CN108009525B - A kind of specific objective recognition methods over the ground of the unmanned plane based on convolutional neural networks - Google Patents
A kind of specific objective recognition methods over the ground of the unmanned plane based on convolutional neural networks Download PDFInfo
- Publication number
- CN108009525B CN108009525B CN201711422364.9A CN201711422364A CN108009525B CN 108009525 B CN108009525 B CN 108009525B CN 201711422364 A CN201711422364 A CN 201711422364A CN 108009525 B CN108009525 B CN 108009525B
- Authority
- CN
- China
- Prior art keywords
- convolutional neural
- neural networks
- training
- unmanned plane
- networks model
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to a kind of unmanned plane based on convolutional neural networks, specific objective recognition methods is as follows over the ground：(1) the collected data set of specific objective over the ground of unmanned plane is labeled according to classification, and proportionally divides training set, verification collection, test set；(2) setting training convolutional neural networks model parameter, training parameter, start to train, according to training, obtain the optimal solution of convolutional neural networks model.(3) change convolutional neural networks model depth, restart (2) training, obtain optimal convolutional neural networks model；(4) test set is tested, obtains recognition accuracy；(5) the convolutional neural networks model parameter that accuracy rate is met the requirements is applied to unmanned plane over the ground in specific objective actual scene, unmanned plane the image collected target is identified.
Description
Technical field
The present invention relates to a kind of based on the unmanned plane of convolutional neural networks to specific objective recognition methods, belong to aviation boat with
Computer vision information processing crossing domain.
Background technology
Unmanned plane (UmannedAir Vehicle, UAV) is to control dress using radio robot and the program provided for oneself
The not manned aircraft of manipulation is set, or fully or is intermittently automatically operated by car-mounted computer.Unmanned plane is led according to application
Domain, can be divided into it is military with it is civilian.Military aspect, unmanned plane are divided into reconnaissance plane and target drone.Civilian aspect, unmanned plane+sector application,
It is unmanned plane really just to have needed；It taking photo by plane at present, agricultural, plant protection, miniature self-timer, express transportation, disaster relief, observing wild move
Object, monitoring infectious disease, mapping, news report, electric inspection process, the disaster relief, movies-making, etc. field application, greatly expand
The purposes of unmanned plane itself.
Traditional target identification, mainly first to expressing this figure with mathematical model after a certain amount of manual features of image zooming-out
Then picture is identified image by grader.For example, object identification uses scale invariant feature (Scale-
Invariant feature transform, SIFT), recognition of face uses Local textural feature (Local Binary
Patterns, LBP), pedestrian detection using histograms of oriented gradients feature (Histogram of Oriented Gradient,
HOG), traditional grader mainly has support vector machines (Support Vector Machine, SVM), K arest neighbors (k-
NearestNeighbor, KNN) etc..But the identification of the machine learning method of this kind of shallow-layer is low, generally can not meet unmanned plane
The engineering demand of target identification over the ground.
With the development of artificial intelligence, the continuous breakthrough of deep learning, in speech recognition, natural language processing, computer
The fields such as vision, video analysis, multimedia all achieve immense success.Convolutional neural networks target identification based on deep learning
Flow is mainly：Image is input in neural network, the propagated forward and reverse propagated error scheduling algorithm of deep learning are utilized
Minimize loss function, after updating weights, obtain a preferably identification model, then using this model to new image come
It is identified.Convolutional neural networks model can from image data automatic learning characteristic, and can be fast from new training data
Speed training arrives new character representation to study.General pattern recognition system all includes that feature extraction and grader two are important
Part, their optimization in traditional method is to be separated from each other, and under the frame of neural network, feature extraction and grader
It is that joint is feedback optimized, the performance of the two integration and cooperation can be played as far as possible.
But it is very strong to the dependence of training dataset based on the target identification method of convolutional neural networks, and unmanned plane is adopted
The image data amount size of the target over the ground collected is uncertain, differs and surely meets the training demand of convolutional neural networks model,
Or network depth not enough causes a degree of waste to the information of data set.Therefore the present invention combines unmanned plane to acquire image
The characteristics of with convolutional neural networks model the advantages of, according to data set adjust network model depth, obtain optimal network model knot
Structure, to propose a kind of convolutional neural networks target identification method that can adapt to unmanned plane and acquire image data set size.
