CN109785361A - Substation's foreign body intrusion detection system based on CNN and MOG - Google Patents
Substation's foreign body intrusion detection system based on CNN and MOG Download PDFInfo
- Publication number
- CN109785361A CN109785361A CN201811575608.1A CN201811575608A CN109785361A CN 109785361 A CN109785361 A CN 109785361A CN 201811575608 A CN201811575608 A CN 201811575608A CN 109785361 A CN109785361 A CN 109785361A
- Authority
- CN
- China
- Prior art keywords
- cnn
- foreign matter
- substation
- mog
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 74
- 238000012549 training Methods 0.000 claims abstract description 45
- 230000002452 interceptive effect Effects 0.000 claims abstract description 14
- 238000000034 method Methods 0.000 claims abstract description 13
- 238000000605 extraction Methods 0.000 claims abstract description 10
- 238000002360 preparation method Methods 0.000 claims abstract description 5
- 238000012544 monitoring process Methods 0.000 claims description 25
- 230000009545 invasion Effects 0.000 claims description 14
- 238000004422 calculation algorithm Methods 0.000 claims description 13
- 230000008859 change Effects 0.000 claims description 10
- 238000012986 modification Methods 0.000 claims description 8
- 230000004048 modification Effects 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 8
- 239000000203 mixture Substances 0.000 claims description 7
- 238000010276 construction Methods 0.000 claims description 4
- 239000000284 extract Substances 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 230000003068 static effect Effects 0.000 claims description 3
- 230000000007 visual effect Effects 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 2
- 238000007689 inspection Methods 0.000 claims 1
- 230000008569 process Effects 0.000 abstract description 6
- 239000000126 substance Substances 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 5
- 230000003993 interaction Effects 0.000 description 4
- 238000012790 confirmation Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000033001 locomotion Effects 0.000 description 1
- 230000007257 malfunction Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
Substation's foreign body intrusion detection system that the invention discloses a kind of based on CNN in conjunction with MOG, comprising: data preparation module is that supervisory control of substation video data is read out and is pre-processed;MOG dynamic foreground extracting module is that background modeling and dynamic foreground extraction are carried out to video data, and the target moved in video is oriented in detection;CNN foreign matter discrimination module is the training for the foreign matter image completion CNN model collected substation's background image and may invade substation, and the MOG target positioned is inputted the model, judges whether moving target belongs to foreign matter target;Man-machine interactively module is to export foreign substance information by interactive interface, realizes artificial mark and utilizes markup information re -training training CNN, obtains new foreign matter and distinguish model.The present invention is able to achieve the automatic detection of foreign body intrusion target in substation's detection video, which has stronger robustness and higher accuracy, and the flexibility of foreign bodies detection is improved by the makeover process of man-machine interactively interface.
Description
Technical field
The present invention relates to substation's foreign bodies detection field, specifically a kind of substation's foreign matter based on CNN in conjunction with MOG
Intruding detection system.
Background technique
The surrounding of substation is generally spacious flat-bottomed land, and capital equipment is placed in outdoor, therefore often receives
To the invasion of alien material, this by substation management and safe operation find out harm.In order to realize substation's main electric power
The intelligent monitoring of equipment, ensures power system security reliability service, and Utilities Electric Co. strengthens the prison to substation operation scene
Survey ways and means.Currently, the panorama monitoring information of substation mainly still takes the mode of manual analysis to handle, this mode
It will lead to the waste of a large amount of human resources, can also cause to malfunction due to lacking objectivity.Therefore, automatic using image recognition technology
The substation's foreign matter for detecting invasion is very necessary, and this mode can not only reduce labor workload, also can avoid artificial detection and leads
The erroneous judgement of cause achievees the purpose that the accuracy for improving detection.
Traditional monitoring system generally uses background modeling method, passes through each pixel in test video data next frame
Whether meet background model, completes the detection to dynamic object and position.But this detection method is to the adaptability of dynamic background
Difference will lead to algorithm for background as before dynamic when the object in background is shaken by a small margin due to light or wind appearance
The case where scape detected.On the other hand, it will appear the replacement or movement of equipment in substation, conventional foreground extraction algorithm does not have
It is standby that ability is adjusted flexibly, it also cannot achieve the function with manually mark interactive learning.Last traditional background modeling method is assumed
The pixel of background obeys same distribution, but when moving target is more in monitoring background complexity or background, background pixel
The form of dynamic change can be presented in the distribution of point, cannot be completely represented using the parameter estimated merely.
