ', w ', h ' are respectively the centre coordinate and Length x Width after anchors is returned, x, y, w, h generation respectively The centre coordinate and Length x Width of table anchors, px、py、ph、pwRepresent the regressand value of RPN neural network forecast；
Obtained frame will be returned cut and size resetting by RoI Pooling layersFeature after cutting Figure is having a size of H × W, having a size of H ' × W ' after adjustment, then characteristic pattern after cutting is divided into H/H ' × W/W ' subgraph, every height Figure uses maximum value pond, localized deep feature area-of-interest required for obtaining；
S4, the matching of feature channel:
Since the number of active lanes of further feature and shallow-layer feature is inconsistent, melt in further feature region and shallow-layer characteristic area Channel matching is carried out using 1 × 1 convolution kernel before closing；
Further feature port number is set as ch, obtained by fortran,
Wherein, xclIt is the shallow-layer feature that vgg16 network extracts, k={ 1,2 ... ..., ch,It is neural network volume Product calculates.
S5, local shallow-layer feature area-of-interest is extracted:
Bounding box regressand value is generated according to RPN network in step S3, it is locally shallow to the input picture after characteristic matching Layer feature extracts, then carries out size adjusting by the pond RoI layer, obtains local shallow-layer feature area-of-interest, such as:
ylThe area-of-interest shallow-layer feature for being, l refer to first of subgraph,It is that maximum pondization calculates；
S6, depth select network:
Local depth layer feature is passed through global poolization operation respectively to even up convolution characteristic pattern, then by multi-layer perception (MLP) point The weight of Fusion Features needs is not obtained, such as:
Wherein ucIt is input convolution feature, C represents the port number of input neuron, and C/r is second layer neuronal quantity, α It is perceptron weight, β is perceptron offset, and multi-layer perception (MLP) first layer neuron prevents gradient disperse using ReLU function, Second layer neuron uses Sigmoid function, by output valve specification (0,1] between region；
The weight for selecting network to obtain by depth in the training process dynamically adjust by autonomous learning, the weight learnt It is whole, to adapt to the test problems under different target different background；
S7, hierarchical regions Fusion Features:
By the weight of step S6 difference generating region further feature and region shallow-layer Fusion Features, being weighted and calculate will The fusion of two parts characteristic pattern；
Specifically byWithPass through y=Ffuse(y, weight)=weightlyl+weighthyhInto Row fusion；
Wherein, yhAnd ylIt is deep layer and shallow-layer region of interest characteristic of field respectively, weight is the power that step S7 learns Weight, local feature is passed to classification respectively after fusion and Recurrent networks are predicted；
S8, classification prediction network:
Fused local feature is inputted into sorter network, sorter network is composed of full articulamentum and softmax layers, Each anchors recurrence frame portion is mitogenetic at n softmax output valve, and the category distribution that softmax output valve is constituted is as last Prediction result.
S9, regression forecasting network: regression forecasting network is made of a fully-connected network, is exported as 4 × n value, and is belonged to In linear activation.
S10, costing bio disturbance:
Use and intersect entropy function as the loss function of classification prediction network, formula is as follows:
In formula, piIt is predicted value, uiIt is true value；
In bounding box recurrence, Smooth L is used1As loss function, formula is as follows:
Wherein, tuIt is predicted value, v is true value.
S11, training dataset:
Be trained transmission line of electricity defective data collection according to neural network backpropagation principle, setting training set and Test set is 7:3 using the quantitative proportion of image, and initial learning rate is that the batchsize in 0.001 and training process is 128；Whole parameters needed for obtaining model extraction feature, obtain trained deep learning model, using model directly to defeated Electric line inspection image carries out defects detection.
The present embodiment is extracted effectively by convolutional neural networks and RPN network respectively from image local feature Localized deep feature and local shallow-layer feature, reinforce model to the complete perception ability of different levels feature；It proposes and establishes depth Degree selection network, the Local Feature Fusion weight of self study is generated using multi-layer perception (MLP), the time of adjusting parameter had both been saved, and had mentioned High efficiency, but the fusion feature for enabling model learning to obtain preferably adapts to the defects detection problem under different complex backgrounds.
Embodiment 2:
Using inspection circuit image as the input of depth detection model, extracted by the vgg16 network that pre-training is completed special Sign, depth model generate various sizes of anchors (size of setting anchor is respectively (4 × 4), (8 × 8), (16 × 16), (32 × 32), length-width ratio are respectively 0.5,1,2), area-of-interest is calculated by RPN neural network forecast offset, by vgg16 Image shallow-layer feature after image further feature and the matching of feature channel that network generates is respectively fed to Pooling layers of RoI and obtains Further feature area-of-interest and shallow-layer feature area-of-interest.The weight of Fusion Features is obtained by depth selection network, is passed through The method of weighted sum generates fused characteristic pattern and visualizes characteristic pattern to obtain region level fusion feature figure, such as Fig. 2 institute Show.
Embodiment 3:
The present embodiment on the basis of embodiment 1 and embodiment 2, using the fusion feature that embodiment 2 obtains as point The input of class network and Recurrent networks, prediction obtain classification score and bounding box offset, by cross entropy loss function and Smooth L1Loss function calculates penalty values, is learnt by backpropagation, and batchsize is set as 128 at this time, study Rate decaying is set as 0.001, and the number of iterations is 40000 times, and candidate frame selection inhibits method using non-maximum value.It, will after the completion of training The inspection image for needing to detect is input to depth model, and obtaining label in Fig. 3 has classification and predict the detection image of frame.
For the present invention on the basis of Faster R-CNN model, the target signature that core network is extracted passes through RPN network Recurrence obtains target area, by carrying out RoI pooling operation and feature channel matching operation generation part to input picture Region depth layer feature generates weight required for Fusion Features as depth selection network and later melts depth layer region feature It closes, last prediction result is generated by sorter network and Recurrent networks.Deep learning model utilizes region level fusion feature The depth selection network predicted, strengthen model to the learning ability of target regional area, while be made of multi-layer perception (MLP) As the method for Fusion Features, keep the learning ability range of model more extensive, dynamic amalgamation mode can adapt to different situations Lower defect characteristic, therefore reduce model under practical circumstances because the complex background and class inherited of transmission line of electricity defect image are led Cause the erroneous detection problem of the inconsistent generation of data distribution.
A specific embodiment of the invention above described embodiment only expresses, the description thereof is more specific and detailed, but simultaneously Limitations on the scope of the patent of the present invention therefore cannot be interpreted as.It should be pointed out that for those of ordinary skill in the art For, without departing from the inventive concept of the premise, various modifications and improvements can be made, these belong to guarantor of the invention Protect range.

