CN114894980B - Unknown gas response signal characteristic extraction and data mining method based on gas sensor array - Google Patents
Unknown gas response signal characteristic extraction and data mining method based on gas sensor array Download PDFInfo
- Publication number
- CN114894980B CN114894980B CN202210574810.2A CN202210574810A CN114894980B CN 114894980 B CN114894980 B CN 114894980B CN 202210574810 A CN202210574810 A CN 202210574810A CN 114894980 B CN114894980 B CN 114894980B
- Authority
- CN
- China
- Prior art keywords
- sensor
- gas
- response
- data
- baseline
- 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
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
- G01N33/0004—Gaseous mixtures, e.g. polluted air
- G01N33/0009—General constructional details of gas analysers, e.g. portable test equipment
- G01N33/0027—General constructional details of gas analysers, e.g. portable test equipment concerning the detector
- G01N33/0031—General constructional details of gas analysers, e.g. portable test equipment concerning the detector comprising two or more sensors, e.g. a sensor array
- G01N33/0034—General constructional details of gas analysers, e.g. portable test equipment concerning the detector comprising two or more sensors, e.g. a sensor array comprising neural networks or related mathematical techniques
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
- G01N33/0001—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00 by organoleptic means
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
- G01N33/0004—Gaseous mixtures, e.g. polluted air
- G01N33/0009—General constructional details of gas analysers, e.g. portable test equipment
- G01N33/0062—General constructional details of gas analysers, e.g. portable test equipment concerning the measuring method, e.g. intermittent, or the display, e.g. digital
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A50/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE in human health protection, e.g. against extreme weather
- Y02A50/20—Air quality improvement or preservation, e.g. vehicle emission control or emission reduction by using catalytic converters
Abstract
The invention discloses an unknown gas response signal characteristic extraction and data mining method based on a gas sensor array, which comprises the following steps: step 1, acquiring multichannel sensor response data from an electronic nose; step 2, constructing a relation network between the volatile compound and the sensor; step 3, first-level judgment; step 4, second-level judgment; and 5, mining out final effective information.
Description
Technical Field
The invention relates to the field of odor detection, in particular to the field of detection and signal analysis of a gas sensor.
Background
With the development of society, gas detection is also beginning to be applied to various industries, such as the field of food safety, the field of industrial safety production, and the like. The technology has the advantages of real-time, quick, nondestructive detection and the like, however, the detection method based on the gas sensor at present often needs a large amount of data to perform labeled supervised learning, and then the technology can be specifically applied to a certain scene. If the data mining can be carried out from the response signals of unknown gas, the effective information can be directly extracted, and the application boundary of the gas detection technology can be greatly expanded. In addition, in the field of odor detection, typical dynamic sample injection methods can be generally classified into a sample injection stage of a gas to be detected and a clean gas cleaning stage, and the method is simple and easy to implement, but the information contained in the method is not particularly abundant. When the smell is distinguished by biology, the smell can be subjected to a certain rapid short-smell and long-time continuous long-smell behaviors, the short-smell process can obtain instant smell stimulus, the long-smell process can obtain more continuous and stable smell stimulus, and the two can be combined to obtain abundant smell sensory information from the two angles of transient state and steady state. The invention is inspired by biological respiration, develops a sample injection mode comprising seven stages of short smell and long smell, and can obtain more abundant response signals and more potential valuable information contained in the response signals.
Disclosure of Invention
The invention is realized by adopting the following technical scheme:
an unknown gas response signal characteristic extraction and data mining method based on a gas sensor array comprises the following steps: step 1, acquiring multichannel sensor response data from an electronic nose; step 2, constructing a relation network between the volatile compound and the sensor; step 3, first-level judgment; step 4, second-level judgment; and 5, mining out final effective information.
The unknown gas response signal characteristic extraction and data mining method based on the gas sensor array comprises the following steps: constructing a relational network W between volatile compounds and sensors by using sensor cross-sensitivity characteristics Vocs-Snrs It includes a set of sensor-sensitive volatile compounds S-V and a set of cross-sensitive volatile compounds-sensors V-S.
The method for extracting unknown gas response signal characteristics and mining data based on the gas sensor array comprises the steps that step 1 comprises the steps that an electronic nose simulates a short-breath process of human being on the rapid smell and a long-breath process which is continuous, and multichannel sensor response data are acquired and obtained, wherein the method specifically comprises seven stages: collecting unknown gas to be tested in a gas collecting bottle, sealing and placing on a constant temperature table, keeping constant temperature at 50 ℃ for 10 minutes, starting an electronic nose device, and performing test operation5 minutes to stabilize the instrument, and simultaneously collecting data c seconds at a frequency of 1 Hz; wherein, collecting clean carrier gasSecond, as data of the cleaning preparation stage; subsequently, connecting a gas collecting bottle, and continuing to collect sample gas +.>Second, as data of the short sniff phase; then, the gas collection bottle is stopped to be charged, and residual gas in the gas collection bottle is collected>Second, as data of the pause stage; subsequently, connecting a gas collecting bottle, and continuing to collect sample gas +.>Second, as data of the secondary short sniffing stage; then, the gas collecting cylinder is stopped to be charged, and residual gas in the gas collecting cylinder is collected>Second, as data of the second pause stage; then, connecting a gas collecting bottle, and continuously collecting sample gas +.>Second, as data of the long sniff phase; finally, clean carrier gas is introduced to collect +.>Second, as data of the purge recovery phase; thereby obtaining D n =[d 1 ,d 2 ,...d c ]Where n represents the number of sensors (i.e., the number of channels) and c represents the acquisition duration of each channel.
The unknown gas response signal characteristic extraction and data mining method based on the gas sensor array comprises the following steps:
(1) The sensor set is established as follows:
Snrs=[S 1 ,S 2 ...S n ]where n is the number of sensors (i.e., the number of channels)
(2) The union set is obtained for the volatile matter sets which are sensitive to all the sensors, and the formula is as follows:
Vocs=(V 11 ,V 12 ...)∪(V 21 ,V 22 ...)∪...(V n1 ,V n2 ...), wherein V 11 Representing a first compound to which the first sensor is sensitive, V n1 Representing the first compound to which the nth sensor corresponds to sensitivity.
(3) Establishing a volatile compound collection:
Vocs=[V 1 ,V 2 ...V m ]wherein m is the number of volatile species that do not repeat.