Invention content
The technology of the present invention solves the problems, such as：Overcoming unmanned plane, conventional method recognition accuracy is inadequate in terms of target identification over the ground
Problem and the implacable problem of deep learning method training dataset, provide a kind of unmanned plane based on convolutional neural networks
Specific objective recognition methods over the ground, according to the collected target data set size over the ground of unmanned plane, to adjust the depth of network model
Degree, realizes higher recognition accuracy, to provide a kind of unmanned plane effective way that specific objective identifies over the ground.It can also answer
In the Pattern Recognition and Intelligent System design complicated for other.
The technology of the present invention solution：A kind of specific objective recognition methods over the ground of the unmanned plane based on convolutional neural networks,
Include the following steps：
(1) unmanned plane acquisition specific objective image data set over the ground is used, is labeled according to classification, and according to having marked
Data set be divided into training set, verification collection, test set, be finally processed into the data class that convolutional neural networks model can identify
Type；
(2) structure identification convolutional neural networks model, setting maximum iteration, learning rate, test frequency, selection are reversed
Transmission method starts to train according to above-mentioned be provided and selected, and then according to training loss function situation of change, obtains this convolution god
Recognition accuracy through network model；
(3) in the structure of convolutional neural networks model, increase or reduce the convolution number of plies, restart step (2)
Training illustrates the convolutional neural networks of this training when the rate of accuracy reached of the identification of training convolutional neural networks model is to highest
Model adapts to current data set size, retains training at this time and obtains the structure and parameter of convolutional neural networks model；
(4) the convolutional neural networks model for utilizing (3) to obtain, tests test set, obtains recognition accuracy, to knowing
Other accuracy rate is judged, if recognition accuracy disclosure satisfy that actual requirement of engineering, the convolutional neural networks model
It is applicable in the actual unmanned plane task that specific objective identifies over the ground, executes step (5)；If not satisfied, then illustrating to instruct
Actual requirement of engineering cannot be met by practicing collection, needed to expand training set, restarted step (1), (2), (3), until meeting practical
Until engineering；
(5) accuracy rate is met to the parameter of the convolutional neural networks model of Practical Project requirement, and to be applied to unmanned plane special over the ground
In the actual scene to set the goal, unmanned plane the image collected target is identified.
In the step (2), the method for obtaining the convolutional neural networks model of parametric optimal solution is：When the loss of training set
Function Loss falls be no more than 0.001 when, and verify collection loss function Loss tend to rise critical point when, that is, obtain
The convolutional neural networks model of parametric optimal solution.
In the step (2), setting maximum iteration, learning rate, test frequency select back-propagation method specifically such as
Under：
Maximum iteration：200000 times；
Initial learning rate：0.001；
Test frequency：1000 iteration/1 time；
Back-propagation method：Stochastic gradient descent algorithm.
Convolutional neural networks model is 5 convolutional layers in the step (2), in addition 3 full articulamentums.
The number of plies is increased or decreased in the step (3) to be no more than two layers.
The advantages of the present invention over the prior art are that：
(1) using convolutional neural networks model, to unmanned plane, specific objective is identified the present invention over the ground, and unmanned plane is over the ground
Specific objective type is more, such as ships, pedestrian, the variation of vehicle these target signatures are greatly, compared to conventional method, engineer's
Feature is difficult to give expression to target information completely, and feedback learning is carried out using convolutional neural networks, can be learnt to more with Shandong
The feature representation of stick, to ensure recognition accuracy requirement；
(2) convolutional neural networks model training method proposed by the present invention is damaged by the variation of training loss function with test
Function situation of change is lost, to obtain globally optimal solution, in contrast to the general target identification method based on convolutional neural networks, is led to
Prior information is crossed fixed number of iterations is arranged, the present invention can improve training effectiveness, while can also improve convolutional neural networks
The recognition accuracy of model；
(3) different depth convolutional neural networks model construction method proposed by the present invention changes for different data sets
Network model depth is trained test, obtains the optimal convolutional neural networks model of depth.General convolutional neural networks model
Target identification method, usual network architecture is fixed, and available data collection cannot meet its training demand well, or
Cause data set information to waste, the present invention can adapt to different data sets, to realize convolutional neural networks model structure with
The optimization of parameter.
Description of the drawings
Fig. 1 is that the present invention is based on the flow diagrams of the unmanned plane of convolutional neural networks specific objective recognition methods over the ground；
Fig. 2 is the convolutional neural networks model structure of the present invention；
Fig. 3 is convolutional neural networks model loss function change curve in training process in the present invention；
Fig. 4 is convolutional neural networks model accuracy rate change curve in training process in the present invention.
Fig. 5 is the test result figure of the present invention.
