CN108664933A - The training method and its sorting technique of a kind of convolutional neural networks for SAR image ship classification, ship classification model - Google Patents
The training method and its sorting technique of a kind of convolutional neural networks for SAR image ship classification, ship classification model Download PDFInfo
- Publication number
- CN108664933A CN108664933A CN201810450109.3A CN201810450109A CN108664933A CN 108664933 A CN108664933 A CN 108664933A CN 201810450109 A CN201810450109 A CN 201810450109A CN 108664933 A CN108664933 A CN 108664933A
- Authority
- CN
- China
- Prior art keywords
- ship
- sar image
- convolutional neural
- neural networks
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000012549 training Methods 0.000 title claims abstract description 105
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 81
- 238000000034 method Methods 0.000 title claims abstract description 64
- 238000013145 classification model Methods 0.000 title claims description 6
- 230000000977 initiatory effect Effects 0.000 claims description 17
- 230000003287 optical effect Effects 0.000 claims description 8
- 230000001537 neural effect Effects 0.000 claims description 7
- 238000013480 data collection Methods 0.000 claims description 5
- 238000002474 experimental method Methods 0.000 description 7
- 238000012795 verification Methods 0.000 description 5
- 238000012986 modification Methods 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 210000002569 neuron Anatomy 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000013136 deep learning model Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- VMXUWOKSQNHOCA-UKTHLTGXSA-N ranitidine Chemical compound [O-][N+](=O)\C=C(/NC)NCCSCC1=CC=C(CN(C)C)O1 VMXUWOKSQNHOCA-UKTHLTGXSA-N 0.000 description 2
- 101100153586 Caenorhabditis elegans top-1 gene Proteins 0.000 description 1
- 101100370075 Mus musculus Top1 gene Proteins 0.000 description 1
- 240000007594 Oryza sativa Species 0.000 description 1
- 235000007164 Oryza sativa Nutrition 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 235000009566 rice Nutrition 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Software Systems (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The present invention provides a kind of training method and its sorting technique of the convolutional neural networks for SAR image ship classification, and training method includes:Obtain the slice that description of ship is carried in SAR image;Based on the slice with description of ship to being trained for the convolutional neural networks of SAR image ship classification, so that it can reach default training precision;Wherein, the convolutional neural networks for being used for SAR image ship classification are to remove the first full articulamentum of the first model, and number the second full articulamentum of addition of the description of ship according to the slice with description of ship is built-up on the basis of the first model.Training method may be implemented that there is only, to being trained for the convolutional neural networks of SAR image ship classification, reach default training precision in the case of a small amount of training data.And the convolutional neural networks for SAR image ship classification for having reached default training precision in the present invention are applied in SAR image ship classification, ship classification precision can reach 97.54%.
Description
Technical field
This application involves remote sensing fields, and in particular to a kind of instruction of convolutional neural networks for SAR image ship classification
Practice method and its sorting technique.
Background technology
From starting in 2007, there is more high resolution SAR satellite launch successes, such as Cosmo-SkyMed, TerraSAR-
X, the resolution ratio of ALOS-2PALSAR-2, Gaofen-3 etc., the high-resolution satellite SAR image obtained based on satellite are more than 3
Rice, it comprises the geometric properties of the information that ground atural object is abundant, such as ship, become so that distinguishing different classes of ship
Obtaining may.
Letter of the deep learning model (for example, convolutional neural networks) since expression atural object in SAR image can be learnt automatically
Breath, and provide it is corresponding handle end to end without manpower intervention, to save feature extraction and selection and
Optimize the time of grader.Therefore, deep learning model is gradually applied to carry out classification task in SAR image.But use depth
The bottleneck of degree learning model is that it needs a large amount of training data, and obtains a large amount of flag data and waste time and be difficult to obtain
.Therefore, there is an urgent need for a kind of completely new methods being trained to deep learning model.