(4) Establishing a two-way relational mapping network W between volatile compounds and sensors Vocs-Snrs In particular comprising a set of sensor-sensitive volatile compounds
S-V={S 1 :{V 1 ,V 2 ,...},S 2 :{V 1 ,V 2 ,...},...S n :{V 1 ,V 2 ,. } meaning the total set of sensitive volatile compounds for each sensor; cross-sensitive volatile compound-sensor set V-s= { V 1 :{S 1 ,S 2 ,...},V 2 :{S 1 ,S 2 ,...},...V m :{S 1 ,S 2 ,., meaning the total set of sensor sets that are simultaneously sensitive to a certain volatile compound.
The unknown gas response signal characteristic extraction and data mining method based on the gas sensor array comprises the following steps: the data of the five stages of short smell, secondary short smell, long smell, cleaning preparation and cleaning recovery of each sensor channel are averaged, and a first-level judging index P is calculated according to the following formula:
wherein p is i Is the first-level judgment index of the ith channel, namely the sensor, d avg1 Is the average value of the cleaning preparation stage of the channel, d avg2 Is the average value of the cleaning recovery phase of the channel, d avg3 Is the average value of the short sniff phase of the channel, d avg4 Is the average value of the secondary short sniffing phase of the channel, d avg5 Is the average value of the long sniff phase of the channel;
further, the first level judgment rule is as follows:
wherein S is i For the ith sensor in the original sensor set Snrs set, S one For one-stage judgment of completion set S two For the secondary judgment alternative set, if the primary judgment index is smaller than or equal to 0.1, the sensor of the channel is put into the primary judgment finished set, otherwise, the sensor is put into the secondary judgment alternative set S two At this time, according to S one And combining the s-v set, a first-order conclusion can be obtained:
Inform 1 in the = { unknown gas, no S is contained one Volatile compounds sensitive to the sensors in the collection }
The unknown gas response signal characteristic extraction and data mining method based on the gas sensor array comprises the following steps:
judging the alternative set S for the second level two The data in (2) is taken to baseline according to the following formula:
R i =d i -d baseline
wherein R is i Is the response value of the ith second after the removal of the baseline, d i Is the original response value of the sensor in the ith second, d baseline Is a baseline value (its size is equal to d avg1 ) Each channel collects c seconds of data altogether, so i E [1, c]。
The unknown gas response signal characteristic extraction and data mining method based on the gas sensor array, wherein the step 4 further comprises the following steps:
extracting a secondary judgment candidate set S from two angles of a time transformation domain and a frequency transformation domain two Time domain features and frequency domain features in (a) including maximum stable difference value F max Steady state mean F stbavg Response energy value F integ Average differential value F ndf Sum of variances F varnc The method comprises the steps of carrying out a first treatment on the surface of the DC component amplitude F after Fourier transform fft0 Amplitude of first order harmonic component F fft1 And a wavelet transformed low frequency component energy value F wdh High frequency component energy value F wdl The specific formulas of the relevant characteristics are as follows:
(1) Maximum stable difference value F max Defined by the following formula:
wherein R is avg1 Is the average value after the cleaning preparation stage of the channel is subjected to baseline removal, R avg2 Is the average value after the cleaning recovery stage of the channel is subjected to baseline removal, R avg2 Is the average value after the short sniffing stage of the channel is de-baselined, R avg4 Is the average value after the secondary short sniffing stage of the channel is subjected to baseline removal, R avg5 Is the average value of the long-sniffing stage of the channel after baseline removal; max []The characteristic is that the maximum function is obtained, and compared with the maximum characteristic or the extreme difference characteristic which is commonly seen, the characteristic can more stably represent the severe condition of the response of the gas sensor;
(2) Steady state mean F stbavg Defined by the following formula:
wherein R is j 、R k And R is l Is the response of the gas sensor when the j, k and l seconds begin to reach steady state, c is the total sampling time, j e (0.1 c,0.2c),k∈(0.3c,0.4c),l∈(0.5c,0.8c);
the parameters j, k and l are determined by two methods, namely, the derivative value of a test stage is obtained, and in the corresponding stage, if the derivative value at a certain moment is smaller than 0.1 (the corresponding tangent angle is smaller than 5 degrees), the response fluctuation is smaller at the moment, and the state is stable, the moment is set as j, k or l; if the conditions of the first method are not satisfied, j=0.175 c, k=0.375 c, l=0.725 c (meaning that the values are taken starting from the last 1/4 part of the three phases of short sniff, secondary short sniff and long sniff). The characteristic can better express the stable response condition of the gas sensor during testing;
(3) Response energy value F integ Defined by the following formula:
wherein R is i Is the response of the gas sensor after the ith second removes the baseline, F integ For integral value characteristics, i takes a value from 0.1c to 0.8c-1. This feature represents the overall degree of response of the gas sensor at the time of testing;
(4) The average differential value, the physical meaning of which reflects the overall dynamic process of the response signal, has the following formula:
wherein R is i Is the response of the gas sensor after the ith second is removed from the baseline, c is the total sampling time, F ndf Is characteristic of average differential value;
(5) The variance value reflects the discrete condition of the response signal in the physical meaning as follows:
F varnc =(R i -Avg(R 1 、R 2 ...R c )) 2 /c
wherein R is i Is the response of the gas sensor after the ith second removes the baseline, c is the total sampling time, avg is a function of the average of the elements in the collection, F varnc Is a variance value feature;
(6) The dc component, the formula is as follows:
F fft0 =FFTO(R 1 、R 2 ...R c )
wherein R is i Is the response of the gas sensor after the i second removes the baseline, c is the total sampling time, and FFT0 (·) is a function of extracting the DC component using the fast Fourier transform;
(7) The first order harmonic component is given by:
F fft1 =FFT1(R 1 、R 2 ...R c )
wherein R is i Is the response of the gas sensor after the i second removes the baseline, c is the total sampling time, and FFT1 (-) is a function of extracting the first order harmonic component by using the fast Fourier transform;
(8) The high frequency component, the formula is as follows:
F wdh =WDH(R 1 、R 2 ...R c )
wherein R is i The response of the gas sensor after the baseline is removed in the ith second, c is the total sampling time, WDH (·) is a function of using orthogonal wavelet Haar as a layer of decomposition to extract high-frequency components;
(9) The low frequency component, the formula is as follows:
F wdl =WDL(R 1 、R 2 ...R c )
wherein R is i Is the response of the gas sensor after the baseline is removed in the ith second, c is the total sampling time, WDL (·) is a function of extracting low-frequency components by using orthogonal wavelet Haar as a layer of decomposition;
characteristic extraction is carried out on response data of each sensor after baseline removal, and a time-frequency characteristic set F is finally obtained i =[F i_max ,F i_stbavg ,F i_integ ,F i_ndf ,F i_varnc ,F i_fft0 ,F i_fft1 ,F i_wdh ,F i_wdl ]Wherein i is the number of the sensor;
and carrying out proportion normalization processing on the time-frequency characteristic set, wherein the formula is as follows:
wherein F is i Is a characteristic value of the ith sensor, z is a secondary judgment candidate set D two I epsilon (1, z);
obtaining normalized time-frequency characteristic matrixWherein i is the sensor number, i ε (1, z); j is the feature class number, j e (1, 9);
calculating a secondary judgment index Q i The formula is as follows:
the second level judgment rule is as follows:
wherein S is two_i Is the ith sensor in the secondary judgment alternative set, z is the secondary judgment alternative set S two The number of the sensors in (S) unsure Is an uncertainty set. The rule means that the second level judgment index is larger thanSensor placement S two Otherwise put into S unsure 。
The unknown gas response signal characteristic extraction and data mining method based on the gas sensor array, wherein the step 4 further comprises the following steps: the heuristic data mining method comprises the following steps: (1) computing into a cross-sensitive set S cross The number of the sensors is recorded as N; (2) if N is equal to 0, directly entering a fifth step, and if N is greater than 0, entering a third step; (3) will S cross And V-S= { V 1 :{S 1 ,S 2 ,...},V 2 :{S 1 ,S 2 ,...},...V m :{S 1 ,S 2 Each subset in the,. set is intersected, and the number of sensors in the intersection is set as M; (4) calculating confidence coefficient C=M/N, and finding out corresponding volatile compounds according to the V-S set; (5) ending the flow; summarizing the analysis results to obtain a second-level conclusion:
Inform 2 in the = { unknown gas, S is contained cross Volatile compounds (confidence M/N) sensitive to sensors shared by the set and the V-S set
The unknown gas response signal characteristic extraction and data mining method based on the gas sensor array comprises the following steps:
valuable information is obtained from unknown gas to be measured through data acquisition, preprocessing, feature extraction, two-stage cascade judgment and heuristic data mining, and the total conclusion is that:
Inform all =Inform 1 +Inform 2
in the = { unknown gas, no S is contained one Volatile compounds sensitive to the sensors in the collection; containing S cross Volatile compounds sensitive to the sensor common to the set and the set V-S (confidence M/N).
Drawings
FIG. 1 is a schematic diagram of a data mining method;
FIG. 2 is an overall schematic diagram of an electronic nose detection hardware system;
FIG. 3 is a hardware wiring diagram;
FIG. 4 is a graph reflecting sensor cross-sensitivity characteristics;
FIG. 5 is a network diagram of the relationship between volatile materials and sensors;
FIG. 6 is a typical electronic nose response data plot;
FIG. 7 is an extracted time domain feature;
FIG. 8 is an extracted frequency domain feature;
fig. 9 is a heuristic data mining method.
Detailed Description
The following describes embodiments of the present invention in detail with reference to fig. 1-9. It should be understood that the examples described herein are for the purpose of illustration and explanation only and are not intended to limit the present invention.
As shown in fig. 1, the unknown gas response signal feature extraction and data mining method of the present invention comprises: firstly, acquiring multichannel sensor response data in a sampling mode inspired by biological respiration and comprising seven stages of short smell and long smell; constructing a relational network W between volatile compounds and sensors by using sensor cross-sensitivity characteristics Vocs-Snrs Comprising a set of sensor-sensitive volatile compounds S-V and a set of cross-sensitive volatile compounds-sensors V-S; defining a first-level judgment index P by using an average value of 5 stages, and extracting volatile compound type information Inform not contained in unknown gas by combining an S-V set if the first-level judgment index P is smaller than or equal to 0.1 1 Otherwise, entering a second stage of judgment; defining a secondary judgment index Q by utilizing the characteristic value after proportion normalization, if the secondary judgment index Q is larger than or equal to the secondary judgment index Q(z is the number of sensors entering the second-level judgment), combining the V-S set, and extracting the type of the volatile compounds possibly contained in the unknown gas and relevant confidence information Inform by using a heuristic data mining and confidence estimation method 2 Otherwise, ending; finally summarize Inform 1 And Inform 2 Dig out the final effective information Inform all 。
The whole electronic nose detection hardware system is schematically shown in fig. 2, and comprises a constant temperature device, an electronic nose device and a computer device, wherein the three devices are sequentially arranged from left to right. The constant temperature device comprises a sample gas collection bottle and a heating constant temperature table; the electronic nose device comprises a sensor array air chamber, an air pump, a data acquisition card, a power supply and various related components; the computer module is connected with the electronic nose module through a data line to realize signal acquisition and analysis. A specific hardware connection is shown in fig. 3. In this embodiment, the sensor types and corresponding sensitive volatile compounds used are shown in the following table:
according to the actual sensor conditions, a sensor set is established, and the formula is as follows:
Snrs=[S 1 (TGS2602),S 2 (TGS2603),S 3 (TGS2612-D00),S 4 (TGS2615-E00),S 5 (NE2-CH2O)]。
a set of volatile compounds is established as follows:
Vocs=(V 11 ,V 12 ...)∪(V 21 ,V 22 ...)∪...(V n1 ,V n2 ...)
wherein V is 11 Representing a first compound to which the first sensor is sensitive, V n1 Represents the first compound to which the nth sensor corresponds to sensitivity, in this embodiment n=5. The formula means that all volatile substances which are sensitive to each sensor are firstly found out to form a subset; and summing all the subsets to obtain the volatile substance set Vocs.
