CN109961460A - A kind of multiple target method for inspecting based on improvement YOLOv3 model - Google Patents
A kind of multiple target method for inspecting based on improvement YOLOv3 model Download PDFInfo
- Publication number
- CN109961460A CN109961460A CN201811603126.2A CN201811603126A CN109961460A CN 109961460 A CN109961460 A CN 109961460A CN 201811603126 A CN201811603126 A CN 201811603126A CN 109961460 A CN109961460 A CN 109961460A
- Authority
- CN
- China
- Prior art keywords
- target
- model
- inspecting
- inspection
- multiple target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
- G06T7/248—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments involving reference images or patches
-
- 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
- G06T2207/10032—Satellite or aerial image; Remote sensing
-
- 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
- G06T2207/30232—Surveillance
-
- 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
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The invention discloses a kind of based on the multiple target method for inspecting for improving YOLOv3 model, belong to line upkeep technical field, include: to design a kind of YOLOv3 network structure for adapting to electric power unmanned plane inspection target visual characteristic, dynamically track is carried out to the target under unmanned plane inspection scene and detection identifies;The video information based on deep learning model is constructed to extract, analyze and target detection model;For the target under the conditions of high-speed motion, analysis is associated to the recognition result of consecutive frame image;Using target effective feature, found in sequence image using matching algorithm with the most like picture position of target template, provide authentic data source for the target inspection of next step.The invention also discloses the step of realizing the above method, equipment and storage medium, the exemplary technical solution of the present invention is used for the real-time inspection of power transmission line route unmanned plane, and real-time is high, result is accurate, compared with currently used manual inspection, greatly reduces cost.
Description
Technical field
The invention belongs to the inspections of power circuit O&M technical field more particularly to transmission line of electricity, specifically a kind of
Image processing techniques based on deep learning.
Background technique
There are significant differences in application demand with other unmanned vehicles for overhead transmission line unmanned plane cruising inspection system, respectively
Item technical requirement is there are larger difference, and detection project and detection method etc. still belong to blank, and unmanned plane is mainly used at present
The terrain environments such as Plain, hills are not yet related to the application study in complicated landforms such as high and cold, High aititude, depopulated zones.
Domestic unit primarily focuses on application feasibility verifying to overhead transmission line unmanned plane inspection operation correlative study
Aspect, technical research in terms of embedded multi-targets recognition are related to less.From the research of constituent parts in recent years, using feelings
From the point of view of condition, unmanned plane inspection operation still has problem following prominent:
(1) unmanned plane inspection background process real-time is insufficient, and routing inspection efficiency is low
The data analysis acquired under unmanned plane inspection scene is mostly backstage processed offline, and real-time is insufficient, for multiple target,
Transmission line of electricity target defect under complex background lacks online tentative diagnosis extremely.Unmanned plane inspection simultaneously lacks power transmission and transforming equipment
Target real-time detection, leading to collected many video image data is not target area, generates mass of redundancy data, and reduction is patrolled
Examine efficiency.
(2) inspection of unmanned plane multiple target cannot achieve
Unmanned plane, can not be effective when carrying out target defect detection to all images of acquisition during inspection
Simultaneously identify more defective locations, be easy to cause and fail to report, lack of wisdom supplementary means, to the validity of inspection result cause compared with
It is big to influence.
(3) patrol unmanned eedle is to small target deteections scarce capacities such as conducting wire, small fittings
Since patrol unmanned machine acquired image is all relatively remote captured, and the fine granularities such as small fitting defect is very
Difficulty, which quickly detects, to be come, and current conventional depth learning model is equally poor to small size target detection capabilities, is unable to satisfy nothing
The requirement of man-machine real-time inspection.
With artificial intelligence, the development of embedded image processing and identification, needed in conjunction with the business of unmanned plane inspection at this stage
It asks, carries out unmanned plane inspection on-line checking, targeted diagnostics and intelligent control technology research and application, provided rationally for the above problem
Solution ensure inspection safety to improve routing inspection efficiency.
Summary of the invention
To solve above-mentioned deficiency of the prior art, the purpose of the present invention is to provide one kind based on improvement YOLOv3 mould
The multiple target method for inspecting of type, this method and be to stay in embedded FPGA chip as carrier using unmanned plane and helicopter are being transmitted electricity
Identification is measured in real time in route multiple target defect inspection, while can be in agricultural, mining industry, traffic infrastructure, exploration and public affairs
The scenes such as safety, which are carried out, altogether promotes and applies.
