CN111680741B - Automatic debugging method of computer-aided interferometer based on deep learning - Google Patents
Automatic debugging method of computer-aided interferometer based on deep learning Download PDFInfo
- Publication number
- CN111680741B CN111680741B CN202010500866.4A CN202010500866A CN111680741B CN 111680741 B CN111680741 B CN 111680741B CN 202010500866 A CN202010500866 A CN 202010500866A CN 111680741 B CN111680741 B CN 111680741B
- Authority
- CN
- China
- Prior art keywords
- training
- data
- network
- interferometer
- data set
- 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.)
- Active
Links
Images
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
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
- G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an automatic debugging method of a computer-aided interferometer based on deep learning. Acquiring an interferogram and a three-dimensional coordinate thereof as training data; classifying the expanded training data based on a K-means clustering algorithm to obtain a data set; dividing a data set into a training set and a verification set by adopting a k-fold cross verification method, and training the VGG-16 network model by using the training set and the verification set obtained by each fold; and inputting the interference pattern acquired in real time into the trained VGG-16 network model to obtain the detuning amount of the interferometer. According to the method, the specific interference pattern deviation amount is not required to be solved, the trained network classification data is fully utilized, the position coordinates of the interferometer are determined by classifying the interference patterns formed by the interferometer, the movement of the lens to be measured to the accurate position is further controlled, the advantages of a deep convolution network are fully played, and the method has the advantages of good robustness and high accuracy.
Description
Technical Field
The invention belongs to the technical field of image processing and computer vision, and particularly relates to an automatic debugging method of a computer-aided interferometer based on deep learning.
Background
Computer aided assembly and debugging technology (CAA) is a new technology which is separated from the traditional assembly and debugging, mainly depends on manual operation and operation experience of people and is based on computers and precise mobile equipment, in the optical field, various precise and complex equipment and devices are widely used, the accuracy is low and the time consumption is long depending on the traditional assembly and debugging method, the position misalignment amount of an optical system can be solved in real time by adopting the computer aided assembly and debugging technology, and then the position of an optical element is adjusted by controlling a mechanical structure. The computer aided installation and adjustment technology makes the optical system combined with electric, mechanical and soft organically, and is more intelligent and automatic.
The computer aided debugging method commonly used in the field of optical element detection at present mainly comprises a reverse optimization method, a downhill simplex method, a sensitivity matrix method and the like. The three algorithms are based on the principle that a simulation model is established by a computer, the detection result of the interferometer is compared with the simulation result of the computer, an evaluation function is adjusted until the value of the evaluation function is 0, and then the position misalignment amount is solved. When the position detuning amount is too much, the calculation efficiency is greatly reduced, the real-time performance cannot be guaranteed, and most of the methods need to research Zernike polynomial and wave aberration theory to solve the system detuning amount, and approximate errors exist in the formula.
The deep learning-based method generally comprises three parts of feature extraction, classifier training and detection. In recent years, with the development of technology, deep learning has become almost one of the hottest directions in the field of image processing. Computer-aided fitting techniques that use deep learning in the optical domain have also begun to develop vigorously. In 2015, the Eseter Oteo and Josepaprsa of Spain propose a new strategy for calculating the misalignment of an optical system by using an artificial neural network, and based on the uncertainty of a mathematical polynomial approximation Zernike coefficient, research is carried out to improve the approximation result of the polynomial by nonlinear function approximation, and the inclination value and the eccentricity in the misalignment of the system are calculated by using the neural network, so as to adjust the position of an optical element. The method improves the fitting precision of the Zernike polynomial, but does not fundamentally solve the error of a computer simulation model.
Disclosure of Invention
The invention aims to provide an automatic debugging method of a computer-aided interferometer based on deep learning, so as to improve the speed and the accuracy of the automatic debugging method of the computer-aided interferometer.
The technical scheme for realizing the purpose of the invention is as follows: an automatic debugging method of a computer-aided interferometer based on deep learning comprises the following specific steps:
step 1: acquiring an interferogram and a three-dimensional coordinate thereof as training data;
step 2: classifying the expanded training data based on a K-means clustering algorithm to obtain a data set;
and 3, step 3: dividing a data set into a training set and a verification set by adopting a k-fold cross verification method, and training the VGG-16 network model by using the training set and the verification set obtained by each fold;
and 4, step 4: and inputting the interference pattern acquired in real time into the trained VGG-16 network model to obtain the detuning amount of the interferometer.