Invention content
The present invention provides a kind of training method of convolutional neural networks for SAR image ship classification and its classification sides
Method.The training method may be implemented there is only in the case of a small amount of training data to described for SAR image ship classification
Convolutional neural networks are trained, and reach default training precision.And default training precision will be had reached in the present invention
The convolutional neural networks for SAR image ship classification be applied in SAR image ship classification, ship classification precision can be with
Reach 97.54%.
In order to solve the above-mentioned technical problem, an embodiment of the present invention provides the following technical solutions:
First aspect present invention provides a kind of training method of the convolutional neural networks for SAR image ship classification,
Including:
Obtain the slice that description of ship is carried in SAR image;
Based on the slice with description of ship to being instructed for the convolutional neural networks of SAR image ship classification
Practice, so that it can reach default training precision;
Wherein, the convolutional neural networks for SAR image ship classification are to remove institute on the basis of the first model
State the first full articulamentum of the first model, and the number addition second of the description of ship according to the slice with description of ship
Full articulamentum is built-up.
Preferably, the first full articulamentum includes the first Softmax layers and two the first fully connected
Layer.
Preferably, the second full articulamentum includes the 2nd Softmax layers and connected layers of the 2nd fully.
Preferably, the slice of description of ship is carried in the acquisition SAR image, including:
On the basis of the SAR image slice with description of ship is obtained according to ship information and preset rules.
Preferably, described refreshing to the convolution for SAR image ship classification based on the slice with description of ship
It is trained through network, including, all articulamentums of the convolutional neural networks for SAR image ship classification are trained,
Or
Second full articulamentum of the convolutional neural networks for SAR image ship classification is trained.
Preferably, described refreshing to the convolution for SAR image ship classification based on the slice with description of ship
It is trained through network, including, all articulamentums of the convolutional neural networks for SAR image ship classification are trained,
Or
The second full articulamentum to the convolutional neural networks for SAR image ship classification and the 2nd Softmax layers
It is trained.
Preferably, based on the slice with description of ship to the convolutional Neural net for SAR image ship classification
Network is trained, so that it can reach default training precision, including:
The initiation parameter trained for the convolutional neural networks of SAR image ship classification is set;
The convolutional neural networks of the slice input for SAR image ship classification with description of ship are trained
Obtain output result;
Judge to export whether result reaches default training precision:
If it is not, then adjustment initiation parameter, continues to execute above-mentioned steps, until output result reaches default training essence
Degree.
Preferably, the initiation parameter that the setting is trained for the convolutional neural networks of SAR image ship classification, packet
It includes,
The pre-training parameter that first model progress pre-training is obtained is as the convolution for being used for SAR image ship classification
The initiation parameter of neural network.
Preferably, first model carries out pre-training on optical data collection obtains pre-training parameter.
Preferably, first model is selected from VGG16, VGG19, InceptionV3 or Xception.
Second aspect of the present invention provides a kind of instruction based on the above-mentioned convolutional neural networks for SAR image ship classification
Practice the method for the SAR image ship classification of the convolutional neural networks obtained after method training comprising:
Obtain the ship slice in SAR image to be sorted;
Ship slice input is had reached to the convolutional Neural net for SAR image ship classification of default training precision
Network obtains the ship classification result of the ship slice.
Third aspect present invention provides a kind of ship classification model, wherein and the model is based on convolutional neural networks realization,
The convolutional neural networks are to remove the first full articulamentum of first model, and according to institute on the basis of the first model
Number the second full articulamentum of addition for stating the description of ship of the slice with description of ship is built-up.
Preferably, the first full articulamentum includes the first Softmax layers and two the first fully connected
Layer.
Preferably, the second full articulamentum includes the 2nd Softmax layers and connected layers of the 2nd fully.
Preferably, first model is selected from VGG16, VGG19, InceptionV3 or Xception.