The technical scheme adopted by the invention is as follows:
On the one hand, it provides a kind of based on the multiple target method for inspecting for improving YOLOv3 model, comprising:
To the multi-target detection based on YOLOv3 model and Multitarget Tracking phase under the motion state of SURF characteristic point
In conjunction with;
Further, inspection multi-target detection is carried out by YOLOv3 model and uses appropriate using target effective feature
With algorithm, searching and the most like picture position of target template in sequence image, according to the SURF feature point set of target image
Carry out the multiple target tracking under motion state.
Improved YOLOv3 model is fired on fpga chip;
Further, it will fire in intelligent chip, meet after the defects detection model optimization trained by YOLOv3
Low-power consumption, high speed, high accurately recognition effect.
Improved YOLOv3 model multiple target method for inspecting is applied in electric power unmanned plane vision inspection;
Dynamically track and detection identification are carried out to the target under unmanned plane inspection scene;
Further, the acquisition original image is carried out by the imaging unit being mounted in line system.
Further, inspection image is pre-processed, obtains target image:
Using the defect sample database of building is denoised based on edge detection weighting guiding filtering algorithm, defogging etc.
Processing.
Further, data set augmentation is carried out to target image, using image decentralization, RGB disturbance, random noise etc.
Method is handled.
The image size in normalization data library;
By classification, the sample size of equalization data library difference defect type constructs reasonable training set and test set.
Further, the calculating of anchors value is carried out to target image:
The anchors of all kinds of defect training samples of database is calculated by kmeans algorithm;
Suitable 9 anchors values are calculated according to defect classification in target image and size situation.
Further, YOLOv3 model parameter is initialized:
Reasonable model training the number of iterations and part learning rate are set;
Modify anchors value and loss function.
Further, training objective defects detection model:
It is realized by the visualization interface Tensorboard in Tensorflow frame to the parameter during model training
Optimization;
Until the loss value of model drops to very low stationary value, it can just stop model training;
Pass through AP and mAP (mean of AP (Average Precision) the script computation model on test set
Average Precision) value analyzes reason, rebuilds data set if result does not reach expected.
Further, the multiple target inspection, which tracks, includes:
After identification model training result, so that it may model is fired into FPGA, later by FPGA module be loaded into inspection without
It is man-machine;
Target window is determined in the object pixel neighborhood of a point;
The gray value that the target pixel points are replaced with the intermediate value of the gray value of all pixels point in the target window, obtains
To filtered image.
Target area characteristic pattern vector in target image is extracted, and is exported;
Target area and high latitude characteristic pattern are read, is ranked up by the probability for target object occur in region is extracted, and defeated
High probability feature graph region out;
It is exported according to region screening layer, regional aim is analyzed, is classified, according to classificating requirement for pixel, area
Domain carries out classification marker;
During unmanned plane inspection, picture can be passed to picture recognition module by detector, after model inspection, output
The result is that the defect type and defective locations coordinate of model prediction.
Further, by the position of model prediction target, the SURF feature point set of target is extracted, and calculates SURF spy
Levy the mapping matrix of point.
Further, by one column vector equivalent representation of the mapping matrix of SURF characteristic point, pass through the column of mapping matrix
The weight coefficient of the adjustable neural network of vector.
Further, by particle filter, the maximum particle confidence level that neural computing goes out is exported as a result, such as
The fruit particle is not local optimal particle, then mobile by the optimum point of particle to part using average drifting, final to can get
Target state estimator position.
It is on the other hand, provided by the invention a kind of based on the multiple target method for inspecting for improving YOLOv3 model further include:
Imaging unit is configured to acquisition original image;
Pretreatment unit is configured to pre-process original image, obtains target image;
Arithmetic element is configured to obtain the defect type of transmission line of electricity;
Output unit is configured to output defect type and location information;
Wherein, arithmetic element includes computing module and database module, and the database module is for storage and output line
The good location information of road actual defects type.
Compared with prior art, the invention has the benefit that
1, the present invention is exemplary a kind of based on the multiple target method for inspecting for improving YOLOv3 model, relatively high for unmanned plane
Target is easy to be lost under fast motion state, is easy to miss inspection, the problems such as diagnostic accuracy is lower, using YOLOv3 model and SURF characteristic point
The unmanned plane inspection multi-object Recognition Model of tracking realizes that unmanned plane load is limited in conjunction with the intelligent chip of embedded type low-power consumption
In the case of inspection scene multi-targets recognition diagnosis, power transmission line unmanned machine inspection fault diagnosis timeliness is improved, through testing
Verification result accuracy is high, and concept feasible is strong.