Claims (8)

1. the transmission line of electricity defect inspection method based on the study of hierarchical regions Fusion Features characterized by comprising
Construct and transfer Faster R-CNN model；
The target signature that core network extracts is obtained into target area by RPN net regression；
Local depth layer region feature is generated by carrying out RoI pooling operation to input picture, network is selected by depth Weight required for Fusion Features is generated to merge further feature region and shallow-layer characteristic area；And pass through sorter network and recurrence Network generates last prediction result.
2. the transmission line of electricity defect inspection method according to claim 1 based on the study of hierarchical regions Fusion Features, feature It is, specifically further include:
S1, the network pre-training based on ImageNet:
Use vgg16 network to extract network as depth characteristic, pre-training is carried out on the data set of ImageNet, pre-training obtains Initial parameter value of the weighted value arrived as model in transmission line of electricity defective data collection；
S2, anchor point frame generate:
It pre-sets and respectively corresponds various sizes of different proportion anchor, the characteristic point in image is extracted, by multiple feature point groups At characteristic pattern, each characteristic point is there are the anchors of k different dimension scales in characteristic pattern, the sum that model anchor is generated Amount is the length and widths that k*w*h, w and h are respectively convolution characteristic pattern；
S3, partial-depth feature area-of-interest is extracted:
The weight that pre-training is obtained extracts the initial parameter of network as the aspect of model, passes through the forward direction biography in vgg16 network Process is broadcast, characteristic pattern of the image after the 5th group of convolutional layer in vgg16 network is obtained；Obtained feature is inputted in RPN again, Regression forecasting is carried out to the anchors generated in step S2, localized deep feature area-of-interest required for obtaining；
S4, the matching of feature channel:
Since the number of active lanes of further feature and shallow-layer feature is inconsistent, before further feature region and the fusion of shallow-layer characteristic area Channel matching is carried out using 1 × 1 convolution kernel；
S5, local shallow-layer feature area-of-interest is extracted:
Bounding box regressand value is generated according to RPN network in step S3, the local shallow-layer feature of original input picture is extracted, Size adjusting is carried out by the pond RoI layer again, obtains local shallow-layer feature area-of-interest, such as:
ylThe area-of-interest shallow-layer feature for being, l refer to first of subgraph,It is that maximum pondization calculates；
S6, depth select network:
Local depth layer feature is passed through global poolization operation to even up convolution characteristic pattern respectively, then is obtained respectively by multi-layer perception (MLP) The weight needed to Fusion Features, such as:
Wherein ucIt is input convolution feature, C represents the port number of input neuron, and C/r is second layer neuronal quantity, and α is perception Machine weight, β are perceptron offsets, and multi-layer perception (MLP) first layer neuron prevents gradient disperse, the second layer using ReLU function Neuron uses Sigmoid function, by output valve specification (0,1] between region；
Autonomous learning, the weight learnt dynamically adjust the weight for selecting network to obtain by depth in the training process, To adapt to the test problems under different target different background；
S7, hierarchical regions Fusion Features:
By the weight of step S6 difference generating region further feature and region shallow-layer Fusion Features, it is weighted two and calculates Divide characteristic pattern fusion；
S8, classification prediction network:
Fused local feature is inputted into sorter network, sorter network is composed of full articulamentum and softmax layers, each Anchors returns that frame portion is mitogenetic at n softmax output valve, and the category distribution of softmax output valve composition is as last pre- Survey result.
