CN109887035A - Based on bat algorithm optimization BP neural network binocular vision calibration - Google Patents
Based on bat algorithm optimization BP neural network binocular vision calibration Download PDFInfo
- Publication number
- CN109887035A CN109887035A CN201811366222.XA CN201811366222A CN109887035A CN 109887035 A CN109887035 A CN 109887035A CN 201811366222 A CN201811366222 A CN 201811366222A CN 109887035 A CN109887035 A CN 109887035A
- Authority
- CN
- China
- Prior art keywords
- bat
- neural network
- hidden layer
- vision calibration
- node
- 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
Abstract
Vision measurement binocular calibration method of the present invention belongs to optical measurement and field of visual inspection；This method including the following steps: determine the initial parameters such as the BP neural network hidden layer number of plies, Inport And Outport Node；Node in hidden layer range is determined using empirical equation, and best node in hidden layer is determined by one-way analysis of variance method；According to bat algorithm echolocation principle, the weight and biasing of BP neural network are most preferably taken, determine optimal weight and bias；Each parameter value for determining BP neural network structure is distributed pixel number according to Distributed learning object space point data, completes calibration.The method of the present invention determines node in hidden layer by one-way analysis of variance method and calculates the error that each neural network generates on training set, fitness function of the error generated using on training set as bat algorithm, the characteristics of, optimization precision few using bat control parameter of algorithm and search precision height, strong robustness, optimal initial weight and biasing are chosen to it, solve the problems, such as that current BP neural network scaling method node in hidden layer can not determine, easily fall into local optimum and convergence rate is slow.
Description
Technical field
The invention belongs to optical measurement and field of visual inspection, and in particular to a kind of bat algorithm optimization BP neural network is double
Mesh vision calibration method.
Background technique
In recent years, machine vision and vision detection technology have been applied in many fields, such as large parts measurement, work
Fields, the vision-based detections such as the detection of industry assembly line not only can be reduced labour cost, while can also improve detection accuracy, avoid by examining
The human error that survey personnel generate.
In the research of common large scale structure vision calibration, research center of gravity focuses primarily on the reason of the complicated mathematical model of building
By upper, and the essence of camera calibration is so that pixel is corresponding with object space point, and the intrinsic parameter in calibration process is non-linear
Function, the method when solving nonlinear function using mathematical model is computationally intensive and stated accuracy is not high.Therefore, in vision mark
In fixed, it is necessary to search out a kind of scaling method that can avoid complicated calculating and can guarantee precision.
In order to realize more effective, fast, accurately scaling method, this invention proposes a kind of based on bat algorithm optimization
The binocular vision calibration of BP neural network, bat algorithm is with structure is simple, parameter is few, strong robustness, should be readily appreciated that and realizes
The advantages that, bat algorithm is applied in Neural Network Optimization, effect is more obvious.
Summary of the invention
In view of the above-mentioned problems, the invention discloses one kind to be based on bat algorithm optimization BP neural network binocular vision calibration side
Method, the invention include measurement task and target analysis, and each link is to the stability of vision calibration, precision and versatility etc.
Aspect has an impact.
The object of the present invention is achieved like this:
Based on bat Optimized BP Neural Network binocular vision calibration method, it is characterised in that the following steps are included:
Step a: the BP neural network hidden layer number of plies, input, output node number are determined；
Step b: determining node in hidden layer range using empirical equation, is determined by one-way analysis of variance method best hidden
Number containing node layer；
Step c: optimal value is chosen to the weight of BP neural network and biasing using bat algorithm；
Step d: determining each parameter value of BP neural network structure, makes pixel number according to Distributed learning object space point data point
Cloth completes calibration.
Above-mentioned bat algorithm optimization BP neural network binocular vision calibration method, the step a specifically:
Four kinds of different layers of BP neural network binocular vision models are established, determine that BP neural network is best using experimental method
The hidden layer number of plies be 3.According to binocular vision calibration model, as follows, pixel is 4 when determining binocular vision calibration,
Object space point is 3.
I is the quantity of video camera in above formula, and 1≤i≤2, i ∈ Z+, f are the focal length of video camera, ri、TiIt is arrived for world coordinate system
The spin matrix and translation matrix of two camera coordinate system conversions, u0i、v0iIt is image coordinate system origin in pixel coordinate system
Coordinate, dx, dy are distance of the pixel coordinate system in the x direction and the y direction between adjacent pixel respectively.
