CN109919950A - A kind of unmanned boat waterborne target image-recognizing method based on genetic neural network - Google Patents
A kind of unmanned boat waterborne target image-recognizing method based on genetic neural network Download PDFInfo
- Publication number
- CN109919950A CN109919950A CN201910186049.3A CN201910186049A CN109919950A CN 109919950 A CN109919950 A CN 109919950A CN 201910186049 A CN201910186049 A CN 201910186049A CN 109919950 A CN109919950 A CN 109919950A
- Authority
- CN
- China
- Prior art keywords
- neural network
- image
- genetic
- target image
- waterborne 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
Landscapes
- Image Analysis (AREA)
Abstract
The present invention relates to unmanned boat waterborne target image identification technical fields, and in particular to a kind of unmanned boat waterborne target image-recognizing method based on genetic neural network.The optimal threshold of foreground image and background image in a differentiation Surface Picture is arranged according to the grey level histogram of Surface Picture first, then learning training is carried out to sample using genetic neural network, finally identify waterborne target image, the waterborne target image after display identification；The present invention guarantees waterborne target in the accuracy after over-segmentation, to keep the complete and clear of waterborne target to greatest extent by the way that waterborne target image segmentation is come out.
Description
Technical field
The present invention relates to unmanned boat waterborne target image identification technical fields, and in particular to one kind is based on genetic neural network
Unmanned boat waterborne target image-recognizing method.
Background technique
Unmanned boat (USV) is a kind of collection contexture by self, and autonomous navigation independently completes environment sensing, the functions such as target acquisition
The small-size water surface motion platform being integrated, it has also become explore the essential equipment of marine resources.
Since the research of waterborne target image-recognizing method is the key point of unmanned boat target identification, key component is
The image information acquired using unmanned boat, identifies naval target.
The waterborne target image recognition technology of unmanned boat reflects maritime affairs unmanned boat intelligent level to a certain extent
It just, is one of the important research content of unmanned boat key technology area.The research enhances sea area to motherland's national defense construction is reinforced
Safety and raising China sea investigation ability have important practical application value and theoretical significance
Summary of the invention
The purpose of the present invention is to provide a kind of unmanned boat waterborne target image-recognizing method based on genetic neural network,
By the way that waterborne target image segmentation is come out, guarantee waterborne target in the accuracy after over-segmentation, to keep to greatest extent
Waterborne target it is complete and clear.
The embodiment of the present invention provides a kind of unmanned boat waterborne target image-recognizing method based on genetic neural network, packet
It includes:
Step 1: setting threshold value: before being arranged in a differentiation Surface Picture according to the grey level histogram of Surface Picture first
Then the optimal threshold of scape image and background image determines foreground image in Surface Picture according to grey level histogram data information
Two pieces of image information is sequentially placed into training sample array P by tonal range and background image tonal range, while being arranged one
A array T identical with its size, array T are used to save the classification of sample data, and indicate foreground image classification with 0, with 1 table
Show background image classification；
Step 2: sample learning: learning training is carried out to sample using genetic neural network, relevant parameter is carried out first
Coding and optimization, then select suitable fitness function, finally choose intermediate recombination interior extrapolation method as crossover operator, select simultaneously
It takes Gaussian mutation method as mutation operator, obtains optimal genetic neural network；
Step 3: identification waterborne target image: waterborne target image is read, the image pixel matrix is obtained, by it
Dimension-reduction treatment is carried out, input vector is obtained；Input vector is trained using trained genetic neural network before, is obtained
Output vector, i.e., the result classified to Surface Picture；The data are sent in genetic neural network and are classified, are obtained
Output valve classifies to the pixel of Surface Picture；Final classification data is reduced to figure from the form of one-dimensional vector array
Waterborne target image as matrix form, after then showing identification；
The step 1, comprising:
Threshold value is set: first according to the grey level histogram of Surface Picture be arranged a differentiation Surface Picture in foreground image with
And the optimal threshold of background image, then foreground image tonal range in Surface Picture is determined according to grey level histogram data information
And background image tonal range, two pieces of image information is sequentially placed into training sample array P, while being arranged one big with it
Small identical array T, array T are used to save the classification of sample data, and indicate that foreground image classification, use 1 indicate Background with 0
As classification；
Wherein, the step method particularly includes:
According to information shown by the grey level histogram of Surface Picture, one-dimension array is formed according to gray level [0,256], and
It is indicated with array P, and array T is the storage vector an of target value, will distinguish the gray scale of foreground image and background image
Value is set as gray level 175, according to the related data of later period learning training, finds and it is expected optimal threshold value；The ash of Surface Picture
Gray level in degree image indicates with 0~255, and using 0~255 grade of gray value as the feature of sample data, i.e. P=[0:1:
255]；Then one target array is set store classification, that is, T=zeros (1,256) of sample, while each sample number is set
According to classification T (175:256)=1, i.e., set 1 more than or equal to 175 classification for gray value, gray value is less than 175
Default is considered as 0 by classification；
The step 2, comprising:
Sample learning: using genetic neural network to sample carry out learning training, first to relevant parameter carry out coding and
Optimization, then selects suitable fitness function, finally chooses intermediate recombination interior extrapolation method as crossover operator, while choosing Gauss
Alternative method obtains optimal genetic neural network as mutation operator；
Wherein,
(a) described that relevant parameter is encoded and is optimized method particularly includes: in genetic neural network, BP nerve net
Network passes the weight, the hidden layer and exporting in the threshold value and hidden layer of node layer that connect between each transport layer using genetic algorithm
The parameter of delivery function optimizes, and in BP neural network, major parameter is converted to the form of coding, selected coding staff
Formula is real coding；
It is encoded using real number using the process of chromosome coding, is retouched respectively using R × 1, S1, S2 × 1 first
The dimension for stating the dimension of input vector in genetic neural network algorithm, the quantity of hidden layer node and output vector, then divides
Input layer in genetic neural network algorithm is not described to the connection weight of hidden layer, input layer to implicit using W1, b1, W2, b2
Threshold value, the connection weight of hidden layer to output layer and the hidden layer of layer obtain following coding mode to the threshold value of output layer:
Using above-mentioned rule, the length of chromosome is described with following formula in genetic neural network:
S=R × S1+S1 × S2+S1+S2
The chromosome coding eventually formed is as shown in table 1:
The signal of 1 chromosome coding of table
(b) selection suitable fitness function method particularly includes: judge that population at individual is using fitness evaluation
It is no successfully to be selected and then be genetic to the next generation, it joined genetic algorithm in BP neural network algorithm, in selection heredity
When the fitness function of neural network algorithm, fitness function is indicated with following formula:
In above formula, fiIndicate fitness value corresponding to individual i, size is the inverse of error sum of squares E, with following formula table
Show:
In above formula, i represents the number of chromosome in network, and p represents its number of training, and q represents the number of nodes of output, Yq
The reality output of the network is represented,The desired output of the network is represented, the fitness function according to defined in above formula is selected
It selects, obtains the smallest network model of error of reality output and desired output；
(c) described to obtain optimal genetic neural network method particularly includes: to choose intermediate recombination interior extrapolation method and be used as and intersect
Operator, while Gaussian mutation method is chosen as mutation operator, determining sample value and characteristic value are put into genetic neural network
Be trained, set in the spatial dimension with threshold value in power, first genetic neural network find out weight that one group is best suitable for condition and
Threshold value, the weight and threshold value used when most starting to calculate as network；Then genetic neural network is trained, Zhi Daoman
One of sufficient the following conditions: first is that mean square error needs to converge to specified numerical value, second is that genetic neural network reaches greatest iteration time
Number；Obtain optimal genetic neural network；
The step 3, comprising:
It identifies waterborne target image: reading waterborne target image, the image pixel matrix is obtained, by carrying out dimensionality reduction to it
Processing, obtains input vector；Input vector is trained using trained genetic neural network before, obtain exporting to
Amount, i.e., the result classified to Surface Picture；The data are sent in genetic neural network and are classified, output valve is obtained
Classify to the pixel of Surface Picture；Final classification data is reduced to image array from the form of one-dimensional vector array
Form, the waterborne target image after then showing identification；
Wherein, the identification waterborne target image method particularly includes:
The output vector classified to Surface Picture is obtained, each data classified are one in image T
A corresponding pixel Tij；It sends the data in genetic neural network gann and classifies, obtain output valve OijTo the water surface
The pixel of target image is classified, i.e.,
Oij=gann (Tij)
Wherein, F indicates the target area of waterborne target image, and B indicates the background area of waterborne target image, and T indicates to know
Waterborne target image after not.
The beneficial effects of the present invention are:
1. the present invention optimizes it using genetic algorithm for deficiency existing for BP neural network algorithm, and
Genetic neural network algorithm after optimization processing is applied in the identification of waterborne target image, utilization is finally realized
Identification of the genetic neural network algorithm to waterborne target image；
2. the present invention identification when not only waterborne target is split, also at the same guarantee waterborne target through over-segmentation with
Accuracy afterwards, to greatest extent keep waterborne target it is complete with it is clear.
Detailed description of the invention
Fig. 1 is the original image of Surface Picture of the invention；
Fig. 2 is the grey level histogram of Surface Picture of the present invention；
Fig. 3 is genetic neural network flow chart of the present invention；
Specific embodiment
In order to make the foregoing objectives, features and advantages of the present invention clearer and more comprehensible, with reference to the accompanying drawing to the present invention
It is described further:
The technical scheme of the present invention is realized as follows:
1. threshold value is arranged
Foreground image and background image can be distinguished by finding one first with the grey level histogram of Surface Picture
Then the optimal threshold come determines the tonal range and back of prospect in Surface Picture according to the data information of grey level histogram
The tonal range of scape is then sequentially placed into the information of foreground and background in array P, and the array of generation is defined as number of training
Group, while an array T identical with its size need to be set again, array T is used to save the classification of sample data, and is indicated with 0
The classification of prospect, use 1 indicate the classification of background.Fig. 1 is original Surface Picture, and Fig. 2 is the histogram of Fig. 1.
The information according to histogram, wherein can according to gray level [0,256] form one-dimension array, and with P into
Row indicates, and T is the storage vector an of target value, and the gray value that prospect is distinguished with background can be set
For gray level 175, then according to the related data of later period learning training, it will be able to find and it is expected optimal threshold value.Gray level image
In gray level indicated with 0~255, and using 0~255 grade of gray value as the feature of sample data, i.e. P=[0:1:255].
Then one target array is set to store the classification T=zeros (1,256) of sample, while needing to be arranged each sample number
According to classification T (175:256)=1, i.e., gray value is arranged to 1 more than or equal to 175 classification, and due to establishing before
Array be one all zero array, therefore classification of the gray value less than 175 will be considered as 0 by default.
2. sample learning
Genetic neural network is trained first.Fig. 3 is genetic neural network flow chart, and training process is as follows:
(a) in genetic neural network, BP neural network is needed using genetic algorithm to the power connected between each transport layer
The relevant parameters such as parameter of transmission function optimize in value, the threshold value of hidden layer and output node layer and hidden layer, but lose
Propagation algorithm needs encode individual to handle related data, therefore in BP neural network, need some major parameters
It is converted to the form of coding.And coding mode selected by this paper is real coding, it is excellent compared with traditional binary coding
Point is that real coding length is shorter and encoding precision is higher.
It illustrates how to encode it using real number below with the process of chromosome coding.Use R respectively first
× 1, S1, S2 × 1 describe the dimension of input vector in genetic neural network algorithm, the quantity of hidden layer node and output
Then the dimension of vector describes in genetic neural network algorithm input layer to the company of hidden layer using W1, b1, W2, b2 respectively
Connect weight, the threshold value of input layer to hidden layer, the connection weight of hidden layer to output layer and hidden layer to output layer threshold value,
Following coding mode then can be obtained:
Using above-mentioned rule, the length of chromosome can be described with following formula in genetic neural network:
S=R × S1+S1 × S2+S1+S2
The chromosome coding eventually formed is as shown in table 1:
The signal of 1 chromosome coding of table
(b) selection of fitness function
Whether population at individual successfully can be selected and then be genetic to the next generation, and sole criterion is exactly to be commented using fitness
Valence is judged.Therefore, selecting the fitness function of which seed type calculate is that genetic neural network algorithm to be faced
One major issue.Due to joined genetic algorithm in BP neural network algorithm, and the basic goal done so be exactly in order to
The operation efficiency of BP neural network is improved, so how will comment when selecting the fitness function of genetic neural network algorithm
Estimate given this key factor of neural network performance to take into account.Therefore, the fitness function that the present invention selects can with following formula into
Row indicates:
Wherein, fiIndicate fitness value corresponding to individual i, size is the inverse of error sum of squares E, with following formula table
Show:
Wherein, i represents the number of chromosome in network, and p represents its number of training, and q represents the number of nodes of output, YqGeneration
The reality output of the table network, andThen represent the desired output of the network.The fitness function according to defined in above formula carries out
Selection, can obtain the smallest network model of error of reality output and desired output.
(c) intermediate recombination interior extrapolation method is chosen as crossover operator；Gaussian mutation method is chosen as mutation operator.
The sample value having determined before is put into genetic neural network with characteristic value and is trained.Genetic neural network
First have to carry out is set in the spatial dimension with threshold value in power, and one group of weight and threshold value for being best suitable for condition is arbitrarily found out, and
Using these data weight and threshold value used when most starting to calculate as network；It then is next exactly to be trained to it,
Until meeting one of the following conditions: first is that mean square error needs to converge to specified numerical value, second is that the network had reached the maximum
The number of iterations.If training can satisfy any of the above conditions, mean that neural network at this time is optimal.
3. waterborne target image recognition
Image recognition based on genetic neural network algorithm mainly includes two parts, first is that utilizing genetic neural network pair
Sample carries out learning training, second is that identified according to training result to waterborne target image, wherein to waterborne target image into
Row identification can be mainly divided into three steps:
(a) waterborne target image is read, obtains the picture element matrix of the image, and dimension-reduction treatment is carried out to the picture element matrix,
Available input vector；
(b) trained genetic neural network is trained input vector before utilizing, and obtained output vector is exactly
To Surface Picture classify as a result, and each data classified are a corresponding pixels in image T
Point Tij.It sends these data in genetic neural network gann and classifies, an available output valve Oij, then utilize
This output valve classifies to the pixel of waterborne target image, i.e.,
Oij=gann (Tij)
Wherein, F indicates the target area of waterborne target image, and B indicates the background area of waterborne target image, and T indicates to know
Waterborne target image after not.
(c) final classification data is reduced to image array form from the form of one-dimensional vector array, then shows knowledge
Waterborne target image after not.
Claims (4)
1. a kind of unmanned boat waterborne target image-recognizing method based on genetic neural network characterized by comprising
Step 1: foreground picture in a differentiation Surface Picture setting threshold value: is arranged according to the grey level histogram of Surface Picture first
Then the optimal threshold of picture and background image determines foreground image gray scale in Surface Picture according to grey level histogram data information
Two pieces of image information is sequentially placed into training sample array P by range and background image tonal range, at the same be arranged one with
Its size identical array T, array T are used to save the classification of sample data, and indicate that foreground image classification, use 1 indicate to carry on the back with 0
Scape image category；
Step 2: sample learning: learning training is carried out to sample using genetic neural network, relevant parameter is encoded first
And optimization, suitable fitness function is then selected, finally chooses intermediate recombination interior extrapolation method as crossover operator, while choosing height
This alternative method obtains optimal genetic neural network as mutation operator；
Step 3: identification waterborne target image: waterborne target image is read, the image pixel matrix is obtained, by carrying out to it
Dimension-reduction treatment obtains input vector；Input vector is trained using trained genetic neural network before, is exported
Vector, i.e., the result classified to Surface Picture；The data are sent in genetic neural network and are classified, are exported
Value classifies to the pixel of Surface Picture；Final classification data is reduced to image moment from the form of one-dimensional vector array
Formation formula, the waterborne target image after then showing identification.
2. a kind of unmanned boat waterborne target image-recognizing method based on genetic neural network according to claim 1,
It is characterized in that, the step 1, comprising:
Threshold value is set: foreground image and back in a differentiation Surface Picture being arranged according to the grey level histogram of Surface Picture first
The optimal threshold of scape image, then according to grey level histogram data information determine in Surface Picture foreground image tonal range and
Two pieces of image information is sequentially placed into training sample array P by background image tonal range, while one and its size phase is arranged
Same array T, array T are used to save the classification of sample data, and indicate that foreground image classification, use 1 indicate background image class with 0
Not；
Wherein, the step method particularly includes:
According to information shown by the grey level histogram of Surface Picture, one-dimension array is formed according to gray level [0,256], and use number
Group P is indicated, and array T is the storage vector an of target value, and the gray value for distinguishing foreground image and background image is set
It is set to gray level 175, according to the related data of later period learning training, finds and it is expected optimal threshold value；The grayscale image of Surface Picture
Gray level as in is indicated with 0~255, and using 0~255 grade of gray value as the feature of sample data, i.e. P=[0:1:255]；
Then one target array is set store classification, that is, T=zeros (1,256) of sample, while each sample data is set
Classification T (175:256)=1, the i.e. classification by gray value more than or equal to 175 are set as 1, classification of the gray value less than 175
Default is considered as 0.
3. a kind of unmanned boat waterborne target image-recognizing method based on genetic neural network according to claim 1,
It is characterized in that: the step 2, comprising:
Sample learning: carrying out learning training to sample using genetic neural network, encoded and optimized to relevant parameter first,
Then suitable fitness function is selected, finally chooses intermediate recombination interior extrapolation method as crossover operator, while choosing Gaussian mutation
Method obtains optimal genetic neural network as mutation operator；
Wherein,
(a) described that relevant parameter encoded and is optimized method particularly includes: in genetic neural network, BP neural network benefit
With genetic algorithm to the weight connected between each transport layer, hidden layer and export in the threshold value and hidden layer of node layer and transmit letter
Several parameters optimize, and in BP neural network, major parameter are converted to the form of coding, selected coding mode is
Real coding；
It is encoded using real number using the process of chromosome coding, is lost respectively using R × 1, the description of S1, S2 × 1 first
The dimension for passing the dimension of input vector in neural network algorithm, the quantity of hidden layer node and output vector, then makes respectively
Connection weight to hidden layer of input layer in genetic neural network algorithm, input layer to hidden layer are described with W1, b1, W2, b2
Threshold value, the connection weight of hidden layer to output layer and hidden layer obtain following coding mode to the threshold value of output layer:
Using above-mentioned rule, the length of chromosome is described with following formula in genetic neural network:
S=R × S1+S1 × S2+S1+S2
The chromosome coding eventually formed is as shown in table 1:
1 chromosome coding schematic diagram of table
(b) the suitable fitness function of the selection method particularly includes: using fitness evaluation judge that population at individual whether can
The next generation is successfully selected and then be genetic to, joined genetic algorithm in BP neural network algorithm, in selection genetic nerve
When the fitness function of network algorithm, fitness function is indicated with following formula:
In above formula, fiIndicating fitness value corresponding to individual i, size is the inverse of error sum of squares E, it is indicated with following formula:
In above formula, i represents the number of chromosome in network, and p represents its number of training, and q represents the number of nodes of output, YqIt represents
The reality output of the network,The desired output of the network is represented, the fitness function according to defined in above formula is selected,
Obtain the smallest network model of error of reality output and desired output；
(c) described to obtain optimal genetic neural network method particularly includes: choose intermediate recombination interior extrapolation method as crossover operator,
Gaussian mutation method is chosen as mutation operator simultaneously, and determining sample value is put into genetic neural network with characteristic value and is instructed
Practice, set in the spatial dimension with threshold value in power, genetic neural network finds out one group of weight and threshold value for being best suitable for condition first, makees
For the network weight and threshold value used when most starting to calculate；Then genetic neural network is trained, it is following until meeting
One of condition: first is that mean square error needs to converge to specified numerical value, second is that genetic neural network reaches maximum number of iterations；It obtains
Optimal genetic neural network.
4. a kind of unmanned boat waterborne target image-recognizing method based on genetic neural network according to claim 1,
It is characterized in that: the step 2, comprising:
It identifies waterborne target image: reading waterborne target image, the image pixel matrix is obtained, by carrying out at dimensionality reduction to it
Reason, obtains input vector；Input vector is trained using trained genetic neural network before, obtains output vector,
The result classified to Surface Picture；The data are sent in genetic neural network and are classified, output valve pair is obtained
The pixel of Surface Picture is classified；Final classification data is reduced to image moment formation from the form of one-dimensional vector array
Formula, the waterborne target image after then showing identification；
Wherein, the identification waterborne target image method particularly includes:
Obtain the output vector classified to Surface Picture, each data classified be in image T one it is right
The pixel T answeredij；It sends the data in genetic neural network gann and classifies, obtain output valve OijTo waterborne target
The pixel of image is classified, i.e.,
Oij=gann (Tij)
Wherein, after F indicates that the target area of waterborne target image, B indicate that the background area of waterborne target image, T indicate identification
Waterborne target image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910186049.3A CN109919950A (en) | 2019-03-12 | 2019-03-12 | A kind of unmanned boat waterborne target image-recognizing method based on genetic neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910186049.3A CN109919950A (en) | 2019-03-12 | 2019-03-12 | A kind of unmanned boat waterborne target image-recognizing method based on genetic neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109919950A true CN109919950A (en) | 2019-06-21 |
Family
ID=66964440
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910186049.3A Pending CN109919950A (en) | 2019-03-12 | 2019-03-12 | A kind of unmanned boat waterborne target image-recognizing method based on genetic neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109919950A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111027568A (en) * | 2019-10-30 | 2020-04-17 | 中国科学院自动化研究所南京人工智能芯片创新研究院 | Iterative neural network batch normalization system |
| CN114092701A (en) * | 2021-12-04 | 2022-02-25 | 特斯联科技集团有限公司 | Intelligent symbol identification method based on neural network |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105488528A (en) * | 2015-11-26 | 2016-04-13 | 北京工业大学 | Improved adaptive genetic algorithm based neural network image classification method |
| CN106023195A (en) * | 2016-05-18 | 2016-10-12 | 河南师范大学 | BP neural network image segmentation method and device based on adaptive genetic algorithm |
| CN106338919A (en) * | 2016-11-02 | 2017-01-18 | 哈尔滨工程大学 | USV (Unmanned Surface Vehicle) track tracking control method based on enhanced learning type intelligent algorithm |
-
2019
- 2019-03-12 CN CN201910186049.3A patent/CN109919950A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105488528A (en) * | 2015-11-26 | 2016-04-13 | 北京工业大学 | Improved adaptive genetic algorithm based neural network image classification method |
| CN106023195A (en) * | 2016-05-18 | 2016-10-12 | 河南师范大学 | BP neural network image segmentation method and device based on adaptive genetic algorithm |
| CN106338919A (en) * | 2016-11-02 | 2017-01-18 | 哈尔滨工程大学 | USV (Unmanned Surface Vehicle) track tracking control method based on enhanced learning type intelligent algorithm |
Non-Patent Citations (1)
| Title |
|---|
| 王汝珣: "基于形状外观的无人船目标识别方法研究", 《中国优秀博硕士学位论文全文数据库(硕士) 信息科技辑》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111027568A (en) * | 2019-10-30 | 2020-04-17 | 中国科学院自动化研究所南京人工智能芯片创新研究院 | Iterative neural network batch normalization system |
| CN114092701A (en) * | 2021-12-04 | 2022-02-25 | 特斯联科技集团有限公司 | Intelligent symbol identification method based on neural network |
| CN114092701B (en) * | 2021-12-04 | 2022-06-03 | 特斯联科技集团有限公司 | Intelligent symbol identification method based on neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111583263B (en) | Point cloud segmentation method based on joint dynamic graph convolution | |
| CN105488528B (en) | Neural network image classification method based on improving expert inquiry method | |
| CN107392919B (en) | Adaptive genetic algorithm-based gray threshold acquisition method and image segmentation method | |
| CN107133690B (en) | Optimal ordering method for river and lake water system communication engineering scheme | |
| CN111723675A (en) | Remote sensing image scene classification method based on multiple similarity measurement deep learning | |
| CN112347970B (en) | Remote sensing image ground object identification method based on graph convolution neural network | |
| CN112115967B (en) | Image increment learning method based on data protection | |
| CN114092697B (en) | Building facade semantic segmentation method with attention fused with global and local depth features | |
| CN108805193A (en) | A kind of power loss data filling method based on mixed strategy | |
| CN112084877B (en) | NSGA-NET-based remote sensing image recognition method | |
| CN112580662A (en) | Method and system for recognizing fish body direction based on image features | |
| CN112819063B (en) | Image identification method based on improved Focal loss function | |
| CN109919950A (en) | A kind of unmanned boat waterborne target image-recognizing method based on genetic neural network | |
| CN113487600A (en) | Characteristic enhancement scale self-adaptive sensing ship detection method | |
| CN115170943A (en) | Improved visual transform seabed substrate sonar image classification method based on transfer learning | |
| CN115292532A (en) | Remote sensing image domain adaptive retrieval method based on pseudo label consistency learning | |
| CN116543269B (en) | Cross-domain small sample fine granularity image recognition method based on self-supervision and model thereof | |
| CN109271833A (en) | Target identification method, device and electronic equipment based on the sparse self-encoding encoder of stack | |
| CN116681921A (en) | Target labeling method and system based on multi-feature loss function fusion | |
| CN116994071A (en) | Multispectral laser radar point cloud classification method based on self-adaptive spectrum residual error | |
| CN116956160A (en) | Data classification prediction method based on self-adaptive tree species algorithm | |
| CN114581470B (en) | Image edge detection method based on plant community behaviors | |
| Wan et al. | RSSM-Net: Remote sensing image scene classification based on multi-objective neural architecture search | |
| CN111191717B (en) | Black box confrontation sample generation algorithm based on hidden space clustering | |
| Jia et al. | Lightweight CNN-Based Image Recognition with Ecological IoT Framework for Management of Marine Fishes |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190621 |