3. the transmission line of electricity defect inspection method according to claim 2 based on the study of hierarchical regions Fusion Features, feature It is, while network processes are predicted in classification, further include regression forecasting network: regression forecasting network is by a fully-connected network It constitutes, exports as 4 × n value, and belong to linear activation.
4. the transmission line of electricity defect inspection method according to claim 1 based on the study of hierarchical regions Fusion Features, feature It is, further includes costing bio disturbance, specifically include:
Use and intersect entropy function as the loss function of classification prediction network, formula is as follows:
In formula, piIt is predicted value, uiIt is true value；
In bounding box recurrence, Smooth L is used1As loss function, formula is as follows:
Wherein, tuIt is predicted value, v is true value.
5. the transmission line of electricity defect inspection method according to claim 1 based on the study of hierarchical regions Fusion Features, feature It is, further includes training dataset:
Be trained transmission line of electricity defective data collection according to neural network backpropagation principle, training set and test are set Collection uses the quantitative proportion of image, the batchsize in initial learning rate and training process；Obtain model extraction feature institute The whole parameters needed, obtain trained deep learning model, directly carry out defect to polling transmission line image using model Detection.
6. the transmission line of electricity defect inspection method according to claim 2 based on the study of hierarchical regions Fusion Features, feature It is, in step s3, also specifically includes:
Regression forecasting, the region suggested position obtained after prediction are carried out to the anchors that step S2 is generated are as follows:
X'=x+w × px
Y'=y+h × py
In formula, x ', y ', w ', h ' are respectively the centre coordinate and Length x Width after anchors is returned, and x, y, w, h are respectively represented The centre coordinate and Length x Width of anchors, px、py、ph、pwRepresent the regressand value of RPN neural network forecast；
Obtained frame will be returned cut and size resetting by RoIPooling layersCharacteristic pattern ruler after cutting Very little is H × W, and having a size of H ' × W ' after adjustment, then characteristic pattern after cutting is divided into H/H ' × W/W ' subgraph, each subgraph makes With maximum value pond, localized deep feature area-of-interest required for obtaining.
7. the transmission line of electricity defect inspection method according to claim 2 based on the study of hierarchical regions Fusion Features, feature It is, the matching of feature channel is carried out by convolution kernel, specifically further include:
Further feature port number is set as ch, obtained by fortran,
Wherein, xclIt is the shallow-layer feature that vgg16 network extracts, k={ 1,2 ... ..., ch,It is neural network convolution meter It calculates.
8. the transmission line of electricity defect inspection method according to claim 2 based on the study of hierarchical regions Fusion Features, feature It is, hierarchical regions Fusion Features are specific further include:
ByWithPass through y=Ffuse(y, weight)=weightlyl+weighthyhIt is merged；
Wherein, yhAnd ylIt is deep layer and shallow-layer region of interest characteristic of field respectively, weight is the weight that step S7 learns, and is melted Local feature is passed to classification respectively after conjunction and Recurrent networks are predicted.
CN201910614536.5A 2019-07-09 2019-07-09 Power transmission line defect detection method based on hierarchical regional feature fusion learning Active CN110335270B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910614536.5A CN110335270B (en) 2019-07-09 2019-07-09 Power transmission line defect detection method based on hierarchical regional feature fusion learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910614536.5A CN110335270B (en) 2019-07-09 2019-07-09 Power transmission line defect detection method based on hierarchical regional feature fusion learning

