CN109190695A - A kind of fish image classification method based on depth convolutional neural networks - Google Patents
A kind of fish image classification method based on depth convolutional neural networks Download PDFInfo
- Publication number
- CN109190695A CN109190695A CN201810984130.1A CN201810984130A CN109190695A CN 109190695 A CN109190695 A CN 109190695A CN 201810984130 A CN201810984130 A CN 201810984130A CN 109190695 A CN109190695 A CN 109190695A
- Authority
- CN
- China
- Prior art keywords
- fish
- image
- fish image
- network
- neural networks
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
Abstract
The embodiment of the present application provides a kind of fish image classification method based on depth convolutional neural networks, is related to underwater technical field of computer vision.This method acquires underwater fish image, constructs fish image data set；Choose pre-training collection ImageNet and fish data set Fish4Knowledge；The convolutional neural networks model based on B-CNNS is constructed, successively three data sets are input in the network model by sample size size, by trained and iterative feedback, obtains the class label of the fish image data set Mesichthyes image.Ideal fish recognition accuracy can be realized by sample size less data collection for this method, provide the foundation for the further research of the fish ecosystem.
Description
Technical field
This application involves underwater technical field of computer vision more particularly to a kind of fishes based on depth convolutional neural networks
Class image classification method.
Background technique
Fish image classification is of great significance to the marine ecosystems research especially fish ecosystem.In fish
In habitat, there is certain difficulty with a large amount of clear images that equipment obtains fish.Since habitat is complicated and fish sheet
The reason of protective coloration of body, all constrains the development of the technologies such as fish image classification.
The correct classification to fish image is completed for the fish ecosystem by the fish image data set of limited quantity
Research be of great significance.However, the method for current fish image classification is less, and generally require more training sample
Amount.
Summary of the invention
This application provides a kind of fish image classification methods based on depth convolutional neural networks, for solving existing skill
Art Mesichthyes image classification method is few, and more demanding to data set condition, is unable to reach the technologies such as higher recognition accuracy
Ideal fish recognition accuracy can be realized by sample size less data collection for problem, this method, be fish ecology
The further research of system provides the foundation.
The first aspect of the embodiment of the present application provides a kind of fish image classification method based on depth convolutional neural networks,
Include the following steps:
S1: acquiring underwater fish image, constructs fish image data set；
Choose pre-training collection ImageNet and fish data set Fish4Knowledge；
S2: convolutional neural networks model of the building based on B-CNNS, by sample size size successively by three data
Collection is input in the network model, by trained and iterative feedback, obtains the fish image data set Mesichthyes image
Class label.
Further, step S2 is specifically included:
S21: the pre-training collection ImageNet is input to training in the network model until convergence, takes a little last
Full articulamentum；
S22: the fish data set Fish4Knowledge is input in the network model after step S21 training and is continued
Training is until convergence；
S23: the fish image data set is input in the network model after step S22 training, network parameter is carried out
Fine tuning.
Further, the network model includes the data enhancement unit and network structure elements after optimization, i.e., will step
Three data sets described in rapid S1 are input to the data enhancement unit after the optimization and obtain the enhanced fish image survey of data
Examination collection, is input to the network structure elements for the fish image measurement collection, obtains the fish image data and concentrates image
Class label.
Further, described three data sets described in step S1 are input to the data enhancement unit after the optimization to obtain
It to the enhanced fish image measurement collection of data, specifically includes: first passing through every piece image that three data are concentrated
SRGAN network carries out super-resolution rebuilding, and the image after the reconstruction is then carried out data enhancing.
Further, data enhancing include fold and or rotation.
Further, the network structure elements include the identical VGG network of two-way, described by the fish image measurement
Collection is input to the network structure elements, obtains the class label that the fish image data concentrates image, specifically includes:
The fish image measurement collection is input in two-way VGG network and carries out feature extraction, later by the two-way VGG
The feature that network extracts carries out bilinearity pond and obtains bilinearity vector, recently enters softmax function and obtains the fish
The class label of image data concentration image.