Figure label and symbol description are as follows：
Jcv(θ) --- cross validation collection loss function change curve；
Jtrain(θ) --- training set loss function change curve；
Iteration --- training iterations；
Test accuracy --- training is test accuracy rate change curve.
Specific implementation mode
The following describes the present invention in detail with reference to the accompanying drawings and embodiments.
One, training algorithm
The identification mission for having supervision for convolutional neural networks is needed due to knowing the classification of all image patterns in advance
Distribution according to its identical image sample in space so that differ the sample distribution of classification on different area of space.Through
Long-time training image data set is crossed, the middle parameter of convolutional neural networks is constantly updated, obtains dividing sample space classification
Boundary position classify to image.Convolutional neural networks are substantially a kind of mappings being input to output, according to spy
Fixed principle learns Function Mapping, which is mapped to an input picture feature vector of one k dimensions.As long as to convolution
Network training obtains the connection weight between network, and by activation primitive, network will learn to reflecting between inputoutput pair
Penetrate ability.
Training algorithm is broadly divided into propagated forward stage and back-propagation phase：
First stage, forward propagation stage：Sample set (X, Y) is input to network, wherein X indicates that sample data, Y indicate
Sample label.It is calculated by the level of network, i.e. input and every layer of weight matrix phase dot product successively obtains after operation corresponding
Output valve hθ(x).When the propagated forward stage, the weights of network random initializtion network connection.
Second stage, back-propagation phase：Calculate reality output hθ(x) error between corresponding ideal output Y, i.e.,
Cost function value.Weight matrix is adjusted by the method backpropagation of minimization error.
Two, cost function (LOSS)：
For network output signal hθ(x) it is the vector that a dimension is k with target desired signal y.Neural network exports
It with the error of actual value, i.e. loss function, is calculated using Euclidean distance, the cost function of neural network is represented by：
It is calculated in layer using Positive Propagation Algorithm is positive since first layer, finds out every layer and obtained by activation primitive
As a result, to the last one layer.But in order to keep the loss function value of whole network minimum, continuous iteration is needed to update nerve
The parameter and deviation of first front.Here back-propagation algorithm is used, i.e., by calculating the error of last layer first, then again successively
The error for reversely finding out each layer is updated weights and biasing using obtained residual error, to calculate network model
Least disadvantage function.
Three, back-propagation algorithm (Back propagation, BP)：
Most of neural network model can use the parameter that network is solved based on gradient descent method, and in network
When training sample is updated parameter, need to use back-propagation algorithm.The first step first initializes the weight parameter of network
For one group of random value；Second step reuses training data and is iterated training.It calculates the output of neural network model and it is expected defeated
Error is propagated to forward input from last layer, and is calculated according to gradient by the error between going out, i.e. loss function value in layer
Method updates the weights of each layer network, until meeting condition or stopping more than maximum iteration.
When back-propagation algorithm is applied to convolutional neural networks, steps are as follows for BP algorithm：Assuming that activation primitive is
Sigmoid functions, neural network one share m layers, and kth layer has skA neuron, for i-th of neuron of kth layer, WijIt indicates
The weight coefficient being attached thereto, then there are s for last layerk-1A weight coefficient Wi1,Wi2,···,Wij,···,Therewith
It is connected, biIndicate biasing.
The step of execution of BP algorithm, is as follows：
(1) to weight coefficient WijIt is random to set initial value, a sample (x, y) is inputted, wherein x indicates that sample, y indicate it is expected defeated
Go out；
(2) every layer of output, the output for k layers of i-th of neuron are calculatedHave：
(3) error of each layer study is sought
There is k=m for output layer,
To other each layers, have
(4) according to errorModified weight coefficient WijWith biasing bi
Wherein,Indicate the of -1 layer of kthjA output, t indicate that iterations, η indicate that learning rate, α indicate random
Gradient momentum.
(5) after having found out the weights of each layer of convolutional neural networks, according to condition judge whether to meet.If demand is full
Foot, then algorithm terminates；If do not met, (2) execution is returned.
Four, solution
The method of the present invention, as shown in Figure 1, being as follows：
(1) the collected target data set over the ground of unmanned plane is labeled according to classification, the image data that will have been marked
Collect according to a certain percentage, i.e., 3:1:1 is divided into training set, verification collection, test set, is then convolutional Neural by image data set processing
The data type that network model can identify.
