CN111652177A - Signal feature extraction method based on deep learning - Google Patents
Signal feature extraction method based on deep learning Download PDFInfo
- Publication number
- CN111652177A CN111652177A CN202010533429.2A CN202010533429A CN111652177A CN 111652177 A CN111652177 A CN 111652177A CN 202010533429 A CN202010533429 A CN 202010533429A CN 111652177 A CN111652177 A CN 111652177A
- Authority
- CN
- China
- Prior art keywords
- time sequence
- deep
- belief network
- deep learning
- training data
- 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
- 238000013135 deep learning Methods 0.000 title claims abstract description 34
- 238000000605 extraction Methods 0.000 title claims abstract description 17
- 238000000034 method Methods 0.000 claims abstract description 52
- 238000001228 spectrum Methods 0.000 claims abstract description 42
- 238000012549 training Methods 0.000 claims abstract description 34
- 238000012545 processing Methods 0.000 claims abstract description 12
- 238000007906 compression Methods 0.000 claims abstract description 7
- 230000006835 compression Effects 0.000 claims abstract description 6
- 230000002068 genetic effect Effects 0.000 claims abstract description 5
- 238000001914 filtration Methods 0.000 claims description 16
- 238000010801 machine learning Methods 0.000 claims description 7
- 230000008859 change Effects 0.000 claims description 3
- 230000004927 fusion Effects 0.000 abstract description 2
- 230000009466 transformation Effects 0.000 description 7
- 230000008569 process Effects 0.000 description 5
- 238000011161 development Methods 0.000 description 4
- 238000013178 mathematical model Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000010187 selection method Methods 0.000 description 3
- 230000003595 spectral effect Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 241001123248 Arma Species 0.000 description 1
- 241000282414 Homo sapiens Species 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000009123 feedback regulation Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- YHXISWVBGDMDLQ-UHFFFAOYSA-N moclobemide Chemical compound C1=CC(Cl)=CC=C1C(=O)NCCN1CCOCC1 YHXISWVBGDMDLQ-UHFFFAOYSA-N 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000008054 signal transmission Effects 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Signal Processing (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a signal feature extraction method based on deep learning, which comprises the following steps: 1) compressing the original sparse spectrum signal to form a time sequence spectrum signal, and completing denoising processing after compression to obtain a reconstructed time sequence spectrum signal serving as training data; 2) marking the unmarked time sequence spectrum signals in the training data by using a semi-supervised learning method; simulating each group of time sequence spectrum signals, and measuring the correlation among the time sequence spectrum signals by using a genetic algorithm; 3) and classifying and predicting each group of time sequence spectrum signals by using the marked training data and a deep learning method. The invention improves the model training efficiency through multi-method fusion, realizes the automatic label adding without label, and has more accurate prediction result.
Description
Technical Field
The invention belongs to the technical field of information, and particularly relates to a signal feature extraction method based on deep learning.
Background
Two common deep learning model training modes are provided: a depth learning model with feedback adjustment and a cascaded depth model without feedback. The depth models including feedback regulation, such as convolutional neural network, depth belief network, automatic coding representation and the like, and the depth models without feedback, such as deep boosting, deep fisher, deep pca and the like. The two deep learning models are regular combinations of feature extraction and feature selection, so how to set up a feature transformation and feature selection method and a combination mode thereof are important links for research on the basis of predecessors.
(1) Feature extraction
Feature extraction is the process of obtaining information from data features. And acquiring information of the data by transforming the characteristics. The simplest feature transformation modes comprise PCA transformation, linear discriminant analysis and the like, wherein the PCA transformation is to search the main shaft direction of the data, so that the data can be projected and then represent as much information as possible by using as few features as possible; the linear discriminant analysis is to enable data in a new space to have larger inter-class distance and smaller intra-class distance through linear projection; the Gabor filtering transformation is an important transformation mode in the field of images, and edge information in the images can be effectively extracted by setting different parameters of Gaussian kernel functions in a filter; the method comprises the steps that key points are detected in a multi-scale space by adopting a DOG operator, and local information of an image is extracted by SIFT (scale invariant feature transform) characteristics, so that the image keeps good invariance to rotation, scale scaling and brightness transformation; by dividing the whole image into a plurality of connected regions and counting the histogram of each region, the HOG feature can effectively deal with the rotation and scale invariance of the image.