Summary of the invention
It is an object of the invention to be directed to the misjudgment phenomenon of existing foreign body intrusion detection system seriously and can not flexibly interact
The problem of change.
Technical solution of the present invention provides a kind of substation's foreign body intrusion detection system based on CNN in conjunction with MOG, packet
It includes:
Data preparation module will be regarded for reading the video data of supervisory control of substation camera storage using wavelet transformation
Frequency evidence is transformed in frequency domain by frame, and removes the anomaly peak in frequency domain, the denoising to video data is completed, to video counts
According to carry out the processing of size scaling, make image data reach be unified for 600 × 800 size;
MOG dynamic foreground extracting module, it is fixed for carrying out moving foreground object detection to monitoring video using MOG algorithm
Position is that monitoring static background models using gauss hybrid models (Gassian Mixture Model, GMM), utilizes several backgrounds
Image data estimates the parameters of GMM, frame image each in pretreated video data is inputted GMM model, according to model
The probability value of output determines the foreground target moved in background and positions extraction;
CNN foreign matter discrimination module, including model training unit and target-recognition unit;Wherein, model training unit is used for
Image block in interception detection electrical substation monitoring background, composition model training positive sample collection, is collected possibly into the different of substation
Object image forms training negative sample collection, and the training of CNN judgment models is completed using positive negative sample;Target-recognition unit is for utilizing
The dynamic object for the CNN judgment models identification MOG detection positioning that training is completed, judges whether moving target belongs to invasion foreign matter mesh
Mark;
Man-machine interactively module is visualized in image, is led to for receiving the foreign matter target information of CNN judgement output
It crosses and manually output location information is modified and is marked again, using artificial markup information re -training CNN judgment models, obtain
Model is distinguished to new foreign matter.
Wherein, in the CNN foreign matter discrimination module, model training unit is when construction invades foreign matter negative sample, to collection
To all negative samples need to carry out size scaling processing;During scaling, every negative sample will according to its script dimension scale
Length and width scaling is within 400;And for the unification of training sample, when positive sample is chosen in sliding from monitoring image, what is utilized is
The square block window that 10 parameters of equal proportion are constituted between 40 to 400.
Wherein, CNN model at last convolutional layer using spatial pyramid pond (Spatial Pyramid Pooling,
SPP) method carries out unification to input data;After input picture extracts feature by convolutional layer, by it according to 3 scale equal parts
It is 1 piece, 4 pieces and 16 pieces, maximum pond is carried out to image block therein respectively, the image for one-dimensional 21 elements unified is special
Sign, to carry out classification based training.
Wherein, if CNN model judgement detection target is really invasion foreign matter, its window location information is output to interaction and is connect
In mouthful, interactive window visualizes location information into monitoring image, exports interactive interface for the positioning of Visual retrieval foreign matter
Window and its classification for belonging to foreign matter and confidence level.
Wherein, if detection target and non-intrusive foreign matter, change the type of output target window by way of manually marking
Modification information is returned to CNN foreign matter discrimination module by interactive interface, utilizes artificial markup information weight by label, man-machine interactively module
New training CNN judgment models.
Wherein, the case where change content manually marked does not detect including at least foreign bodies detection mistake and foreign matter.
Wherein, for the situation of foreign bodies detection mistake, the content of artificial markup information is that target labels modify part, will be former
This foreign matter tag modification is background;In CNN model re -training, it is finely adjusted using the CNN parameter of script as initialization value;
And finely tuning the data used is the wrong background sample of detection by the scaling of the overturning of image, angle offset and scale, is extended for
30 to 50 small samples libraries recycle this small samples library to finely tune CNN parameter again.
Wherein, the invasion foreign matter sample manually marked is then passed through data extending principle by the case where not detecting for foreign matter
Expansion obtains a large amount of negative samples, still uses original CNN model as initialisation structures trim network, completes the re -training of network.
It is different from the prior art, substation's foreign body intrusion detection system of the invention carries out supervisory control of substation video data
It reads and then pretreatment is judged again using MOG algorithm to monitoring video progress dynamic object pre-detection by CNN model, into
Whether one step confirmation detection target is invasion foreign matter, finally completes artificial calibration using man-machine interactively formula interface, and utilize interaction
Information update CNN judgment models parameter.By means of the invention it is possible to the misjudgment phenomenon of foreign body intrusion detection system is effectively reduced, and
And complicated for background or dynamic background electrical substation monitoring data have stronger robustness and flexibility.