Publications (2)

Publication Number Publication Date
CN110335270A true CN110335270A (en) 2019-10-15
CN110335270B CN110335270B (en) 2022-09-13

Family

ID=68143379

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910614536.5A Active CN110335270B (en) 2019-07-09 2019-07-09 Power transmission line defect detection method based on hierarchical regional feature fusion learning

Country Status (1)

Country Link
CN (1) CN110335270B (en)

Cited By (31)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110796039A (en) * 2019-10-15 2020-02-14 北京达佳互联信息技术有限公司 Face flaw detection method and device, electronic equipment and storage medium
CN110827263A (en) * 2019-11-06 2020-02-21 创新奇智(南京)科技有限公司 Magnetic shoe surface defect detection system and detection method based on visual identification technology
CN110866897A (en) * 2019-10-30 2020-03-06 上海联影智能医疗科技有限公司 Image detection method and computer readable storage medium
CN111144304A (en) * 2019-12-26 2020-05-12 上海眼控科技股份有限公司 Vehicle target detection model generation method, vehicle target detection method and device
CN111222467A (en) * 2020-01-08 2020-06-02 燕山大学 Wind driven generator component fault classification detection method
CN111340787A (en) * 2020-02-28 2020-06-26 广东电网有限责任公司 Method and device for detecting and identifying defects of conducting wire of power transmission line and computer equipment
CN111353413A (en) * 2020-02-25 2020-06-30 武汉大学 Low-missing-report-rate defect identification method for power transmission equipment
CN111709415A (en) * 2020-04-29 2020-09-25 北京迈格威科技有限公司 Target detection method, target detection device, computer equipment and storage medium
CN111798447A (en) * 2020-07-18 2020-10-20 太原理工大学 Deep learning plasticized material defect detection method based on fast RCNN
CN111796166A (en) * 2020-08-27 2020-10-20 广东电网有限责任公司电力调度控制中心 Power distribution network single-phase high-resistance earth fault line selection method, system and equipment
CN112001417A (en) * 2020-07-17 2020-11-27 国网宁夏电力有限公司检修公司 Monitoring method, medium and system for transformer oil conservator
CN112308850A (en) * 2020-11-09 2021-02-02 国网山东省电力公司威海供电公司 Multi-scale feature fusion power transmission line detection method and system
CN112329610A (en) * 2020-11-03 2021-02-05 中科九度(北京)空间信息技术有限责任公司 High-voltage line detection method based on edge attention mechanism fusion network
CN112419268A (en) * 2020-11-23 2021-02-26 广东电网有限责任公司广州供电局 Method, device, equipment and medium for detecting image defects of power transmission line
CN112560876A (en) * 2021-02-23 2021-03-26 中国科学院自动化研究所 Single-stage small sample target detection method for decoupling measurement
CN112686831A (en) * 2019-10-17 2021-04-20 神讯电脑(昆山)有限公司 Method for detecting surface form of object based on artificial neural network
CN112990348A (en) * 2021-04-12 2021-06-18 华南理工大学 Small target detection method for self-adjustment feature fusion
CN113205026A (en) * 2021-04-26 2021-08-03 武汉大学 Improved vehicle type recognition method based on fast RCNN deep learning network
CN113298797A (en) * 2021-06-10 2021-08-24 太原科技大学 Solar cell surface defect detection method
CN113420738A (en) * 2021-08-24 2021-09-21 中国地质大学（武汉） Self-adaptive network remote sensing image classification method, computer equipment and storage medium
CN113670929A (en) * 2021-07-05 2021-11-19 国网宁夏电力有限公司电力科学研究院 Power transmission line foreign matter detection method and device, storage medium and terminal equipment
CN113706437A (en) * 2020-05-21 2021-11-26 国网智能科技股份有限公司 Method and system for diagnosing fine-grained bolt defects of power transmission line
CN114002225A (en) * 2021-10-19 2022-02-01 业成科技（成都）有限公司 Optical detection system and method
CN114612443A (en) * 2022-03-16 2022-06-10 南京航空航天大学 Multi-mode data complex defect feature detection method
CN114972721A (en) * 2022-06-13 2022-08-30 中国科学院沈阳自动化研究所 Power transmission line insulator string recognition and positioning method based on deep learning
CN115035108A (en) * 2022-08-10 2022-09-09 四川中电启明星信息技术有限公司 Insulator defect detection method based on deep learning
CN115457553A (en) * 2022-11-11 2022-12-09 南京信息工程大学 Cloth defect detection method based on deep learning
CN115564766A (en) * 2022-11-09 2023-01-03 浙江振兴阿祥集团有限公司 Method and system for preparing volute casing seat ring of water turbine
CN115601363A (en) * 2022-12-14 2023-01-13 中建科技集团有限公司（Cn） Small-target detection algorithm-based assembly type building product defect detection method
CN116525295A (en) * 2023-07-03 2023-08-01 河南华佳新材料技术有限公司 Metallized film for high-frequency pulse capacitor and preparation method thereof
CN116958148A (en) * 2023-09-21 2023-10-27 曲阜师范大学 Method, device, equipment and medium for detecting defects of key parts of power transmission line