(2) Feature selection
Feature selection is a method of picking out a subset of features from data features that are helpful to a task. Common feature selection methods can be classified into 3 categories: filter, Wrapper and Embedded methods. The Filer method selects features one by one according to the size of the corresponding feature estimation indexes of each feature, and the indexes are usually some statistical characteristics of data. The Wrapper method is to select features according to the classification accuracy when the selected feature subspace is classified, and because the feature subspace is uncertain, the Wrapper method needs to train for many times to give the selected feature space. The Embedded method first determines the model using the features and then searches the feature space for a feature subspace that improves the performance of the model. Feature selection can be further divided into supervised learning and unsupervised learning from whether the sample contains class information. In early unsupervised learning, some metric characteristics are defined first, then the metric characteristics of the features are calculated one by one and selected in sequence, the metric characteristics may be clustering effect, information redundancy, and the like, and some representative metrics include Laplacian Score, Traceratio, and the like. However, the feature extraction method relying on search requires a huge amount of calculation, so researchers have begun to consider clustering algorithms that do not require a search space any more, and a series of spectral clustering algorithms have been proposed based on the similar characteristics of features.
Through reasonable combination of proper feature extraction and feature selection methods, deep learning has achieved a series of achievements in various fields. The successful use of convolutional neural networks has made a breakthrough in the field of speech recognition, as have the best current results in the field of target recognition.
In addition to convolutional neural networks, deep belief networks are also an important component of deep learning. The idea of the deep belief network is to learn shallow features by means of a greedy strategy and then obtain more abstract description by means of combination of the shallow features. With the development of the neural network of the deep structure, a plurality of improved algorithms for deep learning appear, and the deep learning is widely applied. Deep learning has been successfully applied to speech recognition, image processing, and the like. Hinton indicates that in order to train the DBN, a restricted boltzmann machine of each layer is trained by unsupervised greedy, a deep belief network DBN is constructed by a combination of a set of restricted boltzmann machines, and then the constructed DBN is finely tuned by a conventional global learning algorithm so that the network is optimal. The core idea of deep learning is that unsupervised learning is used for pre-train of each layer network; only one layer is trained by unsupervised learning each time, and the training result is used as the input of the higher layer; supervised learning is used to adjust all layers.
A Deep Belief Network (DBN) is a typical generative deep structure. Deep learning has been successfully applied to various mode classification problems, which brings great convenience to human beings, but is still only in the development stage at present, and many problems are worth further solving, such as:
(1) the model training time is generally too long, and how to improve the training speed and the practicability of deep learning and the like.
(2) Feature learning of labeled data still dominates, data in real life are unmarked data, and it is obviously unrealistic to add artificial labels to the data one by one, so that a technology for automatically adding the unmarked labels can be researched.
(3) The single deep learning method cannot bring good effect, and how to perfectly apply the deep learning method to practical application by combining with other methods or other methods.
Disclosure of Invention
In view of the above, in order to solve the above problems in the prior art, the present invention provides a deep learning-based signal feature extraction method that improves model training efficiency and realizes multi-method fusion of automatic label addition without label.
The technical scheme of the invention is that a signal feature extraction method based on deep learning is provided, which comprises the following steps:
1) compressing the original sparse spectrum signal to form a time sequence spectrum signal, and completing denoising processing after compression to obtain a reconstructed time sequence spectrum signal serving as training data;
2) marking the unmarked time sequence spectrum signals in the training data by using a semi-supervised learning method; simulating each group of time sequence spectrum signals, and measuring the correlation among the time sequence spectrum signals by using a genetic algorithm;
3) and classifying and predicting each group of time sequence spectrum signals by using the marked training data and a deep learning method.