Detailed description of the invention
Fig. 1 is that a kind of principle of substation's foreign body intrusion detection system based on CNN in conjunction with MOG provided by the invention is shown
It is intended to;
Fig. 2 is that a kind of structure of substation's foreign body intrusion detection system based on CNN in conjunction with MOG provided by the invention is shown
It is intended to;
Fig. 3 is that a kind of logic of substation's foreign body intrusion detection system based on CNN in conjunction with MOG provided by the invention is shown
It is intended to;
Fig. 4 is in a kind of substation's foreign body intrusion detection system based on CNN in conjunction with MOG provided by the invention before MOG
Scape extraction process schematic diagram;
Fig. 5 is SPP behaviour in a kind of substation's foreign body intrusion detection system based on CNN in conjunction with MOG provided by the invention
Make the schematic diagram of process;
Fig. 6 is manually to hand in a kind of substation's foreign body intrusion detection system based on CNN in conjunction with MOG provided by the invention
The structural schematic diagram of mutual module.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with attached drawing and embodiment,
Embodiments of the present invention are further elaborated:
Refering to fig. 1 and Fig. 2, Fig. 1 and Fig. 2 are that a kind of substation's foreign matter based on CNN in conjunction with MOG provided by the invention enters
The schematic illustration and structural schematic diagram of detection system are invaded, which includes:
Data preparation module 110 is that supervisory control of substation video data is read out and is pre-processed;MOG dynamic foreground extraction
Module 120 is that background modeling and dynamic foreground extraction are carried out to video data, and the target moved in video is oriented in detection;CNN
Foreign matter discrimination module 130 is the instruction for the foreign matter image completion CNN model collected substation's background image and may invade substation
Practice, the MOG target positioned is inputted into the model, judges whether moving target belongs to foreign matter target;Man-machine interactively module 140 is logical
Cross interactive interface output foreign substance information, realize it is artificial mark simultaneously using markup information re -training training CNN, obtain new different
Object distinguishes model.
Specifically, data preparation module 110, for reading the video data of supervisory control of substation camera storage, utilization is small
Wave conversion is transformed to video data in frequency domain by frame, and removes the anomaly peak in frequency domain, and video data is removed in completion
Make an uproar, to video data carry out the processing of size scaling, make image data reach be unified for 600 × 800 size.
MOG dynamic foreground extracting module 120, for carrying out moving foreground object detection to monitoring video using MOG algorithm
Positioning is that monitoring static background models using gauss hybrid models (Gassian Mixture Model, GMM), utilizes several back
Scape image data estimates the parameters of GMM, frame image each in pretreated video data is inputted GMM model, according to mould
The probability value of type output determines the foreground target moved in background and positions extraction.
CNN foreign matter discrimination module 130, including model training unit 131 and target-recognition unit 132;Wherein, model training
Unit 131 be used for intercept detection electrical substation monitoring background in image block, composition model training positive sample collection, collection possibly into
The foreign matter image of substation forms training negative sample collection, and the training of CNN judgment models is completed using positive negative sample;Target-recognition list
Whether the dynamic object of CNN judgment models identification MOG detection positioning of the member 132 for being completed using training, judge moving target
Belong to invasion foreign matter target;
Man-machine interactively module 140, the foreign matter target information exported for receiving CNN judgement, is visualized in image,
By being manually modified to output location information and again marking, using artificial markup information re -training CNN judgment models,
It obtains new foreign matter and distinguishes model.The structural schematic diagram of man-machine interactively module 140 is as shown in Figure 6.
Preferably, in CNN foreign matter discrimination module 130, model training unit 131 is right when construction invades foreign matter negative sample
All negative samples collected need to carry out size scaling processing;During scaling, every negative sample is according to its script size ratio
Example will be within length and width scaling to 400;And for the unification of training sample, when positive sample is chosen in sliding from monitoring image, utilize
Be between 40 to 400 10 parameters of equal proportion constitute square block window.
Preferably, CNN model uses spatial pyramid pond (Spatial Pyramid at last convolutional layer
Pooling, SPP) method to input data carry out unification;After input picture extracts feature by convolutional layer, by it according to 3
Scale is divided into 1 piece, 4 pieces and 16 pieces, carries out maximum pond, one-dimensional 21 elements unified to image block therein respectively
Characteristics of image, to carry out classification based training.
Preferably, if CNN model judgement detection target is really invasion foreign matter, its window location information is output to interaction
In interface, interactive window visualizes location information into monitoring image, and output interactive interface determines Visual retrieval foreign matter
Position window and its classification for belonging to foreign matter and confidence level.
