CN110245562A - Ocean based on deep learning produces malicious microalgae type automatic identifying method - Google Patents
Ocean based on deep learning produces malicious microalgae type automatic identifying method Download PDFInfo
- Publication number
- CN110245562A CN110245562A CN201910394170.5A CN201910394170A CN110245562A CN 110245562 A CN110245562 A CN 110245562A CN 201910394170 A CN201910394170 A CN 201910394170A CN 110245562 A CN110245562 A CN 110245562A
- Authority
- CN
- China
- Prior art keywords
- image data
- neural networks
- convolutional neural
- deep learning
- type automatic
- 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
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000013135 deep learning Methods 0.000 title claims abstract description 16
- 241000195493 Cryptophyta Species 0.000 claims abstract description 33
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 26
- 238000012549 training Methods 0.000 claims abstract description 19
- 238000000399 optical microscopy Methods 0.000 claims abstract description 6
- 231100000331 toxic Toxicity 0.000 claims abstract description 3
- 230000002588 toxic effect Effects 0.000 claims abstract description 3
- 238000010606 normalization Methods 0.000 claims description 14
- 238000012545 processing Methods 0.000 claims description 14
- 230000002708 enhancing effect Effects 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 10
- 238000005070 sampling Methods 0.000 claims description 10
- 239000011159 matrix material Substances 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 7
- 230000003247 decreasing effect Effects 0.000 claims description 2
- 241001269238 Data Species 0.000 claims 1
- 238000012360 testing method Methods 0.000 abstract description 7
- 210000002569 neuron Anatomy 0.000 description 7
- 238000010586 diagram Methods 0.000 description 4
- 238000004519 manufacturing process Methods 0.000 description 4
- 238000013507 mapping Methods 0.000 description 4
- 238000000386 microscopy Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 239000002574 poison Substances 0.000 description 2
- 231100000614 poison Toxicity 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 241001474374 Blennius Species 0.000 description 1
- 108090000790 Enzymes Proteins 0.000 description 1
- 102000004190 Enzymes Human genes 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000003935 denaturing gradient gel electrophoresis Methods 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 235000014113 dietary fatty acids Nutrition 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 229930195729 fatty acid Natural products 0.000 description 1
- 239000000194 fatty acid Substances 0.000 description 1
- 150000004665 fatty acids Chemical class 0.000 description 1
- 239000012634 fragment Substances 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 239000004615 ingredient Substances 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000000879 optical micrograph Methods 0.000 description 1
- 238000012797 qualification Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 108091008146 restriction endonucleases Proteins 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 239000013535 sea water Substances 0.000 description 1
- 238000007711 solidification Methods 0.000 description 1
- 230000008023 solidification Effects 0.000 description 1
- 241000894007 species Species 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Classifications
-
- 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
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
- G06V20/698—Matching; Classification
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of, and the ocean based on deep learning produces malicious microalgae type automatic identifying method, the following steps are included: obtaining the image data of toxic algae using optical microscopy, and image data is pre-processed, pretreated image data is fabricated to training data packet；Convolutional neural networks model is established, training data packet is input in convolutional neural networks model and is trained, obtains trained convolutional neural networks model；Malicious microalgae is produced to ocean using trained convolutional neural networks model to identify.The present invention can make the identification of toxiferous algae class get a promotion in time efficiency and testing accuracy.
Description
Technical field
The present invention relates to computer vision recognition technology fields, produce poison more particularly to a kind of ocean based on deep learning
Microalgae type automatic identifying method.
Background technique
Traditional optics microscopy is still the classical way of marine microalgae ecological Studies, and in addition to this, currently there is also other
Two kinds of newer technologies.First is that biochemistry classification.This method be according to the ingredient or content difference of large biological molecule come pair
Microalgae is classified, and the methods of isodynamic enzyme classification, fatty acid classification and base composition point classification (Murphy et is generally included
al,2010；Khotimchenko,1998；Fahrenkrug et al,1992).Second of new method is molecular classification method.