Optionally, in step 1), compressing the original sparse spectrum signal by using a dynamic subspace tracking algorithm in a greedy tracking algorithm, denoising the compressed sparse spectrum signal by using an unscented kalman filter algorithm, or directly predicting the reconstructed time sequence spectrum signal by using the unscented kalman filter algorithm to obtain a first prediction result.
Optionally, the semi-supervised learning method includes an active learning method and a deep belief network in deep learning, the active learning method is used for labeling incompletely labeled training data, the deep belief network is a probability generation model, the probability generation model includes a plurality of restricted boltzmann machines, joint distribution between observation data and labels is established, the restricted boltzmann machines are composed of a visible layer and a hidden layer, connection exists between the layers, no connection exists between the units in the layers, and the deep belief network classifies and predicts the training data to obtain a classification result and a second prediction result.
Optionally, the active learning method is incorporated into the deep belief network as a first hidden layer.
Optionally, firstly, training the restricted boltzmann machine to enable energy of each layer to be set to an extreme value, and after all the restricted boltzmann machines are trained, adjusting weights of each layer by using an actively learned label result to finally obtain a classification surface of the deep belief network; when the depth belief network is used for prediction, all training data are converted into residual errors in the unscented Kalman filtering algorithm model, the depth belief network is used as a machine learning model to predict the change curve of the residual errors in the unscented Kalman filtering algorithm model, and finally, the prediction result of the depth belief network is combined with the prediction result of the unscented Kalman filtering algorithm to obtain a final result.
Compared with the prior art, the invention has the following advantages:
in the invention, the classification and prediction results obtained by the expanded deep belief network (combining active learning and unscented Kalman filtering algorithm) are superior to the deep belief network in the prior art, and the method has great application value in the signal processing field and the machine learning field.
Drawings
FIG. 1 is a flow chart of the present invention.
Detailed Description
Preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, but the present invention is not limited to only these embodiments. The invention is intended to cover alternatives, modifications, equivalents and alternatives which may be included within the spirit and scope of the invention.
In the following description of the preferred embodiments of the present invention, specific details are set forth in order to provide a thorough understanding of the present invention, and it will be apparent to those skilled in the art that the present invention may be practiced without these specific details.
Referring to fig. 1, the specific implementation process of the present invention is as follows:
the signal feature extraction method based on deep learning comprises the following steps:
1) compressing the original sparse spectrum signal to form a time sequence spectrum signal, and completing denoising processing after compression to obtain a reconstructed time sequence spectrum signal serving as training data;
2) marking the unmarked time sequence spectrum signals in the training data by using a semi-supervised learning method; simulating each group of time sequence spectrum signals, and measuring the correlation among the time sequence spectrum signals by using a genetic algorithm;
3) and classifying and predicting each group of time sequence spectrum signals by using the marked training data and a deep learning method.
In the step 1), an original sparse spectrum signal is compressed by adopting a dynamic subspace tracking algorithm in a greedy tracking algorithm, the compressed sparse spectrum signal is subjected to denoising processing by adopting an unscented kalman filtering algorithm, or a reconstructed time sequence spectrum signal is directly predicted by utilizing the unscented kalman filtering algorithm, and a first prediction result is obtained.
The semi-supervised learning method comprises an active learning method and a deep belief network in deep learning, wherein the active learning method is used for marking incompletely marked training data, the deep belief network is a probability generation model, the probability generation model comprises a plurality of limiting type Boltzmann machines, joint distribution between observation data and labels is established, the limiting type Boltzmann machines are composed of a visible layer and a hidden layer, the layers are connected, the units in the layers are not connected, and the deep belief network classifies and predicts the training data to obtain a classification result and a second prediction result.
The idea of the deep belief network is to learn shallow features by means of a greedy strategy and then obtain more abstract description by means of combination of the shallow features. With the development of the neural network of the deep structure, a plurality of improved algorithms for deep learning appear, and the deep learning is widely applied. Deep learning has been successfully applied to speech recognition, image processing, and the like. Hinton indicates that in order to train the DBN, a restricted boltzmann machine of each layer is trained by unsupervised greedy, a deep belief network DBN is constructed by a combination of a set of restricted boltzmann machines, and then the constructed DBN is finely tuned by a conventional global learning algorithm so that the network is optimal. The core idea of deep learning is that unsupervised learning is used for pre-train of each layer network; only one layer is trained by unsupervised learning each time, and the training result is used as the input of the higher layer; supervised learning is used to adjust all layers.