Above-mentioned bat algorithm optimization BP neural network binocular vision calibration method, the step b specifically:
Node in hidden layer range determines that formula is empirical equation in BP neural network vision calibration, is:
Wherein n is input layer number, and m is output layer number of nodes, L is node in hidden layer, and a is constant.
Hidden layer node determines that method is one-way analysis of variance method in BP neural network vision calibration, and distribution function is F distribution,
It is:
Wherein S1、S2For the standard deviation between two samples comparing, k is the freedom degree in first group of sample, and m is second group of sample
Freedom degree.
The sample of one-way analysis of variance is the performance parameter of BP neural network, is:
(1) the number of iterations；(2) runing time；(3) calibrated error.
Above-mentioned bat algorithm optimization BP neural network binocular vision calibration method, the step c specifically:
When optimizing using bat algorithm to neural network vision calibration, training data is selected, generates input layer and hidden
Containing the weight W between layerijWith threshold values bj, setting bat quantity M, bat individual i, maximum impulse frequency R (i) and maximum impulse
Intensity of sound A (i), bat frequency increase coefficient and are set as γ, intensity of sound attenuation coefficient α, maximum number of iterations NmaxWith search essence
Spend ε, random initializtion bat individual position xi(i=1,2 ..., M), and search and be located at optimum position x* bat individual；
Generate random number R1If R1When less than R (i), v is utilizedt+1 i=vt i+(xt i-*x)QiBat current location is updated, wherein QiTable
Show the sound wave that i-th bat issues, vt iIndicate the speed in i-th bat of t moment, xt iIt indicates in i-th bat of t moment
Position；Otherwise bat current location is disturbed, position replaces current location after disturbance；Generate random number R2, it is assumed that R2It is less than
A (i) is optimized and is changed in bat current location simultaneously, then bat individual flies to updated position；If updating position
Bat individual i is better than best bat in individual afterwards, then replacing current best bat individual, and according to formula At+1(i)
=α At(i) and Rt+1(i)=R0(i)×(1-e-γt) adjust bat pulse frequency R (i) and pulse loudness of a sound A (i)；When by one section
Between run, assessment judgement is carried out to new bat group, finds out current location best bat individual and accordingly optimal location；It is defeated
Function globally optimal solution and optimum individual value out, the globally optimal solution of output function, weight and threshold values as BP neural network.
Above-mentioned bat algorithm optimization BP neural network binocular vision calibration method, the step d specifically:
Hidden layer activation primitive is logarithm probability function in bat algorithm optimization BP neural network vision calibration, is:
Training method is steepest descent method in bat algorithm optimization BP neural network vision calibration, and effect is: by reversely passing
The weight and threshold value for constantly to adjust network are broadcast, keeps the error sum of squares of network minimum.
Learning rate selection principle is experience selection principle in bat algorithm optimization BP neural network vision calibration, and learning rate is got over
Small, indoctrination session is finer, but pace of learning can also reduce simultaneously, and learning rate is bigger, and pace of learning can be faster, but learns essence simultaneously
Degree can also reduce.
The utility model has the advantages that
The invention proposes a kind of BP neural network binocular vision calibration methods of bat algorithm optimization, have studied nerve net
The node in hidden layer of network scaling method determines principle and optimal selection of the bat algorithm to the weight and threshold values of BP neural network
Problem, it is contemplated that fitness of the neural network error as bat algorithm, using the spy of intelligent algorithm and BP neural network structure
Point is completed to binocular vision high-precision calibrating.The present invention solves current BP neural network calibration node in hidden layer and can not have
Body is determining, operation when easily fall into local optimum and the slow problem of convergence rate, vision mark may be implemented by means of the present invention
Fixed accuracy and rapidity, to realize that the vision calibration research of large-sized equipment is laid a good foundation.