This is a kind of method of algae specific sequence identification, mainly long by randomly amplified polymorphic DNA, restriction enzyme enzyme fragment
Spend the methods of polymorphism and denaturing gradient gel electrophoresis realize (Nishihara et al, 1997；Ernst etal,1995；
Becker et al,2004).Meanwhile Protocols in Molecular Biology is also applied (Ebenezer in quantitative testing toxiferous algae class
Et al, 2012).But both the above method can only obtain a kind of or a few target species information (Handy et every time
A1,2006).And a large amount of microalgae types are generally comprised in practice, in seawater sample.Simultaneously as biochemistry classification
Professional stronger with molecular biology method, operation difficulty is bigger, and many deficiencies also to receive in practical application
Very big limitation.Therefore, optics microscopy is still current most easy, quick, extensive marine microalgae identification mode.
However optics microscopy needs veteran taxology expert, time and effort consuming, it is even more important that be not easy to distinguish
The extremely similar certain types of shape.Electron microscope can ultra microstructure difference between more subtly distinguishing similar type,
The discrimination degree of different algaes is improved, however its is complicated for operation, it is expensive, sample size and inspection are limited to a certain extent
Degree of testing the speed.In the machine learning techniques of related algae identification, the past utilizes optical microscope image combination shallow-layer machine learning
Method can achieve certain accuracy of identification under given conditions, but in background complexity, the huge situation of sample size, to upload
System method is not met by the needs of practical application.Wherein complicated background generally requires that target zone is manually specified, and leads to algae
Class recognition efficiency is low；Sample size is excessive to will lead to that the training time is too long, and is difficult to train general algae classifier, more closes
Key, traditional classifier often want expert's hand picking feature, often occur feature type it is limited and omit phenomenon, while
Increase cost of labor.
Summary of the invention
Technical problem to be solved by the invention is to provide a kind of, and the malicious microalgae type of ocean production based on deep learning is automatic
Recognition methods can make the identification of toxiferous algae class get a promotion in time efficiency and testing accuracy.
The technical solution adopted by the present invention to solve the technical problems is: providing a kind of ocean production poison based on deep learning
Microalgae type automatic identifying method, comprising the following steps:
(1) image data of toxic algae is obtained using optical microscopy, and image data is pre-processed, and will be located in advance
Image data after reason is fabricated to training data packet；
(2) convolutional neural networks model is established, training data packet is input in convolutional neural networks model and is trained,
Obtain trained convolutional neural networks model；
(3) malicious microalgae is produced to ocean using trained convolutional neural networks model to identify.
Pretreatment is carried out including rejecting ineffective image data to image data in the step (1).
In the step (1) to image data carry out pretreatment further include normalized, the normalized packet
Include gray scale normalization and geometrical normalization, the gray scale normalization by image data carry out Gray (i, j)=0.299R (i,
J) it is realized after the processing of+0.578G (i, j)+0.114B (i, j) gray processing, wherein R, G and B are respectively the red of the i-th row j column pixel
Color, green and blue component；The geometrical normalization is that unification is carried out to the size of image data.
In the step (1) to image data carry out pretreatment further include using Gaussian convolution collecting image data progress
The step of filtering.
Further include the steps that carrying out image data enhancing processing in the step (1), the enhancing processing includes early period
Enhancing and late-enhancement, enhancing early period are the picture numbers obtained under different focal point by the fine adjustment function of optical microscopy
According to；The late-enhancement is to carry out micro-shifting, overturning and rotation process to image data.
The convolutional neural networks model established in the step (2) is based on AlexNet model, includes 5 convolutional layers and 3
Full articulamentum, wherein the operation of down-sampling maximum pondization is added to after the 1st, the 2nd and the 5th convolutional layer.
The convolutional neural networks model established in the step (2) is in convolution, activation and pond process, after pretreatment
Image data as input layer, which is the matrix M of m × m, and convolution kernel is the Matrix C of c × c, is biased to bi, obtained volume
Product characteristic pattern F, uses ReLU as activation primitive S (), and the size of obtained convolution characteristic pattern F is (m-c+1) × (m-c+1),
Calculation formula isWherein, f ∈ F, MijIndicate corresponding with convolution kernel in input layer when convolution
Element, the pond domain of convolution characteristic pattern F is matrix P, behind the pond domain for traversing former characteristic pattern, obtains sub-sampling characteristic pattern S, S
=Po(F)+b2, wherein Po() is pond function, b2For amount of bias.
It is described using monotonic decreasing dynamical learning rate when being trained in the step (2) to convolutional neural networks model
Learning rateWherein, 10minIndexFor
Minimum learning rate, 10maxIndexFor maximum learning rate,CurStep is current
The number of iterations, maxStep are maximum number of iterations.