(2) build convolutional neural networks model, as shown in Fig. 2, five layers of front is convolutional layer, behind two layers be to connect entirely
Layer, last layer are classification output layers, wherein there is 4 times of down-sampled, second convolutional layers between input layer and first convolutional layer
Have between third convolutional layer 2 times it is down-sampled, have between the 5th convolutional layer and first full articulamentum 2 times it is down-sampled.If
Training parameter, training method are set, such as：Maximum iteration, learning rate, test frequency, back-propagation method etc. start to train
This model.According to training loss function situation of change, optimal solution at this time is obtained.When the loss function (Loss) of training set declines
Slowly (amplitude of variation be less than 0.001), when verifying the loss function (Loss) of collection and tending to the critical point risen, network parameter reaches
To global optimum, the convolutional neural networks model parameter that training obtains at this time is preserved.
(3) in the structure of convolutional neural networks model, the appropriate convolution increased or reduce convolutional neural networks model
The number of plies (increases or decreases the number of plies to be no more than two layers), restarts to train, when the rate of accuracy reached of convolutional neural networks Model Identification
When to highest, illustrates that this convolutional neural networks model compares and adapt to current data set size, retain convolutional Neural net at this time
Network model structure and parameter.
(4) the convolutional neural networks model for utilizing (3) to obtain, tests test set, obtains recognition accuracy, work as knowledge
Other accuracy rate meets actual requirement of engineering, this convolutional neural networks model may be used in practical identification mission, if not satisfied,
Then illustrate that training dataset cannot meet actual requirement of engineering, then need dilated data set, restarts (1) (2) (3) step.
(5) the convolutional neural networks model parameter that accuracy rate is met the requirements is applied to unmanned plane specific objective reality over the ground
In scene, unmanned plane the image collected target is identified.
The unmanned plane proposed by the invention based on deep learning specific mesh over the ground is verified by a specific embodiment
The performance of recognition methods is marked, used is the images to be recognized data set that unmanned plane vision system is acquired, in this, as testing
Demonstrate,prove object, the deep learning frame based on caffe1.
As shown in Figure 1, the present invention is as follows：
(1) image is pre-processed：
Unmanned plane the image collected data set is classified as 3 class targets (ship, pedestrian, vehicle), 6000 in total, by data
Image is concentrated to be labeled according to classification, by the image data set marked according to 3:1:1 be divided into training set, cross validation collection,
Test set.Then the data type that image data set processing can be identified for network model, changes into lmdb formats.
(2) convolutional neural networks network model is constructed
According to convolutional neural networks model structure, such as scheme, shown in 2, write the prototxt files of network structure, before 5
Layer network is convolutional layer, behind 3 layer networks be full articulamentum.Wherein, Stride of 4 indicate that convolution step-length is 4, Max
Pooling is indicated using the down-sampled method of maximum value.
(3) training network model：
Training parameter is set first：Maximum iteration is set 200,000 times, initial learning rate is 0.001, using boarding steps
Degree declines back-propagation algorithm, starts to train network model.
Training is observed, the optimal solution of convolutional neural networks model is obtained.As shown in figure 3, convolution god in training process
Through network model loss function change curve, when the loss function (Loss) of training set declines loss letter that is slow, and verifying collection
When number (Loss) tends to the critical point risen, convolutional neural networks model parameter reaches globally optimal solution.As shown in figure 4, final
The test accuracy rate of training pattern illustrates that the target identification accuracy rate of this convolutional neural networks model is higher 0.99 or so.
(4) network model is changed：
Increase by 1 Ge Juan bases in the structure of convolutional neural networks model as shown in Figure 1, restart to train, finds
Final goal recognition accuracy accuracy slightly declines 0.5.1 is reduced in the structure of convolutional neural networks model as shown in Figure 1
Ge Juan bases restart to train, it is found that final goal recognition accuracy slightly declines 0.3.Illustrate the convolution god of 5 convolutional layers
It is more conform with the demand of this data set through network model, need not make an amendment again.
(5) test network model：
The network parameter that step (3) obtains is imported into test network, as shown in figure 5, being three of unmanned plane over the ground
Class target：Ship, vehicle, pedestrian recognition result, test image is identified, it is correct to obtain recognition result all.
Above example is provided just for the sake of the description purpose of the present invention, and is not intended to limit the scope of the present invention.This
The range of invention is defined by the following claims.It does not depart from spirit and principles of the present invention and the various equivalent replacements made and repaiies
Change, should all cover within the scope of the present invention.