Detailed description of the invention
Fig. 1 is that the present invention is based on the algorithm flow charts of bat algorithm optimization BP neural network binocular vision calibration；
Fig. 2 is that the present invention is based on the structure charts of bat algorithm optimization BP neural network binocular vision calibration；
Fig. 3 is that the present invention emulates grid disk angle point grid figure used；
Fig. 4 is the schematic diagram data after average value processing used in one-way analysis of variance method of the present invention；
Fig. 5 is the mean square error comparison diagram of BA-BP (a) of the present invention Yu BP-NN (b).
Specific embodiment
Specific embodiments of the present invention will be described in further detail with reference to the accompanying drawing.
Specific embodiment one
The bat algorithm optimization BP neural network scaling method of the present embodiment, flow chart as shown in Figure 1, this method include with
Lower step:
Step a: the BP neural network hidden layer number of plies, input, output node number are determined；
Step b: determining node in hidden layer range using empirical equation, is determined by one-way analysis of variance method best hidden
Number containing node layer；
Step c: optimal value is chosen to the weight of BP neural network and biasing using bat algorithm；
Step d: determining each parameter value of BP neural network structure, makes pixel number according to Distributed learning object space point data point
Cloth completes calibration.
Specific embodiment two
The bat algorithm optimization BP neural network scaling method of this example, on the basis of specific embodiment one, further
Limit the concrete operation step of step a, step b, step c, step d.Wherein:
The step a specifically:
Four kinds of different layers of BP neural network binocular vision models are established, determine that BP neural network is best using experimental method
The hidden layer number of plies be 3.According to binocular vision calibration model, as follows, pixel is 4 when determining binocular vision calibration,
Object space point is 3.
I is the quantity of video camera in above formula, and 1≤i≤2, i ∈ Z+, f are the focal length of video camera, ri、TiIt is arrived for world coordinate system
The spin matrix and translation matrix of two camera coordinate system conversions, u0i、v0iIt is image coordinate system origin in pixel coordinate system
Coordinate, dx, dy are distance of the pixel coordinate system in the x direction and the y direction between adjacent pixel respectively.
The step b specifically:
Node in hidden layer range determines that formula is empirical equation in BP neural network vision calibration, is:
Wherein n is input layer number, and m is output layer number of nodes, L is node in hidden layer, and a is constant.
Hidden layer node determines that method is one-way analysis of variance method in BP neural network vision calibration, and distribution function is F distribution,
It is:
Wherein S1、S2For the standard deviation between two samples comparing, k is the freedom degree in first group of sample, and m is second group of sample
Freedom degree.
The sample of one-way analysis of variance is the performance parameter of BP neural network, is:
(1) the number of iterations；(2) runing time；(3) calibrated error.
The step c specifically:
When optimizing using bat algorithm to neural network vision calibration, training data is selected, generates input layer and hidden
Containing the weight W between layerijWith threshold values bj, setting bat quantity M, bat individual i, maximum impulse frequency R (i) and maximum impulse sound
Loudness of a sound degree A (i), bat frequency increase coefficient and are set as γ, intensity of sound attenuation coefficient α, maximum number of iterations NmaxAnd search precision
ε, random initializtion bat individual position xi(i=1,2 ..., M), and search and be located at optimum position x* bat individual；It is raw
At random number R1If R1When less than R (i), v is utilizedt+1 i=vt i+(xt i-*x)QiBat current location is updated, wherein QiIt indicates
The sound wave that i-th bat issues, vt iIndicate the speed in i-th bat of t moment, xt iIt indicates in the position of i-th bat of t moment
It sets；Otherwise bat current location is disturbed, position replaces current location after disturbance；Generate random number R2, it is assumed that R2Less than A
(i) bat current location is optimized and is changed simultaneously, then bat individual flies to updated position；If after updating position
Bat individual i is better than best bat in individual, then replacing current best bat individual, and according to formula At+1(i)=α
At(i) and Rt+1(i)=R0(i)×(1-e-γt) adjust bat pulse frequency R (i) and pulse loudness of a sound A (i)；Through after a period of time
Operation, carries out assessment judgement to new bat group, finds out the best bat individual in current location and accordingly optimal location；Output
Function globally optimal solution and optimum individual value, the globally optimal solution of output function, weight and threshold values as BP neural network.
The step d specifically:
Hidden layer activation primitive is logarithm probability function in bat algorithm optimization BP neural network vision calibration, is:
Training method is steepest descent method in bat algorithm optimization BP neural network vision calibration, and effect is: by reversely passing
The weight and threshold value for constantly to adjust network are broadcast, keeps the error sum of squares of network minimum.