A Deep Belief Network (DBN) is a typical generative deep structure. Assuming a DBN has n hidden layers, let gi denote the vector of the ith hidden layer, the model of the DBN can be expressed as:
p(x,g1,g2,…,gn)＝p(x|g1)p(g1|g2)…p(gn-1,gn) (1)
wherein the conditional probability p (g)i|gi+1) Comprises the following steps:
in the formula:
-the jth node of the ith layer, the weight matrix of Wi ith layer, σ is
The active learning method is combined in the deep belief network as a first hidden layer.
Firstly, setting the energy of each layer to an extreme value by utilizing a training restricted Boltzmann machine, and adjusting the weight of each layer by utilizing an actively learned label result after all the restricted Boltzmann machines are trained to finally obtain a classification surface of a deep belief network; when the depth belief network is used for prediction, all training data are converted into residual errors in the unscented Kalman filtering algorithm model, the depth belief network is used as a machine learning model to predict the change curve of the residual errors in the unscented Kalman filtering algorithm model, and finally, the prediction result of the depth belief network is combined with the prediction result of the unscented Kalman filtering algorithm to obtain a final result.
More specific examples of the invention are: the method of the present invention is applied to the Google spectrum database for implementation and verification. First, compression processing will be performed using the obtained sparse data. During the compression process, the compression results are compared between various signal reconstruction techniques. Meanwhile, the accuracy improvement rate of the Dynamic SP on the traditional SP algorithm can be obtained, and the Dynamic SP has profound influence on the whole spectrum signal processing field.
Next, the compressed spectrum signal is filtered using the UKF. The UKF has a better fit to non-linear time series, especially time series with Gaussian distribution. In this step, other mathematical models may be compared, such as fitting results to autoregressive ARX, ARMA, ARMAX and ARIMA models, and so forth. The degree of fit of the UKF model can be predicted to be optimal for the spectral signal. The UKF can be directly used for prediction, and the filtered waveform after filtering by the UKF is predicted through DBNs, so that the effect is better.
Next, in the labeled training dataset, two separate machine learning methods can be compared: marking results of the active learning method and the deep belief network method, and combining the two methods to mark.
Finally, the deep belief network will be used for classification and prediction. The classification and prediction results obtained by the extended deep belief network (combining active learning and UKF) in the invention are superior to those of the traditional deep belief network. The method has considerable research value in the field of signal processing or machine learning.
The main methods used in the present invention are as follows:
1) signal reconstruction for sparse or compressible signals.
The NyKuist sampling theorem determines that the sampling rate of signals in the traditional signal processing method is at least twice of the bandwidth of the signals, however, with the rapid development of telecommunication technology, the bandwidth required by signal transmission is increasingly wide, the existing hardware facilities are difficult to meet the requirements, and the generated signal data lower than the standard adopted frequency is called as sparse signals. Sparse signals require reconstruction of dense samples of the signal by a reconstruction algorithm to form ordered, accurate time series data that can be used for machine learning.
Signal reconstruction techniques can be divided into three categories: a combinatorial optimization algorithm, a convex optimization algorithm, and a greedy tracking algorithm. The greedy tracking method is a relatively novel algorithm, gives consideration to complexity and reconstruction accuracy, and is the key point of the research of the project. The greedy tracing algorithm can be further classified into a non-backtracking-like algorithm represented by an Orthogonal Matching Pursuit (OMP) algorithm and a backtracking-like algorithm represented by a Subspace Pursuit (SP) algorithm. Compared with a non-backtracking algorithm, the backtracking algorithm has higher reconstruction precision. The project adopts a backtracking algorithm to reconstruct sparse radio signals.
2) And denoising the reconstructed signal data.
Denoising using a mathematical model is a recent research result of project team members. For non-linear time series such as radio signal data we can use auto-regressive model with evolution variables (ARX) model or Unscented Kalman Filter (UKF) model for de-noising. The mathematical model generated in the denoising process can also be used for classification and prediction.