Disclosure based on above-described embodiment can know that the embodiment of the present invention has following advantageous effect:The present invention carries
The training method of the convolutional neural networks for SAR image ship classification supplied may be implemented that there is only the feelings of a small amount of training data
The convolutional neural networks for SAR image ship classification are trained under shape, i.e., first by the first model in optics number
According to pre-training is carried out on collection, pre-training parameter is obtained;Using pre-training parameter as to the convolution god for SAR image ship classification
Initiation parameter through network training, rather than random assignment is carried out to it, then to the convolution for SAR image ship classification
Neural network carries out training using two class methods, reaches default training precision.And it is discovered by experiment that using the first kind
The verification ratio of precision that training method obtains is high by 2% using the verification precision that the second class training method obtains, and illustrates that the first kind is trained
Method is more suitable for the convolutional neural networks for SAR image ship classification in the training present invention.
By the convolutional neural networks application for SAR image ship classification for having reached default training precision in the present invention
In SAR image ship classification the experiment has found that under the same conditions, the convolutional neural networks ratio based on VGG16 is based on it
The ship classification precision of the convolutional neural networks of his three kinds of models is high, and ship classification precision can reach 97.54%.
Description of the drawings
Fig. 1 is the flow of the training method of the convolutional neural networks for SAR image ship classification of the embodiment of the present invention
Figure;
Fig. 2 is three kinds of different classes of slices with description of ship of the embodiment of the present invention;
Fig. 3 is that the VGG16 models of the embodiment of the present invention use Gaussian Profile initiation parameter on ImageNet data sets
Carry out the procedure chart of pre-training;
Fig. 4 is that two class methods are respectively adopted to the convolutional Neural that is improved based on VGG16 models in the embodiment of the present invention
The procedure chart that network is trained.
Specific implementation mode
In the following, specific embodiments of the present invention are described in detail in conjunction with attached drawing, but it is not as limiting to the invention.
It should be understood that various modifications can be made to disclosed embodiments.Therefore, description above should not regard
To limit, and only as the example of embodiment.Those skilled in the art will expect within the scope and spirit of this
Other modifications.
The attached drawing being included in the description and forms part of the description shows embodiment of the disclosure, and with it is upper
What face provided is used to explain the disclosure together to the substantially description of the disclosure and the detailed description given below to embodiment
Principle.
It is of the invention by the description of the preferred form of the embodiment with reference to the accompanying drawings to being given as non-limiting examples
These and other characteristic will become apparent.
Although being also understood that invention has been described with reference to some specific examples, people in the art
Member realize with can determine the present invention many other equivalents, they have feature as claimed in claim and therefore all
In the protection domain defined by whereby.
When read in conjunction with the accompanying drawings, in view of following detailed description, above and other aspect, the feature and advantage of the disclosure will become
It is more readily apparent.
The specific embodiment of the disclosure is described hereinafter with reference to attached drawing;It will be appreciated, however, that the disclosed embodiments are only
Various ways implementation can be used in the example of the disclosure.It is known and/or repeat function and structure be not described in detail to avoid
Unnecessary or extra details so that the disclosure is smudgy.Therefore, specific structural and functionality disclosed herein is thin
Section is not intended to restrictions, but as just the basis of claim and representative basis be used to instruct those skilled in the art with
Substantially any appropriate detailed construction diversely uses the disclosure.
This specification can be used phrase " in one embodiment ", " in another embodiment ", " in another embodiment
In " or " in other embodiments ", it can be referred to one or more of the identical or different embodiment according to the disclosure.
The experiment of the present invention carries out in Ubuntu systems, the NVIDAI GPU GTX1070 video cards with 8G memories.And
This experiment is carried out on Keras (being developed on the basis of in tensorflow, CNKT and Theano).First model be from
It downloads and obtains on the official websites Kearas.
In the following, the embodiment of the present invention is described in detail in conjunction with attached drawing:
As shown in Figure 1, the application one embodiment provides a kind of convolutional neural networks for SAR image ship classification
Training method, the present embodiment is described in detail below.
Step S101 obtains the slice that description of ship is carried in SAR image.