Claims (3)
1. a kind of specific objective recognition methods over the ground of the unmanned plane based on convolutional neural networks, it is characterised in that：Including walking as follows
Suddenly：
(1) unmanned plane acquisition specific objective image data set over the ground is used, is labeled according to classification, and according to the number marked
It is divided into training set, verification collection, test set according to collection, is finally processed into the data type that convolutional neural networks model can identify；
(2) structure identification convolutional neural networks model, setting maximum iteration, learning rate, test frequency, selects backpropagation
Method starts to train according to above-mentioned be provided and selected, and then according to training loss function situation of change, obtains this convolutional Neural net
The recognition accuracy of network model；
(3) in the structure of convolutional neural networks model, increase or reduce the convolution number of plies, restart the training of step (2),
When the rate of accuracy reached of the identification of training convolutional neural networks model is to highest, illustrate that the convolutional neural networks model of this training is suitable
Current data set size is answered, retains training at this time and obtains the structure and parameter of convolutional neural networks model；
(4) the convolutional neural networks model for utilizing (3) to obtain, tests test set, obtains recognition accuracy, accurate to identification
True rate is judged, if recognition accuracy disclosure satisfy that actual requirement of engineering, the convolutional neural networks model can
It is applied in the actual unmanned plane task that specific objective identifies over the ground, executes step (5)；If not satisfied, then illustrating training set
Actual requirement of engineering cannot be met, need to expand training set, restart step (1), (2), (3), until meeting Practical Project
Until；
(5) parameter that accuracy rate is met to the convolutional neural networks model of Practical Project requirement is applied to unmanned plane specific mesh over the ground
In target actual scene, unmanned plane the image collected target is identified；
In step (2), the method for obtaining the convolutional neural networks model of parametric optimal solution is：As the loss function Loss of training set
Fall be no more than 0.001 when, and verify collection loss function Loss tend to rise critical point when, that is, obtain parameter it is optimal
The convolutional neural networks model of solution；
In step (2), setting maximum iteration, learning rate, test frequency select back-propagation method specific as follows：
Maximum iteration：200000 times；
Initial learning rate：0.001；
Test frequency：1000 iteration/1 time；
Back-propagation method：Stochastic gradient descent algorithm.
2. the specific objective recognition methods over the ground of the unmanned plane based on convolutional neural networks according to claim 1, feature
It is：Identification convolutional neural networks model is 5 convolutional layers in step (2), in addition 3 full articulamentums.
3. the specific objective recognition methods over the ground of the unmanned plane based on convolutional neural networks according to claim 1, feature
It is：The number of plies is increased or decreased in step (3) to be no more than two layers.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711422364.9A CN108009525B (en) | 2017-12-25 | 2017-12-25 | A kind of specific objective recognition methods over the ground of the unmanned plane based on convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711422364.9A CN108009525B (en) | 2017-12-25 | 2017-12-25 | A kind of specific objective recognition methods over the ground of the unmanned plane based on convolutional neural networks |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108009525A CN108009525A (en) | 2018-05-08 |
| CN108009525B true CN108009525B (en) | 2018-10-12 |
Family
ID=62061187
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711422364.9A Active CN108009525B (en) | 2017-12-25 | 2017-12-25 | A kind of specific objective recognition methods over the ground of the unmanned plane based on convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108009525B (en) |
Families Citing this family (63)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108875593A (en) * | 2018-05-28 | 2018-11-23 | 上海交通大学 | Visible images weather recognition methods based on convolutional neural networks |
