CN115935224A

CN115935224A - Gas identification method, apparatus, device, medium and product supporting online learning

Info

Publication number: CN115935224A
Application number: CN202211418519.2A
Authority: CN
Inventors: 陈虹; 霍德萱; 张吉霖
Original assignee: Tsinghua University
Current assignee: Tsinghua University
Priority date: 2022-11-14
Filing date: 2022-11-14
Publication date: 2023-04-07
Also published as: WO2024103639A1

Abstract

The present application relates to a gas identification method, apparatus, device, medium and product that support online learning. The method comprises the following steps: identifying the gas type of the collected target gas based on prestored network parameters of various sample gases to obtain a gas type identification result of the target gas; wherein each network parameter is generated when the gas types of a plurality of sample gases are learned in advance based on the bionic olfactory neural network. According to the method, the network parameters are obtained by learning various sample gases in advance, so that the identification precision of the target gas is improved.

Description

Gas identification method, device, equipment, medium and product supporting online learning

Technical Field

The present application relates to the field of gas detection technologies, and in particular, to a gas identification method, apparatus, device, medium, and product supporting online learning.

Background

Along with the development of chemical industry, the types of inflammable, explosive and toxic gases are continuously increased, the application range is continuously expanded, and once the gases leak in the processes of production, transportation and use, poisoning, fire and even explosion accidents can be caused, and the life and property safety of people is seriously harmed. Therefore, it is important to detect the gas quickly and accurately.

Generally, the detection of gas is realized by an electronic nose, which is a bionic detection technology simulating the olfactory system of mammals. The electronic nose comprises a gas sensor, a simulation front-end chip and a gas identification processor, wherein the gas identification processor plays a decisive important role in the final gas identification result of the electronic nose.

However, the gas identification processor in the related art has low identification accuracy in identifying gas.

Disclosure of Invention

In view of the foregoing, it is desirable to provide a gas identification method, apparatus, device, medium, and product supporting online learning, which can improve gas identification accuracy.

In a first aspect, the present application provides a gas identification method. The method comprises the following steps:

identifying the gas type of the collected target gas based on prestored network parameters of various sample gases to obtain a gas type identification result of the target gas;

wherein each network parameter is generated when the gas types of a plurality of sample gases are learned in advance based on the bionic olfactory neural network.

In one embodiment, the bionic olfactory nerve network comprises an excitation layer and an inhibition layer, the network parameters of each sample gas comprise standard pulse emission time at the excitation layer and weight information of neuron synapses between the excitation layer and the inhibition layer, and the gas type of the collected target gas is identified based on the pre-stored network parameters of a plurality of sample gases, so as to obtain a gas type identification result of the target gas, including:

acquiring target pulse emission time of the target gas at an excitation layer when the gas identification processor identifies the target gas by using weight information of neuron synapses corresponding to each sample gas;

comparing the target pulse emission time with each standard pulse emission time;

and taking the type of the sample gas corresponding to the standard pulse emission time with the highest similarity to the target pulse emission time as the gas type of the target gas.

In one embodiment, the method further comprises:

quantizing and standardizing the target gas to obtain a gas standard digital signal;

converting the gas standard digital signal into a pulse signal;

in one embodiment, the method further comprises:

and controlling the gas identification processor to execute the process of identifying the gas type of the target gas through the asynchronous logic control circuit.

In one embodiment, the asynchronous logic control circuit comprises a signal control unit and a signal trigger; the gas identification processor comprises a plurality of different functional modules, and each functional module is connected with a signal control unit and a signal trigger;

the asynchronous logic control circuit controls the gas identification processor to execute the process of identifying the gas type of the target gas, and the process comprises the following steps:

converting a starting signal of the current functional module into a control signal through a signal control unit;

and inputting a control signal into the current functional module through the signal trigger to indicate the current functional module to start executing the corresponding function.

In one embodiment, the generating process of the network parameters of the plurality of gases comprises:

acquiring pulse signals of various sample gases;

and updating the network parameters of the bionic olfactory neural network through each pulse signal until the bionic olfactory neural network converges to obtain the network parameters of various sample gases.

In one embodiment, the network parameters include weight information of neuronal synapses between excitation layers and inhibition layers in the bionic olfactory neural network, and pulse emission time of each sample gas at the excitation layer, and the network parameters of the bionic olfactory neural network are updated by each pulse signal until the bionic olfactory neural network converges, including:

for each sample gas, acquiring a first pulse emission time of a pulse signal of the sample gas at an excitation layer and a second pulse emission time at a suppression layer;

acquiring a weight variable quantity according to a time difference between the first pulse emission time and the second pulse emission time;

updating weight information of the neuron synapses based on the weight variation;

and (5) performing iteration for multiple times, and determining the convergence of the bionic olfactory neural network if the weight variation is zero.

In one embodiment, the method further comprises:

and controlling the bionic olfactory neural network to execute the process of updating the network parameters of the bionic olfactory neural network through each pulse signal until the bionic olfactory neural network converges by the asynchronous logic control circuit.

In one embodiment, the asynchronous logic control circuit includes a signal control unit and an enable control unit, and if the weight variation is zero, determining that the bionic olfactory neural network converges includes:

if the weight variation is zero, the enabling control unit sends an enabling signal to a signal control unit connected with the excitation layer to indicate the convergence of the bionic olfactory neural network.

In one embodiment, the method further comprises:

and after the bionic olfactory neural network converges, storing the network parameters of the various sample gases.