Preferably, if detection target and non-intrusive foreign matter, change the kind of output target window by way of manually marking
Modification information is returned to CNN foreign matter discrimination module by interactive interface, utilizes artificial markup information by class label, man-machine interactively module
Re -training CNN judgment models.
Preferably, the case where change content manually marked does not detect including at least foreign bodies detection mistake and foreign matter.
Preferably, for the situation of foreign bodies detection mistake, the content of artificial markup information is that target labels modify part, will
Script foreign matter tag modification is background;In CNN model re -training, carried out using the CNN parameter of script as initialization value micro-
It adjusts;And finely tuning the data used is scaling of the wrong background sample of detection by the overturning of image, angle offset and scale, is expanded
For 30 to 50 small samples libraries, this small samples library is recycled to finely tune CNN parameter again.
Preferably, the invasion foreign matter sample manually marked is then passed through data extending original by the case where not detecting for foreign matter
Reason, which expands, obtains a large amount of negative samples, still uses original CNN model as initialisation structures trim network, completes the instruction again of network
Practice.
Substation's foreign body intrusion detection system treatment process based on CNN and MOG of the invention is shown in Fig. 2, read first to
The video data of the supervisory control of substation camera storage of detection is gone using wavelet transform process video data by the way that threshold value is arranged
Except the anomaly peak in frequency domain completes the denoising to video data.It is pretreated during also need to image set carry out according to
Ratio carry out scaling processing, make image data reach be unified for 600 × 800 size.
Pretreated video data carries out the detection positioning of dynamic foreground target, entire detection algorithm behaviour by MOG algorithm
Fig. 3 is seen as process.MOG algorithm can be that monitoring video background pixel is modeled using GMM model first, in this process
Assuming that background pixel point obeys a stationary distribution in video, variation of the fixed model in time-domain is mostly height with K
This distribution is to simulate, if the sampled value of a certain pixel P (x, y) is { X1,X2,…,Xt, then the current picture observed in t moment
Plain value xtProbability are as follows:
Wherein ωi,t、μi,tAnd ∑i,tThe weight of i-th of Gaussian Profile, mean value and covariance square respectively in t moment model
Battle array.The probability density function of i-th of Gaussian Profile is
K Gaussian Profile is sorted according to the size of ω/σ, the B background GMM model as video before taking utilizes multiframe
Consecutive image data estimate the parameter in GMM model by EM algorithm.It determines in GMM model after parameters, it will be to be checked
Each pixel substitutes into the GMM model in the video data of survey, and detection positioning does not meet the pixel of the probabilistic model wherein simultaneously
It is demarcated, the dynamic object in the calibrating and positioning information, that is, video.MOG foreground extraction process is as shown in Figure 4.
Since MOG algorithm is more sensitive for dynamic background, obtain MOG detection foreign matter target after, need by
Foreign matter target inputs CNN discrimination model, carries out further judgement to reduce misjudgment phenomenon.And CNN judgment models firstly the need of
It is trained, the database of training part is divided into both positive and negative sample type, and positive sample is substation's background image block, negative sample
The foreign matter image block invaded for the possibility of substation.Since the size of intrusion target is any, construction invasion foreign matter is negative
When sample, need to carry out size scaling processing to all negative samples collected.During scaling, by every negative sample according to it
Script dimension scale is by within its length and width scaling to 400, and each sample carries out down-sampling and obtains the figure of the small scale of its at least four
Picture.And for the unification of training sample, when positive sample is chosen in sliding from monitoring image, what is utilized is that ratios are waited between 40 to 400
10 parameters of example are the square block window that side length is constituted.And in window sliding, the sliding step of selection is unified for sample-size
The half of middle side length.
Due to having very big difference between the dynamic object size of MOG algorithm Detection and Extraction, used in this system
CNN model the last one convolutional layer will connect the pond a SPP layer, to complete unified to the size of convolution feature.Wherein
The operation of SPP pondization is as shown in figure 5, mainly by convolution feature according to being divided into 1 piece, 4 pieces and 16 pieces, then respectively to therein
Image block carries out maximum pond and is finally combined into these characteristic values together in order to obtain 1,4 and 16 pond characteristic value
In one vector, the characteristics of image for one-dimensional 21 elements unified.This pond SPP feature can not only be completely unified not
With the scale of size input picture, the multiple dimensioned property of feature is also imparted, is allowed to still have when angle, vision distance change
Stronger robustness.