According to the actual situation of the present embodiment, the first sensor corresponds to 5 compounds, the second sensor corresponds to 5 compounds, the third sensor corresponds to 5 compounds, the fourth sensor corresponds to 1 compound, the fifth sensor corresponds to 1 compound, for the five compound sets, the repetition is removed, the maximum union is obtained, 11 compounds are obtained in total, and finally the volatile compound set is established:
Vocs
=[V 1 (toluene), V 2 (Hydrogen sulfide), V 3 (ethanol), V 4 (Ammonia gas), V 5 (Hydrogen), V 6 (trimethylamine), V 7 (methyl mercaptan), V 8 (propane)),V 9 (isobutane), V 10 (chlorine) V 11 (Formaldehyde)]。
As shown in fig. 4, the sensors have a cross-sensitivity characteristic, i.e., one sensor is sensitive to multiple volatile compounds, and one volatile compound may also be detected by multiple sensors. Constructing a relational network between volatile materials and sensors as shown in FIG. 5The network contains a cross-sensitive volatile compound-sensor set V-s= { V 1 :{S 1 },V 2 :{S 1 ,S 2 },V 3 :{S 1 ,S 2 ,S 3 },V 4 :{S 1 },V 5 :{S 1 ,S 2 ,S 3 },V 6 :{S 2 },V 7 :{S 1 },V 8 :{S 3 },V 9 :{S 3 },V 10 :{S 4 },V 11 :{S 5 A set of sensor-sensitive volatile compounds S-v= { S } }, and 1 :{V 1 ,V 2 ,V 3 ,V 4 ,V 5 },S 2 :{V 2 ,V 3 ,V 5 ,V 6 ,V 7 },S 3 :{V 3 ,V 5 ,V 8 ,V 9 ,V 10 },S 4 :{V 10 },S 5 :{V 11 }}
collecting sensor data: collecting unknown gas to be measured in the gas collecting bottle, sealing and placing the gas on a constant temperature table, and keeping the temperature at 50 ℃ for 10 minutes. Starting an electronic nose device, and performing test operation for 5 minutes to ensure that the instrument is stable, and collecting clean carrier gas at the frequency of 1Hz for 10 seconds as data in a cleaning preparation stage; then, connecting a gas collecting bottle, and continuously collecting sample gas for 10 seconds to serve as data of a short sniffing stage; then, suspending air intake of the gas collecting bottle, and collecting residual gas in the gas chamber for 10 seconds as data of a suspension stage; then, connecting a gas collecting bottle, and continuously collecting sample gas for 10 seconds to serve as data of a secondary short sniffing stage; then, the air inlet of the air collecting bottle is stopped, and the air is collectedResidual gas in the gas collection chamber for 10 seconds is used as data of a secondary suspension stage; then, connecting a gas collecting bottle, and continuously collecting sample gas for 30 seconds to serve as data of a long-sniffing stage; finally, introducing clean carrier gas, and collecting for 20 seconds as data in a cleaning recovery stage; the resulting response curve is shown in fig. 6. Thereby obtaining D 1 =[d 1 ,d 2 ,...d 100 ]、D 2 =[d 1 ,d 2 ,...d 100 ]、D 3 =[d 1 ,d 2 ...d 100 ]、D 4 =[d 1 ,d 2 ,...d 100 ]And D 5 =[d 1 ,d 2 ,...,d 100 ]Raw data of 5 x 100 dimensions total, where d represents data acquired per second, and the 7 phases total 100 data.
And (3) primary judgment: the average value of the 5-stage data of each sensor channel is calculated, and a first-stage judgment index P is calculated according to the following formula:
wherein p is i Is the first-level judgment index of the ith channel (sensor), d avg1 Is the average value of the cleaning preparation stage of the channel, d avg2 Is the average value of the cleaning recovery phase of the channel, d avg3 Is the average value of the short sniff phase of the channel, d avg4 Is the average value of the secondary short sniffing phase of the channel, d avg5 Is the average of the long sniff phase of the channel.
The obtained primary judgment index of each sensor is shown in the following table:
index/channel | S1 | S2 | S3 | S4 | S5 |
First-level judgment index P | 1.538 | 1.186 | 1.04 | 0.418 | 0.05 |
Further, the first level judgment rule is as follows:
wherein S is i For the ith sensor in the original sensor set Snrs set, S one For one-stage judgment of completion set S two For the secondary judgment alternative set, if the primary judgment index P is smaller than or equal to 0.1, the sensor of the channel is put into the primary judgment finished set, otherwise, the sensor is put into the secondary judgment alternative set S two 。
For example, due to p 5 Less than 0.1, putting the data of S5 into a first-level judgment completion set S one And based on the sensor-sensitive volatile compound set S-V, the sensor S5 corresponds to the sensitive compound being V11 (formaldehyde), from which a first order conclusion of Inform can be drawn 1 = { no formaldehyde in the gas to be measured }. The rest p 1 、p 2 、p 3 And p 4 All are larger than 0.1, and the data of S1, S2, S3 and S4 are put into a secondary judgment alternative set S two 。
For secondary judgmentAlternative set S two The baseline was removed according to the following formula:
R i =d i -d baseline
wherein R is i Is the response value of the ith second after the removal of the baseline, d i Is the original response value of the sensor in the ith second, d baseline Is a baseline value (its size is equal to d avg1 ),i∈[1,100]。
In order to avoid the dimension disaster caused by the excessive dimension of the data, the dimension reduction processing is required to be carried out on the original data. Meanwhile, in order to keep more effective information of the original data, the invention extracts the time domain features and the frequency domain features of the data from two angles of a time transformation domain and a frequency transformation domain. The time domain features are shown in FIG. 7 and include a maximum stable differential value F max Steady state mean F stbavg Response energy value F integ Average differential value F ndf Sum of variances F varnc . The frequency domain characteristics are shown in FIG. 8 and include DC component amplitude F after Fourier transform fft0 And first order harmonic component amplitude F fft1 Wavelet transformed low frequency component energy value F wdl High-frequency component energy value F wdh 。
(1) Maximum stable difference value F max Defined by the following formula:
wherein R is avg1 Is the average value after the cleaning preparation stage of the channel is subjected to baseline removal, R avg2 Is the average value after the cleaning recovery stage of the channel is subjected to baseline removal, R avg2 Is the average value after the short sniffing stage of the channel is de-baselined, R avg4 Is the average value after the secondary short sniffing stage of the channel is subjected to baseline removal, R avg5 Is the average value of the long-sniffing stage of the channel after baseline removal; max []The characteristic is that the maximum function is obtained, and compared with the maximum characteristic or the extreme difference characteristic which is commonly seen, the characteristic can more stably represent the severe condition of the response of the gas sensor;
(2) Steady state allValue F stbavg Defined by the following formula:
wherein R is j 、R k And R is l The response of the gas sensor when j, k and l seconds start to reach a steady state is that c is the total sampling time, j epsilon (0.1 c,0.2 c), k epsilon (0.3 c,0.4 c) and l epsilon (0.5 c,0.8 c); the parameters j, k and l are determined by two methods, namely, the derivative value of a test stage is obtained, and in the corresponding stage, if the derivative value at a certain moment is smaller than 0.1 (the corresponding tangent angle is smaller than 5 degrees), the response fluctuation is smaller at the moment, and the state is stable, the moment is set as j, k or l; if the conditions of the first method are not satisfied, j=0.175 c, k=0.375 c, l=0.725 c (meaning that the values are taken starting from the last 1/4 part of the three phases of short sniff, secondary short sniff and long sniff). The characteristic can better express the stable response condition of the gas sensor during testing;
(3) Response energy value F integ Defined by the following formula:
wherein R is i Is the response of the gas sensor after the ith second removes the baseline, F integ For integral value characteristics, i takes a value from 0.1c to 0.8c-1. This feature represents the overall degree of response of the gas sensor at the time of testing;
(4) The average differential value, the physical meaning of which reflects the overall dynamic process of the response signal, has the following formula:
wherein R is i Is the response of the gas sensor after the ith second is removed from the baseline, c is the total sampling time, F ndf Is characteristic of average differential value;
(5) The variance value reflects the discrete condition of the response signal in the physical meaning as follows:
F varnc =(R i -Avg(R 1 、R 2 ...R c )) 2 /c
wherein R is i Is the response of the gas sensor after the ith second removes the baseline, c is the total sampling time, avg is a function of the average of the elements in the collection, F varnc Is a variance value feature;
(6) The dc component, the formula is as follows:
F fft0 =FFTO(R 1 、R 2 ...R c )
wherein R is i Is the response of the gas sensor after the i second removes the baseline, c is the total sampling time, and FFT0 (·) is a function of extracting the DC component using the fast Fourier transform;
(7) The first order harmonic component is given by:
F fft1 =FFT1(R 1 、R 2 ...R c )
wherein R is i Is the response of the gas sensor after the i second removes the baseline, c is the total sampling time, and FFT1 (-) is a function of extracting the first order harmonic component by using the fast Fourier transform;
(8) The high frequency component, the formula is as follows:
F wdh =WDH(R 1 、R 2 ...R c )
wherein R is i The response of the gas sensor after the baseline is removed in the ith second, c is the total sampling time, WDH (·) is a function of using orthogonal wavelet Haar as a layer of decomposition to extract high-frequency components;
(9) The low frequency component, the formula is as follows:
F wdl =WDL(R 1 、R 2 ...R c )
wherein R is i Is the response of the gas sensor after the ith second removes the baseline, c is the total sampling time, WDL (·) is a function of extracting the low frequency components using orthogonal wavelet Haar as a one-layer decomposition.
Response number after baseline removal for each sensorAccording to the feature extraction, finally obtaining a time-frequency feature set F i =[F i_max ,F i_stbavg ,F i_integ ,F i_ndf ,F i_varnc ,F i_fft0 ,F i_fft1 ,F i_wdh ,F i_wdl ]Wherein i takes values of 1, 2, 3 and 4.
And carrying out proportion normalization processing on all four feature sets, wherein the formula is as follows:
wherein F is i Is a characteristic value of the ith sensor, z is a secondary judgment candidate set D two In this embodiment, z=4, so i takes values of 1, 2, 3 and 4.
Wherein i is the sensor number, i e (1, 4); j is the feature class number, j E (1, 9)
The second-level judgment index Q is calculated, and the formula is as follows:
wherein Q is i Is the second-level judgment index of the ith channel. The specific data are shown in the following table:
index/channel | S1 | S2 | S3 | S4 |
NF i_max | 0.39 | 0.25 | 0.27 | 0.09 |
NF i_ndf | 0.38 | 0.27 | 0.26 | 0.08 |
NF i_integ | 0.41 | 0.30 | 0.21 | 0.08 |
NF i_vrnc | 0.37 | 0.28 | 0.32 | 0.03 |
N Fi_stbavg | 0.40 | 0.29 | 0.25 | 0.06 |
NF i_fft0 | 0.36 | 0.27 | 0.29 | 0.08 |
NF i_fft1 | 0.35 | 0.29 | 0.31 | 0.05 |
NF i_wdh | 0.39 | 0.25 | 0.27 | 0.09 |
NF i_wdl | 0.38 | 0.27 | 0.26 | 0.08 |
Second-level judgment index Q | 0.38 | 0.27 | 0.28 | 0.07 |
The second level judgment rule is as follows:
wherein S is two_i Is the ith sensor in the secondary judgment alternative set, z is the secondary judgment alternative set S two The number of the sensors in (S) unsure Is an uncertainty set. The formula means: the sensor with the normalized eigenvalue greater than the average normalized eigenvalue can be considered to have stronger and similar response.
In this embodiment, where z=4, thenIf the second-level judgment index Q is 0.25 or more, putting the sensor corresponding to the channel into a cross-sensitive set S cross Otherwise, put into an uncertain set S unsure (no further analysis was done). Q (Q) 1 、Q 2 And Q 3 The values of S1, S2 and S3 are all placed in the cross-sensitive set S, respectively, greater than 0.25 cross ={S 1 ,S 2 ,S 3 }:Q 4 Less than 0.25, put S4 into an uncertainty set S unsure And no further analysis was performed. Information is further extracted using a heuristic data mining method as shown in fig. 9. The invention herein proposes a way of confidence estimation to cross-sensitize the set S cross The number of sensors contained in the sensor is taken as denominator N, and is combined with cross-sensitive volatile compound-sensor set V-S= { V 1 :{S 1 },V 2 :{S 1 ,S 2 },V 3 :{S 1 ,S 2 ,S 3 },V 4 :{S 1 },V 5 :{S 1 ,S 2 ,S 3 },V 6 :{S 2 },V 7 :{S 1 },V 8 :{S 3 },V 9 :{S 3 },V 10 :{S 4 },V 11 :{S 5 Each subset of the intersection sets is respectively calculated, the number of elements in a certain intersection set is set as M, if M=N, all the sensors sensitive to a certain type of volatile compounds are screened out, and the probability of the substances exists in 100 percent; if m=n/2, it means that only half of all sensors sensitive to a certain class of volatile compounds are screened, that class of substances is 50% likely to be present, and so on. For example, in the present embodiment S cross The number of sensors in the set is 3, the cross-sensitive set is { S1, S2, S3}, the cross-sensitive set is intersected with a subset in the V-S set, and the first subset is V1: { S1}, S1, corresponding to compound V1 (toluene), confidence 33.3% (1/3) as a result of the intersection with the cross-sensitive set { S1, S2, S3 }; the second subset is V2: { S1, S2}, intersecting the cross-sensitive set { S1, S2, S3} results in S1 and S2, corresponding to compound V2 (hydrogen sulfide), with a confidence of 66.6% (2/3); by analogy, the third result is S1, S2 and S3, corresponding to compound V3 (ethanol), with 100% confidence (3/3); the fourth result is S1, corresponding to compound V4 (ammonia), confidence 33.3% (1/3); the fifth result is S1, S2 and S3, corresponding to compound V5 (hydrogen), confidence 100% (3/3); the sixth result is S2, corresponding to compound V6 (trimethylamine), confidence 33.3% (1/3); the seventh result is S1, corresponding to compound V7 (methyl mercaptan), confidence 33.3% (1/3); the eighth result is S3, corresponding to compound V8 (propane), confidence 33.3% (1/3); the ninth result is S3, corresponding to compound V9 (isobutane), with a confidence of 33.3% (1/3). Summarizing the analysis results to obtain a second-level conclusion:
Inform 2
the = { gas to be measured contains ethanol (confidence 100%), hydrogen (confidence 100%, hydrogen sulfide (confidence 66.6%), toluene (confidence 33.3%), ammonia (confidence 33.3%), trimethylamine (confidence 33.3%), methyl mercaptan (confidence 33.3%), propane (confidence 33.3%), isobutane (confidence 33.3%) }
Finally, valuable information is obtained from unknown gas to be measured through data acquisition, preprocessing, feature extraction, two-stage cascade judgment and heuristic data mining, and the total conclusion is that:
Inform all
the = { unknown gas contains no formaldehyde, and contains ethanol (100% confidence), hydrogen sulfide (66.6% confidence), toluene (33.3% confidence), ammonia (33.3% confidence), trimethylamine (33.3% confidence), methyl mercaptan (33.3% confidence), propane (33.3% confidence), isobutane (33.3% confidence) }
The invention firstly collects response signals of the gas sensor in a sample injection mode of seven stages including short smell and long smell, then extracts 5 and 4 characteristics from two angles of a time domain and a frequency domain respectively aiming at original data, and utilizes the original data and time-frequency characteristic data to carry out two-stage cascade judgment and excavate effective information. The method provides an important basis for subsequent further analysis through pre-analysis, pre-judgment and data mining of the unknown component gas, and greatly expands the application boundary of the gas detection technology.
Claims (1)
1. The unknown gas response signal characteristic extraction and data mining method based on the gas sensor array is characterized by comprising the following steps of: step 1, acquiring multichannel sensor response data from an electronic nose; step 2, constructing a relation network between the volatile compound and the sensor; step 3, first-level judgment; step 4, second-level judgment; step 5, excavating final effective information; the method for extracting unknown gas response signal characteristics and mining data based on the gas sensor array comprises the steps that step 1 comprises the steps that an electronic nose simulates a short-breath process of human being on the rapid smell and a long-breath process which is continuous, and multichannel sensor response data are acquired and obtained, wherein the method specifically comprises seven stages: collecting unknown gas to be measured in a gas collecting bottle, sealing and placing the gas on a constant temperature table, keeping the temperature at 50 ℃ for 10 minutes, starting an electronic nose device, and performing test operation for 5 minutes to ensure that the instrument is stable, and simultaneously collecting data c seconds at the frequency of 1 Hz; wherein, collecting clean carrier gasSecond, as data of the cleaning preparation stage; subsequently, connecting a gas collecting bottle, and continuing to collect sample gas +.>Second, as data of the short sniff phase; then, the gas collection bottle is stopped to be charged, and residual gas in the gas collection bottle is collected>Second, as data of the pause stage; subsequently, connecting a gas collecting bottle, and continuing to collect sample gas +.>Second, as data of the secondary short sniffing stage; then, the gas collecting cylinder is stopped to be charged, and residual gas in the gas collecting cylinder is collected>Second, as data of the second pause stage; then, connecting a gas collecting bottle, and continuously collecting sample gas +.>Second, as data of the long sniff phase; finally, clean carrier gas is introduced to collect +.>Second, as data of the purge recovery phase; thereby obtaining D n =[d 1 ,d 2 ,…d c ]N represents the number of sensors, i.e. the number of channels, c represents the acquisition duration of each channel;
wherein step 2 comprises:
(1) The sensor set is established as follows:
Snrs=[S 1 ,S 2 …S n ]wherein n is the number of sensors, i.e., the number of channels;
(2) The union set is obtained for the volatile matter sets which are sensitive to all the sensors, and the formula is as follows:
Vocs=(V 11 ,V 12 …)∪(V 21 ,V 22 …)∪…(V n1 ,V n2 …), wherein V 11 Representing a first compound to which the first sensor is sensitive, V n1 A first compound representing the sensitivity of the nth sensor;
(3) Establishing a volatile compound collection:
Vocs=[V 1 ,V 2 …V m ]whereinm is the number of volatile substance types which are not repeated;
(4) Establishing a two-way relational mapping network W between volatile compounds and sensors Vocs-Snrs Specifically comprising a set of sensor-sensitive volatile compounds S-v= { S 1 :{V 1 ,V 2 ,...},S 2 :{V 1 ,V 2 ,...},...S n :{V 1 ,V 2 ,. }, meaning the total set of sensitive volatile compounds for each sensor; cross-sensitive volatile compound-sensor set V-s= { V 1 :{S 1 ,S 2 ,...},V 2 :{S 1 ,S 2 ,...},...V m :{S 1 ,S 2 ,. }, meaning the total set of sensor sets that are simultaneously sensitive to a certain volatile compound;
wherein step 3 comprises: the data of the five stages of short smell, secondary short smell, long smell, cleaning preparation and cleaning recovery of each sensor channel are averaged, and a first-level judging index P is calculated according to the following formula:
wherein p is i Is the first-level judgment index of the ith channel, namely the sensor, d avg1 Is the average value of the cleaning preparation stage of the channel, d avg2 Is the average value of the cleaning recovery phase of the channel, d avg3 Is the average value of the short sniff phase of the channel, d avg4 Is the average value of the secondary short sniffing phase of the channel, d avg5 Is the average value of the long sniff phase of the channel;
further, the first level judgment rule is as follows:
wherein S is i For the original sensor set Snrs setI-th sensor of (2), S one For one-stage judgment of completion set S two For the secondary judgment alternative set, if the primary judgment index is smaller than or equal to 0.1, the sensor of the channel is put into the primary judgment finished set, otherwise, the sensor is put into the secondary judgment alternative set S two At this time, according to S one And combining the s-v set, a first-order conclusion can be obtained:
Inform 1 in the = { unknown gas, no volatile compound sensitive to the sensor in Sone set }
The unknown gas response signal characteristic extraction and data mining method based on the gas sensor array comprises the following steps:
judging the alternative set S for the second level two The data in (2) is taken to baseline according to the following formula:
R i =d i -d baseline
wherein R is i Is the response value of the ith second after the removal of the baseline, d i Is the original response value of the sensor in the ith second, d baseline Is a baseline value (its size is equal to d avg1 ) Each channel collects c seconds of data altogether, so i E [1, c];