In a second aspect, the present application further provides a gas identification device. The device comprises:

the gas identification processor is used for identifying the gas type of the collected target gas based on the prestored network parameters of various sample gases to obtain a gas type identification result of the target gas; wherein each network parameter is generated when the gas types of a plurality of sample gases are learned in advance based on the bionic olfactory neural network.

In a third aspect, the present application also provides a computer device. The computer arrangement comprises a memory storing a computer program and a processor implementing the steps of any of the above embodiments of the gas identification method of the first aspect when the computer program is executed by the processor.

In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium having stored thereon a computer program which, when executed by a processor, performs the steps in any of the above-described embodiments of the gas identification method of the first aspect.

In a fifth aspect, the present application further provides a computer program product. The computer program product comprising a computer program which, when executed by a processor, performs the steps of any of the above embodiments of the gas identification method of the first aspect:

the gas identification method, the gas identification device, the gas identification equipment, the gas identification medium and the gas identification product supporting online learning are used for identifying the gas type of the collected target gas based on the prestored network parameters of various sample gases to obtain the gas type identification result of the target gas. Wherein each network parameter is generated when the gas types of a plurality of sample gases are learned in advance based on the bionic olfactory neural network. According to the method, the network parameters of the various sample gases are pre-stored by learning the various sample gases in advance, the pre-stored network parameters are called when the target gas is identified, and the learning of the sample gases can be continued after the target gas is identified by identifying the network parameters of the target gas, so that the gas identification precision is improved.

Drawings

FIG. 1 is a diagram of an exemplary gas identification method;

FIG. 2 is a schematic flow diagram of a gas identification method in one embodiment;

FIG. 3 is a schematic flow chart of a gas identification method in another embodiment;

FIG. 4 is a schematic flow chart of a gas identification method in another embodiment;

FIG. 5 is a schematic flow chart of a gas identification method in another embodiment;

FIG. 6 is a schematic diagram of the circuit logic of the gas identification method in one embodiment;

FIG. 7 is a schematic flow chart of a gas identification method in another embodiment;

FIG. 8 is a schematic flow chart of a gas identification method in another embodiment;

FIG. 9 is a schematic circuit logic diagram of a gas identification method in another embodiment;

FIG. 10 is a circuit diagram of a signal control unit of the gas identification method according to an embodiment;

FIG. 11 is a circuit diagram of a signal control unit of a gas identification method in another embodiment;

FIG. 12 is a circuit diagram of a signal control unit of a gas identification method in another embodiment;

FIG. 13 is a schematic flow chart of a gas identification method in another embodiment;

FIG. 14 is a schematic flow chart of a gas identification method in another embodiment;

FIG. 15 is a block diagram showing the structure of a gas identification device according to an embodiment;

FIG. 16 is a diagram illustrating an internal structure of a computer device according to an embodiment.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of and not restrictive on the broad application.

The gas identification method provided by the embodiment of the application can be applied to the application environment shown in fig. 1. Wherein collection facility 102 communicates with gas identification device 104 via a network. The collection device 102 may be a sensor array composed of a plurality of sensors, the gas identification device 104 is used for identifying gas, and the collection device 102 transmits collected data to the gas identification device 104. Optionally, the gas identification device 104 is disposed in an electronic nose, and the acquisition device 102 interacts with the electronic nose and transmits the acquired data to the electronic nose. Alternatively, the acquisition device 102 and the gas identification device may be provided in an electronic nose at the same time, and the acquisition of data and the identification of gas may be accomplished electronically.

With the increase of toxic gases, the rapid and accurate detection of the gas types is very important, so that in case of gas leakage accidents, remedial measures can be taken as soon as possible through real-time rapid detection, and the accident loss is reduced to the minimum.

The electronic nose has been developed for many years as a bionic detection technology for simulating the olfactory system of mammals, and is widely applied to scenes such as food safety, environmental monitoring, agriculture, industry, public safety and the like. And the early electronic nose has the problems of heavy volume, single function and limited use scene, and is difficult to realize the field real-time monitoring of gas, so that the integrated and intelligent portable electronic nose appears.

The portable electronic nose comprises a gas sensor, an analog front-end chip and a gas identification processor. Among these, the gas identification methods used by conventional gas identification processors include, but are not limited to, support vector machines, K-nearest neighbors and decision trees. However, these common gas identification methods have the disadvantages of low identification accuracy and few identification types, and can only identify 5 mixed gases at most simultaneously, and more importantly, the common gas identification processors have no learning ability, and thus do not have the function of dynamically adjusting parameters and structures, and under the influence of the drift phenomenon of the gas sensor (the sensitivity and selectivity of the sensor change gradually with the use time), the parameters stored in the gas identification processors are often not matched with the actual situation, which results in the substantial reduction of the actual identification accuracy. In addition, the conventional gas identification processor has high computational power requirement, the power consumption is often in the milliwatt level, and under the condition of no independent power supply, the portable electronic nose with the built-in chip can only work for several hours and is difficult to apply to tasks of long-time standby such as post-disaster search and rescue.

Based on the above, embodiments of the present application provide a gas identification method, apparatus, device, medium, and product supporting online learning, which can identify a gas type of a target gas, and simultaneously adjust network parameters in real time before identifying the gas type of the target gas each time. The gas type identification device can identify various gas types, improves identification precision, and overcomes the defect of high power consumption by adopting an asynchronous logic circuit.

The gas identification method provided in the examples of the present application is explained below with specific examples.