2, the present invention is exemplary based on the multiple target method for inspecting for improving YOLOv3 model, is located in advance to original image
Reason eliminates the noise in original image, improves the clarity of original image, while carrying out data augmentation to original image, improves
Model training effect directly improves the accuracy of inspection result.
3, the present invention is exemplary based on the multiple target method for inspecting for improving YOLOv3 model, uses embedded FPGA etc.
Intelligent chip front end recognition technology reduces unmanned plane picture amount of storage and picture transfer pressure, directly in front end by defect recognition
As a result it is transferred to background system, effectively increases unmanned plane routing inspection efficiency.
Detailed description of the invention
By reading a detailed description of non-restrictive embodiments in the light of the attached drawings below, the application's is other
Feature, objects and advantages will become more apparent upon:
Fig. 1 is the flow chart of the embodiment of the present invention;
Fig. 2 is multiple target inspection algorithm frame figure of the invention;
Fig. 3 is training pattern algorithm flow chart of the invention;
Fig. 4 is that the embodiment of the present invention improves YOLOv3 network parameter figure;
Fig. 5 is the image preprocessing and augmentation flow chart of the embodiment of the present invention;
Fig. 6 is gray scale schematic diagram of the invention.
Specific embodiment
The application is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched
The specific embodiment stated is used only for explaining related invention, rather than the restriction to the invention.It also should be noted that in order to
Convenient for description, part relevant to invention is illustrated only in attached drawing.
It should be noted that in the absence of conflict, the features in the embodiments and the embodiments of the present application can phase
Mutually combination.The application is described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.
As shown in Figure 1, An embodiment provides a kind of based on the multiple target inspection for improving YOLOv3 model
Method, comprising:
S1: target area image is acquired by unmanned aerial vehicle onboard inspection detector;
S2: pretreatment and partial data augmentation are carried out to acquired image;
S3: improved YOLOv3 model parameter is initialized;
S4: the defective data of acquisition is subjected to disaggregated model training, and carries out parameter optimization, it is ensured that output model effect is most
It is excellent;
S5: by the trained all kinds of airborne fpga chips of defects detection model burning;
S6: airborne inspection device is according to the defects of the acquired image of front end recognition model inspection position and type;
S7: carrying out SURF feature point tracking to the target image of acquisition, gives defect inspection model feedback in real time, and will know
Other result is transmitted back to background system.
In S1, the real-time acquisition of image is carried out to region of patrolling and examining by airborne imaging equipment, early period, the data of acquisition passed through
It is used for the training of defect inspection model after artificial mark and verification, airborne equipment will be adopted after the completion of the training of all kinds of defect models
The image of collection carries out intelligent recognition analysis.
In S2, pretreatment operation is carried out to the image for defects detection of acquisition, process is as shown in Figure 5.By being based on
Edge detection weighting guiding filtering algorithm denoises the defect sample database of building, the processing such as defogging, augmentation;Normalization
The image size of database, by classification, the sample size of equalization data library difference defect type, construct reasonable training set and
Test set.
In S3, using the network structure and parameter of Fig. 4, all kinds of defect training samples of database are calculated by kmeans algorithm
This anchors, is arranged reasonable model training the number of iterations and part learning rate, and initialization model parameter starts
Multiple target defects detection model of the training based on YOLOv3 on Tensorflow frame, optimizes loss function, realizes end-to-end
Training and detection, overcome to mutually from the poor disadvantage of close object and microcommunity detection effect, improve and detect position
Accuracy.
Improved YOLOv3 loss function:
Wherein, formula (1), (2) are the predictions to defect coordinate, and formula (3) is the confidence level to the box containing target object
Prediction, formula (4) are the confidence level predictions to the box without target object, and formula (5) is the prediction to target defect classification.
It is real by the visualization interface Tensorboard in Tensorflow frame using the training process of Fig. 3 in S4
Now to the parameter optimization during model training, until the loss value of model drops to very low stationary value, it can just stop model instruction
Practice;Pass through AP and mAP (mean Average of AP (Average Precision) the script computation model on test set
Precision) value analyzes reason, rebuilds database if result does not reach expected.