The 6 scape SAR images that this experiment is obtained using Cosmo-Skymed satellites is research objects, and the SAR images are specifically
It is bright as shown in table 1.
1 SAR image information of table
On the basis of above-mentioned SAR image the slice with description of ship is obtained according to ship information and preset rules.
In the specific embodiment of the present invention, refer to according to AIS information and expert on the basis of above-mentioned SAR image
Lead the slice with description of ship of the interception pixel of 256 pixels × 256.Wherein, 446 cutting with description of ship are obtained altogether
The classification of piece, ship is mainly freighter, container ship and oil tanker, three kinds of different classes of ship particular number distribution situation such as tables
Shown in 2.Three kinds of different classes of slices with description of ship are as shown in Figure 2.
The particular number table of the different classes of ship of table 2
Convolutional neural networks of the step S102 structures for SAR image ship classification.
Convolutional neural networks for SAR image ship classification are to remove first mould on the basis of the first model
First full articulamentum of type, and the number of the description of ship according to the slice with description of ship adds the second full articulamentum
It is built-up.
Wherein, the described first full articulamentum includes the first Softmax layers and two connected layers of the first fully.Institute
It is 1000 to state the first Softmax layers of output classification, and the number of the neuron of connected layer of the first fully is 4096
It is a.
Wherein, the described second full articulamentum includes the 2nd Softmax layers and connected layers of the 2nd fully.Described
Two Softmax layers of output classification is 3, and the number of the neuron of connected layers of the 2nd fully is 128.
In the other embodiment of the present invention, the second full articulamentum includes the 2nd Softmax layers, the 2nd fully
Connected layers and first " Dropout " layer.
The present invention another embodiment in, first model be selected from VGG16, VGG19,
InceptionV3 or Xception.The specifying information of four the first models of difference is as shown in table 3.
The first different model information of 3 four kinds of table
In a specific implementation mode in the present invention, the number of the description of ship can be two or more, for example,
2-10 kinds.If the description of ship of the slice with description of ship is freighter, container ship and oil tanker, at this time description of ship
Number be 3.
As shown in Figures 3 and 4, it shows with the structure for the convolutional neural networks that VGG16 models are the first model, it can from figure
To find out, convolutional neural networks and VGG16 models based on VGG16 models are maximum the difference is that top layer, that is,
It is 4096 to have two neuron numbers that output classification is 1000 in VGG16 models, and based on VGG16 model constructions
Have in convolutional neural networks output classification be 3 the 2nd Softmax layer with a neuron number be 128 the 2nd fully
Connected layers.
Convolutional neural networks of the step S103 training for SAR image ship classification.
Step S103-1 obtains the initiation parameter trained for the convolutional neural networks of SAR image ship classification:
Because the training data for training for the convolutional neural networks of SAR image ship classification (refers in the present invention
Be the slice with description of ship) it is limited, therefore, it is necessary to first by the first model in the optical data collection with mass data
Pre-training on (such as ImageNet or Coco), and the pre-training parameter that pre-training obtains is used as to being used for SAR image ship
The initiation parameter that the convolutional neural networks of oceangoing ship classification are trained, rather than random assignment is carried out to it.Wherein, Fig. 3 is shown
The process for carrying out pre-training using Gaussian Profile initiation parameter on ImageNet data sets by taking VGG16 models as an example.
That is, the detailed process for obtaining the initiation parameter trained for the convolutional neural networks of SAR image ship classification is:
First model is subjected to pre-training on optical data collection, obtains pre-training parameter;
Using pre-training parameter as the initiation parameter to the convolutional neural networks training for SAR image ship classification.
Step S103-2 is by the slice input with description of ship obtained in step S101 for SAR image ships point
The convolutional neural networks of class are trained acquisition output result.