Beneficial effect
Due to the adoption of the above technical solution, compared with prior art, the present invention having the following advantages that and actively imitating
Fruit: the identification precision of ocean toxiferous algae class can be improved in the present invention, improves determination rates, reduces the difficulty manually identified.
Detailed description of the invention
Fig. 1 is the unsharp original image schematic diagram of content；
Fig. 2 is the clear original image schematic diagram of algae blur margin；
Fig. 3 is the algae original image schematic diagram containing bubble.
The data of Fig. 4 sample enhance flow chart；
Fig. 5 is the basic block diagram of model；
Precision trend chart during Fig. 6 model training；
The practical precision test result figure called of Fig. 7 model.
Specific embodiment
Present invention will be further explained below with reference to specific examples.It should be understood that these embodiments are merely to illustrate the present invention
Rather than it limits the scope of the invention.In addition, it should also be understood that, after reading the content taught by the present invention, those skilled in the art
Member can make various changes or modifications the present invention, and such equivalent forms equally fall within the application the appended claims and limited
Range.
Embodiments of the present invention are related to a kind of malicious microalgae type automatic identifying method of ocean production based on deep learning, should
Method is started with from the sampling of the microscopy of marine microalgae image, in conjunction with the priori knowledge of expert, establishes algae image data base, building volume
Product neural network, screens optimal models, and research model parameter etc. changes the influence to discrimination precision.Specifically include with
Lower step:
1. the pre-processing of data
(1) screening of algae image
Training data includes image data and label data two parts, before production, it to clean, pick to data
Except invalid image.Ineffective image data mainly includes following 3 kinds: a. content unsharp image (such as Fig. 1)；B. blur margin is clear
(such as Fig. 2)；C. image containing bubble (such as Fig. 3).
(2) gray scale normalization and geometrical normalization of microalgae image
Normalization includes two operations.First is that gray scale normalization, gray scale normalization is used to the different intensities of light source, light source side
The microalgae image obtained downwards compensates, to weaken the variation of the picture signal as caused by illumination variation merely.Herein it
Before should first by microalgae picture carry out gray processing processing.It is usually weighted and averaged as the following formula, to obtain reasonable gray level image
Gray (i, j)=0.299R (i, j)+0.578G (i, j)+0.114B (i, j)
Wherein, R, G and B are respectively the red of the i-th row j column pixel, green and blue component.
Microalgae image is unified for phase its object is to solve microalgae dimensional variation for geometrical normalization by another operation
Same size, convenient for the input of model.
(3) reparation and reconstruction of image
Since image preprocessing may excessively influence whether the training effect of later period model, reparation to image and again
Build to carry out with caution.Consider that this point, present embodiment will take more conservative method, is carried out using Gaussian convolution collecting image
Filtering.This method only considers the one week image pixel value in damaged area edge, therefore speed is faster, and the width of restoring area
Also it can control between 2-3 pixel.
The picture size of microalgae is M × M pixel (M=127) after pretreatment, determines classification locating for it through expert appraisal
Status is learned, as class label.
(4) data enhancement process
Present embodiment enhances data in such a way that early period enhances and late-enhancement combines.
Enhancing early period refers to data enhancing when sampling, is real to the sampling of same algae image under different focus distance
It is existing.When to the sampling of same algae, by the fine adjustment function of optical microscopy, the algae picture under different focal point can be obtained, into
Row repeatedly focusing, can obtain different algae image structures.
In addition, algae type should not locate the influence of present position in the picture and rotation angle, therefore, late-enhancement is logical
Micro-shifting, overturning and the rotation process for crossing image can also carry out data set enhancing processing (such as Fig. 4).
(5) file saves training dataset in a binary format, including two parts:
I label data file: tag data structure is as shown in the table
Wherein L indicates label mark length (byte), and n indicates that tag class number, Namei indicate i-th of bookmark name, N table
Show and record number in training sample, Label Index indicates algae index value.
Ii. image data file: image data structure is as shown in the table:
| Row | Col | N | Img1 | … | ImgN-1 | ImgN |
| 2 bytes | 2 bytes | 4 bytes | The byte of Row × Col × 4 | The byte of Row × Col × 4 | The byte of Row × Col × 4 |
Wherein Row and Col respectively indicates the height and width (pixel number) of every algae image.N is indicated in training sample
Number is recorded, Img indicates each image content.Wherein 4 bytes are taken up space by a real-coded GA.