Further, the VGG network includes five groups of convolutional layers, wherein optimization is inserted into behind preceding four groups of convolutional layers
SE block afterwards.
Further, the SE block concrete operations after the optimization are as follows:
It is C that input picture x is obtained port number after convolutional layer1Characteristic pattern；
By each channel segmentation at four regions, wherein k-th of channel is by four real number zk1、zk2、zk3、zk4To characterize；
The weight in each channel is obtained by excitation operation study；
By each region in each channel of the characteristic pattern and the multiplied by weight learnt；
Dimension is adjusted, so that the dimension of output image is equal to the dimension of input picture.
Further, four real numbers are respectively as follows:
In formula, W × H is the characteristic dimension in k-th of channel in the characteristic spectrum for input SE block, and i is the change from (1, H)
Amount, j are the variable from (1, W), uk(i, j) is the k-th channel i for being input to the characteristic spectrum of SE block.
It is further, described that the weight in each channel is obtained by excitation operation study, specifically: s=σ (L2δ(L1,r
(z)) reshape operation r (z)), i.e., is carried out first, will characterize each channel dimension by C1* 4 switch to 4C1* 1, then pass through according to this
Full articulamentum L1, ReLU activation primitive δ, full articulamentum L2And sigmoid function σ operation, after trained and iteration, study
Obtain the weight in each channel.
Fish image classification method provided by the embodiments of the present application based on depth convolutional neural networks, it is existing for solving
Technology Mesichthyes image classification method is few, and more demanding to data set condition, is unable to reach the skills such as higher recognition accuracy
Ideal fish recognition accuracy can be realized by sample size less data collection for art problem, this method, raw for fish
The further research of state system provides the foundation.
Detailed description of the invention
In order to illustrate more clearly of the technical solution of the application, letter will be made to attached drawing needed in the embodiment below
Singly introduce, it should be apparent that, for those of ordinary skills, without any creative labor,
It is also possible to obtain other drawings based on these drawings.
Fig. 1 is flow diagram provided by the embodiments of the present application；
Fig. 2 is SE block comparison diagram after SE block provided by the embodiments of the present application and optimization.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Whole description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Embodiment
Referring to Fig. 1, the fish image classification method based on depth convolutional neural networks shown in the embodiment of the present application
Flow diagram includes the following steps:
S1: acquiring underwater fish image, constructs fish image data set；
Choose pre-training collection ImageNet and fish data set Fish4Knowledge；
Wherein, the fish image data set of building is that (Crotian data set is to adopt in Croatia to Crotian data set
The fish data set of collection is a public data collection), ImageNet is the large size for being used for the research of visual object identification software
Visible database, Fish4Knowledge are common fish databases (abbreviation F4K).
S2: building is based on B-CNNS (Bilinear Convolution Neural Network) bilinearity convolutional Neural
Three data sets are successively input in the network model by network model by sample size size, by training and repeatedly
Generation feedback, obtains the class label of the fish image data set Mesichthyes image.
Specifically:
S21: the pre-training collection ImageNet is input to training in the network model until convergence, removes last
Full articulamentum；
S22: the fish data set Fish4Knowledge is input in the network model after step S21 training and is continued
Training is until convergence；
S23: the fish image data set is input in the network model after step S22 training, network parameter is carried out
Fine tuning.
In the embodiment of the present application, the network model includes the data enhancement unit and network structure list after optimization
Three data sets described in step S1 are input to the data enhancement unit after the optimization and obtain the enhanced fish of data by member
The fish image measurement collection is input to the network structure elements, obtains the fish image data by class image measurement collection
Concentrate the class label of image.
Enhance flow chart, working principle as Fig. 1 left-half show the data after optimization specifically: first will be described
Every piece image that three data are concentrated carries out super-resolution rebuilding by SRGAN network, then by the image after the reconstruction
Carry out data enhancing.