In above-mentioned steps, although the first model is carried out on optical data collection the pre-training parameter that pre-training obtains
As the initiation parameter to being trained for the convolutional neural networks of SAR image ship classification, it is contemplated that optical picture
The factors such as the difference of picture and SAR image, such as the different information differences with about target of image-forming mechanism, therefore, in optical data
The pre-training parameter that pre-training obtains on collection may be not suitable for SAR image, therefore, it still will be to for SAR image ship classification
Convolutional neural networks are trained.
Using two class methods to being trained for the convolutional neural networks of SAR image ship classification in the present invention:
First kind training method is:All articulamentums are trained;
Second class training method is:Only the second full articulamentum is trained.
In an embodiment of the invention, for the convolutional neural networks for SAR image ship classification, Ke Yiyong
Two class training methods are trained it, and the first kind is, to all companies of the convolutional neural networks for SAR image ship classification
Layer is connect to be trained;Second class is only to be carried out to the second full articulamentum of the convolutional neural networks for SAR image ship classification
Training.
As shown in figure 4, indicate to be respectively adopted two class methods to the convolutional neural networks that are improved based on VGG16 models into
The process of row training, it can be seen from the figure that first kind method is trained all articulamentums, and the second class method is only right
" Fully connected " articulamentum and " Softmax " in " network layer newly increased " this sash shown in figure connect
Layer is trained.
In another embodiment of the present invention, first kind training method be 0.0001 by learning rate and moment is
The training method of 0.99 stochastic gradient descent.
In other embodiments of the invention, the second class training method be 0.001 by learning rate and moment is
0.9 training method.
Convolutional neural networks of the step S103-3 training for SAR image ship classification reach default training precision.
Judge to export whether result reaches default training precision;
If it is not, then adjustment initiation parameter, continues to execute above-mentioned steps, until output result reaches default training essence
Degree.
In an embodiment of the invention, judge that it is output result to export result to reach the condition of default training precision
Or loss hardly changes.
Step S104 tests the convolutional neural networks for SAR image ship classification for having reached default training precision
Card, is verified precision.
(described 4 different to be used for the different convolutional neural networks for SAR image ship classification of the present invention couple 4
The convolutional neural networks of SAR image ship classification are known respectively as convolutional neural networks based on InceptionV3, are based on
The convolutional neural networks of VGG16, the convolutional neural networks based on VGG19 and the convolutional neural networks based on Xception) respectively
Use two above-mentioned class training methods (that is, first kind training method for:All articulamentums are trained;Second class training side
Method is:Only the second full articulamentum is trained.) it is trained, each convolutional neural networks is obtained in two classes difference
Default training precision under training method and verification precision, the results are shown in Table 4 for specific experiment.
Table 4 uses the corresponding different accuracy table of convolutional neural networks of different training methods
The experimental results showed that, on the one hand, first kind training method is stablized than the second class training method process, on the other hand from
Table can be seen that this four different convolutional neural networks and use second using the verification ratio of precision that first kind training method obtains
The verification precision that class training method obtains is high by 2%.The reason of causing above-mentioned phenomenon may be to be trained by first kind training method
Obtained convolutional neural networks are more suitable for SAR image.
Second embodiment of the application provides a kind of based on the above-mentioned convolutional neural networks for SAR image ship classification
The method of the SAR image ship classification of the convolutional neural networks obtained after training method training:
Obtain the ship slice in SAR image to be sorted;
Ship slice input is had reached to the convolutional Neural net for SAR image ship classification of default training precision
Network obtains the ship classification result of the ship slice.
Experiment is found, compared with VGG16 models, although the Top-1 accuracys rate of other models, Top-5 accuracys rate and model
Depth is above VGG16 models, still, higher ship classification is obtained than other models based on the convolutional neural networks of VGG16
Precision, ship classification precision can reach 97.54%.Its reason may be the feature of the model extraction with deeper model depth
It is more and more abstract, thus be more suitable for optical imagery and be not suitable for SAR image.