2. convolutional neural networks
In numerous deep learning algorithms, and convolutional neural networks (Convolutional Neural Networks, referred to as
It CNNs) is one of the important method for handling image recognition.In the present embodiment, TensorFlow is based under Python environment
Open source machine learning library, constructs CNNs model.It collects main ocean and produces malicious 8 kinds of microalgae, first progress image sampling, then through data
35240 records are obtained altogether after enhancing technical treatment, are classified as training data and test data two parts, are trained entirely
Cheng Zhong compares error by algae label, residual error is then solved, then by chain type Rule for derivation, by residual error by asking
Solution partial derivative is gradually communicated up, and is updated to weight, then layer-by-layer adjustment weight and bias.
(1) convolution, activation and pond process
Using pretreated algae gray level image as input layer, which is the matrix M of m × m, and convolution kernel is the square of c × c
Battle array C, is biased to bi, obtained convolution characteristic pattern F uses ReLU as activation primitive S (), and obtained convolution characteristic pattern F's is big
Small is (m-c+1) × (m-c+1).Calculation formula is
Wherein f ∈ F, MijElement corresponding with convolution kernel in input layer when expression convolution, and the i row j column of non-matrix M
Corresponding value.
Calculate algae image regional characteristic value when, usually the feature in the region is counted, with new spy
It levies to represent the general characteristic in the region.This process is called pond (Pooling).Through pond first is that former characteristic pattern can be reduced
Dimension and resolution ratio, second is that avoiding over-fitting.If the pond domain of convolution characteristic pattern F is matrix P, the pond of former characteristic pattern is traversed
Behind domain, sub-sampling characteristic pattern S is obtained, i.e.,
S=Po(F)+b2
Wherein, Po() is pond function, b2For amount of bias.
(2) model structure
CNN basic structure includes two layers, and one is characterized extract layer, and the input of each neuron and the part of preceding layer connect
It is connected by domain, and extracts the feature of the part.After the local feature is extracted, its positional relationship between other feature
It decides therewith；The second is Feature Mapping layer, each computation layer of network is made of multiple Feature Mappings, each Feature Mapping
It is a plane, the weight of all neurons is equal in plane.Further, since the neuron on a mapping face shares weight,
Thus reduce the number of network freedom parameter.Each of convolutional neural networks convolutional layer all followed by one is used to ask office
Portion is averagely and the computation layer of second extraction, this distinctive structure of feature extraction twice reduce feature resolution.Marine algae
Image can be used as two dimension input variable, directly input network.The reference of convolutional neural networks used by present embodiment
The overall architecture of AlexNet model, network is as shown in Figure 5.Model includes that 5 convolutional layers and 3 full articulamentums (including export
Layer), down-sampled (Max-pooling, maximum pond) operation is added to after the 1st, 2,5 convolutional calculation.Network is mainly held
Row process and design parameter information are as follows.
In the 1st convolutional layer, the gray scale of Input is inputted as specification is that (M is algae image side length pixel number, this model to M × M
Middle M=127), the convolution kernel for being 5 × 5 using 96 size specifications carries out algae image characteristics extraction.Altogether using 2 GPU clothes
Business device is handled, and is responsible for 48 convolution kernels on each server.The convolution when carrying out convolution using filter filter and data
Core size is 11 × 11, connects 11 × 11 size areas every time, obtains a feature, on this basis repeatedly above procedure, then
Obtain new feature.Then some neuron in next hidden layer is multiplied by multiple neurons in a upper network layer
It is shared plus the weight obtained after biasing with weight, finally using full link.ReLU makes characteristic value in model as activation primitive
It is trapped among within zone of reasonableness and carries out down-sampled processing again.
It is carried out in the 2nd convolutional layer using the filter filter of 256 5 × 5 sizes to 96 × 27 × 27 characteristic patterns
Feature is further extracted, filter is to region corresponding in certain several characteristic pattern in 96 characteristic patterns multiplied by corresponding power
Then weight carries out convolution plus region acquired after biasing.Equally carried out again through ReLU function and down-sampled processing.
3rd and the 4th convolutional layer respectively obtains 384 13 × 13 new feature figures, but all without down-sampled processing.