Wherein, SRGAN (Photo-Realistic Single Image Super-Resolution Using a
Generative Adversarial Network, arxiv, 21Nov, 2016) production confrontation network (GAN) is asked for SR
Topic, super-resolution technique (Super-Resolution, abbreviation SR) refers to be reconstructed accordingly from the low-resolution image observed
High-definition picture.Because of the too low feature extraction that will affect image of resolution ratio, the embodiment of the present application utilizes SRGAN net
Network carries out super-resolution rebuilding, promotes the resolution ratio of image.
In the embodiment of the present application, by after the reconstruction image carry out data enhancing mainly include fold and or rotation,
Wherein, the fold includes horizontal fold and vertical fold, and the rotation includes rotating clockwise 90 °, 180 °, 270 °.
The right half part of Fig. 1 feeds the flow chart of network structure part, and the network structure elements include the identical VGG of two-way
Network, it is described that the fish image measurement collection is input to the network structure elements, it obtains the fish image data and concentrates
The class label of image.
After the image of the fish image measurement collection of data enhancing is input to network structure part, because VGG network
Entrance fixed size is 224*224, so, it is first the image of 224*224 size by the Image Adjusting of input, it is defeated respectively later
Enter to two-way VGG network, wherein VGG network has a very strong ability in feature extraction, entire VGG network by 16 convolutional layers and
Full articulamentum, including five groups of convolutional layers and 3 full articulamentums, i.e. 16=2+2+3+3+3+3.
In order to improve the ability in feature extraction of network, the embodiment of the present application is after insertion optimization behind preceding four groups of convolutional layers
SE block.
SE block after being illustrated in figure 2 original SE block (Squeeze-and-Excitation network structure block) and optimization
(Refined SE Block) comparison diagram.In figure, Conv Layer indicates that convolutional layer, Golbal Pooling indicate global pool
Change, FC indicates that full articulamentum, ReLU indicate that ReLU activation primitive, Sigmoid indicate that Sigmoid function, Scale indicate that dimension becomes
Change, Reshape expression is straightened, and Quarter&Pooling indicates to divide each channel cross, and then each square carries out
Pondization operation.
Wherein, the SE block concrete operations after optimization are as follows:
It is C that input picture x is obtained port number after convolutional layer1Characteristic pattern；
It is C that input picture x obtains port number after convolutional layer1Characteristic pattern, what is carried out first is squeeze operation, along
The sequence of feature space is compressed, and in original SE block, the method for use is that each two-dimensional feature channel is converted to one
A real number indicates, is distributed with it come the overall situation on characteristic feature channel, and the SE block after optimizing in the embodiment of the present application uses
Be the feature that a feature channel is indicated with four real numbers.
By each channel segmentation at four regions, wherein k-th of channel is by four real number zk1、zk2、zk3、zk4It characterizes,
It averages after summing respectively to all the points in each region；
In formula, W × H is the characteristic dimension in k-th of channel in the characteristic spectrum for input SE block, and i is the change from (1, H)
Amount, j are the variable from (1, W), uk(i, j) is k-th of channel i for being input to the characteristic spectrum of SE block.
The weight in each channel is obtained by excitation operation study；
Specifically: s=σ (L2δ(L1, r (z))), i.e., reshape operation r (z) is carried out first, will characterize each channel dimension
By C1* 4 switch to 4C1* 1, then pass through full articulamentum L according to this1, ReLU activation primitive δ, full articulamentum L2And sigmoid function
σ operation, after trained and iteration, study obtains the weight in each channel.
By each region in each channel of the characteristic pattern and the multiplied by weight learnt；
Dimension is adjusted, so that the dimension of output image is equal to the dimension of input picture.
SE block after optimization is embedded into after every group of convolution of VGG network, can effectively promote the feature extraction of network
Ability, and then network is promoted for the recognition capability of fish picture.