Learning rate selection principle is experience selection principle in bat algorithm optimization BP neural network vision calibration, and learning rate is got over
Small, indoctrination session is finer, but pace of learning can also reduce simultaneously, and learning rate is bigger, and pace of learning can be faster, but learns essence simultaneously
Degree can also reduce.
Claims (5)
1. being based on bat Optimized BP Neural Network binocular vision calibration method, it is characterised in that the following steps are included:
Step a: the BP neural network hidden layer number of plies, input, output node number are determined；
Step b: determining node in hidden layer range using empirical equation, determines best hidden layer by one-way analysis of variance method
Number of nodes；
Step c: optimal value is chosen to the weight of BP neural network and biasing using bat algorithm；
Step d: determining each parameter value of BP neural network structure, is distributed pixel number according to Distributed learning object space point data, complete
At calibration.
2. being based on bat algorithm optimization BP neural network binocular vision calibration method according to one kind described in right 1, feature exists
In the step a specifically:
Four kinds of different layers of BP neural network binocular vision models are established, determine that BP neural network is optimal hidden using experimental method
It is 3 containing number layer by layer.According to binocular vision calibration model, as follows, pixel is 4 when determining binocular vision calibration, object space
Point is 3.
I is the quantity of video camera in above formula, and 1≤i≤2, i ∈ Z+, f are the focal length of video camera, ri、TiIt is arrived for world coordinate system
The spin matrix and translation matrix of two camera coordinate system conversions, u0i、v0iIt is image coordinate system origin in pixel coordinate system
Coordinate, dx, dy are distance of the pixel coordinate system in the x direction and the y direction between adjacent pixel respectively.
3. being based on bat algorithm optimization BP neural network binocular vision calibration method according to one kind described in right 1, feature exists
In the step b specifically:
Node in hidden layer range determines that formula is empirical equation in BP neural network vision calibration, is:
Wherein n is input layer number, and m is output layer number of nodes, L is node in hidden layer, and a is constant.
Hidden layer node determines that method is one-way analysis of variance method in BP neural network vision calibration, and distribution function is F distribution,
It is:
Wherein S1、S2For the standard deviation between two samples comparing, k is the freedom degree in first group of sample, and m is second group of sample
Freedom degree.
The sample of one-way analysis of variance is the performance parameter of BP neural network, is:
(1) the number of iterations；(2) runing time；(3) calibrated error.
4. being based on bat algorithm optimization BP neural network binocular vision calibration method according to one kind described in right 1, feature exists
In the step c specifically:
When optimizing using bat algorithm to neural network vision calibration, training data is selected, generates input layer and hidden layer
Between weight WijWith threshold values bj, setting bat quantity M, bat individual i, maximum impulse frequency R (i) and maximum impulse sound are strong
It spends A (i), bat frequency increases coefficient and is set as γ, intensity of sound attenuation coefficient α, maximum number of iterations NmaxWith search precision ε, with
Machine initializes bat individual position xi(i=1,2 ..., M), and search and be located at optimum position x*Bat individual；It generates random
Number R1If R1When less than R (i), v is utilizedt+1 i=vt i+(xt i-x*)QiBat current location is updated, wherein QiIndicate i-th bat
The sound wave that bat issues, vt iIndicate the speed in i-th bat of t moment, xt iIt indicates in the position of i-th bat of t moment；Otherwise
Bat current location is disturbed, position replaces current location after disturbance；Generate random number R2, it is assumed that R2Simultaneously less than A (i)
Bat current location is optimized and is changed, then bat individual flies to updated position；If updating bat behind position
Body i is better than best bat in individual, then replacing current best bat individual, and according to formula At+1(i)=α At(i) with
Rt+1(i)=R0(i)×(1-e-γt) adjust bat pulse frequency R (i) and pulse loudness of a sound A (i)；It is run after a period of time, it is right
New bat group carries out assessment judgement, finds out the best bat individual in current location and accordingly optimal location；Output function is complete
Office's optimal solution and optimum individual value, the globally optimal solution of output function, weight and threshold values as BP neural network.