5th convolutional layer obtains 256 13 × 13 characteristic patterns, and down-sampled layer pool, prevents over-fitting.
In this model, it is simplified model structure, the full articulamentum of the 6th layer and the 7th layer is reduced respectively to 2048 minds
Through member.
8th full articulamentum represents 8 algaes using 8 neurons, carries out to 2048 neurons in the 7th layer complete
Connection, obtains the value of N number of float type, using the probability value of 8 algaes of Softmax function representation.
(3) training process and model accuracy
In present embodiment, it is used as training data by the 70% of data set, 30% is used as test data, the number of iterations setting
It is 10000, gradually records trained Loss value, training precision and measuring accuracy, and foundationBy learning rate from 10-4 to 10-6
Power learning rate transition, wherein 10minIndexFor minimum learning rate, 10maxIndexFor maximum learning rate,CurStep is current iteration number, and maxStep is maximum number of iterations,
From 255506 downward trends always to 2.99184, training precision rises to 99.18% from 9.745% for model loss, measuring accuracy from
9.459% rises to 84.752%.After the completion of iteration, model structure and parameters figure is obtained, training precision is as shown in Figure 6.
(3) solidification and calling of model parameter
The graph structure and parameter of stress model under Python environment, randomly select 50 width seaweed sample graphs, input model,
The calling of implementation model removes the loading time of data and model when starting, and identifies that total used time is 199 millis to 20 sample algaes
Second, single recognition speed is 9.95 milliseconds, and error number is 2 in 20 algaes, and total accuracy rate is 90%.Specific qualification result
As shown in fig. 7, being differentiation probability value in its bracket.
Claims (8)
1. a kind of ocean based on deep learning produces malicious microalgae type automatic identifying method, which comprises the following steps:
(1) image data of toxic algae is obtained using optical microscopy, and image data is pre-processed, after pretreatment
Image data be fabricated to training data packet；
(2) convolutional neural networks model is established, training data packet is input in convolutional neural networks model and is trained, is obtained
Trained convolutional neural networks model；
(3) malicious microalgae is produced to ocean using trained convolutional neural networks model to identify.
2. the ocean according to claim 1 based on deep learning produces malicious microalgae type automatic identifying method, feature exists
In, in the step (1) to image data carry out pretreatment include reject ineffective image data.
3. the ocean according to claim 1 based on deep learning produces malicious microalgae type automatic identifying method, feature exists
In, in the step (1) to image data carry out pretreatment further include normalized, the normalized includes gray scale
Normalization and geometrical normalization, the gray scale normalization by image data carry out Gray (i, j)=0.299R (i, j)+
0.578G (i, j)+0.114B (i, j) gray processing processing after realize, wherein R, G and B be respectively the i-th row j column pixel red,
Green and blue component；The geometrical normalization is that unification is carried out to the size of image data.
4. the ocean according to claim 1 based on deep learning produces malicious microalgae type automatic identifying method, feature exists
In, in the step (1) to image data carry out pretreatment further include being filtered using Gaussian convolution collecting image data
The step of.
5. the ocean according to claim 1 based on deep learning produces malicious microalgae type automatic identifying method, feature exists
In, further include the steps that carrying out image data enhancing processing in the step (1), the enhancing processing include enhancing early period with
Late-enhancement, enhancing early period are the image datas obtained under different focal point by the fine adjustment function of optical microscopy；It is described
Late-enhancement is to carry out micro-shifting, overturning and rotation process to image data.
6. the ocean according to claim 1 based on deep learning produces malicious microalgae type automatic identifying method, feature exists
In the convolutional neural networks model established in the step (2) is based on AlexNet model, connects entirely comprising 5 convolutional layers and 3
Connect layer, wherein the operation of down-sampling maximum pondization is added to after the 1st, the 2nd and the 5th convolutional layer.
7. the ocean according to claim 6 based on deep learning produces malicious microalgae type automatic identifying method, feature exists
In the convolutional neural networks model established in the step (2) is in convolution, activation and pond process, by pretreated figure
For picture data as input layer, which is the matrix M of m × m, and convolution kernel is the Matrix C of c × c, is biased to bi, obtained convolution is special
Sign figure F, uses ReLU as activation primitive S (), and the size of obtained convolution characteristic pattern F is (m-c+1) × (m-c+1), calculates
Formula isWherein, f ∈ F, MijMember corresponding with convolution kernel in input layer when expression convolution
Element, the pond domain of convolution characteristic pattern F are matrix P, behind the pond domain for traversing former characteristic pattern, obtain sub-sampling characteristic pattern S, S=Po
(F)+b2, wherein Po() is pond function, b2For amount of bias.