In one embodiment, as shown in fig. 2, there is provided a gas identification method comprising the steps of:

s201, identifying the gas type of the collected target gas based on the pre-stored network parameters of the multiple sample gases to obtain a gas type identification result of the target gas.

The target gas is a gas whose gas type needs to be identified by a gas identifier, and the target gas may be a flammable, explosive, or toxic gas, and exemplarily can identify gases such as ammonia gas, nitrogen dioxide, sulfur dioxide, or hydrogen sulfide. The target gas can be collected by a sensor, for example, a metal oxide semiconductor conductivity type gas sensor, a surface acoustic wave gas sensor or a quartz crystal microbalance gas sensor. The gas identification processor is a module in the electronic nose, and the gas type of the target gas can be obtained by inputting the collected target gas into the gas identification processor.

The gas identification processor comprises a bionic olfactory nerve network, and each network parameter is generated when the bionic olfactory nerve network learns the gas types of multiple sample gases in advance. The bionic olfactory neural network is a mathematical model simulating some mechanisms of biological olfaction, namely, a process that in biological neurons, when one neuron is activated, a signal is generated and transmitted to other neurons is simulated. The sample gas is adopted when the bionic olfactory nerve network is studied, and the network parameters of the multiple sample gases pre-stored in the gas identification processor are obtained under the study mode of the gas identification processor. In the process of identifying the target gas, based on the network parameters of the plurality of pre-stored sample gases, the identification result of the target gas is output, the target result is one of the sample gases, and the identification of the gas type of the target gas is performed in the working mode of the gas identifier. Illustratively, two sample gases, ammonia gas and nitrogen dioxide, are learned, and the gas type of the target gas is identified based on the network parameters of the two sample gases, specifically, the identification result may be that the target gas is nitrogen dioxide.

In the gas identification method provided by the embodiment of the application, the gas type of the collected target gas is identified based on the prestored network parameters of various sample gases, and the gas type identification result of the target gas is obtained. Wherein each network parameter is generated when the gas types of a plurality of sample gases are learned in advance based on the bionic olfactory neural network. According to the method, the network parameters of the various sample gases are pre-stored by learning the various sample gases in advance, the pre-stored network parameters are called when the target gas is identified, and the learning of the sample gases can be continued after the target gas is identified by identifying the network parameters of the target gas, so that the gas identification precision is improved.

In this embodiment of the application, based on the network parameters of the plurality of pre-stored sample gases, the gas type of the target gas is identified, and first the network parameters of the plurality of pre-stored sample gases need to be obtained, based on which, in an embodiment, the bionic olfactory nerve network includes an excitation layer and an inhibition layer, the network parameters of each sample gas include a standard pulse emission time at the excitation layer and weight information of a neuron synapse between the excitation layer and the inhibition layer, and then based on the network parameters of the plurality of pre-stored sample gases, the gas type of the collected target gas is identified, and a gas type identification result of the target gas is obtained, as shown in fig. 3, including:

s301, acquiring target pulse emission time of the target gas at the excitant layer when the gas identification processor identifies the target gas by the weight information of the neuron synapse corresponding to each sample gas.

The neuron is a basic information processing unit in a bionic olfactory neural network, and the bionic olfactory neural network is composed of all neurons of an excitation layer and a suppression layer and is used for simulating the transmission characteristics of biological neuron information, namely information transmission is carried out under the pulse triggering by means of synapses connected with the neurons. Specifically, the excitation layer neurons in this application use integrated emission neuron (IF) circuits, and the excitation layer neurons use Leaky-integrated-and-Fire (LIF) circuits. The synaptic connection between the two is excitatory synapse, each excitatory layer IF neuron corresponds to five inhibitory layer LIF neurons, specifically, the six neurons are in one group, each group of five inhibitory layer LIF neurons are connected with different groups of excitatory layer IF neurons, and the connection between the two is regarded as inhibitory synapse. Specifically, the number of neuron groups in the present application is at least twelve, that is, a minimum of twelve excitable layer IF neurons are required.

The target pulse emission time is the pulse transmission time of the pulse information of the target gas through IF neuron pulses of the excitation layer, and after learning various sample gases by adopting a gas identifier of a bionic olfactory nerve network, the weight information of corresponding neuron synapses of the various sample gases can be obtained, wherein w is used _n The weight information of corresponding neuronal synapses for multiple sample gases obtained is shown, illustratively, as w, if two sample gases are learned ₁ And w ₂ Specifically, if there are twelve groups of neurons at this time, each w _n Is a set of weight information corresponding to twelve groups of neurons.

When learning the sample gas, the standard pulse emission time of the sample gas at the excitation layer is also obtained, here by S _Ln Illustratively, if two sample gases are learned, the weight information w of the corresponding neuronal synapses of the two sample gases is utilized ₁ And w ₂ Obtaining the standard pulse emission time S of two sample gases at the excitation layer _L1 And S _L2 Each sample gas corresponds to weight information of a neuronal synapse and a standard pulse firing time. Specifically, if there are twelve groups of neurons, each neuron has its own pulse firing time, at which time each S _Ln All are standard pulse sent by twelve groups of neurons at excitation layerA set of firing times.

Calling weight information w of corresponding neuron synapses of multiple sample gases respectively when identifying target gas _n Acquiring target pulse emission times of multiple target gases at the excitable layer, here S _tn Shows, illustratively, learning two sample gases, then obtaining the sum w ₁ And w ₂ Corresponding S _t1 And S _t2 Specifically, each obtained S _tn Are a set of multiple target pulse firing times.

And S302, comparing the target pulse emission time with each standard pulse emission time.