| CN108957418A (en) * | 2018-05-30 | 2018-12-07 | 西安电子科技大学 | A kind of radar target identification method based on Recognition with Recurrent Neural Network model |
| CN109086790A (en) * | 2018-06-19 | 2018-12-25 | 歌尔股份有限公司 | A kind of alternative manner of disaggregated model, device and electronic equipment |
| CN109063824B (en) * | 2018-07-25 | 2023-04-07 | 深圳市中悦科技有限公司 | Deep three-dimensional convolutional neural network creation method and device, storage medium and processor |
| CN109033095B (en) * | 2018-08-01 | 2022-10-18 | 苏州科技大学 | Target transformation method based on attention mechanism |
| CN109144099B (en) * | 2018-08-28 | 2020-10-20 | 北京航空航天大学 | Fast evaluation method for unmanned aerial vehicle group action scheme based on convolutional neural network |
| TWI709188B (en) | 2018-09-27 | 2020-11-01 | 財團法人工業技術研究院 | Fusion-based classifier, classification method, and classification system |
| CN110969072B (en) * | 2018-09-30 | 2023-05-02 | 杭州海康威视系统技术有限公司 | Model optimization method, device and image analysis system |
| CN110163236B (en) * | 2018-10-15 | 2023-08-29 | 腾讯科技（深圳）有限公司 | Model training method and device, storage medium and electronic device |
| CN109657541A (en) * | 2018-11-09 | 2019-04-19 | 南京航空航天大学 | A kind of ship detecting method in unmanned plane image based on deep learning |
| CN109657596A (en) * | 2018-12-12 | 2019-04-19 | 天津卡达克数据有限公司 | A kind of vehicle appearance component identification method based on deep learning |
| CN109815798A (en) * | 2018-12-17 | 2019-05-28 | 广东电网有限责任公司 | Unmanned plane image processing method and system |
| CN109685017B (en) * | 2018-12-26 | 2023-02-03 | 中山大学 | Real-time target detection system and detection method based on lightweight neural network |
| CN109726761B (en) * | 2018-12-29 | 2023-03-31 | 青岛海洋科学与技术国家实验室发展中心 | CNN evolution method, CNN-based AUV cluster working method, CNN evolution device and CNN-based AUV cluster working device and storage medium |
| CN109934088A (en) * | 2019-01-10 | 2019-06-25 | 海南大学 | Sea ship discrimination method based on deep learning |
| CN109740560B (en) * | 2019-01-11 | 2023-04-18 | 山东浪潮科学研究院有限公司 | Automatic human body cell protein identification method and system based on convolutional neural network |
| CN109829898B (en) * | 2019-01-17 | 2023-05-16 | 柳州康云互联科技有限公司 | Measurement detection system and method based on neural network for Internet detection |
| CN109886923B (en) * | 2019-01-17 | 2023-05-02 | 柳州康云互联科技有限公司 | Measurement detection system and method based on machine learning for Internet detection |
| CN109886141B (en) * | 2019-01-28 | 2023-06-06 | 同济大学 | Pedestrian re-identification method based on uncertainty optimization |
| CN111507128A (en) * | 2019-01-30 | 2020-08-07 | 北京沃东天骏信息技术有限公司 | Face recognition method and device, electronic equipment and readable medium |
| CN110059538B (en) * | 2019-02-27 | 2021-07-09 | 成都数之联科技有限公司 | Water body identification method based on deep dense neural network |
| CN109934333A (en) * | 2019-02-28 | 2019-06-25 | 天津大学 | Object localization method based on full Connection Neural Network |
| CN109871906B (en) * | 2019-03-15 | 2023-03-28 | 西安获德图像技术有限公司 | Cop appearance defect classification method based on deep convolutional neural network |
| CN110517219B (en) * | 2019-04-01 | 2023-04-07 | 中山大学中山眼科中心 | Corneal topography distinguishing method and system based on deep learning |
| CN109977902A (en) * | 2019-04-03 | 2019-07-05 | 刘西 | A kind of construction vehicle identification method based on deep learning |
| CN110245582A (en) * | 2019-05-25 | 2019-09-17 | 天津大学 | A method of based on classification component single in deep learning for identification bitmap |
| CN110163177B (en) * | 2019-05-28 | 2022-12-09 | 李峥嵘 | Unmanned aerial vehicle automatic sensing and identifying method for wind turbine generator blades |
| CN110347600B (en) * | 2019-07-11 | 2023-04-07 | 中国人民解放军陆军工程大学 | Convolutional neural network-oriented variation coverage testing method and computer storage medium |
| CN110458079A (en) * | 2019-08-05 | 2019-11-15 | 黑龙江电力调度实业有限公司 | A kind of Image Acquisition and target identification method based on FPGA and convolutional neural networks |
| CN110544249A (en) * | 2019-09-06 | 2019-12-06 | 华南理工大学 | Convolutional neural network quality identification method for arbitrary-angle case assembly visual inspection |
| CN110717260A (en) * | 2019-09-26 | 2020-01-21 | 杭州电子科技大学 | Unmanned aerial vehicle maneuvering capability model establishing method |
| CN110782078B (en) * | 2019-10-18 | 2023-04-25 | 武汉科技大学 | Learning method for predicting mud yield of trailing suction hopper dredger |