In the deep learning framework, input terminal is original power transmission line inspection video image, and output end is the inspection of electric power target
Survey result.Wherein input picture is analyzed without empirical model, directly extracts high latitude characteristic pattern by depth convolutional layer.Output
Characteristic pattern is for supporting three aspect functions:
(1) target area extract layer extracts potential target area in characteristic pattern according to characteristic pattern vector, and by target area
Domain output;
(2) region screening layer, which extracts, reads target area extraction once output and high dimensional feature figure, occurs by extracting in region
The probability of target object is ranked up, and exports high probability feature graph region;
(3) target analysis layer is exported according to region screening layer, and regional aim is analyzed, is classified, can be according to classification
It is required that carrying out classification marker for pixel, region.
Target area extract layer, region screening layer and target analysis layer are based on deep learning network in the deep learning network
Layer exploitation, is connected with existing convolutional layer, full articulamentum.The end-to-end deep learning network architecture is trained via great amount of images,
Wherein target area extract layer and screening layer are according to label results area training network layer parameter, convolutional layer and target analysis layer root
It is adjusted according to label result loss function.
In S5, after identification model training result, so that it may model be fired FPGA, FPGA module is loaded into patrols later
Examine unmanned plane.
In S6, during unmanned plane inspection, picture can be passed to picture recognition module by detector, by model inspection it
Afterwards, output the result is that the defect type and defective locations coordinate of model prediction.
In S7, during tracking inspection, the high level that multiple target inspection technology obtains training under SURF feature and line is special
Sign combines, and using the scale invariability of SURF feature, improves mentioned algorithm to the accuracy of the target following of dimensional variation,
The tracking result of particle filter is verified and corrected later by average drifting track algorithm.
(1) improved BP neural network feature learning
Improved BP neural network, by the fine tuning to neural network, may finally be obtained for calculating particle confidence level
Maximum particle confidence level.Target in first frame can obtain the prediction target position of model by the algorithm of target detection of S6
It sets, so as to obtain target sample and background sample and initialize neural network.During tracking, if the maximum of particle
Confidence level is less than threshold value, then resampling background template, and uses new template set re -training neural network.Extract present frame
SURF feature point set, and matched with the target SURF feature point set of first frame, the SURF feature point set meter that will match to
Calculate SURF characteristic point mapping matrix.In SURF characteristic point matrix, if certain point is SURF characteristic point, corresponding in matrix
Element value be set as 0.3;Otherwise, it is set as 0.0.3 this value is obtained by experiment experience, and the other values that compare effect is more
It is good.Then, the weight between the first hidden layer of neural network and input layer, the first hidden layer are finely tuned according to SURF characteristic point mapping matrix
Activation primitive such as formula (6), (7) indicate.
H (x)=f (x) * ((W0+Ws)*x+b) (7)
Wherein, x indicates input, and h indicates output, and b indicates bias, initialization of the W0 between input layer and the first hidden layer
Weight, WS are the SURF characteristic point mapping matrix that column vector indicates.
(2) multiple target inspection tracking technique
During the inspection of multiple target tracking, the new frame that unmanned plane inspection is transmitted is input to mind by sampling particle
Through in network, obtaining its confidence level, detailed process is as follows:
Sampling particle is indicated with X, with (W, b)=(W(1),b(1),W(2),b(2),...,W(6),b(6)) indicate BP neural network
Weight and offset parameter collection between each layer, the calculation formula of the output of the first hidden layer are as follows:
a(1)=f (W(1)X+b(1)) (8)
Wherein, W(1)And b(1)Respectively indicate the weight and biasing between the first hidden layer and input layer, W(1)=W0 (1)+Ws, f
() indicates activation primitive, calculation formula such as formula (6).
For the 2nd hidden layer to the 5th hidden layer, m layers of output is m+1 layers of input, carries out propagated forward calculating, is calculated
Shown in formula such as formula (9):
a(m)=f (W(m)*a(m-1)+b(m)) (9)
For output layer, the confidence level that result is corresponding particle is exported, calculation formula is as follows:
C=f (W(6)*a(5)+b(6)) (10)
The result is that possessing the particle of maximum confidence, the particle is verified further according to average drifting is for final target following
No is the particle of local optimum, if it is not, then being iterated calculating using the particle centre as the origin center of average drifting
Mean shift vectors, particle sample is constantly close to local best points, until obtaining more accurate tracking result,
Multiple target tracking algorithm steps are as follows.
Initialization: the future position of model is obtained by the algorithm of target detection of 2.1 sections, so as to obtain target
Sample T and background sample B, and initialize BP neural network.