The target pulse emission time S obtained when identifying the gas _tn And a standard pulse transmission time S obtained when learning the sample gas _Ln By contrast, for example, the weight information w of a sample gas at the synapse of a neuron ₁ Standard pulse emission time S obtained _L1 And using the weight information w in the target gas ₁ Acquired target pulse emission time S _t1 By contrast, in particular, the standard pulse emission time S _L1 And target pulse emission time S _t1 All are sets with the same number, and whether the pulse emission time of the corresponding positions in the two sets is the same or not is compared.

And S303, taking the type of the sample gas corresponding to the standard pulse emission time with the highest similarity to the target pulse emission time as the gas type of the target gas.

Comparing the standard pulse emission time S _L1 And target pulse emission time S _t1 The same number of pulse emission times of corresponding positions in the two sets is compared, and the type of the sample gas corresponding to the standard pulse emission time with the highest similarity to the target pulse emission time is used as the gas type of the target gas. Learning how many samples to make several comparisons, illustratively, learning two sample gases results in a standard pulse transmission time S _L1 、S _L2 And target pulse emission time S _t1 And S _t2 A 1, S _L1 And S _t1 The comparison is carried out, and the comparison is carried out,then the S is mixed _L2 And S _t2 Comparing if S is _L1 And S _t1 When the comparison is carried out, the number of the pulse emission time at the corresponding positions in the two sets is the largest, and the corresponding learned sample gas type is taken as the gas type of the target gas.

In the embodiment of the application, when the gas identification processor identifies the target gas by the weight information of the neuron synapse corresponding to each sample gas, the emission time of the target pulse of the target gas at the excitant layer is obtained; comparing the target pulse emission time with each standard pulse emission time; and taking the type of the sample gas corresponding to the standard pulse emission time with the highest similarity to the target pulse emission time as the gas type of the target gas. Before the target gas is identified, a plurality of sample gases are learned in advance in a learning mode, and compared with network parameters of the target gas in a working mode, the gas identification precision is improved.

In this embodiment of the present application, a network parameter of each sample gas is obtained, a gas type of a target gas may be identified through prestored network parameters of multiple sample gases, and before learning the sample gas or identifying the target gas, data preprocessing needs to be performed on digital signals of the sample gas and the target gas output by an analog front end input into a bionic olfactory neural network, and based on this, in an embodiment, as shown in fig. 4, the method further includes:

s401, quantizing and standardizing the target gas to obtain a gas standard digital signal.

And S402, converting the gas specification digital signal into a pulse signal.

In the embodiment of the application, the digital signal of the target gas output by the analog front end is subjected to data preprocessing, specifically, the digital signal of the target gas can be subjected to minimum-maximum normalization, the original digital signal is subjected to linear transformation, the data is mapped to a 4-bit digital signal interval of 0-15, and the data is sorted from large to small to obtain the gas-normalized digital signal.

The pulse coding is to convert the gas standard digital signal into a random pulse, specifically, the pulse coding adopts an event triggering mechanism expressed by an address event, and when data is transmitted, the gas standard digital signal is digitally coded first. The encoded digital signal is then pulse decoded to generate random pulses, which are abrupt changes in the level state of the electronic circuit, either abruptly rising (rising edge of the pulse) or abruptly falling (falling edge of the pulse). Generally, after the level is suddenly changed, the original level state can be restored in a short time.

In the real world, data is mostly incomplete and inconsistent, the data lacks uniform standards and definitions, and the data structure has larger differences and cannot be directly mined. The data preprocessing is carried out on the digital signals, so that the data mining quality can be improved, the time required by actual mining is reduced, and valuable information resources are obtained.

The method also comprises the following steps of controlling the target gas through an asynchronous logic control circuit when the function of identifying the gas type of the target gas is realized, and based on the control result, in one embodiment, the method further comprises the following steps: and controlling the gas identification processor to execute the process of identifying the gas type of the target gas through the asynchronous logic control circuit.

The asynchronous logic circuit is not driven by the same clock or a clock with an effect relationship for part of input and output of the circuit, and the asynchronous circuit is also called a clockless or self-timing circuit, namely a circuit without a global clock, and whether a party can accept data is informed by a Request signal and acknowledgement sent by a front stage and a back stage. Specifically, in the asynchronous logic circuit, a control signal is input to trigger the gas identification processor to identify the gas type of the target gas.

An asynchronous logic circuit is adopted to replace a synchronous logic circuit, the synchronous circuit works under the same global clock, and the working frequency of the clock must meet the requirement of the maximum load, so that power consumption waste is caused. The asynchronous logic circuit is driven by data, clock signals are removed, meanwhile, the gas identification processor is directly caused to act only through input control signals, dynamic power consumption is reduced, and application requirements of high speed and low power consumption are met.

In the embodiment of the application, the gas identification processor is controlled by the asynchronous logic control circuit to identify the gas type of the target gas, which is specifically realized by a plurality of different functional modules included in the gas identification processor, and in one embodiment, the asynchronous logic control circuit includes a signal control unit and a signal trigger; the gas identification processor comprises a plurality of different functional modules, and each functional module is connected with a signal control unit and a signal trigger; the process of identifying the gas type of the target gas is executed by controlling the gas identification processor through the asynchronous logic control circuit, as shown in fig. 5, and comprises the following steps:

s501, converting the starting signal of the current functional module into a control signal through a signal control unit.

The signal control unit is used for converting the starting signal into a control signal. Specifically, when the start signal is inverted, an output pulse, that is, a control signal is generated, and the rising edge of the control signal drives the flip-flop, thereby completing the logic of the circuit. Wherein, the signal control unit is a Click control unit.