Can be by the final information input man-machine interactively interface of judgement after CNN judgement, final information will be with the mesh of recognition detection
It marks window and corresponding foreign matter differentiates label and its confidence level composition, therefore the entire form for differentiating output result is similar:Then it is visualized in interactive interface window, which can realize to output location information
Visualization, change and mark again, user can change according to the actual situation detection foreign matter window position or mark again
Foreign matter window and corresponding label.The type label that output target window can be changed if detection target and non-intrusive foreign matter, by it
Label is changed to background, then re-enters CNN discrimination module;If output interface and mesh is not detected there are foreign matter in image
Markup information then can manually re-scale position and the label information of foreign matter, then is returned to CNN foreign matter discrimination module by mark,
Utilize artificial markup information re -training CNN judgment models.These man-machine interactivelies mark after sample by by the overturning of image,
The scaling of angle offset and scale is extended for 30 to 50 small samples libraries, finally using the CNN parameter of script as initialization
Value, finely tunes CNN parameter by this small samples library again.
Substation's foreign body intrusion detection system based on CNN and MOG of the invention can be by depth CNN model to originally
The dynamic foreground target of MOG detection further judges, reduces the misjudgment phenomenon due to caused by other reasons.Simultaneously in CNN model
In joined SPP method, increase system to the discrimination of multiscale target.Man-machine interactively formula interface in last this system can
To guarantee the flexibility of the foreign body intrusion detection system, when the target object in substation's background is changed, this system is still
It can operate normally.
It is different from the prior art, substation's foreign body intrusion detection system of the invention carries out supervisory control of substation video data
It reads and then pretreatment is judged again using MOG algorithm to monitoring video progress dynamic object pre-detection by CNN model, into
Whether one step confirmation detection target is invasion foreign matter, finally completes artificial calibration using man-machine interactively formula interface, and utilize interaction
Information update CNN judgment models parameter.By means of the invention it is possible to the misjudgment phenomenon of foreign body intrusion detection system is effectively reduced, and
And complicated for background or dynamic background electrical substation monitoring data have stronger robustness and flexibility.
The above is only presently preferred embodiments of the present invention, not does limitation in any form, this field to the present invention
The technology contents of technical staff according to the above description make simple modification, equivalent variations or modification, all fall within protection of the invention
In range.
Claims (8)
1. a kind of substation's foreign body intrusion detection system based on CNN in conjunction with MOG characterized by comprising
Data preparation module, for reading the video data of supervisory control of substation camera storage, using wavelet transformation by video counts
Transformed in frequency domain according to by frame, and remove the anomaly peak in frequency domain, complete denoising to video data, to video data into
Row size scaling processing, make image data reach be unified for 600 × 800 size;
MOG dynamic foreground extracting module, for carrying out moving foreground object detection positioning, benefit to monitoring video using MOG algorithm
It is monitoring static background modeling with gauss hybrid models (Gassian Mixture Model, GMM), utilizes several background images
Data estimate the parameters of GMM, and frame image each in pretreated video data is inputted GMM model, is exported according to model
Probability value determine the foreground target moved in background and position extraction;
CNN foreign matter discrimination module, including model training unit and target-recognition unit;Wherein, model training unit is for intercepting
The image block in electrical substation monitoring background is detected, composition model training positive sample collection collects the foreign matter figure possibly into substation
As forming training negative sample collection, the training of CNN judgment models is completed using positive negative sample;Target-recognition unit is used to utilize training
The dynamic object of the CNN judgment models identification MOG detection positioning of completion, judges whether moving target belongs to invasion foreign matter target;
Man-machine interactively module is visualized in image for receiving the foreign matter target information of CNN judgement output, passes through people
Work is modified output location information and marks again, using artificial markup information re -training CNN judgment models, obtains new
Foreign matter distinguish model.
2. substation's foreign body intrusion detection system according to claim 1 based on CNN and MOG, it is characterised in that: described
In CNN foreign matter discrimination module, model training unit is when construction invades foreign matter negative sample, to all negative sample needs collected
Carry out the processing of size scaling;During scaling, every negative sample will be within length and width scaling to 400 according to its script dimension scale;
And for the unification of training sample, when positive sample is chosen in sliding from monitoring image, what is utilized is equal proportion 10 between 40 to 400
The square block window that a parameter is constituted.