1) the target template T1 of first frame F1 is established;
2) the SURF feature point set P1 for extracting F1 calculates SURF characteristic point mapping matrix, and according to formula (7) to BP mind
It is finely adjusted through network weight;
3) it is directed to the i-th frame Fi:
(a) sampling particle is obtained according to Particle filtering theory;
(b) according to formula (8)-(10), the confidence level of each particle is calculated using BP neural network, finally obtains confidence level
Maximum particle qi max;
(c) according to mean shift theory, more accurate tracking result result is finally obtained by iterationi;
(d) by resultiAs Fi final tracking result and export.
4) if necessary to update, then resampling and BP neural network and training are initialized, returns to step 2;Otherwise, i=i
+ 1, return to step 3.
Algorithm of target detection performance comparison based on different neural networks is as shown in table 1, it can be seen that this patent proposes
It is a kind of based on improve YOLOv3 model multiple target method for inspecting in single picture recognition time and discrimination and multiple target
There is preferable effect in detection.
Algorithm of target detection performance comparison of the table 1 based on different neural networks
As can be seen that a kind of multiple target method for inspecting based on improvement YOLOv3 model is to transmission line of electricity from test result
The preparation verification and measurement ratio of defect is all larger than 80%, and average response time meets unmanned plane inspection real-time and standard within 250ms
True property requirement.By the test, basic verification is of the invention exemplary based on the multiple target method for inspecting for improving YOLOv3 model
Feasibility.
Above description is only the preferred embodiment of the application and the explanation to institute's application technology principle.Those skilled in the art
Member is it should be appreciated that invention scope involved in the application, however it is not limited to technology made of the specific combination of above-mentioned technical characteristic
Scheme, while should also cover in the case where not departing from the inventive concept, it is carried out by above-mentioned technical characteristic or its equivalent feature
Any combination and the other technical solutions formed.Such as features described above has similar function with (but being not limited to) disclosed herein
Can technical characteristic replaced mutually and the technical solution that is formed.
Except for the technical features described in the specification, remaining technical characteristic is the known technology of those skilled in the art, is prominent
Innovative characteristics of the invention out, details are not described herein for remaining technical characteristic.
Claims (10)
1. a kind of based on the multiple target method for inspecting for improving YOLOv3 model, characterized in that include:
Multi-target detection based on YOLOv3 model is combined with Multitarget Tracking under the motion state of SURF characteristic point;
Improved YOLOv3 model multiple target method for inspecting is applied in electric power unmanned plane vision inspection;
Improved YOLOv3 model is fired on fpga chip;
Dynamically track and detection identification are carried out to the target under unmanned plane inspection scene;
According to multi-object Recognition Model, accuracy is further increased using context frame matching technology to adjacent frame sequence;
Pass unmanned plane defects detection result back background system;
Wherein, on the basis of moving object detection, using target effective feature, using matching algorithm, target is not only provided
Motion profile and preparation positioning also provide reliable data source with understanding for the goal behavior analysis of next step.
2. according to claim 1 based on the multiple target method for inspecting for improving YOLOv3 model, characterized in that transported to convolution
It calculates module, sampling computing module, activation primitive module and carries out targeted design.
3. according to claim 2 based on the multiple target method for inspecting for improving YOLOv3 model, characterized in that examine target
Survey problem is converted into regression problem, realizes end-to-end training and detection.
4. according to claim 1 based on the multiple target method for inspecting for improving YOLOv3 model, characterized in that adopting in real time
The image collected carries out at the normalized of size 416*416 and the guiding filtering defogging denoising based on edge detection weighting
Reason, is transported to model identification module for pretreated image, obtains the prediction defect type and predicted position of model.
5. according to claim 1 based on the multiple target method for inspecting for improving YOLOv3 model, characterized in that use more mesh
Mark tracking technique, which combines SURF feature with the high-level characteristic that training obtains under line, carries out inspection tracking.
6. according to claim 1 based on the multiple target method for inspecting for improving YOLOv3 model, characterized in that extract current
The SURF feature point set of frame, and matched with the target SURF feature point set of first frame, the SURF feature point set that will match to
Calculate SURF characteristic point mapping matrix.
7. according to claim 1 based on the multiple target method for inspecting for improving YOLOv3 model, characterized in that target following
The result is that possess the particle of maximum confidence, verified further according to average drifting the particle whether be local optimum particle, such as
Fruit is not, then using the particle centre as the origin center of average drifting, to be iterated calculating mean shift vectors, by particle sample
This is constantly close to local best points, until obtaining more accurate tracking result.
8. a kind of based on the multiple target method for inspecting for improving YOLOv3 model, comprising:
Imaging unit is configured to unmanned plane acquisition original image;
Pretreatment unit is configured to pre-process original image, obtains target image;
Arithmetic element is configured to obtain transmission line of electricity defect information;
Output unit is configured to the result of output transmission line of electricity defect type, anchor point;
Wherein, arithmetic element includes computing module and database module, and the database module is real with outlet line for storing
Border defect type and anchor point information.
9. a kind of equipment, characterized in that the equipment includes:
One or more processors;
Memory, for storing one or more programs,
When one or more of programs are executed by one or more of processors, so that one or more of processors
It executes a method as claimed in any one of claims 1-8 based on the multiple target method for inspecting for improving YOLOv3 model.
10. a kind of computer readable storage medium for being stored with computer program, characterized in that when the program is executed by processor
It realizes a method as claimed in any one of claims 1-8 based on the multiple target method for inspecting for improving YOLOv3 model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811603126.2A CN109961460A (en) | 2018-12-26 | 2018-12-26 | A kind of multiple target method for inspecting based on improvement YOLOv3 model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811603126.2A CN109961460A (en) | 2018-12-26 | 2018-12-26 | A kind of multiple target method for inspecting based on improvement YOLOv3 model |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109961460A true CN109961460A (en) | 2019-07-02 |
Family
ID=67023379
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811603126.2A Pending CN109961460A (en) | 2018-12-26 | 2018-12-26 | A kind of multiple target method for inspecting based on improvement YOLOv3 model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109961460A (en) |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110889841A (en) * | 2019-11-28 | 2020-03-17 | 江苏电力信息技术有限公司 | YOLOv 3-based bird detection algorithm for power transmission line |
CN110992307A (en) * | 2019-11-04 | 2020-04-10 | 华北电力大学(保定) | Insulator positioning and identifying method and device based on YOLO |
CN110990658A (en) * | 2019-11-15 | 2020-04-10 | 西安交通大学 | Method for realizing image processing algorithm of power transmission line on embedded system |
CN111027388A (en) * | 2019-11-12 | 2020-04-17 | 国网天津市电力公司 | Method and system for monitoring safety operation behaviors of constructors based on image recognition |
CN111046785A (en) * | 2019-12-10 | 2020-04-21 | 长讯通信服务有限公司 | Method for identifying key target of unmanned aerial vehicle routing inspection video based on convolutional neural network |
CN111104906A (en) * | 2019-12-19 | 2020-05-05 | 南京工程学院 | Transmission tower bird nest fault detection method based on YOLO |
CN111210408A (en) * | 2019-12-30 | 2020-05-29 | 南京航空航天大学 | Ray image-based composite material defect identification method |
CN111240355A (en) * | 2020-01-10 | 2020-06-05 | 哈尔滨工业大学 | Cruise formation planning system of multi-target communication unmanned aerial vehicle based on secondary clustering |
CN111474953A (en) * | 2020-03-30 | 2020-07-31 | 清华大学 | Multi-dynamic-view-angle-coordinated aerial target identification method and system |
CN111709308A (en) * | 2020-05-25 | 2020-09-25 | 五邑大学 | Unmanned aerial vehicle-based maritime distress personnel detection and tracking method and system |
CN111984034A (en) * | 2020-08-24 | 2020-11-24 | 广东电网有限责任公司 | Unmanned aerial vehicle electric power patrols line system |
CN113205063A (en) * | 2021-05-19 | 2021-08-03 | 云南电网有限责任公司电力科学研究院 | Visual identification and positioning method for defects of power transmission conductor |
CN113297910A (en) * | 2021-04-25 | 2021-08-24 | 云南电网有限责任公司信息中心 | Distribution network field operation safety belt identification method |
CN113312991A (en) * | 2021-05-14 | 2021-08-27 | 华能阜新风力发电有限责任公司 | Front-end intelligent recognition system based on unmanned aerial vehicle |
CN113534832A (en) * | 2021-08-03 | 2021-10-22 | 国网江苏省电力有限公司泰州供电分公司 | Unmanned aerial vehicle inspection tracking distribution network line flying method based on edge calculation |