| CN111008641B (en) * | 2019-10-24 | 2023-04-07 | 云南电网有限责任公司昆明供电局 | Power transmission line tower external force damage detection method based on convolutional neural network |
| CN110749324A (en) * | 2019-10-28 | 2020-02-04 | 深圳市赛为智能股份有限公司 | Unmanned aerial vehicle rescue positioning method and device, computer equipment and storage medium |
| CN110764064A (en) * | 2019-11-08 | 2020-02-07 | 哈尔滨工业大学 | Radar interference signal identification method based on deep convolutional neural network integration |
| CN111046785B (en) * | 2019-12-10 | 2023-07-11 | 长讯通信服务有限公司 | Method for identifying key targets of inspection video of unmanned aerial vehicle based on convolutional neural network |
| CN111160226B (en) * | 2019-12-26 | 2023-03-31 | 华侨大学 | Pedestrian gender identification method based on visual angle adaptive feature learning |
| JP7421363B2 (en) * | 2020-02-14 | 2024-01-24 | 株式会社Ｓｃｒｅｅｎホールディングス | Parameter update device, classification device, parameter update program, and parameter update method |
| CN111325143A (en) * | 2020-02-18 | 2020-06-23 | 西北工业大学 | Underwater target identification method under unbalanced data set condition |
| CN111539178B (en) * | 2020-04-26 | 2023-05-05 | 成都市深思创芯科技有限公司 | Chip layout design method and system based on neural network and manufacturing method |
| CN113642592A (en) * | 2020-04-27 | 2021-11-12 | 武汉Tcl集团工业研究院有限公司 | Training method of training model, scene recognition method and computer equipment |
| CN113743168B (en) * | 2020-05-29 | 2023-10-13 | 北京机械设备研究所 | Urban flyer identification method based on micro-depth neural network search |
| CN111652326B (en) * | 2020-06-30 | 2023-04-28 | 华南农业大学 | Fruit maturity identification method and system based on MobileNet v2 network improvement |
| CN111860278B (en) * | 2020-07-14 | 2024-05-14 | 陕西理工大学 | Human behavior recognition algorithm based on deep learning |
| CN111897333B (en) * | 2020-07-31 | 2022-12-20 | 常州码库数据科技有限公司 | Robot walking path planning method |
| CN111899091B (en) * | 2020-08-06 | 2023-03-24 | 华院计算技术(上海)股份有限公司 | Overdue risk identification method based on robust algorithm |
| CN112153615B (en) * | 2020-09-15 | 2022-07-12 | 南京信息工程大学滨江学院 | Deep learning-based user association method in multi-cell cellular D2D equipment |
| CN112418266A (en) * | 2020-10-15 | 2021-02-26 | 南昌大学 | Pile foundation integrity classification and identification method based on convolutional neural network |
| CN112699928B (en) * | 2020-12-25 | 2022-09-20 | 南京理工大学 | Non-motor vehicle detection and identification method based on deep convolutional network |
| CN113505628A (en) * | 2021-04-02 | 2021-10-15 | 上海师范大学 | Target identification method based on lightweight neural network and application thereof |
| CN113076895B (en) * | 2021-04-09 | 2022-08-02 | 太原理工大学 | Conveyor belt longitudinal damage vibration sensing method based on infrared computer vision |
| CN113361682B (en) * | 2021-05-08 | 2022-08-16 | 南京理工大学 | Reconfigurable neural network training with IP protection and using method |
| CN113313021B (en) * | 2021-05-27 | 2023-05-30 | 云南电网有限责任公司电力科学研究院 | Deep learning model construction method based on low-quality image recognition |
| CN113671161B (en) * | 2021-07-13 | 2023-04-18 | 郑州大学 | Unmanned aerial vehicle pavement disease detection method based on LSTM neural network algorithm |
| CN113505851B (en) * | 2021-07-27 | 2023-01-31 | 电子科技大学 | Multitasking method for intelligent aircraft |
| CN113327461B (en) * | 2021-08-03 | 2021-11-23 | 杭州海康威视数字技术股份有限公司 | Cooperative unmanned aerial vehicle detection method, device and equipment |
| CN113702719B (en) * | 2021-08-03 | 2022-11-29 | 北京科技大学 | Broadband near-field electromagnetic positioning method and device based on neural network |
| CN113627611A (en) * | 2021-08-06 | 2021-11-09 | 苏州科韵激光科技有限公司 | Model training method and device, electronic equipment and storage medium |
| CN113673031B (en) * | 2021-08-11 | 2024-04-12 | 中国科学院力学研究所 | Flexible airship service attack angle identification method integrating strain response and deep learning |
| CN113887636B (en) * | 2021-10-09 | 2023-04-07 | 四川大学 | Selectable data enhancement method and system based on genetic algorithm |
| CN114359546B (en) * | 2021-12-30 | 2024-03-26 | 太原科技大学 | Day lily maturity identification method based on convolutional neural network |
| CN115859056B (en) * | 2022-12-29 | 2023-09-15 | 华诺星空技术股份有限公司 | Unmanned aerial vehicle target detection method based on neural network |