Preferably, the specific steps of collecting the interferogram and its three-dimensional coordinates as training data are as follows:
step 11: fixing a to-be-measured mirror on an adjusting frame, adjusting the to-be-measured mirror to enable an interferometer image acquisition window to generate concentric circular interference fringes, adjusting the to-be-measured mirror to move in the X, Y, Z axis direction respectively, and acquiring an interference pattern corresponding to each movement and a three-dimensional coordinate of the position of the to-be-measured mirror;
step 12: and carrying out data enhancement on the acquired interference image to realize data expansion.
Preferably, the specific method for classifying the expanded training data based on the K-means clustering algorithm is as follows:
dividing three-dimensional coordinate data into K groups, randomly selecting K objects as initial clustering centers, calculating Euclidean distances between each coordinate and each clustering center, and allocating each coordinate to the clustering center closest to the coordinate; every time one coordinate is allocated, the clustering center is recalculated according to the existing coordinate in the clustering until the clustering center can not be calculated or does not change any more; and clustering the interference graph corresponding to the coordinates according to the clustering result of the coordinates to obtain a data set, and taking the category as a label value of the data set.
Preferably, the specific method for dividing the data set into a training set and a verification set by using a k-fold cross verification method and training the VGG-16 network model by using the training set and the verification set obtained by each fold comprises the following steps:
step 31: preprocessing a data set, including equalizing and adaptively cutting the data set;
step 32: dividing each type of data set at equal intervals;
step 33: randomly selecting a data set from each type of data set and using the data set as a verification set, and using the rest data sets as training sets;
step 34: freezing the fully-connected network part, inputting the training set and the network loss function into the feature extraction convolutional neural network part of the VGG-16 network, and training the feature extraction convolutional neural network by taking the corresponding category as a label value;
step 35: the freezing of the full connection layer is removed, the network weight is changed into the weight obtained by the training in the step 34, and the training set and the network loss function are input into the VGG-16 network model for the second training;
step 36: returning to the step 33 to train the next discount, and taking the discount weight with the minimum total loss function as the optimal weight of the VGG-16 network model.
Preferably, the initial learning rate is set to 0.01 in step 34, and Dropout is added between the last layer of the feature extraction convolutional neural network and the previous layer to selectively inactivate connections between neurons.
Preferably, the initial learning rate in step 35 is set to 1e -4 。
Compared with the prior art, the invention has the following remarkable advantages: according to the method, the specific interference pattern deviation amount does not need to be solved, the interference patterns are classified by means of a deep learning algorithm, the range of the system misadjustment amount is determined, and the speed and the accuracy of the automatic debugging method of the computer-aided interferometer are improved.
Drawings
FIG. 1 is a flow chart of an automatic installation and adjustment method of a computer-aided interferometer based on deep learning.
FIG. 2 is a schematic diagram of the detailed structure of VGG-16.
FIG. 3 is a flow chart of a k-fold cross-validation method.
Detailed Description
As shown in fig. 1, an automatic tuning method of a computer-aided interferometer based on deep learning includes two stages of training and testing, and includes the following specific steps:
step 1: and acquiring the interferogram and the three-dimensional coordinate thereof by using the interferometer as training data. Light rays are emitted from the interferometer, enter the interferometer after being reflected by the to-be-measured mirror, form interference inside the interferometer, collect images through the CCD and display an interference image in a computer matched with the interferometer. The method for acquiring the interferogram and the three-dimensional coordinate thereof as training data by using the interferometer comprises the following steps of:
step 11: fixing the lens to be measured on an adjusting frame, adjusting the lens to be measured to enable concentric circular interference fringes to appear on an image acquisition window of the interferometer, then finely adjusting X, Y, Z coordinates of the lens to be measured until the fringes are located at the center and are most sparse, and taking the position as an original point. And adjusting the to-be-measured mirror to move in the direction of the X, Y, Z axis respectively, and acquiring an interference pattern corresponding to each movement and a three-dimensional coordinate of the position of the to-be-measured mirror. At least 8000 interferograms with X, Y, Z different coordinates are acquired.
Step 12: carrying out data enhancement on the acquired interference image to realize data expansion: in consideration of the limitation of the data of the picture set, in order to avoid network overfitting, more data is obtained for learning and processing, and data enhancement operations such as turning, rotating, zooming, clipping, noise adding and the like are carried out on the image to expand the picture set, so that the neural network has better generalization effect.