Citations (6)

* 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
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
CN108510012A (en) * 2018-05-04 2018-09-07 四川大学 A kind of target rapid detection method based on Analysis On Multi-scale Features figure
CN108710826A (en) * 2018-04-13 2018-10-26 燕山大学 A kind of traffic sign deep learning mode identification method
CN109034245A (en) * 2018-07-27 2018-12-18 燕山大学 A kind of object detection method merged using characteristic pattern
CN109711399A (en) * 2018-11-05 2019-05-03 北京三快在线科技有限公司 Shop recognition methods based on image, device, electronic equipment

Patent Citations (6)

* 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
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
CN108710826A (en) * 2018-04-13 2018-10-26 燕山大学 A kind of traffic sign deep learning mode identification method
CN108510012A (en) * 2018-05-04 2018-09-07 四川大学 A kind of target rapid detection method based on Analysis On Multi-scale Features figure
CN109034245A (en) * 2018-07-27 2018-12-18 燕山大学 A kind of object detection method merged using characteristic pattern
CN109711399A (en) * 2018-11-05 2019-05-03 北京三快在线科技有限公司 Shop recognition methods based on image, device, electronic equipment

Cited By (47)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110796039A (en) * 2019-10-15 2020-02-14 北京达佳互联信息技术有限公司 Face flaw detection method and device, electronic equipment and storage medium
CN112686831A (en) * 2019-10-17 2021-04-20 神讯电脑(昆山)有限公司 Method for detecting surface form of object based on artificial neural network
CN110866897A (en) * 2019-10-30 2020-03-06 上海联影智能医疗科技有限公司 Image detection method and computer readable storage medium
CN110827263A (en) * 2019-11-06 2020-02-21 创新奇智(南京)科技有限公司 Magnetic shoe surface defect detection system and detection method based on visual identification technology
CN110827263B (en) * 2019-11-06 2021-01-29 创新奇智(南京)科技有限公司 Magnetic shoe surface defect detection system and detection method based on visual identification technology
CN111144304A (en) * 2019-12-26 2020-05-12 上海眼控科技股份有限公司 Vehicle target detection model generation method, vehicle target detection method and device
CN111222467A (en) * 2020-01-08 2020-06-02 燕山大学 Wind driven generator component fault classification detection method
CN111222467B (en) * 2020-01-08 2022-03-25 燕山大学 Wind driven generator component fault classification detection method
CN111353413A (en) * 2020-02-25 2020-06-30 武汉大学 Low-missing-report-rate defect identification method for power transmission equipment
CN111353413B (en) * 2020-02-25 2022-04-15 武汉大学 Low-missing-report-rate defect identification method for power transmission equipment
CN111340787A (en) * 2020-02-28 2020-06-26 广东电网有限责任公司 Method and device for detecting and identifying defects of conducting wire of power transmission line and computer equipment
CN111340787B (en) * 2020-02-28 2023-09-29 广东电网有限责任公司 Method and device for detecting and identifying wire defects of power transmission line and computer equipment
CN111709415B (en) * 2020-04-29 2023-10-27 北京迈格威科技有限公司 Target detection method, device, computer equipment and storage medium
CN111709415A (en) * 2020-04-29 2020-09-25 北京迈格威科技有限公司 Target detection method, target detection device, computer equipment and storage medium
CN113706437B (en) * 2020-05-21 2024-03-15 国网智能科技股份有限公司 Method and system for diagnosing defects of fine-granularity bolts of power transmission line
CN113706437A (en) * 2020-05-21 2021-11-26 国网智能科技股份有限公司 Method and system for diagnosing fine-grained bolt defects of power transmission line
CN112001417B (en) * 2020-07-17 2022-07-05 国网宁夏电力有限公司检修公司 Monitoring method, medium and system for transformer oil conservator
CN112001417A (en) * 2020-07-17 2020-11-27 国网宁夏电力有限公司检修公司 Monitoring method, medium and system for transformer oil conservator