And S502, inputting a control signal into the current functional module through the signal trigger, and indicating the current functional module to start to execute the corresponding function.

In the embodiment of the application, the signal flip-flop is a flip-flop without a clock, and data is not transmitted and processed by triggering of a clock signal. The signal triggers are all controlled by control signals generated by the signal control unit of the asynchronous logic circuit.

Specifically, the acquired digital signal of the target gas output by the analog front end is temporarily stored in a trigger, and after receiving the control signal, the trigger transmits and processes the temporarily stored data after being triggered. For example, as shown in fig. 6, which is a flowchart of a process in which an asynchronous logic control circuit controls a gas identification processor to identify a gas type of a target gas, in fig. 6, an acquired digital signal of the target gas output by an analog front end is temporarily stored in a trigger, a start signal 1 is input into a signal control unit 1, the start signal 1 is converted into a control signal 1, and then the control signal is sent to the trigger, so that the trigger transmits and preprocesses the temporarily stored data, and the preprocessed data is continuously temporarily stored in a next trigger. And then transmitting the starting signal 1 of the previous step to the next signal control unit 2 to serve as the starting signal 2, and so on until all the signal control units generate the control signals, and triggering to finish data transmission and processing.

In practical application, the Flip-Flop may be a D-type Flip-Flop (DFF), the start signal may be represented as Reqin, the control signal may be represented as fire, and the control unit represents a Click.

In the process of identifying the gas type of the target gas by controlling the gas identification processor through the asynchronous logic control circuit, the identification is mainly completed through the functional modules, and data can be triggered to be transmitted and processed only after the control signal of the signal control unit is received, namely, the input of a control signal is allowed at one moment, so that the competition risk caused by input signals is avoided.

The gas type of the target gas is identified based on the network parameters of the multiple sample gases prestored in the gas identification processor. Based on this, in one embodiment, the network parameters of the plurality of gases are generated as shown in fig. 7, which includes:

s701, pulse signals of various sample gases are obtained.

S702, updating the network parameters of the bionic olfactory neural network through each pulse signal until the bionic olfactory neural network converges to obtain the network parameters of various sample gases.

It should be noted that the pulse signal of the sample gas corresponds to the type of the sample gas, and since the embodiment of the present application identifies different types of gases through the bionic olfactory neural network, the pulse signals of a plurality of sample gases need to be acquired.

After pulse signals of various sample gases are obtained, the obtained pulse signals of various sample navel patches are used as a training set to train the bionic olfactory nerve network. The network parameters of the bionic olfactory nerve network can be updated along with the training process, meanwhile, the training process of the bionic olfactory nerve network can be visually displayed, and the training convergence degree of the bionic olfactory nerve network is judged from the display result. When the training process of the bionic olfactory nerve network tends to converge, the completion of the training of the bionic olfactory nerve network can be judged, and the network parameters of the trained bionic olfactory nerve network are used as the network parameters of various sample gases.

In the embodiment of the application, the network parameters are updated through the pulse signals of various sample gases, so that the generalization capability of the olfactory neural network can be improved while the training precision of the model is improved. In addition, the network parameters of the sample gas obtained in the embodiment of the application are obtained by pre-training, which means that in the actual gas identification process, the network parameters of various gases do not need to be retrained, so that the gas identification process is faster.

Based on the above network parameters of multiple sample gases prestored in the gas identification processor, identifying the gas type of the target gas, where the network parameters of each sample gas need to be obtained through calculation of the neural network, in one embodiment, the network parameters include weight information of neuronal synapses between an excitation layer and an inhibition layer in the bionic olfactory neural network, and pulse emission time of each sample gas at the excitation layer, and the network parameters of the bionic olfactory neural network are updated through each pulse signal until the bionic olfactory neural network converges, as shown in fig. 8, including:

s801, for each sample gas, acquiring a first pulse emission time of a pulse signal of the sample gas at the excitation layer and a second pulse emission time at the inhibition layer.

The pulse signal is a signal transmitted to neurons in the excitation layer and is obtained through a target signal. When learning each sample gas, the bionic olfactory nerve network finishes convergence, namely learning is finished after the weight of synapses among neurons stops updating.

In the bionic olfactory nerve network, the pulse time emitted by excitation layer neuron is the first pulse emission time, and t is used ₁ The pulse time of the neuron in the inhibition layer is represented as a second pulse emission time, which is represented by t ₂ And (4) showing.

S802, acquiring the weight variation according to the time difference between the first pulse emission time and the second pulse emission time.

From the first pulse transmission time and the second pulse transmission time, a time difference can be derived, denoted Δ t = t ₂ -t ₁ The obtained time difference Δ t is a set of a plurality of time differences. The weight variation Δ w can be calculated according to the time difference, the weight variation Δ w is a set of weight variations corresponding to synapses of a plurality of groups of neurons, and synapses between each neuron calculate the corresponding weight variation. Illustratively, if there are twelve groups of neurons, there are twelve time differences forming the set Δ t and twelve weight changes Δ w.

Specifically, the method includes performing an exponential operation according to a time difference Δ t between a first pulse transmission time and a second pulse transmission time and a synapse coefficient between corresponding neurons to obtain a weight variation Δ w of synapses, outputting the weight variation according to a required precision, where the weight coefficient of each synapse is given autonomously, for example, the method may also include performing an exponential operation according to the time difference between the first pulse transmission time and the second pulse transmission time and the synapse coefficient to obtain a corresponding weight variation, storing the time difference between the first pulse transmission time and the second pulse transmission time, the synapse coefficient and the corresponding weight variation in a lookup table, and after obtaining the time difference, finding the corresponding weight variation in the lookup table according to a segmented LUT method to update weight information.