CN113657551A (en) * | 2021-09-01 | 2021-11-16 | 陕西工业职业技术学院 | Robot grabbing posture task planning method for sorting and stacking multiple targets |
CN114782805A (en) * | 2022-03-29 | 2022-07-22 | 中国电子科技集团公司第五十四研究所 | Unmanned aerial vehicle patrol-oriented man-in-loop hybrid enhanced target identification method |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20090080158A1 (en) * | 2007-09-24 | 2009-03-26 | Research Triangle Institute | Comb-shaped power bus bar assembly structure having integrated capacitors |
CN106529587A (en) * | 2016-08-26 | 2017-03-22 | 北京航空航天大学 | Visual course identification method based on target point identification |
CN207133937U (en) * | 2017-07-26 | 2018-03-23 | 南京信息工程大学 | A kind of forest fire early-warning system based on FPGA and unmanned plane |
CN108053714A (en) * | 2017-11-10 | 2018-05-18 | 广东电网有限责任公司教育培训评价中心 | Multi-rotor unmanned aerial vehicle based on polling transmission line makes an inspection tour operation simulation training system |
CN108320510A (en) * | 2018-04-03 | 2018-07-24 | 深圳市智绘科技有限公司 | One kind being based on unmanned plane video traffic information statistical method and system |
CN108416299A (en) * | 2018-03-09 | 2018-08-17 | 重庆览辉信息技术有限公司 | A kind of power circuit intelligent inspection system |
-
2018
- 2018-12-26 CN CN201811603126.2A patent/CN109961460A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20090080158A1 (en) * | 2007-09-24 | 2009-03-26 | Research Triangle Institute | Comb-shaped power bus bar assembly structure having integrated capacitors |
CN106529587A (en) * | 2016-08-26 | 2017-03-22 | 北京航空航天大学 | Visual course identification method based on target point identification |
CN207133937U (en) * | 2017-07-26 | 2018-03-23 | 南京信息工程大学 | A kind of forest fire early-warning system based on FPGA and unmanned plane |
CN108053714A (en) * | 2017-11-10 | 2018-05-18 | 广东电网有限责任公司教育培训评价中心 | Multi-rotor unmanned aerial vehicle based on polling transmission line makes an inspection tour operation simulation training system |
CN108416299A (en) * | 2018-03-09 | 2018-08-17 | 重庆览辉信息技术有限公司 | A kind of power circuit intelligent inspection system |
CN108320510A (en) * | 2018-04-03 | 2018-07-24 | 深圳市智绘科技有限公司 | One kind being based on unmanned plane video traffic information statistical method and system |
Non-Patent Citations (1)
Title |
---|
任楠: "基于深度学习的空间运动目标检测与跟踪研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110992307A (en) * | 2019-11-04 | 2020-04-10 | 华北电力大学(保定) | Insulator positioning and identifying method and device based on YOLO |
CN111027388A (en) * | 2019-11-12 | 2020-04-17 | 国网天津市电力公司 | Method and system for monitoring safety operation behaviors of constructors based on image recognition |
CN110990658A (en) * | 2019-11-15 | 2020-04-10 | 西安交通大学 | Method for realizing image processing algorithm of power transmission line on embedded system |
CN110889841A (en) * | 2019-11-28 | 2020-03-17 | 江苏电力信息技术有限公司 | YOLOv 3-based bird detection algorithm for power transmission line |
CN111046785A (en) * | 2019-12-10 | 2020-04-21 | 长讯通信服务有限公司 | Method for identifying key target of unmanned aerial vehicle routing inspection video based on convolutional neural network |
CN111104906A (en) * | 2019-12-19 | 2020-05-05 | 南京工程学院 | Transmission tower bird nest fault detection method based on YOLO |
CN111210408A (en) * | 2019-12-30 | 2020-05-29 | 南京航空航天大学 | Ray image-based composite material defect identification method |
CN111240355A (en) * | 2020-01-10 | 2020-06-05 | 哈尔滨工业大学 | Cruise formation planning system of multi-target communication unmanned aerial vehicle based on secondary clustering |
CN111240355B (en) * | 2020-01-10 | 2022-04-12 | 哈尔滨工业大学 | Cruise formation planning system of multi-target communication unmanned aerial vehicle based on secondary clustering |
CN111474953A (en) * | 2020-03-30 | 2020-07-31 | 清华大学 | Multi-dynamic-view-angle-coordinated aerial target identification method and system |
CN111709308B (en) * | 2020-05-25 | 2021-12-28 | 未来航空科技(江门)有限公司 | Unmanned aerial vehicle-based maritime distress personnel detection and tracking method and system |
CN111709308A (en) * | 2020-05-25 | 2020-09-25 | 五邑大学 | Unmanned aerial vehicle-based maritime distress personnel detection and tracking method and system |
CN111984034A (en) * | 2020-08-24 | 2020-11-24 | 广东电网有限责任公司 | Unmanned aerial vehicle electric power patrols line system |