Feature that the two-way VGG network extracts is subjected to bilinearity pond and obtains bilinearity vector, enables network
The second order information for enough noticing characteristics of image recently enters softmax function and obtains the fish image data concentration image
Class label.
In the following, will be by testing further verifying effectiveness of the invention, validity, optimization SE including pre-training twice
The validity that the validity and SE block and B-CNNs of block combine.
1, the validity of pre-training twice:
Whether the several frequently seen convolutional network of table 1 is compared using the results of comparison of pre-training twice
As shown in table 1, this group experiment has chosen four relatively conventional convolutional neural networks: AlexNet, VGG-16,
Inception-v4, ResNet-50, and more detailed control experiment has been done with them.Specifically, each network is done
Four experiments are respectively from top to bottom:
(1) two datasets never use, and are directly trained with target data set Croatian fish data set.
(2) pre-training is carried out only with ImageNet data set, does not use F4K data set, finally use Croatian fish
The fine tuning of data set progress network parameter.
(3) ImageNet data set is not used, F4K data set is only used and carries out pre-training, finally use Croatian fish
The fine tuning of class data set progress network parameter.
(4) a pre-training first is carried out using ImageNet, carries out secondary pre-training with F4K again on this basis, finally
The fine tuning of network parameter is carried out with Croatian fish data set.
Can be seen that compared with without using training set from the accuracy rate result in table, no matter using which data set into
Row pre-training can all promote the accuracy rate of network, but it is not obvious enough to promote effect.Obviously, on each network, two have been used
Relatively high accuracy rate is all reached after secondary pre-training.Network obtains identification generic features by first time pre-training
Ability.The professional knowledge of fish characteristic aspect has been grasped by second of pre-training, therefore uses target data set Croatian again
Training can obtain higher accuracy rate.
2, optimize the validity of SE block.
Table 2 is inserted into the accuracy rate Comparative result of the SE block after original SE block and optimization over different networks
In order to illustrate the validity of the SE block after optimization, the embodiment of the present application is using very common in computer vision
Two datasets CIRAR10 and the flower data set of Google tested.In order to absolutely prove the present invention for SE
The effect of optimization of block, the embodiment of the present application have used three classical networks: Inception-v4, Inception_resnet_v2,
ResNext_v1, the SE block (rSE) after being wherein inserted into SE block and optimization, then illustrates it effectively by control experiment respectively
Property.
From Table 2, it can be seen that on three networks, after original SE block to be replaced with to the SE block after optimization, three nets
The accuracy rate of network on both data sets has certain promotion, and the SE block after illustrating optimization further improves network
Ability in feature extraction, to improve network for the classification capacity of fish picture.
3, validity of the SE block in conjunction with B-CNNs after optimizing.
The case where 3 heterogeneous networks module of table and different data Enhancement Method, compares
After the validity of SE block after illustrating optimization.In order to further apply it in fish classification problem, this
Application embodiment has selected to combine it and B-CNNs, because fish classification belongs to fine grit classification, and B-CNNs is exactly
It is specifically used to carry out the network of fine grit classification.So being inserted into behind every group of convolution of B-CNNs.In addition, in order to test
It whether effective demonstrate,proves the optimization enhanced for data, has done four groups of control experiments, on this also to illustrate the important of super-resolution
Property.
As can be seen from the table, B-CNNs accuracy rate after it joined SE block has certain promotion, further, will
Original SE block is changed to after the SE block after optimization, and network has certain promotion for the classification capacity of fish picture again.In addition, with
Before method use general data enhancing unlike, for the not high problem of target data set resolution ratio propose first by
SRGAN promotes image resolution ratio, then carries out the method for general data enhancing again.It is counted after having used optimization as can be seen from the table
It is all relatively preferable according to the effect of Enhancement Method network.
In addition, in order to illustrate the versatility that combines of SE block after B-CNNs and optimization, in addition to Croatian data set it
Outside, the embodiment of the present application has also done identical experiment on the fish data set of QUT:
4 heterogeneous networks module situation of table compares
Because the image resolution ratio of QUT fish data set is higher.So not using SRGAN technology, and only adopt
With common data enhancement methods.Here after SE block and B-CNNs after main or verifying SE block and optimization combine
Effect.As can be seen from the table, the SE block after optimization has especially been used with SE block to the promotion effect of network accuracy rate the most
Obviously.
It should be noted that when the embodiment of the present application refers to ordinal numbers such as " first ", " second ", " third " or " the 4th "
When, unless the based on context meaning of its certain order of representation, it is appreciated that being only to distinguish to be used.Term " packet
Include ", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion so that including the object of a series of elements
Product or equipment not only include those elements, but also including other elements that are not explicitly listed, or further include for this
The intrinsic element of kind process, method, article or equipment.In the absence of more restrictions, by sentence " including one
It is a ... " limit element, it is not excluded that there is also in addition in the process, method, article or apparatus that includes the element
Identical element.
The above is only the specific embodiment of the application, is made skilled artisans appreciate that or realizing this Shen
Please.Various modifications to these embodiments will be apparent to one skilled in the art, as defined herein
General Principle can be realized in other embodiments without departing from the spirit or scope of the application.Therefore, the application
It is not intended to be limited to the embodiments shown herein, and is to fit to and the principles and novel features disclosed herein phase one
The widest scope of cause.
It should be understood that the application is not limited to the content being described above, and its model can not departed from
It encloses and carry out various modifications and change.Scope of the present application is only limited by the accompanying claims.
Claims (10)
1. a kind of fish image classification method based on depth convolutional neural networks, which comprises the steps of:
S1: acquiring underwater fish image, constructs fish image data set；
Choose pre-training collection ImageNet and fish data set Fish4Knowledge；
S2: convolutional neural networks model of the building based on B-CNNS, it is successively that three data sets is defeated by sample size size
Enter into the network model, by trained and iterative feedback, obtains the classification of the fish image data set Mesichthyes image
Label.
2. the fish image classification method according to claim 1 based on depth convolutional neural networks, which is characterized in that step
Rapid S2 is specifically included:
S21: the pre-training collection ImageNet is input to training in the network model until convergence；
S22: the fish data set Fish4Knowledge is input in the network model after step S21 training and continues to train
Until convergence；
S23: the fish image data set is input to micro- to network parameter progress in the network model after step S22 training
It adjusts.
3. the fish image classification method according to claim 1 or 2 based on depth convolutional neural networks, feature exist
In the network model includes the data enhancement unit and network structure elements after optimization, i.e., by three described in step S1
Data set is input to the data enhancement unit after the optimization and obtains the enhanced fish image measurement collection of data, by the fish
Image measurement collection is input to the network structure elements, obtains the class label that the fish image data concentrates image.
4. the fish image classification method according to claim 3 based on depth convolutional neural networks, which is characterized in that institute
It states and three data sets described in step S1 are input to the data enhancement unit after the optimization obtain the enhanced fish of data
Image measurement collection, specifically includes: the every piece image for first concentrating three data carries out super-resolution by SRGAN network
Rate is rebuild, and the image after the reconstruction is then carried out data enhancing.
5. the fish image classification method according to claim 4 based on depth convolutional neural networks, which is characterized in that institute
State data enhancing include fold and or rotation.
6. the fish image classification method according to claim 3 based on depth convolutional neural networks, which is characterized in that institute
Stating network structure elements includes the identical VGG network of two-way, described that the fish image measurement collection is input to the network knot
Structure unit obtains the class label that the fish image data concentrates image, specifically includes:
The fish image measurement collection is input in two-way VGG network and carries out feature extraction, later by the two-way VGG network
The feature extracted carries out bilinearity pond and obtains bilinearity vector, recently enters softmax function and obtains the fish image
The class label of image in data set.
7. the fish image classification method according to claim 6 based on depth convolutional neural networks, which is characterized in that institute
Stating VGG network includes five groups of convolutional layers, wherein the SE block after insertion optimization behind preceding four groups of convolutional layers.
8. the fish image classification method according to claim 6 based on depth convolutional neural networks, which is characterized in that institute
SE block concrete operations after stating optimization are as follows:
It is C that input picture x is obtained port number after convolutional layer1Characteristic pattern；
By each channel segmentation at four regions, wherein k-th of channel is by four real number zk1、zk2、zk3、zk4To characterize；
The weight in each channel is obtained by excitation operation study；
By each region in each channel of the characteristic pattern and the multiplied by weight learnt；
Dimension is adjusted, so that the dimension of output image is equal to the dimension of input picture.
9. the fish image classification method according to claim 8 based on depth convolutional neural networks, which is characterized in that institute
Four real numbers are stated to be respectively as follows:
In formula, W × H is the characteristic dimension in k-th of channel in the characteristic spectrum for input SE block, and i is the variable from (1, H), and j is
Variable from (1, W), uk(i, j) is the characteristic spectrum for being input to SE block.
10. the fish image classification method according to claim 8 based on depth convolutional neural networks, which is characterized in that
It is described that the weight in each channel is obtained by excitation operation study, specifically: s=σ (L2δ(L1, r (z))), i.e., it carries out first
Reshape operates r (z), will characterize each channel dimension by C1* 4 switch to 4C1* 1, then pass through full articulamentum L according to this1、ReLU
Activation primitive δ, full articulamentum L2And sigmoid function σ operation, after trained and iteration, study obtains each channel
Weight.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810984130.1A CN109190695B (en) | 2018-08-28 | 2018-08-28 | Fish image classification method based on deep convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810984130.1A CN109190695B (en) | 2018-08-28 | 2018-08-28 | Fish image classification method based on deep convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109190695A true CN109190695A (en) | 2019-01-11 |
| CN109190695B CN109190695B (en) | 2021-08-03 |
Family
ID=64916074
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810984130.1A Active CN109190695B (en) | 2018-08-28 | 2018-08-28 | Fish image classification method based on deep convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109190695B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109886925A (en) * | 2019-01-19 | 2019-06-14 | 天津大学 | A kind of aluminium material surface defect inspection method that Active Learning is combined with deep learning |
| CN110298824A (en) * | 2019-06-17 | 2019-10-01 | 浙江省农业科学院 | Squid automatic classification method based on color image and convolutional neural networks technology |
| CN110322418A (en) * | 2019-07-11 | 2019-10-11 | 北京航空航天大学 | A kind of super-resolution image generates the training method and device of confrontation network |
| CN110543879A (en) * | 2019-08-20 | 2019-12-06 | 高新兴科技集团股份有限公司 | SSD target detection method based on SE module and computer storage medium |
| CN110766013A (en) * | 2019-09-25 | 2020-02-07 | 浙江农林大学 | Fish identification method and device based on convolutional neural network |
| CN111046967A (en) * | 2019-12-18 | 2020-04-21 | 江苏科技大学 | Underwater image classification method based on convolutional neural network and attention mechanism |
| CN111666852A (en) * | 2020-05-28 | 2020-09-15 | 天津大学 | Micro-expression double-flow network identification method based on convolutional neural network |
| CN112488003A (en) * | 2020-12-03 | 2021-03-12 | 深圳市捷顺科技实业股份有限公司 | Face detection method, model creation method, device, equipment and medium |
| CN112559791A (en) * | 2020-11-30 | 2021-03-26 | 广东工业大学 | Cloth classification retrieval method based on deep learning |
| CN112906002A (en) * | 2021-03-26 | 2021-06-04 | 山西三友和智慧信息技术股份有限公司 | Malicious software identification method based on deep learning |
| CN113052227A (en) * | 2021-03-22 | 2021-06-29 | 山西三友和智慧信息技术股份有限公司 | Pulmonary tuberculosis identification method based on SE-ResNet |
| US11157811B2 (en) | 2019-10-28 | 2021-10-26 | International Business Machines Corporation | Stub image generation for neural network training |
| CN114240686A (en) * | 2022-02-24 | 2022-03-25 | 深圳市旗扬特种装备技术工程有限公司 | Wisdom fishery monitoring system |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040184657A1 (en) * | 2003-03-18 | 2004-09-23 | Chin-Teng Lin | Method for image resolution enhancement |
| CN106096535A (en) * | 2016-06-07 | 2016-11-09 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | A kind of face verification method based on bilinearity associating CNN |
| CN106372648A (en) * | 2016-10-20 | 2017-02-01 | 中国海洋大学 | Multi-feature-fusion-convolutional-neural-network-based plankton image classification method |
| CN106650721A (en) * | 2016-12-28 | 2017-05-10 | 吴晓军 | Industrial character identification method based on convolution neural network |
| CN107292333A (en) * | 2017-06-05 | 2017-10-24 | 浙江工业大学 | A kind of rapid image categorization method based on deep learning |
| CN108038823A (en) * | 2017-12-06 | 2018-05-15 | 厦门美图之家科技有限公司 | Image-type becomes the training method of network model, image-type becomes method and computing device |
-
2018
- 2018-08-28 CN CN201810984130.1A patent/CN109190695B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040184657A1 (en) * | 2003-03-18 | 2004-09-23 | Chin-Teng Lin | Method for image resolution enhancement |
| CN106096535A (en) * | 2016-06-07 | 2016-11-09 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | A kind of face verification method based on bilinearity associating CNN |
| CN106372648A (en) * | 2016-10-20 | 2017-02-01 | 中国海洋大学 | Multi-feature-fusion-convolutional-neural-network-based plankton image classification method |
| CN106650721A (en) * | 2016-12-28 | 2017-05-10 | 吴晓军 | Industrial character identification method based on convolution neural network |
| CN107292333A (en) * | 2017-06-05 | 2017-10-24 | 浙江工业大学 | A kind of rapid image categorization method based on deep learning |
| CN108038823A (en) * | 2017-12-06 | 2018-05-15 | 厦门美图之家科技有限公司 | Image-type becomes the training method of network model, image-type becomes method and computing device |
Non-Patent Citations (5)
| Title |
|---|
| QINGHE ZHENG 等: "A Bilinear Multi-Scale Convolutional Neural Network for Fine-grained Object Classification", 《INTERNATIONAL JOURNAL OF COMPUTER SCIENCE》 * |
| TSUNG-YU LIN 等: "Bilinear CNN Models for Fine-grained Visual Recognition", 《2015 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION》 * |
| ZHENGUO NIE 等: "Deep Learning for Stress Field Prediction Using Convolutional Neural Networks", 《ARXIV》 * |
| 张潜 等: "基于Xception的细粒度图像分类", 《重庆大学学报》 * |
| 沈海鸿 等: "分类错误指导的分层B-CNN模型用于细粒度分类", 《中国图象图形学报》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109886925A (en) * | 2019-01-19 | 2019-06-14 | 天津大学 | A kind of aluminium material surface defect inspection method that Active Learning is combined with deep learning |
| CN110298824A (en) * | 2019-06-17 | 2019-10-01 | 浙江省农业科学院 | Squid automatic classification method based on color image and convolutional neural networks technology |
| CN110322418A (en) * | 2019-07-11 | 2019-10-11 | 北京航空航天大学 | A kind of super-resolution image generates the training method and device of confrontation network |
| CN110543879A (en) * | 2019-08-20 | 2019-12-06 | 高新兴科技集团股份有限公司 | SSD target detection method based on SE module and computer storage medium |
| CN110766013A (en) * | 2019-09-25 | 2020-02-07 | 浙江农林大学 | Fish identification method and device based on convolutional neural network |
| US11157811B2 (en) | 2019-10-28 | 2021-10-26 | International Business Machines Corporation | Stub image generation for neural network training |