S803, the weight information of the neuron synapse is updated based on the weight variation.

The weight information is updated by the synaptic weight change Δ w between corresponding neurons, specifically, there is a corresponding weight change in the synaptic weight change Δ w between each neuron, and the synaptic weight information of the synaptic neurons is updated by adjusting the synaptic weight between each neuron up or down according to the corresponding weight change.

Specifically, the first time the weight information of the neuron synapse is updated, the time t is emitted by the first pulse ₁ And a second pulse transmission time t ₂ Calculating the weight variation of the weight information of the neuron synapse updated for the first time by the obtained time difference delta t, and when the weight information of the neuron synapse is updated for the second time, utilizing the weight information of the neuron synapse updated for the first time, wherein the first pulse emission time emitted by the neuron of the excitation layer is t ₃ The time difference used for the updating of the round is the second pulse emission time t obtained in the previous round ₂ And the first pulse transmission time t of the round ₃ Time difference Δ t.

And S804, performing iteration for multiple times, and determining the convergence of the bionic olfactory neural network if the weight variation is zero.

And updating the weight information of the synapse of the neuron according to the weight variation obtained after the time difference between the first pulse emission time and the second pulse emission time is transmitted each time, and ending the cycle until the obtained weight variation is zero, which indicates that the learning of the sample gas is finished.

When a sample gas is studied, the network parameters of the bionic olfactory nerve network are updated through each pulse signal until the nerve network converges to indicate that the sample gas is studied, and the obtained network parameters are accurate and are compared with the network parameters of the target gas in a working mode, so that the gas identification precision is improved.

The generation process of the network parameters of the multiple gases also needs to be controlled by an asynchronous logic control circuit to complete the training of the bionic olfactory neural network, and based on this, in one embodiment, the method further comprises the following steps: and controlling the bionic olfactory neural network to execute the process of updating the network parameters of the bionic olfactory neural network through each pulse signal until the bionic olfactory neural network converges by the asynchronous logic control circuit.

Before the gas identification processor executes the process of identifying the gas type of the target gas, the sample gas is required to control the bionic olfactory neural network through the asynchronous logic control circuit to update the network parameters of the bionic olfactory neural network through each pulse signal until the bionic olfactory neural network converges.

As shown in fig. 9, the signal control units 3 to 8 control the bionic olfactory neural network through the asynchronous logic control circuit to update the network parameters of the bionic olfactory neural network through each pulse signal until the bionic olfactory neural network converges. Specifically, as shown in fig. 10 and 11, the signal control unit 3 and the signal control unit 3 are both multi-input multi-output signal control units, specifically, the signal control unit 3 transmits the start signal 3 to the signal control unit 4 and the signal control unit 6 respectively, which represent that the pulse signal is transmitted to the neuron of the inhibition layer and the first pulse emission time of the pulse signal of the sample gas at the excitation layer is sent to the weight updating module, the signal control unit 6 will not convert to the control signal 6 when receiving the two start signals at the same time, so that the trigger triggers the weight updating, and the signal control unit 3 continues to transmit the pulse signal after receiving the start signal 3 after the updated weight until the weight updating is finished. Similarly, the signal control unit may be a Click control unit, the start signal may be denoted by Reqin, and the control signal may be denoted by fire, and the specific reference of each part is not limited in this embodiment of the application.

When a sample gas is learned, in the process of updating the network parameters of the bionic olfactory nerve network by each pulse signal, the gas identification processor needs to be controlled by the asynchronous logic control circuit to update the network parameters, and data can be triggered to be transmitted and processed only after the control signal of the signal control unit is received, that is, the input of a control signal is allowed at one moment, so that the competition risk caused by input signals is avoided.

The signal control unit 3 is an enable control unit with an enable signal, and as shown in fig. 12, if the weight change amount calculated in the weight update is zero, the enable control unit sends an enable signal to the signal control unit connected to the excitation layer at this time, and the rising edge is turned off, so that the control signal is not converted into the control signal. And the network parameters stored by the weight storage module are final results.

When a sample gas is studied, in the process of updating the network parameters of the bionic olfactory neural network by each pulse signal, the signal control unit with an enabling signal in the gas identification processor is controlled by an asynchronous logic control circuit to indicate the convergence of the bionic olfactory neural network. The end of the sample gas learning can be judged in time, and the network parameters of the sample gas are stored.

Before identifying the gas type of the target gas, pre-stored network parameters of a plurality of sample gases need to be called, and the identification parameters need to be stored in advance for later calling, and based on the identification, in one embodiment, the method further comprises the following steps:

After the bionic olfactory neural network converges, that is, after learning a plurality of sample gases, the network parameters of each sample gas are obtained. The network parameters of each sample gas include a standard pulse emission time at the excitation layer and weight information of the neuron synapses between the excitation layer and the inhibition layer, and specifically, when the weight information of the neuron synapses between the excitation layer and the inhibition layer is not updated, the network parameters at this time are temporarily stored in a buffer group until the update of the weight information is completed, and the final network parameters are stored. When the target gas is identified, the target pulse emission time of the target gas is obtained based on weight information of neuron synapses between an excitable layer and a suppressive layer, and then is compared with the standard pulse emission time of each sample gas at the excitable layer. Thus, the network parameters of the plurality of sample gases are stored.