CN111798447B (en) * 2020-07-18 2023-03-10 太原理工大学 Deep learning plasticized material defect detection method based on fast RCNN
CN111798447A (en) * 2020-07-18 2020-10-20 太原理工大学 Deep learning plasticized material defect detection method based on fast RCNN
CN111796166B (en) * 2020-08-27 2022-02-15 广东电网有限责任公司电力调度控制中心 Power distribution network single-phase high-resistance earth fault line selection method, system and equipment
CN111796166A (en) * 2020-08-27 2020-10-20 广东电网有限责任公司电力调度控制中心 Power distribution network single-phase high-resistance earth fault line selection method, system and equipment
CN112329610B (en) * 2020-11-03 2024-07-12 中科九度(北京)空间信息技术有限责任公司 High-voltage line detection method based on edge attention mechanism fusion network
CN112329610A (en) * 2020-11-03 2021-02-05 中科九度(北京)空间信息技术有限责任公司 High-voltage line detection method based on edge attention mechanism fusion network
CN112308850A (en) * 2020-11-09 2021-02-02 国网山东省电力公司威海供电公司 Multi-scale feature fusion power transmission line detection method and system
CN112419268A (en) * 2020-11-23 2021-02-26 广东电网有限责任公司广州供电局 Method, device, equipment and medium for detecting image defects of power transmission line
CN112560876A (en) * 2021-02-23 2021-03-26 中国科学院自动化研究所 Single-stage small sample target detection method for decoupling measurement
CN112990348A (en) * 2021-04-12 2021-06-18 华南理工大学 Small target detection method for self-adjustment feature fusion
CN112990348B (en) * 2021-04-12 2023-08-22 华南理工大学 Small target detection method based on self-adjusting feature fusion
CN113205026A (en) * 2021-04-26 2021-08-03 武汉大学 Improved vehicle type recognition method based on fast RCNN deep learning network
CN113298797A (en) * 2021-06-10 2021-08-24 太原科技大学 Solar cell surface defect detection method
CN113298797B (en) * 2021-06-10 2022-12-02 太原科技大学 Solar cell surface defect detection method
CN113670929A (en) * 2021-07-05 2021-11-19 国网宁夏电力有限公司电力科学研究院 Power transmission line foreign matter detection method and device, storage medium and terminal equipment
CN113420738A (en) * 2021-08-24 2021-09-21 中国地质大学（武汉） Self-adaptive network remote sensing image classification method, computer equipment and storage medium
CN113420738B (en) * 2021-08-24 2021-11-09 中国地质大学（武汉） Self-adaptive network remote sensing image classification method, computer equipment and storage medium
CN114002225A (en) * 2021-10-19 2022-02-01 业成科技（成都）有限公司 Optical detection system and method
CN114002225B (en) * 2021-10-19 2023-05-12 业成科技（成都）有限公司 Optical detection system and method
CN114612443A (en) * 2022-03-16 2022-06-10 南京航空航天大学 Multi-mode data complex defect feature detection method
CN114972721A (en) * 2022-06-13 2022-08-30 中国科学院沈阳自动化研究所 Power transmission line insulator string recognition and positioning method based on deep learning
CN115035108A (en) * 2022-08-10 2022-09-09 四川中电启明星信息技术有限公司 Insulator defect detection method based on deep learning
CN115564766A (en) * 2022-11-09 2023-01-03 浙江振兴阿祥集团有限公司 Method and system for preparing volute casing seat ring of water turbine
CN115457553A (en) * 2022-11-11 2022-12-09 南京信息工程大学 Cloth defect detection method based on deep learning
CN115601363A (en) * 2022-12-14 2023-01-13 中建科技集团有限公司（Cn） Small-target detection algorithm-based assembly type building product defect detection method
CN116525295B (en) * 2023-07-03 2023-09-08 河南华佳新材料技术有限公司 Metallized film for high-frequency pulse capacitor and preparation method thereof
CN116525295A (en) * 2023-07-03 2023-08-01 河南华佳新材料技术有限公司 Metallized film for high-frequency pulse capacitor and preparation method thereof
CN116958148A (en) * 2023-09-21 2023-10-27 曲阜师范大学 Method, device, equipment and medium for detecting defects of key parts of power transmission line
CN116958148B (en) * 2023-09-21 2023-12-12 曲阜师范大学 Method, device, equipment and medium for detecting defects of key parts of power transmission line