Step 2: classifying the expanded training data based on a K-means clustering algorithm, which specifically comprises the following steps:
dividing three-dimensional coordinate data into K groups, randomly selecting K objects as initial clustering centers, then calculating Euclidean distances between each coordinate and each clustering center, and allocating each coordinate to the clustering center closest to the coordinate. The cluster centers and the coordinates assigned to them represent a cluster. Every time a coordinate is assigned, the cluster center is recalculated based on the existing coordinates in the cluster. This process will be repeated until no (or a minimum number of) cluster centers change again. And according to the clustering result of the coordinates, clustering the interferograms corresponding to the coordinates, namely clustering the interferograms according to the coordinates to form a data set for subsequent network training, and recording the categories as label values of the data set.
And step 3: dividing a data set into a training set and a verification set by adopting a k-fold cross verification method, and training the VGG-16 network model by using the training set and the verification set obtained by each fold;
the VGG-16 network model is implemented based on a VGG-16 framework. VGG Networks are models of Convolutional neural Networks proposed by Simony and Zisserman in the document "Very Deep conditional Networks for Large Scale Image Recognition". The prominent characteristic of the VGG-16 is conciseness and is shown in the following steps: (1) the convolutional layers all use the same convolutional kernel parameters. The convolutional layers are all denoted as conv3-XXX, where conv3 indicates that the convolutional kernel used by the convolutional layer has a size of 3, i.e., both width and height are 3, 3 × 3 is a very small convolutional kernel size, and in combination with other parameters (stride 1, padding same), it is possible to keep the width and height of each convolutional layer (tensor) the same as that of the previous layer (tensor). XXX represents the number of channels in the convolutional layer. (2) The pooling layers all use the same pooling nuclear parameters. The parameters of the pooling layers are 2 × 2, and the stride is 2, so that the width and height of each pooling layer are 1/2 of the previous layer. (3) The model is formed by stacking a plurality of convolution layers and pooling layers, and a deeper network structure is easier to form. The specific structure of the VGG-16 is shown in figure 2.
The VGG-16 network model includes: the Convolutional Layer and the pooling Layer form a feature extraction Convolutional neural network. Adopting a transfer learning strategy: the principle of transfer learning is that the VGG-16 network model of the invention is initialized by means of the relevant parameters of the training network of the general VGG-16, so that the network is prevented from sinking into a poor local minimum value in the subsequent training, and the convergence of the network is accelerated; improving the network structure and the weight number in the pre-training weights according to the characteristics of the training set: the number of output layer nodes of the network is determined according to the X, Y, Z axis adjustable range of the test mirror, and the number of hidden layer nodes is increased properly due to the fact that the number of output layer nodes is large;
in a further embodiment, the specific training process is as follows:
step 31: preprocessing the data set, and balancing the data set by considering the characteristic of data set unbalance, namely after the data set is subjected to K-means clustering, the number of images contained in each class is different, so that the consistency of the final classification result is ensured; the images with different sizes are subjected to self-adaptive cutting processing, so that the images meet the input standard of a general VGG-16 network;
step 32: dividing each type of data set at equal intervals;
step 33: randomly selecting a data set from each type of data set and using the data set as a verification set, and using the rest data sets as training sets;
step 34: and freezing the fully-connected network part, inputting the training set and the network loss function into the feature extraction convolutional neural network part of the VGG-16 network, and training the feature extraction convolutional neural network by taking the corresponding category as a label value. In the training process, the initial learning rate is set to be 0.01, Dropout is added between the last layer and the previous layer of the characteristic extraction convolutional neural network, connection between neurons is selectively inactivated, so that the total quantity of parameters is reduced, the overfitting probability of the network is reduced, and then an Earlystop mechanism, namely a mechanism for terminating training when the verification error is continuously increased for 3 times in the iteration process, is added, so that the optimal training effect is ensured. In the iteration process, the learning rate is slowly attenuated so as to reduce the parameter updating speed, and the lower limit of the learning rate is set so as to prevent the training progress from being influenced by the excessively low learning rate; the weight after each iteration is stored, so that the optimal weight file can be conveniently selected for subsequent fine adjustment;
step 35: the freezing of the full connection layer is removed, the network weight is changed into the weight obtained by the training in the step 34, and then the second training is carried out. The second training sets the initial learning rate to 1e -4 Meanwhile, the learning rate is slowly attenuated in each iteration, the attenuation is 0.85 of the previous iteration, and the fine adjustment effect is achieved; simultaneously, an Earlystop mechanism is added to ensure the optimal training effect;
step 36: returning to step 33 for training the next fold. And after all training is finished, taking a discounted weight file with a smaller total loss function (val _ loss) as the optimal weight of the VGG-16 network model. In the present invention, the total loss function (val _ loss) is defined as the sum of the loss function of the training set and the loss function of the validation set in the present reduction.
The invention adopts a k-fold cross validation method to ensure that the characteristics contained in the training set and the validation set can be covered in each training.