5. being based on bat algorithm optimization BP neural network binocular vision calibration method according to one kind described in right 1, feature exists
In the step d specifically:
Hidden layer activation primitive is logarithm probability function in bat algorithm optimization BP neural network vision calibration, is:
Training method is steepest descent method in bat algorithm optimization BP neural network vision calibration, and effect is: by reversely passing
The weight and threshold value for constantly to adjust network are broadcast, keeps the error sum of squares of network minimum.
Learning rate selection principle is experience selection principle in bat algorithm optimization BP neural network vision calibration, and learning rate is got over
Small, indoctrination session is finer, but pace of learning can also reduce simultaneously, and learning rate is bigger, and pace of learning can be faster, but learns essence simultaneously
Degree can also reduce.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811366222.XA CN109887035A (en) | 2018-12-27 | 2018-12-27 | Based on bat algorithm optimization BP neural network binocular vision calibration |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811366222.XA CN109887035A (en) | 2018-12-27 | 2018-12-27 | Based on bat algorithm optimization BP neural network binocular vision calibration |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109887035A true CN109887035A (en) | 2019-06-14 |
Family
ID=66924919
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811366222.XA Pending CN109887035A (en) | 2018-12-27 | 2018-12-27 | Based on bat algorithm optimization BP neural network binocular vision calibration |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109887035A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112700500A (en) * | 2020-12-08 | 2021-04-23 | 中大检测(湖南)股份有限公司 | Binocular camera calibration method and device and readable storage medium |
| CN113221954A (en) * | 2021-04-15 | 2021-08-06 | 长春工业大学 | BP (Back propagation) classification algorithm based on improved bat algorithm |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2003099196A2 (en) * | 2002-05-23 | 2003-12-04 | Cure Tech Ltd. | Humanized immunomodulatory monoclonal antibodies for the treatment of neoplastic disease or immunodeficiency |
| TW200919210A (en) * | 2007-07-18 | 2009-05-01 | Steven Kays | Adaptive electronic design |
| CN101907448A (en) * | 2010-07-23 | 2010-12-08 | 华南理工大学 | Depth measurement method based on binocular three-dimensional vision |
| CN103157772A (en) * | 2011-12-14 | 2013-06-19 | 鞍钢股份有限公司 | Continuously casting blank internal quality forecasting method |
| CN106127789A (en) * | 2016-07-04 | 2016-11-16 | 湖南科技大学 | Stereoscopic vision scaling method in conjunction with neutral net Yu virtual target |
| CN106384153A (en) * | 2016-09-18 | 2017-02-08 | 河海大学常州校区 | WSAN actuator task distribution method based on BA-BPNN data fusion |
| CN206162876U (en) * | 2016-07-22 | 2017-05-10 | 福州大学 | Road speed limit prison bat system based on visibility detection |
| CN108010085A (en) * | 2017-11-30 | 2018-05-08 | 西南科技大学 | Target identification method based on binocular Visible Light Camera Yu thermal infrared camera |
| CN108830418A (en) * | 2018-06-14 | 2018-11-16 | 东北电力大学 | A kind of Short-Term Load Forecasting Method |
-
2018
- 2018-12-27 CN CN201811366222.XA patent/CN109887035A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2003099196A2 (en) * | 2002-05-23 | 2003-12-04 | Cure Tech Ltd. | Humanized immunomodulatory monoclonal antibodies for the treatment of neoplastic disease or immunodeficiency |
| TW200919210A (en) * | 2007-07-18 | 2009-05-01 | Steven Kays | Adaptive electronic design |
| CN101907448A (en) * | 2010-07-23 | 2010-12-08 | 华南理工大学 | Depth measurement method based on binocular three-dimensional vision |
| CN103157772A (en) * | 2011-12-14 | 2013-06-19 | 鞍钢股份有限公司 | Continuously casting blank internal quality forecasting method |
| CN106127789A (en) * | 2016-07-04 | 2016-11-16 | 湖南科技大学 | Stereoscopic vision scaling method in conjunction with neutral net Yu virtual target |