3) Incomplete labeling signals are labeled and classified.
The radio signal data sets obtained from the raw database are often incompletely labeled, with some data sets having even few signals labeled. At this point, the training data set needs to be labeled and classified. Traditional semi-supervised classification methods include Active learning (Active learning) and deep belief networks in deep learning. In the project, the two methods are combined, and the genetic algorithm is used for evaluating the correlation between the spectrum signals to label and classify the training data so as to obtain a more perfect classifier.
4) The spectral signals are classified and predicted.
By using the classifier formed in the previous step, the unknown spectrum signal can be classified and predicted. In the prediction process, the influence of a mathematical model generated in the denoising process can be added, so that the prediction result is more accurate.
Although the embodiments have been described and illustrated separately, it will be apparent to those skilled in the art that some common techniques may be substituted and integrated between the embodiments, and reference may be made to one of the embodiments not explicitly described, or to another embodiment described.
The above-described embodiments do not limit the scope of the present invention. Any modification, equivalent replacement, and improvement made within the spirit and principle of the above-described embodiments should be included in the protection scope of the technical solution.
Claims (5)
1. A signal feature extraction method based on deep learning is characterized in that: the method comprises the following steps:
1) compressing the original sparse spectrum signal to form a time sequence spectrum signal, and completing denoising processing after compression to obtain a reconstructed time sequence spectrum signal serving as training data;
2) marking the unmarked time sequence spectrum signals in the training data by using a semi-supervised learning method; simulating each group of time sequence spectrum signals, and measuring the correlation among the time sequence spectrum signals by using a genetic algorithm;
3) and classifying and predicting each group of time sequence spectrum signals by using the marked training data and a deep learning method.
2. The signal feature extraction method based on deep learning of claim 1, characterized in that: in the step 1), an original sparse spectrum signal is compressed by adopting a dynamic subspace tracking algorithm in a greedy tracking algorithm, the compressed sparse spectrum signal is subjected to denoising processing by adopting an unscented kalman filtering algorithm, or a reconstructed time sequence spectrum signal is directly predicted by utilizing the unscented kalman filtering algorithm, and a first prediction result is obtained.
3. The signal feature extraction method based on deep learning according to claim 2, characterized in that: the semi-supervised learning method comprises an active learning method and a deep belief network in deep learning, wherein the active learning method is used for marking incompletely marked training data, the deep belief network is a probability generation model, the probability generation model comprises a plurality of limiting type Boltzmann machines, joint distribution between observation data and labels is established, the limiting type Boltzmann machines are composed of a visible layer and a hidden layer, the layers are connected, the units in the layers are not connected, and the deep belief network classifies and predicts the training data to obtain a classification result and a second prediction result.
4. The signal feature extraction method based on deep learning according to claim 3, characterized in that: the active learning method is combined in the deep belief network as a first hidden layer.
5. The signal feature extraction method based on deep learning of claim 4, characterized in that: firstly, setting the energy of each layer to an extreme value by utilizing a training restricted Boltzmann machine, and adjusting the weight of each layer by utilizing an actively learned label result after all the restricted Boltzmann machines are trained to finally obtain a classification surface of a deep belief network; when the depth belief network is used for prediction, all training data are converted into residual errors in the unscented Kalman filtering algorithm model, the depth belief network is used as a machine learning model to predict the change curve of the residual errors in the unscented Kalman filtering algorithm model, and finally, the prediction result of the depth belief network is combined with the prediction result of the unscented Kalman filtering algorithm to obtain a final result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010533429.2A CN111652177A (en) | 2020-06-12 | 2020-06-12 | Signal feature extraction method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010533429.2A CN111652177A (en) | 2020-06-12 | 2020-06-12 | Signal feature extraction method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111652177A true CN111652177A (en) | 2020-09-11 |
Family
ID=72344737
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010533429.2A Pending CN111652177A (en) | 2020-06-12 | 2020-06-12 | Signal feature extraction method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111652177A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112332853A (en) * | 2020-11-02 | 2021-02-05 | 重庆邮电大学 | Time sequence data compression and recovery method based on power system |
| CN112420023A (en) * | 2020-11-26 | 2021-02-26 | 杭州音度人工智能有限公司 | Music infringement detection method |
| CN115965080A (en) * | 2022-11-07 | 2023-04-14 | 河海大学 | New energy power generation unit operation state identification method and device and storage medium |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106679672A (en) * | 2017-01-15 | 2017-05-17 | 吉林大学 | AGV (Automatic Guided Vehicle) location algorithm based on DBN (Dynamic Bayesian Network) and Kalman filtering algorithm |
| CN106971410A (en) * | 2017-03-27 | 2017-07-21 | 华南理工大学 | A kind of white matter fiber tract method for reconstructing based on deep learning |
| CN107463890A (en) * | 2017-07-20 | 2017-12-12 | 浙江零跑科技有限公司 | A kind of Foregut fermenters and tracking based on monocular forward sight camera |
| CN107808661A (en) * | 2017-10-23 | 2018-03-16 | 中央民族大学 | A kind of Tibetan voice corpus labeling method and system based on collaborative batch Active Learning |
| CN107958216A (en) * | 2017-11-27 | 2018-04-24 | 沈阳航空航天大学 | Based on semi-supervised multi-modal deep learning sorting technique |
| CN108509859A (en) * | 2018-03-09 | 2018-09-07 | 南京邮电大学 | A kind of non-overlapping region pedestrian tracting method based on deep neural network |
| CN109978041A (en) * | 2019-03-19 | 2019-07-05 | 上海理工大学 | A kind of hyperspectral image classification method based on alternately update convolutional neural networks |
| CN110728230A (en) * | 2019-10-10 | 2020-01-24 | 江南大学 | Signal modulation mode identification method based on convolution limited Boltzmann machine |
| CN111156987A (en) * | 2019-12-18 | 2020-05-15 | 东南大学 | Inertia/astronomical combined navigation method based on residual compensation multi-rate CKF |
| CN111160139A (en) * | 2019-12-13 | 2020-05-15 | 中国科学院深圳先进技术研究院 | Electrocardiosignal processing method and device and terminal equipment |
-
2020
- 2020-06-12 CN CN202010533429.2A patent/CN111652177A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106679672A (en) * | 2017-01-15 | 2017-05-17 | 吉林大学 | AGV (Automatic Guided Vehicle) location algorithm based on DBN (Dynamic Bayesian Network) and Kalman filtering algorithm |
| CN106971410A (en) * | 2017-03-27 | 2017-07-21 | 华南理工大学 | A kind of white matter fiber tract method for reconstructing based on deep learning |
| CN107463890A (en) * | 2017-07-20 | 2017-12-12 | 浙江零跑科技有限公司 | A kind of Foregut fermenters and tracking based on monocular forward sight camera |
| CN107808661A (en) * | 2017-10-23 | 2018-03-16 | 中央民族大学 | A kind of Tibetan voice corpus labeling method and system based on collaborative batch Active Learning |
| CN107958216A (en) * | 2017-11-27 | 2018-04-24 | 沈阳航空航天大学 | Based on semi-supervised multi-modal deep learning sorting technique |
| CN108509859A (en) * | 2018-03-09 | 2018-09-07 | 南京邮电大学 | A kind of non-overlapping region pedestrian tracting method based on deep neural network |