3. substation's foreign body intrusion detection system according to claim 1 based on CNN and MOG, it is characterised in that: CNN
Model is at last convolutional layer using spatial pyramid pond (Spatial Pyramid Pooling, SPP) method to input number
According to carry out unification;After input picture extracts feature by convolutional layer, it is divided into 1 piece, 4 pieces and 16 pieces according to 3 scales,
Maximum pond, the characteristics of image for one-dimensional 21 elements unified, to carry out classification instruction are carried out to image block therein respectively
Practice.
4. substation's foreign body intrusion detection system according to claim 1 based on CNN and MOG, it is characterised in that: if
CNN model judgement detection target is really invasion foreign matter, then its window location information is output in interactive interface, interactive window will
Location information is visualized into monitoring image, and the anchor window of Visual retrieval foreign matter is belonged to foreign matter with it by output interactive interface
Classification and confidence level.
5. substation's foreign body intrusion detection system according to claim 4 based on CNN and MOG, it is characterised in that: if inspection
Target and non-intrusive foreign matter are surveyed, changes the type label of output target window, man-machine interactively module by way of manually marking
Modification information is returned into CNN foreign matter discrimination module by interactive interface, judges mould using artificial markup information re -training CNN
Type.
6. substation's foreign body intrusion detection system according to claim 5 based on CNN and MOG, it is characterised in that: artificial
The case where change content of mark does not detect including at least foreign bodies detection mistake and foreign matter.
7. substation's foreign body intrusion detection system according to claim 6 based on CNN and MOG, it is characterised in that: be directed to
The situation of foreign bodies detection mistake, the content of artificial markup information are that target labels modify part, are by script foreign matter tag modification
Background;In CNN model re -training, it is finely adjusted using the CNN parameter of script as initialization value;And finely tune the data used
It is the wrong background sample of detection by the scaling of the overturning of image, angle offset and scale, is extended for 30 to 50 small samples
Library recycles this small samples library to finely tune CNN parameter again.
8. substation's foreign body intrusion detection system according to claim 6 based on CNN and MOG, it is characterised in that: be directed to
The case where foreign matter does not detect then expands the invasion foreign matter sample manually marked by data extending principle to obtain a large amount of negative samples
This, still uses original CNN model as initialisation structures trim network, completes the re -training of network.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811575608.1A CN109785361A (en) | 2018-12-22 | 2018-12-22 | Substation's foreign body intrusion detection system based on CNN and MOG |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811575608.1A CN109785361A (en) | 2018-12-22 | 2018-12-22 | Substation's foreign body intrusion detection system based on CNN and MOG |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109785361A true CN109785361A (en) | 2019-05-21 |
Family
ID=66498090
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811575608.1A Pending CN109785361A (en) | 2018-12-22 | 2018-12-22 | Substation's foreign body intrusion detection system based on CNN and MOG |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109785361A (en) |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110378903A (en) * | 2019-09-16 | 2019-10-25 | 广东电网有限责任公司佛山供电局 | A kind of transmission line of electricity anti-accident measures Intelligent statistical method |
CN110691224A (en) * | 2019-10-31 | 2020-01-14 | 上海电力大学 | Transformer substation perimeter video intelligent detection system |
CN110705360A (en) * | 2019-09-05 | 2020-01-17 | 上海零眸智能科技有限公司 | Method for efficiently processing classified data by human-computer combination |
CN111028452A (en) * | 2019-12-20 | 2020-04-17 | 云南电网有限责任公司保山供电局 | Transformer substation foreign matter intrusion monitoring system and method |
CN111161740A (en) * | 2019-12-31 | 2020-05-15 | 中国建设银行股份有限公司 | Intention recognition model training method, intention recognition method and related device |
CN111310635A (en) * | 2020-02-10 | 2020-06-19 | 上海应用技术大学 | Security inspection contraband identification system and method based on TensorFlow |