8. the ocean according to claim 1 based on deep learning produces malicious microalgae type automatic identifying method, feature exists
In, when being trained in the step (2) to convolutional neural networks model, using monotonic decreasing dynamical learning rate, the study
RateWherein, 10minIndexFor minimum
Learning rate, 10maxIndexFor maximum learning rate,CurStep is current iteration
Number, maxStep are maximum number of iterations.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910394170.5A CN110245562A (en) | 2019-05-13 | 2019-05-13 | Ocean based on deep learning produces malicious microalgae type automatic identifying method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910394170.5A CN110245562A (en) | 2019-05-13 | 2019-05-13 | Ocean based on deep learning produces malicious microalgae type automatic identifying method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110245562A true CN110245562A (en) | 2019-09-17 |
Family
ID=67884496
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910394170.5A Pending CN110245562A (en) | 2019-05-13 | 2019-05-13 | Ocean based on deep learning produces malicious microalgae type automatic identifying method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110245562A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113435389A (en) * | 2021-07-09 | 2021-09-24 | 大连海洋大学 | Chlorella and chrysophyceae classification and identification method based on image feature deep learning |
| CN114418995A (en) * | 2022-01-19 | 2022-04-29 | 生态环境部长江流域生态环境监督管理局生态环境监测与科学研究中心 | Cascade algae cell statistical method based on microscope image |
| TWI786711B (en) * | 2021-07-07 | 2022-12-11 | 台灣電力股份有限公司 | Intelligent microalgae cultivation system and method thereof |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104398252A (en) * | 2014-11-05 | 2015-03-11 | 深圳先进技术研究院 | Electrocardiogram signal processing method and device |
| CN105279556A (en) * | 2015-11-05 | 2016-01-27 | 国家卫星海洋应用中心 | Enteromorpha detection method and enteromorpha detection device |
| CN106485251A (en) * | 2016-10-08 | 2017-03-08 | 天津工业大学 | Egg embryo classification based on deep learning |
| CN107730473A (en) * | 2017-11-03 | 2018-02-23 | 中国矿业大学 | A kind of underground coal mine image processing method based on deep neural network |
| CN108073075A (en) * | 2017-12-21 | 2018-05-25 | 苏州大学 | Silicon micro accerometer temperature-compensation method, system based on GA Optimized BP Neural Networks |
| CN108596038A (en) * | 2018-03-28 | 2018-09-28 | 电子科技大学 | Erythrocyte Recognition method in the excrement with neural network is cut in a kind of combining form credit |
| US20180286038A1 (en) * | 2015-09-23 | 2018-10-04 | The Regents Of The University Of California | Deep learning in label-free cell classification and machine vision extraction of particles |
| CN108985455A (en) * | 2018-07-09 | 2018-12-11 | 肖朝晖 | A kind of computer application neural net prediction method and system |
| CN109284765A (en) * | 2018-07-18 | 2019-01-29 | 成都信息工程大学 | The scene image classification method of convolutional neural networks based on negative value feature |
-
2019
- 2019-05-13 CN CN201910394170.5A patent/CN110245562A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104398252A (en) * | 2014-11-05 | 2015-03-11 | 深圳先进技术研究院 | Electrocardiogram signal processing method and device |
| US20180286038A1 (en) * | 2015-09-23 | 2018-10-04 | The Regents Of The University Of California | Deep learning in label-free cell classification and machine vision extraction of particles |
| CN105279556A (en) * | 2015-11-05 | 2016-01-27 | 国家卫星海洋应用中心 | Enteromorpha detection method and enteromorpha detection device |
| CN106485251A (en) * | 2016-10-08 | 2017-03-08 | 天津工业大学 | Egg embryo classification based on deep learning |
| CN107730473A (en) * | 2017-11-03 | 2018-02-23 | 中国矿业大学 | A kind of underground coal mine image processing method based on deep neural network |
| CN108073075A (en) * | 2017-12-21 | 2018-05-25 | 苏州大学 | Silicon micro accerometer temperature-compensation method, system based on GA Optimized BP Neural Networks |