When the sample gas is learned, the previously learned data is not required to be discarded, only new sample gas data is required to be input into the processor again, network parameters of various sample gases are stored, a learning-working-reckoning mode can be performed, and the algorithm challenge of catastrophic forgetting is overcome.

In one embodiment, as shown in fig. 13, which is another application environment diagram of the gas identification method, corresponding specific steps are shown in fig. 14, and the gas identification method includes the following steps:

s901, quantizing and standardizing the sample gas through a preprocessing module to obtain a gas standard digital signal;

s902, converting the gas standard digital signals into pulse signals through a pulse coding module, and inputting the pulse signals of various sample gases into a gas identification processor;

s903, aiming at each sample gas, acquiring first pulse emission time of a pulse signal of the sample gas at an excitation layer;

s904, aiming at each sample gas, acquiring second pulse emission time of a pulse signal of the sample gas at a suppression layer;

s905, acquiring a weight variable quantity according to a time difference between the first pulse emission time and the second pulse emission time;

specifically, the connections of the excitation layer and the inhibition layer are described by a synaptic address generator, which is composed of a buffer and a lookup table.

S906, updating weight information of the synapse of the neuron based on the weight variation;

s907, performing iteration for multiple times, and if the weight variation is zero, determining that the bionic olfactory nerve network converges, and finishing the learning of the sample gas;

s908, storing network parameters of various sample gases;

s909, obtaining the target pulse emission time of the target gas at the excitant layer when the gas identification processor identifies the target gas by the weight information of the neuron synapses corresponding to each sample gas;

in particular, the classifier is used to invoke stored weight information for the neuronal synapses. The above-described S901 to S907 are operations performed by the gas identifier in the learning mode as a whole.

S910, calling the pulse emission time recorded when the sample gas is learned;

in particular, the classifier is used to recall the stored standard pulse firing times.

S911, comparing the target pulse emission time with each standard pulse emission time;

specifically, a comparator is used to compare the target pulse emission time with each standard pulse emission time.

And S912, taking the type of the sample gas corresponding to the standard pulse emission time with the highest similarity to the target pulse emission time as the gas type of the target gas.

Specifically, the operations of the gas identifier in the operation mode are performed as a whole in S910 to S912.

The problems of few identification types, low identification precision and high power consumption of the portable electronic nose are solved. The asynchronous circuit based on multiple Click control units realizes online learning of the gas recognizer, realizes the design of the low-power-consumption gas recognition processor, can solve the problem of performance reduction and even failure caused by sensor drift, improves recognition accuracy, prolongs service life, meets the actual application requirements of multiple recognition types and low power consumption, and widens the application field of gas recognition.

The above steps are all input control signals into the function module through the signal trigger to indicate the current function module to start executing the corresponding function

It should be understood that, although the steps in the flowcharts related to the embodiments as described above are sequentially displayed as indicated by arrows, the steps are not necessarily performed sequentially as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least a part of the steps in the flowcharts related to the embodiments described above may include multiple steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, and the execution order of the steps or stages is not necessarily sequential, but may be rotated or alternated with other steps or at least a part of the steps or stages in other steps.

Based on the same inventive concept, the embodiment of the application also provides a gas identification device for realizing the gas identification method. The solution to the problem provided by the apparatus is similar to the solution described in the above method, so the specific limitations in one or more embodiments of the gas identification apparatus provided below can be referred to the limitations on the gas identification method in the above, and are not described herein again.

In one embodiment, as shown in fig. 15, there is provided a gas identification device including: an acquisition module 10, wherein:

the acquisition module 10 is configured to identify a gas type of the acquired target gas based on prestored network parameters of multiple sample gases, so as to obtain a gas type identification result of the target gas; wherein each network parameter is generated when the gas types of a plurality of sample gases are learned in advance based on the bionic olfactory neural network.

In one embodiment, the obtaining module 10 includes: a first acquisition unit, a comparison unit and a determination unit, wherein:

and the first acquisition unit is used for acquiring the emission time of the target pulse of the target gas at the excitable layer when the gas identification processor identifies the target gas by the weight information of the neuron synapse corresponding to each sample gas.

And the comparison unit is used for comparing the target pulse emission time with each standard pulse emission time.

And the determining unit is used for taking the type of the sample gas corresponding to the standard pulse emission time with the highest similarity to the target pulse emission time as the gas type of the target gas.

In one embodiment, the above apparatus further comprises: processing module and conversion module, wherein:

the processing module is used for carrying out quantization and standardization processing on the target gas to obtain a gas standard digital signal;

and the conversion module is used for converting the gas standard digital signal into a pulse signal.

In one embodiment, the above apparatus further comprises: and the identification module is used for controlling the gas identification processor to execute the process of identifying the gas type of the target gas through the asynchronous logic control circuit.

In one embodiment, the identification module further comprises: a conversion unit and an indication unit, wherein:

the conversion unit is used for converting the starting signal of the current functional module into a control signal through the signal control unit;

and the indicating unit is used for inputting the control signal into the current functional module through the signal trigger and indicating the current functional module to start to execute the corresponding function.

In one embodiment, the apparatus further comprises a learning module, the learning module comprising: a second acquisition unit and an update unit, wherein:

and the second acquisition unit is used for acquiring pulse signals of the multiple sample gases.

And the updating unit is used for updating the network parameters of the bionic olfactory neural network through each pulse signal until the bionic olfactory neural network converges to obtain the network parameters of various sample gases.