CN111325708A (en) * | 2019-11-22 | 2020-06-23 | 济南信通达电气科技有限公司 | Power transmission line detection method and server |
CN111340843A (en) * | 2020-02-19 | 2020-06-26 | 山东大学 | Power scene video detection method based on environment self-adaption and small sample learning |
AU2020201897A1 (en) * | 2019-05-31 | 2020-12-17 | Raytheon Company | Labeling using interactive assisted segmentation |
CN112102443A (en) * | 2020-09-15 | 2020-12-18 | 国网电力科学研究院武汉南瑞有限责任公司 | Marking system and marking method suitable for substation equipment inspection image |
CN113191339A (en) * | 2021-06-30 | 2021-07-30 | 南京派光智慧感知信息技术有限公司 | Track foreign matter intrusion monitoring method and system based on video analysis |
CN113822240A (en) * | 2021-11-22 | 2021-12-21 | 广东电网有限责任公司中山供电局 | Method and device for extracting abnormal behaviors from power field operation video data |
CN113837063A (en) * | 2021-10-15 | 2021-12-24 | 中国石油大学(华东) | Curling motion field analysis and decision-making assisting method based on reinforcement learning |
CN114157829A (en) * | 2020-09-08 | 2022-03-08 | 顺丰科技有限公司 | Model training optimization method and device, computer equipment and storage medium |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102955940A (en) * | 2012-11-28 | 2013-03-06 | 山东电力集团公司济宁供电公司 | System and method for detecting power transmission line object |
CN104318242A (en) * | 2014-10-08 | 2015-01-28 | 中国人民解放军空军工程大学 | High-efficiency SVM active half-supervision learning algorithm |
CN104463191A (en) * | 2014-10-30 | 2015-03-25 | 华南理工大学 | Robot visual processing method based on attention mechanism |
CN104463904A (en) * | 2014-09-22 | 2015-03-25 | 国家电网公司 | High-voltage line foreign matter invasion target detection method |
CN104574340A (en) * | 2013-10-18 | 2015-04-29 | 上海睿穹信息科技有限公司 | Video intrusion detection method based on historical images |
CN105744232A (en) * | 2016-03-25 | 2016-07-06 | 南京第五十五所技术开发有限公司 | Method for preventing power transmission line from being externally broken through video based on behaviour analysis technology |
CN106355162A (en) * | 2016-09-23 | 2017-01-25 | 江西洪都航空工业集团有限责任公司 | Method for detecting intrusion on basis of video monitoring |
CN106446926A (en) * | 2016-07-12 | 2017-02-22 | 重庆大学 | Transformer station worker helmet wear detection method based on video analysis |
CN106778472A (en) * | 2016-11-17 | 2017-05-31 | 成都通甲优博科技有限责任公司 | The common invader object detection and recognition method in transmission of electricity corridor based on deep learning |
-
2018
- 2018-12-22 CN CN201811575608.1A patent/CN109785361A/en active Pending
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102955940A (en) * | 2012-11-28 | 2013-03-06 | 山东电力集团公司济宁供电公司 | System and method for detecting power transmission line object |
CN104574340A (en) * | 2013-10-18 | 2015-04-29 | 上海睿穹信息科技有限公司 | Video intrusion detection method based on historical images |
CN104463904A (en) * | 2014-09-22 | 2015-03-25 | 国家电网公司 | High-voltage line foreign matter invasion target detection method |
CN104318242A (en) * | 2014-10-08 | 2015-01-28 | 中国人民解放军空军工程大学 | High-efficiency SVM active half-supervision learning algorithm |
CN104463191A (en) * | 2014-10-30 | 2015-03-25 | 华南理工大学 | Robot visual processing method based on attention mechanism |
CN105744232A (en) * | 2016-03-25 | 2016-07-06 | 南京第五十五所技术开发有限公司 | Method for preventing power transmission line from being externally broken through video based on behaviour analysis technology |
CN106446926A (en) * | 2016-07-12 | 2017-02-22 | 重庆大学 | Transformer station worker helmet wear detection method based on video analysis |
CN106355162A (en) * | 2016-09-23 | 2017-01-25 | 江西洪都航空工业集团有限责任公司 | Method for detecting intrusion on basis of video monitoring |
CN106778472A (en) * | 2016-11-17 | 2017-05-31 | 成都通甲优博科技有限责任公司 | The common invader object detection and recognition method in transmission of electricity corridor based on deep learning |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11170264B2 (en) | 2019-05-31 | 2021-11-09 | Raytheon Company | Labeling using interactive assisted segmentation |
AU2020201897A1 (en) * | 2019-05-31 | 2020-12-17 | Raytheon Company | Labeling using interactive assisted segmentation |
AU2020201897B2 (en) * | 2019-05-31 | 2021-09-09 | Raytheon Company | Labeling using interactive assisted segmentation |
CN110705360A (en) * | 2019-09-05 | 2020-01-17 | 上海零眸智能科技有限公司 | Method for efficiently processing classified data by human-computer combination |
CN110378903A (en) * | 2019-09-16 | 2019-10-25 | 广东电网有限责任公司佛山供电局 | A kind of transmission line of electricity anti-accident measures Intelligent statistical method |