| CN108596038A (en) * | 2018-03-28 | 2018-09-28 | 电子科技大学 | Erythrocyte Recognition method in the excrement with neural network is cut in a kind of combining form credit |
| CN108985455A (en) * | 2018-07-09 | 2018-12-11 | 肖朝晖 | A kind of computer application neural net prediction method and system |
| CN109284765A (en) * | 2018-07-18 | 2019-01-29 | 成都信息工程大学 | The scene image classification method of convolutional neural networks based on negative value feature |
Non-Patent Citations (5)
| Title |
|---|
| ANIBAL PEDRAZA等: "Automated Diatom Classification (Part B): A Deep Learning Approach", 《APPLIED SCIENCES》 * |
| IAGO CORRÊA等: "Deep Learning for Microalgae Classification", 《2017 16TH IEEE INTERNATIONAL CONFERENCE ON MACHINE LEARNING AND APPLICATIONS》 * |
| 万永清等: "深度学习在藻类分类识别中的应用", 《传感器世界》 * |
| 杨森等: "应用自组织特征映射神经网络技术实现分布式入侵检测", 《计算机应用》 * |
| 龙满生等: "基于卷积神经网络与迁移学习的油茶病害图像识别", 《农业工程学报》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI786711B (en) * | 2021-07-07 | 2022-12-11 | 台灣電力股份有限公司 | Intelligent microalgae cultivation system and method thereof |
| CN113435389A (en) * | 2021-07-09 | 2021-09-24 | 大连海洋大学 | Chlorella and chrysophyceae classification and identification method based on image feature deep learning |
| CN113435389B (en) * | 2021-07-09 | 2024-03-01 | 大连海洋大学 | Chlorella and golden algae classification and identification method based on image feature deep learning |
| CN114418995A (en) * | 2022-01-19 | 2022-04-29 | 生态环境部长江流域生态环境监督管理局生态环境监测与科学研究中心 | Cascade algae cell statistical method based on microscope image |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20230127698A1 (en) | Automated stereology for determining tissue characteristics | |
| CN111448582B (en) | System and method for single channel whole cell segmentation | |
| Nie et al. | Strawberry verticillium wilt detection network based on multi-task learning and attention | |
| Kumar et al. | Image based tomato leaf disease detection | |
| CN107016405B (en) | A kind of pest image classification method based on classification prediction convolutional neural networks | |
| Tahir et al. | A fungus spores dataset and a convolutional neural network based approach for fungus detection | |
| CN110349126A (en) | A kind of Surface Defects in Steel Plate detection method based on convolutional neural networks tape label | |
| CN110245562A (en) | Ocean based on deep learning produces malicious microalgae type automatic identifying method | |
| CN108564589A (en) | A kind of plant leaf blade dividing method based on the full convolutional neural networks of improvement | |
| CN106204642B (en) | A kind of cell tracker method based on deep neural network | |
| CN103984939B (en) | A kind of sample visible component sorting technique and system | |
| CN112037219A (en) | Metal surface defect detection method based on two-stage convolution neural network | |
| CN109635806A (en) | Ammeter technique for partitioning based on residual error network | |
| CN108647695A (en) | Soft image conspicuousness detection method based on covariance convolutional neural networks | |
| CN112464983A (en) | Small sample learning method for apple tree leaf disease image classification | |
| CN103177266A (en) | Intelligent stock pest identification system | |
| CN113160136B (en) | Wood defect identification and segmentation method based on improved Mask R-CNN | |
| CN116612472B (en) | Single-molecule immune array analyzer based on image and method thereof | |
| Sundara et al. | Segmentation and evaluation of white blood cells using segmentation algorithms | |
| Viertel et al. | Pattern recognition methodologies for pollen grain image classification: a survey | |
| Baykal et al. | Modern convolutional object detectors for nuclei detection on pleural effusion cytology images | |
| Kayasandik et al. | Automated sorting of neuronal trees in fluorescent images of neuronal networks using NeuroTreeTracer | |
| Rebelo et al. | A fully automatic classification of bee species from wing images | |
| Pushpa et al. | Deep learning model for plant species classification using leaf vein features | |
| CN116757998A (en) | Screening method and device for CTC cells and CTC-like cells based on AI |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190917 |