The application third embodiment provides a kind of ship classification model, and the structure of the ship classification model is:
On the basis of the first model, remove the first full articulamentum of first model, and ship is carried according to described
It is built-up that the number of the description of ship of the slice of classification adds the second full articulamentum.
In one embodiment of the invention, the described first full articulamentum include the first Softmax layer with two first
Connected layers of fully.
In another embodiment of the present invention, the described second full articulamentum includes the 2nd Softmax layers and the 2nd fully
Connected layers.
In a preferred embodiment of the present invention, first model is selected from VGG16, VGG19, InceptionV3
Or Xception.
Above example is only exemplary embodiment of the present invention, is not used in the limitation present invention, protection scope of the present invention
It is defined by the claims.Those skilled in the art can within the spirit and scope of the present invention make respectively the present invention
Kind modification or equivalent replacement, this modification or equivalent replacement also should be regarded as being within the scope of the present invention.
Claims (10)
1. a kind of training method of convolutional neural networks for SAR image ship classification comprising:
Obtain the slice that description of ship is carried in SAR image;
Based on the slice with description of ship to being trained for the convolutional neural networks of SAR image ship classification, with
So as to reach default training precision;
Wherein, the convolutional neural networks for SAR image ship classification are to remove described the on the basis of the first model
First full articulamentum of one model, and the number addition second of the description of ship according to the slice with description of ship connects entirely
Layer building is connect to form.
2. training method according to claim 1, wherein the first full articulamentum include the first Softmax layer with two
A connected layers of first fully, the second full articulamentum include the 2nd Softmax layers and the 2nd fully
Connected layers.
3. training method according to claim 1, wherein described to obtain the slice for carrying description of ship in SAR image, packet
It includes:
On the basis of the SAR image slice with description of ship is obtained according to ship information and preset rules.
4. training method according to claim 1, wherein it is described based on the slice with description of ship to being used for
The convolutional neural networks of SAR image ship classification are trained, including, to the convolutional Neural net for SAR image ship classification
All articulamentums of network are trained, or
Second full articulamentum of the convolutional neural networks for SAR image ship classification is trained.
5. training method according to claim 1, wherein based on the slice with description of ship to being used for SAR figures
As the convolutional neural networks of ship classification are trained, so that it can reach default training precision, including:
The initiation parameter trained for the convolutional neural networks of SAR image ship classification is set;
The convolutional neural networks of the slice input for SAR image ship classification with description of ship are trained acquisition
Export result;
Judge to export whether result reaches default training precision:
If it is not, then adjustment initiation parameter, continues to execute above-mentioned steps, until output result reaches default training precision.
6. training method according to claim 5, wherein convolutional Neural of the setting for SAR image ship classification
The initiation parameter of network training, including,
The pre-training parameter that first model progress pre-training is obtained is as the convolutional Neural for being used for SAR image ship classification
The initiation parameter of network.
7. training method according to claim 6, wherein first model carries out pre-training on optical data collection and obtains
To pre-training parameter.
8. training method according to claim 1, wherein first model be selected from VGG16, VGG19,
InceptionV3 or Xception.
9. a kind of training of convolutional neural networks based on claim 1-8 any one of them for SAR image ship classification
The method of the SAR image ship classification of the convolutional neural networks obtained after method training comprising:
Obtain the ship slice in SAR image to be sorted;
The convolutional neural networks for SAR image ship classification that ship slice input is had reached to default training precision obtain
Take the ship classification result of the ship slice.