| CN117292283B (en) * | 2023-11-24 | 2024-02-13 | 成都庆龙航空科技有限公司 | Target identification method based on unmanned aerial vehicle |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103955702A (en) * | 2014-04-18 | 2014-07-30 | 西安电子科技大学 | SAR image terrain classification method based on depth RBF network |
| CN105512676A (en) * | 2015-11-30 | 2016-04-20 | 华南理工大学 | Food recognition method at intelligent terminal |
| CN106682569A (en) * | 2016-09-28 | 2017-05-17 | 天津工业大学 | Fast traffic signboard recognition method based on convolution neural network |
| CN107330405A (en) * | 2017-06-30 | 2017-11-07 | 上海海事大学 | Remote sensing images Aircraft Target Recognition based on convolutional neural networks |
| CN107423760A (en) * | 2017-07-21 | 2017-12-01 | 西安电子科技大学 | Based on pre-segmentation and the deep learning object detection method returned |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9940544B2 (en) * | 2016-06-08 | 2018-04-10 | Adobe Systems Incorporated | Event image curation |
-
2017
- 2017-12-25 CN CN201711422364.9A patent/CN108009525B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103955702A (en) * | 2014-04-18 | 2014-07-30 | 西安电子科技大学 | SAR image terrain classification method based on depth RBF network |
| CN105512676A (en) * | 2015-11-30 | 2016-04-20 | 华南理工大学 | Food recognition method at intelligent terminal |
| CN106682569A (en) * | 2016-09-28 | 2017-05-17 | 天津工业大学 | Fast traffic signboard recognition method based on convolution neural network |
| CN107330405A (en) * | 2017-06-30 | 2017-11-07 | 上海海事大学 | Remote sensing images Aircraft Target Recognition based on convolutional neural networks |
| CN107423760A (en) * | 2017-07-21 | 2017-12-01 | 西安电子科技大学 | Based on pre-segmentation and the deep learning object detection method returned |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108009525A (en) | 2018-05-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108009525B (en) | A kind of specific objective recognition methods over the ground of the unmanned plane based on convolutional neural networks | |
| US20220101127A1 (en) | Automatic optimization of machine learning algorithms in the presence of target datasets | |
| KR102641116B1 (en) | Method and device to recognize image and method and device to train recognition model based on data augmentation | |
| KR102595399B1 (en) | Detection of unknown classes and initialization of classifiers for unknown classes | |
| US10275719B2 (en) | Hyper-parameter selection for deep convolutional networks | |
| CN111462130B (en) | Method and apparatus for detecting lane lines included in input image using lane mask | |
| JP6236296B2 (en) | Learning device, learning program, and learning method | |
| CN111507469B (en) | Method and device for optimizing super parameters of automatic labeling device | |
| US8379994B2 (en) | Digital image analysis utilizing multiple human labels | |
| WO2016175925A1 (en) | Incorporating top-down information in deep neural networks via the bias term | |
| KR102349854B1 (en) | System and method for tracking target | |
| Binguitcha-Fare et al. | Crops and weeds classification using convolutional neural networks via optimization of transfer learning parameters | |
| CN113743417B (en) | Semantic segmentation method and semantic segmentation device | |
| CN111833322B (en) | Garbage multi-target detection method based on improved YOLOv3 | |
| CN107423747A (en) | A kind of conspicuousness object detection method based on depth convolutional network | |
| WO2020198173A1 (en) | Subject-object interaction recognition model | |
| CN116524183A (en) | Camouflage target detection method based on multitask adapter fine adjustment | |
| CN117292283B (en) | Target identification method based on unmanned aerial vehicle | |
| JP2019023801A (en) | Image recognition device, image recognition method and image recognition program | |
| Batool et al. | Ielmnet: An application for traffic sign recognition using cnn and elm | |
| CN116630828A (en) | Unmanned aerial vehicle remote sensing information acquisition system and method based on terrain environment adaptation | |
| CN114359653A (en) | Attack resisting method, defense method and device based on reinforced universal patch | |
| CN114067155A (en) | Image classification method, device, product and storage medium based on meta learning | |
| CN112233071A (en) | Multi-granularity hidden danger detection method and system based on power transmission network picture in complex environment | |
| CN115116054B (en) | Multi-scale lightweight network-based pest and disease damage identification method |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |