CN103346799A - Method for identifying gas based on compressed sensing theory - Google Patents
Method for identifying gas based on compressed sensing theory Download PDFInfo
- Publication number
- CN103346799A CN103346799A CN2013102205872A CN201310220587A CN103346799A CN 103346799 A CN103346799 A CN 103346799A CN 2013102205872 A CN2013102205872 A CN 2013102205872A CN 201310220587 A CN201310220587 A CN 201310220587A CN 103346799 A CN103346799 A CN 103346799A
- Authority
- CN
- China
- Prior art keywords
- data
- sparse matrix
- matrix
- compressed sensing
- sparse
- 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
Images
Abstract
The invention discloses a method for identifying gas based on the compressed sensing theory. The method includes the steps that compressed data are collected in an undersampled mode; the collected compressed data are reconstructed to acquire reconstructed data; the reconstructed data are utilized to train a back-propagation neural network, and the trained back-propagation neural network is stored; data to be detected are input to the trained back-propagation neural network, and the trained back-propagation neural network identifies the data to be detected to achieve qualitative identification of the gas. By the adoption of the method for identifying gas based on the compressed sensing theory, the problems that a gas detecting method in the prior art is large in transmission and storage data amount and inaccurate in identification are solved, and the goal of achieving accurate qualitative identification through small data amount is achieved.
Description
Technical field
The present invention relates to the sensor array signal processing technology field, particularly relate to a kind of gas recognition methods based on the compressed sensing theory.
Background technology
Gas sensor is because the ubiquity cross-sensitivity, its output signal is subjected to the influence of factors such as temperature, humidity and environmental condition, stability and selectivity are relatively poor, make its application be confined to the not high or better simply occasion of gas ingredients of accuracy of detection, as being harmful to gas leakage alarm etc.Traditional solution is to eliminate above-mentioned various adverse effect by seeking new sensitive material, device architecture and compensating circuit, yet this solution not only makes device architecture complicated, and the device manufacturing cost is improved.
The compressed sensing theory is that of occurring in recent years takes full advantage of the sparse property of signal or compressible brand-new signals collecting, encoding and decoding theory.These theoretical basic ideas are from few extracting data information as much as possible of trying one's best.This theory is pointed out, as long as signal is sparse in certain territory, so just can utilize an incoherent matrix that it is carried out projection, these a spot of projection values of recycling are found the solution an optimization problem, finally with a high probability reconstruct primary signal, that is to say that a compressible signal can carry out undistorted sampling with a speed more much lower than Nyquist sampling rate, utilize these a spot of data can accurately reconstruct initial data then.
Neural net is with its Nonlinear Mapping, parallel processing and height self study, self-organizing, adaptive ability, can solve the nonlinear problem that the gas sensor cross-sensitivity produces effectively, and suppress drift or the noise of transducer to a certain extent, help the raising of gas accuracy of detection.Therefore, high-performance, the detection technique that combines with the intelligent identification technology that with the neural net is representative of gas sensor has become the fashion trend that gas at present detects cheaply.
In neural net, backpropagation (Back Propagation, BP) neural net is a kind of Multi-layered Feedforward Networks according to the training of error Back-Propagation algorithm, generally constituted by a plurality of network layers, comprising an input layer, one or several hidden layer, an output layer, adopt totally interconnected connecing between layer and the layer, do not interconnect with not existing between the layer neuron.Hidden neuron adopts Sigmoid type transfer function usually, and output layer adopts purelin type transfer function.The learning process of BP neural net is made up of propagated forward and backpropagation, and in the propagated forward process, input pattern is successively handled through input layer, hidden layer, and passes to output layer.If the output in that output layer can not obtain expecting then changes back-propagation process over to, error amount is successively oppositely transmitted along connecting path, and revise each layer connection weights, up to the training error that reaches expection.
Summary of the invention
(1) technical problem that will solve
In order to overcome above-mentioned the deficiencies in the prior art, the invention provides a kind of gas recognition methods based on the compressed sensing theory, with solve present gas detect in transmission storage data volume big, identify coarse problem, reach the purpose that reaches accurate qualitative identification with less data volume.
(2) technical scheme
For achieving the above object, the invention provides a kind of gas recognition methods based on the compressed sensing theory, this method comprises: step 1: gather data after the compression in the mode of owing to sample; Step 2: the data after this compression of gathering are reconstructed obtain reconstruct data; Step 3: utilize this reconstruct data that reverse transmittance nerve network is trained, and preserve the reverse transmittance nerve network that trains; Step 4: with the reverse transmittance nerve network that the testing data input trains, this reverse transmittance nerve network that trains is identified testing data, realizes the qualitative identification to gas.
In the such scheme, gather data after the compression in the mode of owing to sample described in the step 1, specifically comprise: the array node of sensor network is gathered compressible initial data; This initial data is carried out sparse decomposition, obtain first sparse matrix, this first sparse matrix is a sparse matrix relevant with initial data; This first sparse matrix is carried out non-linear projection handle, obtain second sparse matrix, the element in this second sparse matrix is the combination at random of element in this first sparse matrix; Owe sampling to the data that coefficient is bigger in this second sparse matrix to carry out low speed less than the frequency of Nyquist sampling frequency.
In the such scheme, described this initial data is carried out sparse decomposition, obtain first sparse matrix, comprising: construct the sparse matrix that a random Gaussian distributes, the initial data of the gathering sparse matrix with the random Gaussian distribution of this structure is multiplied each other, obtain first sparse matrix.The sparse matrix that random Gaussian of described structure distributes, comprise: select Gauss's matrix that dimension is M * N, each element of this Gauss's matrix is Gaussian distributed all, then each row of this Gauss's matrix is carried out normalized and obtains sparse matrix Ψ.
In the such scheme, described the initial data of the gathering sparse matrix with the random Gaussian distribution of this structure is multiplied each other, obtain first sparse matrix, comprise: the sparse matrix Ψ that the compressible initial data X that will gather and the random Gaussian of this structure distribute multiplies each other, obtain first sparse matrix, this first sparse matrix is the rarefaction representation of compressible initial data X, if the number K of nonzero element is less than the number of the nonzero element among the compressible initial data X in first sparse matrix, then first sparse matrix is compressible.
In the such scheme, described in the step 2 data after this compression of gathering are reconstructed and obtain reconstruct data, comprising: select an observing matrix, this observing matrix is uncorrelated with first sparse matrix; Data and this observing matrix after this compression of gathering are multiplied each other, obtain observation data; This observation data is carried out inverse transformation obtain reconstruct data.
In the such scheme, the observing matrix of described selection is the observing matrix P of M * N dimension, and this observing matrix P is uncorrelated with first sparse matrix; Data X and observing matrix P after this compression of gathering are multiplied each other, obtain observation data Y, Y is exactly the linear combination of column vector among the corresponding observing matrix P of non-vanishing vector in first sparse matrix.Described this observation data being carried out inverse transformation, is the X among the solving equation P Ψ X=Y, wherein Θ=Ψ X, because the number of this equation unknown number is more than the number of equation group, so the solution of X is not unique, adopt minimum 1 norm to approach at this and find the solution, the result who tries to achieve is exactly the data after the reconstruct.
In the such scheme, utilize this reconstruct data that reverse transmittance nerve network is trained described in the step 3, and preserve the reverse transmittance nerve network train, specifically comprise: this reconstruct data is input to reverse transmittance nerve network as the input sample, this reverse transmittance nerve network carries out iterative processing to this reconstruct data, and each step iteration error and previous step iteration error compare, when iteration error reaches the initial setting threshold value, training finishes, and preserves this reverse transmittance nerve network.
In the such scheme, the reverse transmittance nerve network that this trains described in the step 4 is identified testing data, be this train the reverse transmittance nerve network comparative training time each connects weights, with maximum close weights output binary quantization numbers, the binary quantization number of this output represents different types of gas, and then realizes the qualitative identification to gas.
(3) beneficial effect
From technique scheme as can be seen, the present invention has following beneficial effect:
1, this gas recognition methods based on the compressed sensing theory provided by the invention, the compressed sensing technology is applied to the transmission course of array signal in the sensor network, and in receiving terminal BP neural metwork training identification, computation complexity is transferred to the network training of off-line from online detection, thereby can significantly improve real-time and the accuracy of the online detection of detection system, solved present gas detect in transmission storage data volume big, identify coarse problem, reached the purpose that reaches accurate qualitative identification with less data volume.
2, this gas recognition methods based on the compressed sensing theory provided by the invention, compared with prior art, theoretical and the neural net combination with compressed sensing, simultaneously data are compressed and transmitted, utilize the signal of BP neural metwork training identification compression reconfiguration then in the rear end, can effectively improve data storage capacities and bandwidth availability ratio, and can accomplish accurate identification, help to improve the precision that gas detects.
Description of drawings
Fig. 1 is the gas recognition methods flow chart based on the compressed sensing theory provided by the invention;
Fig. 2 to Fig. 4 shows the gas recognition methods based on the compressed sensing theory according to the embodiment of the invention, wherein:
Fig. 2 is primary signal and reconstruction signal comparison diagram;
Fig. 3 is the physical training condition figure of BP neural net;
Fig. 4 is the training error figure of qualitative identification.
Embodiment
For making the purpose, technical solutions and advantages of the present invention clearer, below in conjunction with specific embodiment, and with reference to accompanying drawing, the present invention is described in more detail.
Gas recognition methods based on the compressed sensing theory provided by the invention, the compressed sensing technology is applied to the transmission course of array signal in the sensor network, and identify with the BP neural metwork training, computation complexity is transferred to the network training of off-line from online detection, thereby can significantly improve real-time and the accuracy of the online detection of detection system.
As shown in Figure 1, Fig. 1 is the gas recognition methods flow chart based on the compressed sensing theory provided by the invention, this method is at first gathered data after the compression in the mode of owing to sample, data after the compression are reconstructed and obtain reconstruct data, utilize this reconstruct data that the BP network is trained then, and preserve the network train, the neural net that testing data input is trained at last, the neural net that trains is identified testing data, output represents the binary quantization number of gas with various, realization is to the qualitative identification of gas, and this method specifically may further comprise the steps:
Step 1: gather data after the compression in the mode of owing to sample;
Step 2: the data after this compression of gathering are reconstructed obtain reconstruct data;
Step 3: utilize this reconstruct data that reverse transmittance nerve network is trained, and preserve the reverse transmittance nerve network that trains;
Step 4: with the reverse transmittance nerve network that the testing data input trains, this reverse transmittance nerve network that trains is identified testing data, realizes the qualitative identification to gas.
Wherein, gather data after the compression in the mode of owing to sample described in the step 1, specifically comprise: the array node of sensor network is gathered compressible initial data; This initial data is carried out sparse decomposition, obtain first sparse matrix, this first sparse matrix is a sparse matrix relevant with initial data; This first sparse matrix is carried out non-linear projection handle, obtain second sparse matrix, the element in this second sparse matrix is the combination at random of element in this first sparse matrix; Owe sampling to the data that coefficient is bigger in this second sparse matrix to carry out low speed less than the frequency of Nyquist sampling frequency.
Described this initial data is carried out sparse decomposition, obtain first sparse matrix, comprising: construct the sparse matrix that a random Gaussian distributes, the initial data of the gathering sparse matrix with the random Gaussian distribution of this structure is multiplied each other, obtain first sparse matrix.
The sparse matrix that random Gaussian of described structure distributes, comprise: select Gauss's matrix that dimension is M * N, each element of this Gauss's matrix is Gaussian distributed all, then each row of this Gauss's matrix is carried out normalized and obtains sparse matrix Ψ.
Described the initial data of the gathering sparse matrix with the random Gaussian distribution of this structure is multiplied each other, obtain first sparse matrix, comprise: the sparse matrix Ψ that the compressible initial data X that will gather and the random Gaussian of this structure distribute multiplies each other, obtain first sparse matrix, this first sparse matrix is the rarefaction representation of compressible initial data X, if the number K of nonzero element is less than the number of the nonzero element among the compressible initial data X in first sparse matrix, then first sparse matrix is compressible.
Described in the step 2 data after this compression of gathering are reconstructed and obtain reconstruct data, comprising: select an observing matrix, this observing matrix is uncorrelated with first sparse matrix; Data and this observing matrix after this compression of gathering are multiplied each other, obtain observation data; This observation data is carried out inverse transformation obtain reconstruct data.The observing matrix of described selection is the observing matrix P of M * N dimension, and this observing matrix P is uncorrelated with first sparse matrix; Data X and observing matrix P after this compression of gathering are multiplied each other, obtain observation data Y, Y is exactly the linear combination of column vector among the corresponding observing matrix P of non-vanishing vector in first sparse matrix.Described this observation data being carried out inverse transformation, is the X among the solving equation P Ψ X=Y, wherein Θ=Ψ X, because the number of this equation unknown number is more than the number of equation group, so the solution of X is not unique, adopt minimum 1 norm to approach at this and find the solution, the result who tries to achieve is exactly the data after the reconstruct.
Utilize this reconstruct data that reverse transmittance nerve network is trained described in the step 3, and preserve the reverse transmittance nerve network train, specifically comprise: this reconstruct data is input to reverse transmittance nerve network as the input sample, this reverse transmittance nerve network carries out iterative processing to this reconstruct data, each step iteration error and previous step iteration error compare, when iteration error reached the initial setting threshold value, training finished, and preserves this reverse transmittance nerve network.
The reverse transmittance nerve network that this trains described in the step 4 is identified testing data, be this train the reverse transmittance nerve network comparative training time each connects weights, with maximum close weights output binary quantization numbers, the binary quantization number of this output represents different types of gas, and then realizes the qualitative identification to gas.
Based on the gas recognition methods flow chart based on the compressed sensing theory provided by the invention shown in Figure 1, Fig. 2 to Fig. 4 shows the gas recognition methods based on the compressed sensing theory according to the embodiment of the invention, specifically may further comprise the steps:
Step 1, gather data after the compression in the mode of owing to sample;
With the node of sensor array as sensor network, the output voltage that node records is X[n], n=1,2,3 ... 180, namely obtain our required compressible initial data X.This compressible initial data X is carried out rarefaction representation that discrete Fourier transform obtains initial data Θ as a result, carry out the non-linear projection operation with pseudorandom Gauss matrix as the rarefaction representation result of observing matrix P then, namely two matrix multiples obtain the observation data Y[m after sparse compression], m=1,2,3 ... 30, can see that after treatment data volume is less than 20% of original data volume.
Step 2: the data after this compression of gathering are reconstructed obtain reconstruct data;
Utilize 30 data after compressing to rebuild primary signals, be the X among the solving equation P Ψ X=Y (wherein Θ=Ψ X), because the number of unknown number is less than the number of equation group, so the solution of X is not unique, need to adopt non-linear approaching, adopt minimum 1 norm to approach in this present invention and find the solution, specific algorithm is to adopt orthogonal matching pursuit algorithm (OMP) reconstruct primary signal.Fig. 2 is the contrast that data that sensor array collects are utilized data and initial data after the compressed sensing reconstruct, and approximate error is 0.87031.
Step 3: utilize this reconstruct data that reverse transmittance nerve network is trained, and preserve the reverse transmittance nerve network that trains;
Data after reconstruct reduction become one 6 * 30 vector matrix as the input of BP neural net, object vector also is defined by a three-dimensional matrice simultaneously.Each object vector contains 3 elements, and vector is representing certain gas, be 1 with the element value of its correspondence position, and the element value of other positions is 0.For example, the vector of CO correspondence, the element value of its first position are 1, and the element value of two other position is 0, i.e. [1,0,0].
Based on above-mentioned consideration, the structure that the present invention designs the BP neural net of employing is 6: 7: 3, and namely input layer has 6 inputs (number of sensors of node), and middle single hidden node number is 7, and output layer then needs 3 neurons (dimension of object vector).Adopt the neural net of the neural net of single hidden layer and many hidden layers can finish the task of qualitative detection, difference only is: the training time of many hidden layers neural net is apparently higher than the training time of single hidden layer neural net.According to the practical problem of this paper, adopt single hidden layer BP network, the reconstruction signal of different degree of rarefications is trained.In the present embodiment, adopt the BP network of single hidden layer, i.e. input layer, a hidden layer, an output layer.
When training, earlier sample data is divided into two groups, wherein the odd number group data are training sample, and the even number set data are used for checking the actual performance of the network that trains.Use rapid bp algorithm to come training network, elect the hidden neuron number of network as 7.In order to make network to new input good generalization ability be arranged, elect the training function of network as trainbr, this function has used the Bayesian frame structure, supposes the weights of network and the stochastic variable that threshold value is special distribution, obtains estimated value with statistical method then.At the training part of network, training sample is divided into training, checking, test three parts, use mean square error to assess the training result of network, physical training condition as shown in Figure 3:
Fig. 4 is the error curve of training process, and iterations is 24 times as can be seen from Figure, and training goal reaches, and error extension converges on expectation index, and training stops, and preserves the network that trains, and waits until the back prediction and uses.
Step 4, with the reverse transmittance nerve network that testing data input trains, this reverse transmittance nerve network that trains is identified testing data, realizes the qualitative identification to gas.
The testing data sample is input in the network that trains, guarantee that pattern of the input follows when training form identical, during network comparative training that this trains each connects weights, with maximum close weights output binary quantization numbers, the binary quantization number of output then represents different types of gas, the accurate rate of the identification of network prediction output reaches 100%, and its output mean error only is 6.0206e
-6
In above-described embodiment, the compressible initial data that the X representative collects, Ψ represents the sparse matrix that rarefaction representation is used, M, N represent the dimension of sparse matrix, and Θ is the rarefaction representation result of initial data, and K represents the number of nonzero element among the sparse result, the degree of rarefication that also is called X, what represent is the degree of compressibility of initial data, and P represents the observing matrix that non-linear projection is used, and Y represents the observed result of initial data.
Because being different from first sampling of conventional method, compressed sensing afterwards reduces the method for redundant information wherein, but the information after directly " collection " compressed, make the data volume of sampling reduce, saved the step of compression simultaneously, simultaneously in conjunction with the simple advantage of single hidden layer neural network structure, take all factors into consideration transfer of data memory space and accuracy of identification, the gas recognition methods best performance based on the compressed sensing theory that the present invention proposes.
Above-described specific embodiment; purpose of the present invention, technical scheme and beneficial effect are further described; institute is understood that; the above only is specific embodiments of the invention; be not limited to the present invention; within the spirit and principles in the present invention all, any modification of making, be equal to replacement, improvement etc., all should be included within protection scope of the present invention.
Claims (10)
1. gas recognition methods based on the compressed sensing theory is characterized in that this method comprises:
Step 1: gather data after the compression in the mode of owing to sample;
Step 2: the data after this compression of gathering are reconstructed obtain reconstruct data;
Step 3: utilize this reconstruct data that reverse transmittance nerve network is trained, and preserve the reverse transmittance nerve network that trains;
Step 4: with the reverse transmittance nerve network that the testing data input trains, this reverse transmittance nerve network that trains is identified testing data, realizes the qualitative identification to gas.
2. the gas recognition methods based on the compressed sensing theory according to claim 1 is characterized in that, gathers data after the compression in the mode of owing to sample described in the step 1, specifically comprises:
The array node of sensor network is gathered compressible initial data;
This initial data is carried out sparse decomposition, obtain first sparse matrix, this first sparse matrix is a sparse matrix relevant with initial data;
This first sparse matrix is carried out non-linear projection handle, obtain second sparse matrix, the element in this second sparse matrix is the combination at random of element in this first sparse matrix;
Owe sampling to the data that coefficient is bigger in this second sparse matrix to carry out low speed less than the frequency of Nyquist sampling frequency.
3. the gas recognition methods based on the compressed sensing theory according to claim 2 is characterized in that, described this initial data is carried out sparse decomposition, obtains first sparse matrix, comprising:
Construct the sparse matrix that random Gaussian distributes, the initial data of the gathering sparse matrix with the random Gaussian distribution of this structure is multiplied each other, obtain first sparse matrix.
4. the gas recognition methods based on the compressed sensing theory according to claim 3 is characterized in that, the sparse matrix that random Gaussian of described structure distributes comprises:
Select Gauss's matrix that dimension is M * N, each element of this Gauss's matrix is Gaussian distributed all, then each row of this Gauss's matrix is carried out normalized and obtains sparse matrix Ψ.
5. the gas recognition methods based on the compressed sensing theory according to claim 4 is characterized in that, described the initial data of the gathering sparse matrix with the random Gaussian distribution of this structure is multiplied each other, and obtains first sparse matrix, comprising:
The compressible initial data X that gathers and the sparse matrix Ψ of the random Gaussian distribution of this structure are multiplied each other, obtain first sparse matrix, this first sparse matrix is the rarefaction representation of compressible initial data X, if the number K of nonzero element is less than the number of the nonzero element among the compressible initial data X in first sparse matrix, then first sparse matrix is compressible.
6. the gas recognition methods based on the compressed sensing theory according to claim 1 is characterized in that, described in the step 2 data after this compression of gathering is reconstructed and obtains reconstruct data, comprising:
Select an observing matrix, this observing matrix is uncorrelated with first sparse matrix;
Data and this observing matrix after this compression of gathering are multiplied each other, obtain observation data;
This observation data is carried out inverse transformation obtain reconstruct data.
7. the gas recognition methods based on the compressed sensing theory according to claim 6 is characterized in that, the observing matrix of described selection is the observing matrix P of M * N dimension, and this observing matrix P is uncorrelated with first sparse matrix; Data X and observing matrix P after this compression of gathering are multiplied each other, obtain observation data Y, Y is exactly the linear combination of column vector among the corresponding observing matrix P of non-vanishing vector in first sparse matrix.
8. the gas recognition methods based on the compressed sensing theory according to claim 7, it is characterized in that, described this observation data is carried out inverse transformation, be the X among the solving equation P Ψ X=Y, Θ=Ψ X wherein is because the number of this equation unknown number is more than the number of equation group, so the solution of X is not unique, adopt minimum 1 norm to approach at this and find the solution, the result who tries to achieve is exactly the data after the reconstruct.
9. the gas recognition methods based on the compressed sensing theory according to claim 1 is characterized in that, utilizes this reconstruct data that reverse transmittance nerve network is trained described in the step 3, and preserves the reverse transmittance nerve network that trains, and specifically comprises:
This reconstruct data is input to reverse transmittance nerve network as the input sample, this reverse transmittance nerve network carries out iterative processing to this reconstruct data, each step iteration error and previous step iteration error compare, when iteration error reaches the initial setting threshold value, training finishes, and preserves this reverse transmittance nerve network.