10. a kind of ship classification model, wherein the model realizes that the convolutional neural networks are based on convolutional neural networks
On the basis of the first model, remove the first full articulamentum of first model, and according to the cutting with description of ship
It is built-up that the number of the description of ship of piece adds the second full articulamentum.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810450109.3A CN108664933B (en) | 2018-05-11 | 2018-05-11 | Training method of convolutional neural network for SAR image ship classification, classification method of convolutional neural network and ship classification model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810450109.3A CN108664933B (en) | 2018-05-11 | 2018-05-11 | Training method of convolutional neural network for SAR image ship classification, classification method of convolutional neural network and ship classification model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108664933A true CN108664933A (en) | 2018-10-16 |
CN108664933B CN108664933B (en) | 2021-12-28 |
Family
ID=63779348
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810450109.3A Active CN108664933B (en) | 2018-05-11 | 2018-05-11 | Training method of convolutional neural network for SAR image ship classification, classification method of convolutional neural network and ship classification model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108664933B (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109583412A (en) * | 2018-12-07 | 2019-04-05 | 中国科学院遥感与数字地球研究所 | A kind of training method and its ship detecting method carrying out ship detecting using convolutional neural networks |
CN109766780A (en) * | 2018-12-20 | 2019-05-17 | 武汉理工大学 | A kind of ship smog emission on-line checking and method for tracing based on deep learning |
CN110415224A (en) * | 2019-07-22 | 2019-11-05 | 北京金交信息通信导航设计院 | A kind of marine ships remote sense monitoring system and platform and method |
CN110427981A (en) * | 2019-07-11 | 2019-11-08 | 四川大学 | SAR ship detecting system and method based on deep neural network |
CN110633353A (en) * | 2019-07-29 | 2019-12-31 | 南京莱斯网信技术研究院有限公司 | Ship type counterfeit monitoring method based on ensemble learning |
CN116310516B (en) * | 2023-02-20 | 2023-11-21 | 交通运输部水运科学研究所 | Ship classification method and device |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5057843A (en) * | 1990-06-25 | 1991-10-15 | The United States Of America As Represented By The Administrator Of The National Aeronautics And Space Administration | Method for providing a polarization filter for processing synthetic aperture radar image data |
CN106408030A (en) * | 2016-09-28 | 2017-02-15 | 武汉大学 | SAR image classification method based on middle lamella semantic attribute and convolution neural network |
CN107563999A (en) * | 2017-09-05 | 2018-01-09 | 华中科技大学 | A kind of chip defect recognition methods based on convolutional neural networks |
CN107563422A (en) * | 2017-08-23 | 2018-01-09 | 西安电子科技大学 | A kind of polarization SAR sorting technique based on semi-supervised convolutional neural networks |
CN107886123A (en) * | 2017-11-08 | 2018-04-06 | 电子科技大学 | A kind of synthetic aperture radar target identification method based on auxiliary judgement renewal learning |
-
2018
- 2018-05-11 CN CN201810450109.3A patent/CN108664933B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5057843A (en) * | 1990-06-25 | 1991-10-15 | The United States Of America As Represented By The Administrator Of The National Aeronautics And Space Administration | Method for providing a polarization filter for processing synthetic aperture radar image data |
CN106408030A (en) * | 2016-09-28 | 2017-02-15 | 武汉大学 | SAR image classification method based on middle lamella semantic attribute and convolution neural network |
CN107563422A (en) * | 2017-08-23 | 2018-01-09 | 西安电子科技大学 | A kind of polarization SAR sorting technique based on semi-supervised convolutional neural networks |
CN107563999A (en) * | 2017-09-05 | 2018-01-09 | 华中科技大学 | A kind of chip defect recognition methods based on convolutional neural networks |
CN107886123A (en) * | 2017-11-08 | 2018-04-06 | 电子科技大学 | A kind of synthetic aperture radar target identification method based on auxiliary judgement renewal learning |
Non-Patent Citations (2)
Title |
---|
JINXIN LI 等: "《Classification of very high resolution SAR image based on convolutional neural network》", 《2017 INTERNATIONAL WORKSHOP ON REMOTE SENSING WITH INTELLIGENT PROCESSING (RSIP)》 * |