CN113297910A (en) * | 2021-04-25 | 2021-08-24 | 云南电网有限责任公司信息中心 | Distribution network field operation safety belt identification method |
CN113297910B (en) * | 2021-04-25 | 2023-04-18 | 云南电网有限责任公司信息中心 | Distribution network field operation safety belt identification method |
CN113312991A (en) * | 2021-05-14 | 2021-08-27 | 华能阜新风力发电有限责任公司 | Front-end intelligent recognition system based on unmanned aerial vehicle |
CN113205063A (en) * | 2021-05-19 | 2021-08-03 | 云南电网有限责任公司电力科学研究院 | Visual identification and positioning method for defects of power transmission conductor |
CN113534832A (en) * | 2021-08-03 | 2021-10-22 | 国网江苏省电力有限公司泰州供电分公司 | Unmanned aerial vehicle inspection tracking distribution network line flying method based on edge calculation |
CN113534832B (en) * | 2021-08-03 | 2024-03-15 | 国网江苏省电力有限公司泰州供电分公司 | Unmanned aerial vehicle routing inspection tracking distribution network line flight method based on edge calculation |
CN113657551A (en) * | 2021-09-01 | 2021-11-16 | 陕西工业职业技术学院 | Robot grabbing posture task planning method for sorting and stacking multiple targets |
CN113657551B (en) * | 2021-09-01 | 2023-10-20 | 陕西工业职业技术学院 | Robot grabbing gesture task planning method for sorting and stacking multiple targets |
CN114782805A (en) * | 2022-03-29 | 2022-07-22 | 中国电子科技集团公司第五十四研究所 | Unmanned aerial vehicle patrol-oriented man-in-loop hybrid enhanced target identification method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109961460A (en) | A kind of multiple target method for inspecting based on improvement YOLOv3 model | |
CN110363140B (en) | Human body action real-time identification method based on infrared image | |
CN107862705B (en) | Unmanned aerial vehicle small target detection method based on motion characteristics and deep learning characteristics | |
Wu et al. | Hybrid deep learning architecture for rail surface segmentation and surface defect detection | |
Carletti et al. | An intelligent flying system for automatic detection of faults in photovoltaic plants | |
CN112380952A (en) | Power equipment infrared image real-time detection and identification method based on artificial intelligence | |
CN111368690B (en) | Deep learning-based video image ship detection method and system under influence of sea waves | |
CN102799903B (en) | High-spectrum automatic cloud detection method based on space and spectral information | |
CN103235830A (en) | Unmanned aerial vehicle (UAV)-based electric power line patrol method and device and UAV | |
CN109657581A (en) | Urban track traffic gate passing control method based on binocular camera behavioral value | |
CN109086803B (en) | Deep learning and personalized factor-based haze visibility detection system and method | |
CN113111727A (en) | Method for detecting rotating target in remote sensing scene based on feature alignment | |
CN103353988A (en) | Method for evaluating performance of heterogeneous SAR (synthetic aperture radar) image feature matching algorithm | |
CN114241511B (en) | Weak supervision pedestrian detection method, system, medium, equipment and processing terminal | |
CN109492700A (en) | A kind of Target under Complicated Background recognition methods based on multidimensional information fusion | |
CN114627447A (en) | Road vehicle tracking method and system based on attention mechanism and multi-target tracking | |
CN116704273A (en) | Self-adaptive infrared and visible light dual-mode fusion detection method | |
Zhang et al. | Surface defect detection of wind turbine based on lightweight YOLOv5s model | |
Wang et al. | Intrusion detection for high-speed railways based on unsupervised anomaly detection models | |
CN113962973A (en) | Power transmission line unmanned aerial vehicle intelligent inspection system and method based on satellite technology | |
CN112261719B (en) | Area positioning method combining SLAM technology with deep learning | |
CN114218999A (en) | Millimeter wave radar target detection method and system based on fusion image characteristics | |
Scharf et al. | A semi-automated multi-sensor data labeling process for deep learning in automotive radar environment | |
CN111898427A (en) | Multispectral pedestrian detection method based on feature fusion deep neural network | |
xi Wang et al. | Detection of MMW radar target based on doppler characteristics and deep learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190702 |
|
RJ01 | Rejection of invention patent application after publication |