The unknown gas response signal characteristic extraction and data mining method based on the gas sensor array, wherein the step 4 further comprises the following steps:
extracting a secondary judgment candidate set S from two angles of a time transformation domain and a frequency transformation domain two Time domain features and frequency domain features in (a) including maximum stable difference value F max Steady state mean F stbavg Response energy value F integ Average differential value F ndf Sum of variances F varnc The method comprises the steps of carrying out a first treatment on the surface of the DC component amplitude F after Fourier transform fft0 Amplitude of first order harmonic component F fft1 And a wavelet transformed low frequency component energy value F wdh High frequency component energy value F wdl The specific formulas of the relevant characteristics are as follows:
(1) Maximum stable difference value F max Defined by the following formula:
wherein R is avg1 Is the average value after the cleaning preparation stage of the channel is subjected to baseline removal, R avg2 Is the average value after the cleaning recovery stage of the channel is subjected to baseline removal, R avg2 Is the average value after the short sniffing stage of the channel is de-baselined, R avg4 Is the average value after the secondary short sniffing stage of the channel is subjected to baseline removal, R avg5 Is the average value of the long-sniffing stage of the channel after baseline removal; max []The characteristic is that the maximum function is obtained, and compared with the maximum characteristic or the extreme difference characteristic which is commonly seen, the characteristic can more stably represent the severe condition of the response of the gas sensor;
(2) Steady state mean F stbavg Defined by the following formula:
wherein R is j 、R k And R is l The response of the gas sensor when j, k and l seconds start to reach a steady state is that c is the total sampling time, j epsilon (0.1 c,0.2 c), k epsilon (0.3 c,0.4 c) and l epsilon (0.5 c,0.8 c);
the parameters j, k and l are determined by two methods, namely, the derivative value of a test stage is obtained, and in the corresponding stage, if the derivative value at a certain moment is smaller than 0.1, the response fluctuation is smaller at the moment, and the state is stable, the moment is set as j, k or l; if the conditions of the first method are not met, j=0.175 c, k=0.375 c, l=0.725 c; the characteristic can better express the stable response condition of the gas sensor during testing;
(3) Response energy value F integ Defined by the following formula:
wherein R is i Is the response of the gas sensor after the ith second has removed the baseline,F integ for integral value characteristics, i takes values from 0.1c to 0.8c-1;
(4) The average differential value, the physical meaning of which reflects the overall dynamic process of the response signal, has the following formula:
wherein R is i Is the response of the gas sensor after the ith second is removed from the baseline, c is the total sampling time, F ndf Is characteristic of average differential value;
(5) The variance value reflects the discrete condition of the response signal in the physical meaning as follows:
F varnc =(R i -Avg(R 1 、R 2 ...R c )) 2 /c
wherein R is i Is the response of the gas sensor after the ith second removes the baseline, c is the total sampling time, avg is a function of the average of the elements in the collection, F varnc Is a variance value feature;
(6) The dc component, the formula is as follows:
F fft0 =FFT0(R 1 、R 2 ...R c )
wherein R is i Is the response of the gas sensor after the i second removes the baseline, c is the total sampling time, and FFT0 (·) is a function of extracting the DC component using the fast Fourier transform;
(7) The first order harmonic component is given by:
F fft1 =FFT1(R 1 、R 2 ...R c )
wherein R is i Is the response of the gas sensor after the i second removes the baseline, c is the total sampling time, and FFT1 (-) is a function of extracting the first order harmonic component by using the fast Fourier transform;
(8) The high frequency component, the formula is as follows:
F wdh =WDH(R 1 、R 2 ...R c )
wherein R is i Is the gas sensor to remove in the ith secondExcept for the response after the baseline, c is the total sampling time, WDH (·) is a function of extracting high-frequency components by using orthogonal wavelet Haar as a layer of decomposition;
(9) The low frequency component, the formula is as follows:
F wdl =WDL(R 1 、R 2 ...R c )
wherein R is i Is the response of the gas sensor after the baseline is removed in the ith second, c is the total sampling time, WDL (·) is a function of extracting low-frequency components by using orthogonal wavelet Haar as a layer of decomposition;
characteristic extraction is carried out on response data of each sensor after baseline removal, and a time-frequency characteristic set F is finally obtained i =[F i_max ,F i_stbavg ,F i_integ ,F i_ndf ,F i_varnc ,F i_fft0 ,F i_fft1 ,F i_wdh ,F i_wdl ]Wherein i is the number of the sensor;
and carrying out proportion normalization processing on the time-frequency characteristic set, wherein the formula is as follows:
wherein F is i Is a characteristic value of the ith sensor, z is a secondary judgment candidate set S two I epsilon (1, z);
obtaining normalized time-frequency characteristic matrixWherein i is the sensor number, i ε (1, z); j is the feature class number, j e (1, 9);
calculating a secondary judgment index Q i The formula is as follows:
the second level judgment rule is as follows:
wherein S is two_i Is the ith sensor in the secondary judgment alternative set, z is the secondary judgment alternative set S two The number of the sensors in (S) unsure Is an uncertainty set; the rule means that the second level judgment index is larger thanPut the sensor of (2) into the scanner, otherwise put into S unsure ;
Wherein step 4 further comprises: the heuristic data mining method comprises the following steps: (1) computing into a cross-sensitive set S cross The number of the sensors is recorded as N; (2) if N is equal to 0, directly entering a fifth step, and if N is greater than 0, entering a third step; (3) will S cross And V-S= { V 1 :{S 1 ,S 2 ,...},V 2 :{S 1 ,S 2 ,...},...V m :{S 1 ,S 2 Each subset in the,. set is intersected, and the number of sensors in the intersection is set as M; (4) calculating confidence coefficient C=M/N, and finding out corresponding volatile compounds according to the V-S set; (5) ending the flow; summarizing the analysis results to obtain a second-level conclusion:
Inform 2 in the = { unknown gas, S is contained cross Volatile compounds (confidence M/N) sensitive to sensors shared by the set and the V-S set
Wherein step 5 comprises:
valuable information is obtained from unknown gas to be measured through data acquisition, preprocessing, feature extraction, two-stage cascade judgment and heuristic data mining, and the total conclusion is that:
Inform all =Inform 1 +Inform 2 in the = { unknown gas, no S is contained one
Volatile compounds sensitive to the sensors in the collection; containing S cross Volatility of sensitivity corresponding to sensors shared by set and V-S setCompound (confidence M/N).