吴樊 等: "《SAR图像船只分类识别研究进展》", 《遥感技术与应用》 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109583412A (en) * | 2018-12-07 | 2019-04-05 | 中国科学院遥感与数字地球研究所 | A kind of training method and its ship detecting method carrying out ship detecting using convolutional neural networks |
CN109766780A (en) * | 2018-12-20 | 2019-05-17 | 武汉理工大学 | A kind of ship smog emission on-line checking and method for tracing based on deep learning |
CN110427981A (en) * | 2019-07-11 | 2019-11-08 | 四川大学 | SAR ship detecting system and method based on deep neural network |
CN110427981B (en) * | 2019-07-11 | 2023-01-31 | 四川大学 | SAR ship detection system and method based on deep neural network |
CN110415224A (en) * | 2019-07-22 | 2019-11-05 | 北京金交信息通信导航设计院 | A kind of marine ships remote sense monitoring system and platform and method |
CN110633353A (en) * | 2019-07-29 | 2019-12-31 | 南京莱斯网信技术研究院有限公司 | Ship type counterfeit monitoring method based on ensemble learning |
CN110633353B (en) * | 2019-07-29 | 2020-05-19 | 南京莱斯网信技术研究院有限公司 | Ship type counterfeit monitoring method based on ensemble learning |
CN116310516B (en) * | 2023-02-20 | 2023-11-21 | 交通运输部水运科学研究所 | Ship classification method and device |
Also Published As
Publication number | Publication date |
---|---|
CN108664933B (en) | 2021-12-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108664933A (en) | The training method and its sorting technique of a kind of convolutional neural networks for SAR image ship classification, ship classification model | |
CN108399362A (en) | A kind of rapid pedestrian detection method and device | |
CN104834933B (en) | A kind of detection method and device in saliency region | |
CN108960135B (en) | Dense ship target accurate detection method based on high-resolution remote sensing image | |
CN109840560B (en) | Image classification method based on clustering in capsule network | |
CN107341518A (en) | A kind of image classification method based on convolutional neural networks | |
CN107403141A (en) | Method for detecting human face and device, computer-readable recording medium, equipment | |
CN106874956A (en) | The construction method of image classification convolutional neural networks structure | |
CN111222519B (en) | Construction method, method and device of hierarchical colored drawing manuscript line extraction model | |
CN109271990A (en) | A kind of semantic segmentation method and device for RGB-D image | |
CN108961220B (en) | Image collaborative saliency detection method based on multilayer convolution feature fusion | |
CN110188863B (en) | Convolution kernel compression method of convolution neural network suitable for resource-limited equipment | |
CN112668630B (en) | Lightweight image classification method, system and equipment based on model pruning | |
CN106855996B (en) | Gray-scale image coloring method and device based on convolutional neural network | |
CN103761526B (en) | Urban area detecting method based on feature position optimization and integration | |
CN110059539A (en) | A kind of natural scene text position detection method based on image segmentation | |
CN110245711A (en) | The SAR target identification method for generating network is rotated based on angle | |
CN111062441A (en) | Scene classification method and device based on self-supervision mechanism and regional suggestion network | |
Doi et al. | The effect of focal loss in semantic segmentation of high resolution aerial image | |
CN109919252A (en) | The method for generating classifier using a small number of mark images | |
CN111161201A (en) | Infrared and visible light image fusion method based on detail enhancement channel attention | |
CN113160062A (en) | Infrared image target detection method, device, equipment and storage medium | |
CN110135446A (en) | Method for text detection and computer storage medium | |
CN109636764A (en) | A kind of image style transfer method based on deep learning and conspicuousness detection | |
CN116740344A (en) | Knowledge distillation-based lightweight remote sensing image semantic segmentation method and device |
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 | ||
TA01 | Transfer of patent application right | ||
TA01 | Transfer of patent application right |
Effective date of registration: 20211124 Address after: 100190 No. 19 West North Fourth Ring Road, Haidian District, Beijing Applicant after: Aerospace Information Research Institute,Chinese Academy of Sciences Address before: No. 9 Dengzhuang South Road, Haidian District, Beijing 100094 Applicant before: Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences |
|
GR01 | Patent grant | ||
GR01 | Patent grant |