Also Published As

Publication number Publication date
CN110335270B (en) 2022-09-13

Similar Documents

Publication Publication Date Title
CN110335270A (en) Transmission line of electricity defect inspection method based on the study of hierarchical regions Fusion Features
CN106022237B (en) A kind of pedestrian detection method of convolutional neural networks end to end
CN106326937B (en) Crowd density distribution estimation method based on convolutional neural networks
CN109765462A (en) Fault detection method, device and the terminal device of transmission line of electricity
CN107392901A (en) A kind of method for transmission line part intelligence automatic identification
CN110852347B (en) Fire detection method using improved YOLO v3
CN106991666B (en) A kind of disease geo-radar image recognition methods suitable for more size pictorial informations
CN108021947B (en) A kind of layering extreme learning machine target identification method of view-based access control model
CN107818302A (en) Non-rigid multi-scale object detection method based on convolutional neural network
CN110263705A (en) Towards two phase of remote sensing technology field high-resolution remote sensing image change detecting method
CN113160062B (en) Infrared image target detection method, device, equipment and storage medium
CN105787501B (en) Power transmission line corridor region automatically selects the vegetation classification method of feature
CN108805070A (en) A kind of deep learning pedestrian detection method based on built-in terminal
CN108629367A (en) A method of clothes Attribute Recognition precision is enhanced based on depth network
CN110378297A (en) A kind of Remote Sensing Target detection method based on deep learning
CN108154072A (en) Insulator breakdown of taking photo by plane based on depth convolutional neural networks detects automatically
CN109636846B (en) Target positioning method based on cyclic attention convolution neural network
CN109818929A (en) Based on the unknown threat cognitive method actively from step study, system, storage medium, terminal
CN109063719A (en) A kind of image classification method of co-ordinative construction similitude and category information
CN109492665A (en) Detection method, device and the electronic equipment of growth period duration of rice
CN109214426A (en) A kind of method and deep neural network model of the detection of object appearance
CN115240069A (en) Real-time obstacle detection method in full-fog scene
Yang et al. Research on subway pedestrian detection algorithms based on SSD model
Cai et al. Cloud classification of satellite image based on convolutional neural networks
Jia et al. Polar-Net: Green fruit instance segmentation in complex orchard environment

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
CB03 Change of inventor or designer information

Inventor after: Zhao Zhenbing

Inventor after: Nie Liqiang

Inventor after: Xiong Jianping

Inventor after: Zhang Wanzheng

Inventor after: Luo Wang

Inventor after: Zhao Yanqing

Inventor after: Li Yanxu

Inventor after: Qi Yincheng

Inventor after: Zhao Wenqing

Inventor before: Zhao Zhenbing

Inventor before: Li Yanxu

Inventor before: Qi Yincheng

Inventor before: Zhao Wenqing

CB03 Change of inventor or designer information
TA01 Transfer of patent application right

Effective date of registration: 20220215

Address after: 071003 Hebei province Baoding Yonghua No. 619 North Street

Applicant after: NORTH CHINA ELECTRIC POWER University (BAODING)

Applicant after: SHANDONG University

Applicant after: ZHEJIANG DAHUA TECHNOLOGY Co.,Ltd.

Applicant after: Zhiyang Innovation Technology Co.,Ltd.

Applicant after: NARI Group Corp.

Address before: 071003 Hebei province Baoding Yonghua No. 619 North Street

Applicant before: NORTH CHINA ELECTRIC POWER University (BAODING)

TA01 Transfer of patent application right
GR01 Patent grant
GR01 Patent grant