| CN206162876U (en) * | 2016-07-22 | 2017-05-10 | 福州大学 | Road speed limit prison bat system based on visibility detection |
| CN106384153A (en) * | 2016-09-18 | 2017-02-08 | 河海大学常州校区 | WSAN actuator task distribution method based on BA-BPNN data fusion |
| CN108010085A (en) * | 2017-11-30 | 2018-05-08 | 西南科技大学 | Target identification method based on binocular Visible Light Camera Yu thermal infrared camera |
| CN108830418A (en) * | 2018-06-14 | 2018-11-16 | 东北电力大学 | A kind of Short-Term Load Forecasting Method |
Non-Patent Citations (4)
| Title |
|---|
| JEFF LONG: "Tools of the Trade: Testing the Axe Bat’s Claims", 《HTTPS://WWW.BASEBALLPROSPECTUS.COM/NEWS/ARTICLE/29179/TOOLS-OF-THE-TRADE-TESTING-THE-AXE-BATS-CLAIMS/》 * |
| JEFF LONG: "Tools of the Trade: Testing the Axe Bat’s Claims", 《HTTPS://WWW.BASEBALLPROSPECTUS.COM/NEWS/ARTICLE/29179/TOOLS-OF-THE-TRADE-TESTING-THE-AXE-BATS-CLAIMS/》, 13 May 2016 (2016-05-13), pages 1 * |
| 姜玮: "均匀设计-BP神经网络优化超临界CO_2提取杜香挥发油工艺", 《农业工程学报》 * |
| 姜玮: "均匀设计-BP神经网络优化超临界CO_2提取杜香挥发油工艺", 《农业工程学报》, vol. 27, no. 2, 15 December 2011 (2011-12-15), pages 293 - 296 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112700500A (en) * | 2020-12-08 | 2021-04-23 | 中大检测(湖南)股份有限公司 | Binocular camera calibration method and device and readable storage medium |
| CN113221954A (en) * | 2021-04-15 | 2021-08-06 | 长春工业大学 | BP (Back propagation) classification algorithm based on improved bat algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104978947B (en) | Adjusting method, dispaly state regulating device and the display device of dispaly state | |
| CN108520197A (en) | A kind of Remote Sensing Target detection method and device | |
| CN105741267B (en) | The multi-source image change detecting method of cluster guidance deep neural network classification | |
| CN104850865B (en) | A kind of Real Time Compression tracking of multiple features transfer learning | |
| CN107169435A (en) | A kind of convolutional neural networks human action sorting technique based on radar simulation image | |
| CN103679641B (en) | Depth image enhancement method and device | |
| CN104320617B (en) | A kind of round-the-clock video frequency monitoring method based on deep learning | |
| CN109584290A (en) | A kind of three-dimensional image matching method based on convolutional neural networks | |
| CN106485227A (en) | A kind of Evaluation of Customer Satisfaction Degree method that is expressed one's feelings based on video face | |
| CN109102515A (en) | A kind of method for cell count based on multiple row depth convolutional neural networks | |
| CN107886157A (en) | A kind of new bat optimized algorithm system | |
| CN109887035A (en) | Based on bat algorithm optimization BP neural network binocular vision calibration | |
| CN109035292A (en) | Moving target detecting method and device based on deep learning | |
| CN109711401A (en) | A kind of Method for text detection in natural scene image based on Faster Rcnn | |
| CN104899866A (en) | Intelligent infrared small target detection method | |
| CN108362284A (en) | A kind of air navigation aid based on bionical hippocampus cognitive map | |
| CN108229589A (en) | A kind of ground cloud atlas sorting technique based on transfer learning | |
| CN112560619B (en) | Multi-focus image fusion-based multi-distance bird accurate identification method | |
| CN110009696A (en) | It is demarcated based on ant colony algorithm Optimized BP Neural Network trinocular vision | |
| CN115017591A (en) | Building structure group earthquake-resistant performance evaluation method | |
| CN107273799A (en) | A kind of indoor orientation method and alignment system | |
| CN111611889A (en) | Miniature insect pest recognition device in farmland based on improved convolutional neural network | |
| CN107329233B (en) | A kind of droplet type PCR instrument Atomatic focusing method neural network based | |
| CN111368900A (en) | Image target object identification method | |
| CN111126220A (en) | Real-time positioning method for video monitoring target |
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 |
Application publication date: 20190614 |
|
| WD01 | Invention patent application deemed withdrawn after publication |