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210574810.2A CN114894980B (en) | 2022-05-25 | 2022-05-25 | Unknown gas response signal characteristic extraction and data mining method based on gas sensor array |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210574810.2A CN114894980B (en) | 2022-05-25 | 2022-05-25 | Unknown gas response signal characteristic extraction and data mining method based on gas sensor array |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114894980A CN114894980A (en) | 2022-08-12 |
CN114894980B true CN114894980B (en) | 2023-06-20 |
Family
ID=82726575
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210574810.2A Active CN114894980B (en) | 2022-05-25 | 2022-05-25 | Unknown gas response signal characteristic extraction and data mining method based on gas sensor array |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114894980B (en) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6649416B1 (en) * | 2000-02-18 | 2003-11-18 | Trustees Of Tufts College | Intelligent electro-optical sensor array and method for analyte detection |
CN106338597A (en) * | 2016-10-18 | 2017-01-18 | 哈尔滨工业大学深圳研究生院 | Respiratory gas measuring method and system |
CN108108437A (en) * | 2017-12-20 | 2018-06-01 | 胡明建 | The design method that a kind of computer vision sense of hearing smell is mutually mapped with the time |
CN112903811A (en) * | 2021-01-22 | 2021-06-04 | 湖北第二师范学院 | Acoustic sensing method for detecting concentration ratio of mixed gas |
WO2022006419A1 (en) * | 2020-07-01 | 2022-01-06 | President And Fellows Of Harvard College | Gas analysis through sniffing sequences |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7062385B2 (en) * | 2002-11-25 | 2006-06-13 | Tufts University | Intelligent electro-optical nucleic acid-based sensor array and method for detecting volatile compounds in ambient air |
CN101493431A (en) * | 2009-03-06 | 2009-07-29 | 天津商业大学 | Method for detecting fresh degree of chicken meat by electronic nose |
US9500563B2 (en) * | 2013-12-05 | 2016-11-22 | General Electric Company | System and method for detecting an at-fault combustor |
CN105954412B (en) * | 2016-05-07 | 2018-05-22 | 浙江大学 | For the sensor array optimization method of hickory nut freshness detection |
-
2022
- 2022-05-25 CN CN202210574810.2A patent/CN114894980B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6649416B1 (en) * | 2000-02-18 | 2003-11-18 | Trustees Of Tufts College | Intelligent electro-optical sensor array and method for analyte detection |
CN106338597A (en) * | 2016-10-18 | 2017-01-18 | 哈尔滨工业大学深圳研究生院 | Respiratory gas measuring method and system |
CN108108437A (en) * | 2017-12-20 | 2018-06-01 | 胡明建 | The design method that a kind of computer vision sense of hearing smell is mutually mapped with the time |
WO2022006419A1 (en) * | 2020-07-01 | 2022-01-06 | President And Fellows Of Harvard College | Gas analysis through sniffing sequences |
CN112903811A (en) * | 2021-01-22 | 2021-06-04 | 湖北第二师范学院 | Acoustic sensing method for detecting concentration ratio of mixed gas |
Non-Patent Citations (3)
Title |
---|
电子鼻判别挥发性气体的实验研究;邹小波,吴守一,方如明;江苏理工大学学报(第02期);全文 * |
电子鼻对干酪识别的数据预处理和特征提取;芦筱菲;郑丽敏;贾宗艳;朱虹;;计算机应用(第S1期);全文 * |
简述信号特征提取使用小波变换的优点;肖绪桐;虞天遥;;今日科苑(第12期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN114894980A (en) | 2022-08-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Van Dijck et al. | Speeding up the wrapper feature subset selection in regression by mutual information relevance and redundancy analysis | |
CN110786850B (en) | Electrocardiosignal identity recognition method and system based on multi-feature sparse representation | |
CN112508105B (en) | Fault detection and retrieval method for oil extraction machine | |
CN111238843B (en) | Fan health evaluation method based on rapid spectrum kurtosis analysis | |
CN113008559B (en) | Bearing fault diagnosis method and system based on sparse self-encoder and Softmax | |
CN106446777A (en) | Noninvasive blood sugar data processing method and noninvasive blood sugar data processing system based on convolutional neural network | |
CN108107086A (en) | A kind of gas detection method and gas sensor based on array gas sensor | |
CN113588847B (en) | Biological metabonomics data processing method, analysis method, device and application | |
CN113128567A (en) | Abnormal electricity consumption behavior identification method based on electricity consumption data | |
Liu et al. | A general-purpose signal processing algorithm for biological profiles using only first-order derivative information | |
CN113836704A (en) | Time series noise reduction method and device based on improved singular spectrum analysis | |
CN114894980B (en) | Unknown gas response signal characteristic extraction and data mining method based on gas sensor array | |
Ambika et al. | Vibration signal based condition monitoring of mechanical equipment with scattering transform | |
CN108345289B (en) | Industrial process stability detection method based on alternative data method | |
Haddad et al. | A feature extraction algorithm for multi-peak signals in electronic noses | |
CN113343869A (en) | Electroencephalogram signal automatic classification and identification method based on NTFT and CNN | |
CN109840386B (en) | Damage identification method based on factor analysis | |
CN110222390B (en) | Gear crack identification method based on wavelet neural network | |
CN112945556A (en) | Bearing fault diagnosis method based on wavelet packet decomposition and optimal rejection classification strategy | |
CN110378229B (en) | Electronic nose data feature selection method based on filter-wrapper frame | |
KR101110535B1 (en) | Method for recognizing plural smells | |
CN115420726A (en) | Method for rapidly identifying target object by using reconstructed SERS spectrum | |
Yeung et al. | Measuring correlation between microarray time-series data using dominant spectral component | |
CN114444218A (en) | Small sample rolling bearing fault diagnosis method under multiple working conditions | |
CN107977680A (en) | A kind of method of the sparse reconstruct of signal |
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 |