| CN109978041A (en) * | 2019-03-19 | 2019-07-05 | 上海理工大学 | A kind of hyperspectral image classification method based on alternately update convolutional neural networks |
| CN110728230A (en) * | 2019-10-10 | 2020-01-24 | 江南大学 | Signal modulation mode identification method based on convolution limited Boltzmann machine |
| CN111160139A (en) * | 2019-12-13 | 2020-05-15 | 中国科学院深圳先进技术研究院 | Electrocardiosignal processing method and device and terminal equipment |
| CN111156987A (en) * | 2019-12-18 | 2020-05-15 | 东南大学 | Inertia/astronomical combined navigation method based on residual compensation multi-rate CKF |
Non-Patent Citations (6)
| Title |
|---|
| SHUSEN ZHOU ET.: "active deep network for semi-supervised sentiment classification", 《PROCEEDINGS OF THE 23RD INTERNATIONAL CONFERENCE ON COMPUTATIONAL LINGUISTICS》, pages 1515 - 1523 * |
| 任珈民;宫宁生;韩镇阳;: "基于YOLOv3与卡尔曼滤波的多目标跟踪算法", 计算机应用与软件, no. 05 * |
| 任珈民;宫宁生;韩镇阳;: "基于YOLOv3与卡尔曼滤波的多目标跟踪算法", 计算机应用与软件, no. 05, 12 May 2020 (2020-05-12) * |
| 徐时怀 等: "基于云平台和深度学习的 软件ＧＵＩ自动测试系统", vol. 29, no. 4, pages 398 * |
| 田金鹏;薛莹;闵天;: "基于卡尔曼滤波的稀疏流信号动态压缩感知", 电子测量技术, no. 19, 8 October 2018 (2018-10-08) * |
| 荆楠;毕卫红;胡正平;王林;: "动态压缩感知综述", 自动化学报, no. 01, 15 January 2015 (2015-01-15) * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112332853A (en) * | 2020-11-02 | 2021-02-05 | 重庆邮电大学 | Time sequence data compression and recovery method based on power system |
| CN112420023A (en) * | 2020-11-26 | 2021-02-26 | 杭州音度人工智能有限公司 | Music infringement detection method |
| CN112420023B (en) * | 2020-11-26 | 2022-03-25 | 杭州音度人工智能有限公司 | Music infringement detection method |
| CN115965080A (en) * | 2022-11-07 | 2023-04-14 | 河海大学 | New energy power generation unit operation state identification method and device and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| McConville et al. | N2d:(not too) deep clustering via clustering the local manifold of an autoencoded embedding | |
| Sun et al. | A particle swarm optimization-based flexible convolutional autoencoder for image classification | |
| Jiang et al. | A survey on artificial intelligence in Chinese sign language recognition | |
| CN109671102B (en) | Comprehensive target tracking method based on depth feature fusion convolutional neural network | |
| CN103425996B (en) | A kind of large-scale image recognition methods of parallel distributed | |
| CN108875816A (en) | Merge the Active Learning samples selection strategy of Reliability Code and diversity criterion | |
| CN111652177A (en) | Signal feature extraction method based on deep learning | |
| CN109190544B (en) | Human identity recognition method based on sequence depth image | |
| CN113283282B (en) | Weak supervision time sequence action detection method based on time domain semantic features | |
| CN112990282B (en) | Classification method and device for fine-granularity small sample images | |
| Li et al. | Multiple VLAD encoding of CNNs for image classification | |
| CN103985143A (en) | Discriminative online target tracking method based on videos in dictionary learning | |
| Yang et al. | Structurally enhanced incremental neural learning for image classification with subgraph extraction | |
| Wang et al. | Deep generative mixture model for robust imbalance classification | |
| CN117198468A (en) | Intervention scheme intelligent management system based on behavior recognition and data analysis | |
| Chen et al. | Sample balancing for deep learning-based visual recognition | |
| Zarbakhsh et al. | Low-rank sparse coding and region of interest pooling for dynamic 3D facial expression recognition | |
| Terzopoulos | Multi-adversarial variational autoencoder networks | |
| Wistuba | Bayesian optimization combined with incremental evaluation for neural network architecture optimization | |
| CN114048843A (en) | Small sample learning network based on selective feature migration | |
| CN116630816B (en) | SAR target recognition method, device, equipment and medium based on prototype comparison learning | |
| Jia et al. | Latent task adaptation with large-scale hierarchies | |
| CN117093924A (en) | Rotary machine variable working condition fault diagnosis method based on domain adaptation characteristics | |
| Novakovic et al. | Classification accuracy of neural networks with pca in emotion recognition | |
| Kacem et al. | Disentangled face identity representations for joint 3D face recognition and neutralisation |
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 |