CN110691224A (en) * | 2019-10-31 | 2020-01-14 | 上海电力大学 | Transformer substation perimeter video intelligent detection system |
CN111325708A (en) * | 2019-11-22 | 2020-06-23 | 济南信通达电气科技有限公司 | Power transmission line detection method and server |
CN111028452A (en) * | 2019-12-20 | 2020-04-17 | 云南电网有限责任公司保山供电局 | Transformer substation foreign matter intrusion monitoring system and method |
CN111161740A (en) * | 2019-12-31 | 2020-05-15 | 中国建设银行股份有限公司 | Intention recognition model training method, intention recognition method and related device |
CN111310635A (en) * | 2020-02-10 | 2020-06-19 | 上海应用技术大学 | Security inspection contraband identification system and method based on TensorFlow |
CN111310635B (en) * | 2020-02-10 | 2024-04-19 | 上海应用技术大学 | TensorFlow-based security inspection contraband identification system and TensorFlow-based security inspection contraband identification method |
CN111340843A (en) * | 2020-02-19 | 2020-06-26 | 山东大学 | Power scene video detection method based on environment self-adaption and small sample learning |
CN114157829A (en) * | 2020-09-08 | 2022-03-08 | 顺丰科技有限公司 | Model training optimization method and device, computer equipment and storage medium |
CN112102443A (en) * | 2020-09-15 | 2020-12-18 | 国网电力科学研究院武汉南瑞有限责任公司 | Marking system and marking method suitable for substation equipment inspection image |
CN113191339A (en) * | 2021-06-30 | 2021-07-30 | 南京派光智慧感知信息技术有限公司 | Track foreign matter intrusion monitoring method and system based on video analysis |
CN113837063B (en) * | 2021-10-15 | 2024-05-10 | 中国石油大学(华东) | Reinforcement learning-based curling motion field analysis and auxiliary decision-making method |
CN113837063A (en) * | 2021-10-15 | 2021-12-24 | 中国石油大学(华东) | Curling motion field analysis and decision-making assisting method based on reinforcement learning |
CN113822240A (en) * | 2021-11-22 | 2021-12-21 | 广东电网有限责任公司中山供电局 | Method and device for extracting abnormal behaviors from power field operation video data |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109785361A (en) | Substation's foreign body intrusion detection system based on CNN and MOG | |
CN106897670B (en) | Express violence sorting identification method based on computer vision | |
CN112380952B (en) | Power equipment infrared image real-time detection and identification method based on artificial intelligence | |
CN110310264A (en) | A kind of large scale object detection method, device based on DCNN | |
CN106778757B (en) | Scene text detection method based on text conspicuousness | |
Yu et al. | A landslide intelligent detection method based on CNN and RSG_R | |
CN106127204B (en) | A kind of multi-direction meter reading Region detection algorithms of full convolutional neural networks | |
CN110472597A (en) | Rock image rate of decay detection method and system based on deep learning | |
CN106408015A (en) | Road fork identification and depth estimation method based on convolutional neural network | |
CN105590099B (en) | A kind of more people's Activity recognition methods based on improvement convolutional neural networks | |
CN109657581A (en) | Urban track traffic gate passing control method based on binocular camera behavioral value | |
CN112560675B (en) | Bird visual target detection method combining YOLO and rotation-fusion strategy | |
CN108960124B (en) | Image processing method and device for pedestrian re-identification | |
CN104268528A (en) | Method and device for detecting crowd gathered region | |
CN110008877B (en) | Substation disconnecting switch detection and identification method based on Faster RCNN | |
CN110287798A (en) | Vector network pedestrian detection method based on characteristic module and context fusion | |
CN111914634A (en) | Complex-scene-interference-resistant automatic manhole cover type detection method and system | |
CN111145222A (en) | Fire detection method combining smoke movement trend and textural features | |
CN112749738B (en) | Zero sample object detection method for performing superclass reasoning by fusing context | |
Liang et al. | Methods of moving target detection and behavior recognition in intelligent vision monitoring. | |
CN111814696A (en) | Video ship target detection method based on improved YOLOv3 | |
CN113989683A (en) | Ship detection method for synthesizing synchronous orbit sequence optical image space-time information | |
CN115661932A (en) | Fishing behavior detection method | |
CN105631410B (en) | A kind of classroom detection method based on intelligent video processing technique | |
CN114821647A (en) | Sleeping post identification method, device, equipment and medium |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190521 |