Step 37: and predicting the category of the existing picture by calling the trained weight file, and taking the predicted accuracy as a final model evaluation index.
Therefore, the training phase of the automatic debugging method of the computer-aided interferometer based on deep learning is completed.
And 4, step 4: in the testing stage, an interferogram acquired by the interferometer in real time is input into a trained model, so that a coordinate corresponding to the interferogram, namely the detuning amount of a lens to be tested of the interferometer can be obtained, and the lens to be tested is adjusted to move to a position with three-dimensional coordinates of 0 according to the detuning amount, so that automatic adjustment is realized.
The method is implemented based on an open source tool pycharm for deep learning.
In conclusion, the invention discloses an automatic debugging method of a computer-aided interferometer based on deep learning, which mainly explains a network training method. Firstly, data enhancement is carried out on the acquired and imported interferogram image, a data set is expanded, then the data set is classified by using a K-means clustering algorithm, and then training is carried out based on an improved VGG-16 network. The invention adopts a non-complex general network, can achieve good experimental effect, fully exerts the advantages of the deep convolution network, and has the advantages of simple design and high efficiency.
According to the method, the specific interference pattern deviation amount is not required to be solved, the trained network classification data is fully utilized, the position coordinates of the interferometer are determined by classifying the interference patterns formed by the interferometer, the movement of the lens to be measured to the accurate position is further controlled, the advantages of a deep convolution network are fully played, and the method has the advantages of good robustness and high accuracy.
The method directly learns the relationship between the image characteristics and the position misalignment amount of the interferogram from the training sample to obtain the model, then introduces the interferogram acquired by the interferometer in real time into the model, solves the position misalignment amount according to the image characteristics, does not need to solve the specific interference pattern deviation amount, and can improve the accuracy and efficiency of debugging.
Claims (5)
1. An automatic debugging method of a computer-aided interferometer based on deep learning is characterized by comprising the following specific steps:
step 1: acquiring an interferogram and a three-dimensional coordinate thereof as training data;
step 2: classifying the expanded training data based on a K-means clustering algorithm to obtain a data set;
and step 3: dividing a data set into a training set and a verification set by adopting a k-fold cross verification method, and training the VGG-16 network model by using the training set and the verification set obtained by each fold, wherein the specific method comprises the following steps:
step 31: preprocessing a data set, including equalizing and adaptively cutting the data set;
step 32: dividing each type of data set at equal intervals;
step 33: randomly selecting a data set from each type of data set and using the data set as a verification set, and using the rest data sets as training sets;
step 34: freezing the fully-connected network part, inputting the training set and the network loss function into the feature extraction convolutional neural network part of the VGG-16 network, and training the feature extraction convolutional neural network by taking the corresponding category as a label value;
step 35: the freezing of the full connection layer is removed, the network weight is changed into the weight obtained by the training in the step 34, and the training set and the network loss function are input into the VGG-16 network model for the second training;
step 36: returning to the step 33 to train the next fold, and taking the weight of the fold with the minimum total loss function as the optimal weight of the VGG-16 network model;
and 4, step 4: and inputting the interference pattern acquired in real time into the trained VGG-16 network model to obtain the detuning amount of the interferometer.
2. The method for automatically adjusting the computer-aided interferometer based on the deep learning of claim 1, wherein the specific steps of collecting the interferogram and the three-dimensional coordinates thereof as training data are as follows:
step 11: fixing a to-be-detected mirror on an adjusting frame, adjusting the to-be-detected mirror to enable an image acquisition window of the interferometer to generate concentric circular interference fringes, adjusting the to-be-detected mirror to move in the direction of an X, Y, Z axis respectively, and acquiring an interference pattern corresponding to each movement and a three-dimensional coordinate of the position of the to-be-detected mirror;
step 12: and carrying out data enhancement on the acquired interference image to realize data expansion.
3. The method for automatically adjusting the computer-aided interferometer based on the deep learning of claim 1, wherein the specific method for classifying the extended training data based on the K-means clustering algorithm is as follows:
dividing three-dimensional coordinate data into K groups, randomly selecting K objects as initial clustering centers, calculating Euclidean distances between each coordinate and each clustering center, and allocating each coordinate to the clustering center closest to the coordinate; every time one coordinate is allocated, the clustering center is recalculated according to the existing coordinate in the clustering until the clustering center can not be calculated or does not change any more; and clustering the interference graph corresponding to the coordinates according to the clustering result of the coordinates to obtain a data set, and taking the category as a label value of the data set.