10. the gas recognition methods based on the compressed sensing theory according to claim 1, it is characterized in that, the reverse transmittance nerve network that this trains described in the step 4 is identified testing data, be this train the reverse transmittance nerve network comparative training time each connects weights, with maximum close weights output binary quantization numbers, the binary quantization number of this output represents different types of gas, and then realizes the qualitative identification to gas.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2013102205872A CN103346799A (en) | 2013-06-05 | 2013-06-05 | Method for identifying gas based on compressed sensing theory |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2013102205872A CN103346799A (en) | 2013-06-05 | 2013-06-05 | Method for identifying gas based on compressed sensing theory |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103346799A true CN103346799A (en) | 2013-10-09 |
Family
ID=49281581
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2013102205872A Pending CN103346799A (en) | 2013-06-05 | 2013-06-05 | Method for identifying gas based on compressed sensing theory |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103346799A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014194482A1 (en) * | 2013-06-05 | 2014-12-11 | 中国科学院微电子研究所 | Gas recognition method based on compressive perception theory |
| CN105628868A (en) * | 2015-12-18 | 2016-06-01 | 北京航空航天大学 | Undersampled signal impact positioning processing method and system for compound material structure |
| CN107092961A (en) * | 2017-03-23 | 2017-08-25 | 中国科学院计算技术研究所 | A kind of neural network processor and design method based on mode frequency statistical coding |
| CN109060892A (en) * | 2018-06-26 | 2018-12-21 | 西安交通大学 | SF based on graphene composite material sensor array6Decompose object detecting method |
| CN109239649A (en) * | 2018-04-04 | 2019-01-18 | 唐晓杰 | A kind of relatively prime array DOA under the conditions of array error estimates new method |
| CN109784390A (en) * | 2019-01-03 | 2019-05-21 | 西安交通大学 | A kind of artificial intelligence smell dynamic response map gas detection recognition methods |
| CN111398188A (en) * | 2020-02-27 | 2020-07-10 | 深圳市安车检测股份有限公司 | Smoke blackness and spatial distribution detection method based on compressed sensing algorithm |
| CN111542819A (en) * | 2017-09-26 | 2020-08-14 | 地质探索***公司 | Apparatus and method for improved subsurface data processing system |
| CN112187282A (en) * | 2020-09-02 | 2021-01-05 | 北京电子工程总体研究所 | Compressed sensing signal reconstruction method and system based on dictionary double learning |
| CN113261932A (en) * | 2021-06-28 | 2021-08-17 | 山东大学 | Heart rate measurement method and device based on PPG signal and one-dimensional convolutional neural network |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100246651A1 (en) * | 2009-03-31 | 2010-09-30 | Qualcomm Incorporated | Packet loss mitigation in transmission of biomedical signals for healthcare and fitness applications |
| CN102590335A (en) * | 2012-01-06 | 2012-07-18 | 电子科技大学 | SAW (Surface Acoustic Wave) sensor based embedded electronic nose testing system and testing method |
| CN102665221A (en) * | 2012-03-26 | 2012-09-12 | 南京邮电大学 | Cognitive radio frequency spectrum perception method based on compressed sensing and BP (back-propagation) neural network |
-
2013
- 2013-06-05 CN CN2013102205872A patent/CN103346799A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100246651A1 (en) * | 2009-03-31 | 2010-09-30 | Qualcomm Incorporated | Packet loss mitigation in transmission of biomedical signals for healthcare and fitness applications |
| CN102590335A (en) * | 2012-01-06 | 2012-07-18 | 电子科技大学 | SAW (Surface Acoustic Wave) sensor based embedded electronic nose testing system and testing method |
| CN102665221A (en) * | 2012-03-26 | 2012-09-12 | 南京邮电大学 | Cognitive radio frequency spectrum perception method based on compressed sensing and BP (back-propagation) neural network |
Non-Patent Citations (2)
| Title |
|---|
| 张宗念 等: "压缩感知信号盲稀疏度重构算法", 《电子学报》, vol. 39, no. 01, 31 January 2011 (2011-01-31), pages 18 - 22 * |
| 龚静: "无线传感器网络中基于压缩感知技术的数据压缩方法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》, no. 4, 15 April 2012 (2012-04-15), pages 140 - 272 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014194482A1 (en) * | 2013-06-05 | 2014-12-11 | 中国科学院微电子研究所 | Gas recognition method based on compressive perception theory |
| CN105628868A (en) * | 2015-12-18 | 2016-06-01 | 北京航空航天大学 | Undersampled signal impact positioning processing method and system for compound material structure |