| CN111046967A (en) * | 2019-12-18 | 2020-04-21 | 江苏科技大学 | Underwater image classification method based on convolutional neural network and attention mechanism |
| CN111666852A (en) * | 2020-05-28 | 2020-09-15 | 天津大学 | Micro-expression double-flow network identification method based on convolutional neural network |
| CN112559791A (en) * | 2020-11-30 | 2021-03-26 | 广东工业大学 | Cloth classification retrieval method based on deep learning |
| CN112488003A (en) * | 2020-12-03 | 2021-03-12 | 深圳市捷顺科技实业股份有限公司 | Face detection method, model creation method, device, equipment and medium |
| CN113052227A (en) * | 2021-03-22 | 2021-06-29 | 山西三友和智慧信息技术股份有限公司 | Pulmonary tuberculosis identification method based on SE-ResNet |
| CN112906002A (en) * | 2021-03-26 | 2021-06-04 | 山西三友和智慧信息技术股份有限公司 | Malicious software identification method based on deep learning |
| CN114240686A (en) * | 2022-02-24 | 2022-03-25 | 深圳市旗扬特种装备技术工程有限公司 | Wisdom fishery monitoring system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109190695B (en) | 2021-08-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109190695A (en) | A kind of fish image classification method based on depth convolutional neural networks | |
| Dong et al. | Remote sensing image super-resolution using novel dense-sampling networks | |
| CN110119780B (en) | Hyper-spectral image super-resolution reconstruction method based on generation countermeasure network | |
| CN108596248B (en) | Remote sensing image classification method based on improved deep convolutional neural network | |
| Lin et al. | Hyperspectral image denoising via matrix factorization and deep prior regularization | |
| CN109035260A (en) | A kind of sky areas dividing method, device and convolutional neural networks | |
| CN109063753A (en) | A kind of three-dimensional point cloud model classification method based on convolutional neural networks | |
| CN111325165B (en) | Urban remote sensing image scene classification method considering spatial relationship information | |
| CN107358293A (en) | A kind of neural network training method and device | |
| CN109871830A (en) | Spatial-spectral fusion hyperspectral image classification method based on three-dimensional depth residual error network | |
| CN108961245A (en) | Picture quality classification method based on binary channels depth parallel-convolution network | |
| CN106250931A (en) | A kind of high-definition picture scene classification method based on random convolutional neural networks | |
| CN106874956A (en) | The construction method of image classification convolutional neural networks structure | |
| CN107194426A (en) | A kind of image-recognizing method based on Spiking neutral nets | |
| CN107679462A (en) | A kind of depth multiple features fusion sorting technique based on small echo | |
| CN108765425A (en) | Image partition method, device, computer equipment and storage medium | |
| CN110502987A (en) | A kind of plant pest recognition methods and system based on deep learning | |
| CN113887517B (en) | Crop remote sensing image semantic segmentation method based on parallel attention mechanism | |
| CN110363770A (en) | A kind of training method and device of the infrared semantic segmentation model of margin guide formula | |
| CN114004847A (en) | Medical image segmentation method based on graph reversible neural network | |
| Xiao et al. | TTST: A top-k token selective transformer for remote sensing image super-resolution | |
| Cao et al. | Adversarial and adaptive tone mapping operator for high dynamic range images | |
| CN107679539A (en) | A kind of single convolutional neural networks local message wild based on local sensing and global information integration method | |
| CN116052016A (en) | Fine segmentation detection method for remote sensing image cloud and cloud shadow based on deep learning | |
| CN116935043A (en) | Typical object remote sensing image generation method based on multitasking countermeasure network |
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 |