4. The method of claim 1, wherein the initial learning rate is set to 0.01 and Dropout is added between the last layer and the previous layer of the feature extraction convolutional neural network to selectively inactivate connections between neurons in step 34.
5. The method of claim 1, wherein the initial learning rate is set to 1e in step 35 -4 。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010500866.4A CN111680741B (en) | 2020-06-04 | 2020-06-04 | Automatic debugging method of computer-aided interferometer based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010500866.4A CN111680741B (en) | 2020-06-04 | 2020-06-04 | Automatic debugging method of computer-aided interferometer based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111680741A CN111680741A (en) | 2020-09-18 |
| CN111680741B true CN111680741B (en) | 2022-09-20 |
Family
ID=72434934
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010500866.4A Active CN111680741B (en) | 2020-06-04 | 2020-06-04 | Automatic debugging method of computer-aided interferometer based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111680741B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116593137B (en) * | 2023-07-14 | 2023-09-29 | 苏州然玓光电科技有限公司 | Interferometer-based optical lens quality testing method and system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20170140473A (en) * | 2016-06-10 | 2017-12-21 | 주식회사 토모큐브 | Method and Apparatus for Identifying Cell Species Using 3D Refractive Index Tomography and Machine Learning Algorithm |
| CN107564025A (en) * | 2017-08-09 | 2018-01-09 | 浙江大学 | A kind of power equipment infrared image semantic segmentation method based on deep neural network |
| CN109726743A (en) * | 2018-12-12 | 2019-05-07 | 苏州大学 | A kind of retina OCT image classification method based on Three dimensional convolution neural network |
-
2020
- 2020-06-04 CN CN202010500866.4A patent/CN111680741B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20170140473A (en) * | 2016-06-10 | 2017-12-21 | 주식회사 토모큐브 | Method and Apparatus for Identifying Cell Species Using 3D Refractive Index Tomography and Machine Learning Algorithm |
| CN107564025A (en) * | 2017-08-09 | 2018-01-09 | 浙江大学 | A kind of power equipment infrared image semantic segmentation method based on deep neural network |
| CN109726743A (en) * | 2018-12-12 | 2019-05-07 | 苏州大学 | A kind of retina OCT image classification method based on Three dimensional convolution neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111680741A (en) | 2020-09-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109145939B (en) | Semantic segmentation method for small-target sensitive dual-channel convolutional neural network | |
| CN111563706A (en) | Multivariable logistics freight volume prediction method based on LSTM network | |
| CN114841257B (en) | Small sample target detection method based on self-supervision comparison constraint | |
| CN108665005B (en) | Method for improving CNN-based image recognition performance by using DCGAN | |
| WO2022083009A1 (en) | Customized product performance prediction method based on heterogeneous data error compensation fusion | |
| CN114092832B (en) | High-resolution remote sensing image classification method based on parallel hybrid convolutional network | |
| CN106228185A (en) | A kind of general image classifying and identifying system based on neutral net and method | |
| CN110543906B (en) | Automatic skin recognition method based on Mask R-CNN model | |
| CN113554156B (en) | Multitask image processing method based on attention mechanism and deformable convolution | |
| CN113228096A (en) | Optical correction by machine learning | |
| CN116992779B (en) | Simulation method and system of photovoltaic energy storage system based on digital twin model | |
| CN112598683A (en) | Frequency sweep OCT (optical coherence tomography) human eye image segmentation method based on frequency sweep optical coherence tomography | |
| CN111694977A (en) | Vehicle image retrieval method based on data enhancement | |
| CN111680741B (en) | Automatic debugging method of computer-aided interferometer based on deep learning | |
| CN115423145A (en) | Photovoltaic power prediction method based on Kmeans-VMD-WT-LSTM method | |
| CN109993208A (en) | A kind of clustering processing method having noise image | |
| CN114372926A (en) | Traditional Chinese medicine tongue tenderness identification method based on image restoration and convolutional neural network | |
| CN113762370A (en) | Depth network set generation method combined with Gaussian random field | |
| CN112651499A (en) | Structural model pruning method based on ant colony optimization algorithm and interlayer information | |
| CN112862063A (en) | Complex pipe network leakage positioning method based on deep belief network | |
| CN112508183A (en) | Method and device for constructing simplex neural network for image classification | |
| CN117034060A (en) | AE-RCNN-based flood classification intelligent forecasting method | |
| CN116933090A (en) | PM2.5 prediction method and device based on deep neural network | |
| CN117152427A (en) | Remote sensing image semantic segmentation method and system based on diffusion model and knowledge distillation | |
| CN113657438B (en) | Drainage pipeline disease detection method of VGG neural network under thermal infrared mode |
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 |