| CN105628868B (en) * | 2015-12-18 | 2018-08-21 | 北京航空航天大学 | A kind of undersampled signal im-pact location processing method and system of composite structure |
| CN107092961A (en) * | 2017-03-23 | 2017-08-25 | 中国科学院计算技术研究所 | A kind of neural network processor and design method based on mode frequency statistical coding |
| CN111542819A (en) * | 2017-09-26 | 2020-08-14 | 地质探索***公司 | Apparatus and method for improved subsurface data processing system |
| CN109239649A (en) * | 2018-04-04 | 2019-01-18 | 唐晓杰 | A kind of relatively prime array DOA under the conditions of array error estimates new method |
| CN109060892B (en) * | 2018-06-26 | 2020-12-25 | 西安交通大学 | SF based on graphene composite material sensor array6Method for detecting decomposition product |
| CN109060892A (en) * | 2018-06-26 | 2018-12-21 | 西安交通大学 | SF based on graphene composite material sensor array6Decompose object detecting method |
| CN109784390A (en) * | 2019-01-03 | 2019-05-21 | 西安交通大学 | A kind of artificial intelligence smell dynamic response map gas detection recognition methods |
| CN111398188A (en) * | 2020-02-27 | 2020-07-10 | 深圳市安车检测股份有限公司 | Smoke blackness and spatial distribution detection method based on compressed sensing algorithm |
| CN111398188B (en) * | 2020-02-27 | 2023-01-06 | 深圳市安车检测股份有限公司 | Smoke blackness and spatial distribution detection method based on compressed sensing algorithm |
| CN112187282A (en) * | 2020-09-02 | 2021-01-05 | 北京电子工程总体研究所 | Compressed sensing signal reconstruction method and system based on dictionary double learning |
| CN113261932A (en) * | 2021-06-28 | 2021-08-17 | 山东大学 | Heart rate measurement method and device based on PPG signal and one-dimensional convolutional neural network |
| CN113261932B (en) * | 2021-06-28 | 2022-03-04 | 山东大学 | Heart rate measurement method and device based on PPG signal and one-dimensional convolutional neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103346799A (en) | Method for identifying gas based on compressed sensing theory | |
| WO2014194482A1 (en) | Gas recognition method based on compressive perception theory | |
| Zhao et al. | Multiscale inverted residual convolutional neural network for intelligent diagnosis of bearings under variable load condition | |
| CN107065545B (en) | Distributed event trigger filtering system based on Markov jump and design method | |
| Shoaib et al. | Signal processing with direct computations on compressively sensed data | |
| CN101221213A (en) | Analogue circuit fault diagnosis neural network method based on particle swarm algorithm | |
| CN110443724B (en) | Electric power system rapid state estimation method based on deep learning | |
| CN113113030B (en) | High-dimensional damaged data wireless transmission method based on noise reduction self-encoder | |
| CN103124179A (en) | Electric power system data reconfiguration decompressing method based on orthogonal matching pursuit | |
| CN109756740B (en) | Half tensor image compression method and image recovery method based on optimal measurement matrix | |
| CN111783336B (en) | Uncertain structure frequency response dynamic model correction method based on deep learning theory | |
| Yu et al. | Solar radio spectrum classification with LSTM | |
| Poh et al. | LSTM and HMM comparison for home activity anomaly detection | |
| Song et al. | An improved reconstruction method based on auto-adjustable step size sparsity adaptive matching pursuit and adaptive modular dictionary update for acoustic emission signals of rails | |
| CN117633448A (en) | Cloud-edge collaboration-based equipment cluster health monitoring method | |
| Prono et al. | Deep neural oracle with support identification in the compressed domain | |
| Dejman et al. | Exploring the disorders of brain effective connectivity network in ASD: a case study using EEG, transfer entropy, and graph theory | |
| Jawali et al. | A learning approach for wavelet design | |
| CN116338322A (en) | Power grid line impedance prediction method and system | |
| Han et al. | A new multilayer LSTM method of reconstruction for compressed sensing in acquiring human pressure data | |
| Xu et al. | A Graph Convolutional Shrinkage Network-based Fault Diagnosis Method for Industrial Process | |
| CN112179503A (en) | Deep learning wavefront restoration method based on sparse subaperture shack-Hartmann wavefront sensor | |
| CN111008584A (en) | Electric energy quality measurement deficiency repairing method of fuzzy self-organizing neural network | |
| Yang et al. | A neural network pruning approach based on compressive sampling | |
| CN114118163B (en) | Optical fiber intrusion detection method based on machine learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20131009 |