In one embodiment, the device second obtaining unit includes a first obtaining subunit and a second obtaining subunit, and the device updating unit includes an updating subunit and a determining subunit. Wherein:

the first acquisition subunit is used for acquiring a first pulse emission time of a pulse signal of the sample gas at the excitation layer and a second pulse emission time at the inhibition layer for each sample gas.

And the second acquiring subunit is used for acquiring the weight variation according to the time difference between the first pulse emission time and the second pulse emission time.

And the updating subunit is used for updating the weight information of the neuron synapse based on the weight variation.

And determining the subunit, which is used for executing iteration for multiple times, and determining the convergence of the bionic olfactory neural network if the weight variation is zero.

In one embodiment, the apparatus further comprises: a storage module, wherein:

and the storage module is used for storing the network parameters of the various sample gases after the bionic olfactory nerve network converges.

The modules in the gas identification device can be wholly or partially realized by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent of a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.

In one embodiment, a computer device is provided, which may be a server, and the internal structure thereof may be as shown in fig. 16. The computer device includes a processor, a memory, an Input/Output interface (I/O for short), and a communication interface. The processor, the memory and the input/output interface are connected through a system bus, and the communication interface is connected to the system bus through the input/output interface. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The database of the computer device is used to store gas identification data. The input/output interface of the computer device is used for exchanging information between the processor and an external device. The communication interface of the computer device is used for connecting and communicating with an external terminal through a network. The computer program is executed by a processor to implement a gas identification method.

Those skilled in the art will appreciate that the architecture shown in fig. 16 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.

In one embodiment, a computer device is provided, comprising a memory and a processor, the memory having a computer program stored therein, the processor implementing the following steps when executing the computer program:

In one embodiment, the processor when executing the computer program further performs the steps of:

the block converts the gas specification digital signal to a pulse signal.

In one embodiment, the processor, when executing the computer program, further performs the steps of:

acquiring pulse signals of various sample gases;

if the weight variation is zero, an enabling signal is sent to a signal control unit connected with the excitation layer through an enabling control unit, and the bionic olfactory neural network is indicated to converge.

The implementation scheme for solving the problem provided by the processor in the computer device provided by the embodiment of the present application is similar to the implementation scheme described in the above method, so specific limitations in one or more embodiments of the computer device provided below may refer to the limitations on the gas identification method in the foregoing, and details are not described here again.

In one embodiment, a computer-readable storage medium is provided, having a computer program stored thereon, which when executed by a processor, performs the steps of:

In one embodiment, the computer program when executed by the processor further performs the steps of:

the gas specification digital signal is converted to a pulse signal.

acquiring pulse signals of various sample gases;

The implementation scheme for solving the problem provided by the processor in the computer-readable storage medium provided by the embodiments of the present application is similar to the implementation scheme described in the above method, so specific limitations in one or more of the computer-readable storage medium embodiments provided below may refer to the limitations in the above gas identification method, and details are not repeated here.

In one embodiment, a computer program product is provided, comprising a computer program which, when executed by a processor, performs the steps of:

the gas specification digital signal is converted to a pulse signal.

acquiring pulse signals of various sample gases;

The implementation of the solution to the problem provided by the processor in the computer program product provided in the embodiments of the present application is similar to the implementation described in the above method, so specific limitations in one or more of the computer program product embodiments provided below may refer to the limitations on the gas identification method in the foregoing, and details are not repeated here.

It should be noted that, the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) referred to in the present application are information and data authorized by the user or sufficiently authorized by each party, and the collection, use and processing of the related data need to comply with the relevant laws and regulations and standards of the relevant country and region.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above may be implemented by hardware instructions of a computer program, which may be stored in a non-volatile computer-readable storage medium, and when executed, may include the processes of the embodiments of the methods described above. Any reference to memory, databases, or other media used in the embodiments provided herein can include at least one of non-volatile and volatile memory. The nonvolatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, high-density embedded nonvolatile Memory, resistive Random Access Memory (ReRAM), magnetic Random Access Memory (MRAM), ferroelectric Random Access Memory (FRAM), phase Change Memory (PCM), graphene Memory, and the like. Volatile Memory can include Random Access Memory (RAM), external cache Memory, and the like. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM), among others. The databases referred to in various embodiments provided herein may include at least one of relational and non-relational databases. The non-relational database may include, but is not limited to, a block chain based distributed database, and the like. The processors referred to in the embodiments provided herein may be general purpose processors, central processing units, graphics processors, digital signal processors, programmable logic devices, quantum computing based data processing logic devices, etc., without limitation.

The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.

The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present application. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, and these are all within the scope of protection of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A method of gas identification, the method comprising:

each network parameter is generated when the gas types of various sample gases are learned in advance based on the bionic olfactory neural network.

2. The method of claim 1, wherein the biomimetic olfactory neural network comprises an excitation layer and a suppression layer, and the network parameters for each sample gas comprise a standard pulse firing time at the excitation layer and weight information for neuronal synapses between the excitation layer and the suppression layer;

the method comprises the following steps of identifying the gas type of the collected target gas based on the prestored network parameters of various sample gases to obtain the gas type identification result of the target gas, wherein the identification result comprises the following steps:

acquiring a target pulse emission time of the target gas at the excitant layer when the gas identification processor identifies the target gas by weight information of a neuron synapse corresponding to each sample gas;

3. The method of claim 2, further comprising:

converting the gas specification digital signal to a pulse signal.

4. A method according to any one of claims 1 to 3, characterized in that the method further comprises:

and controlling a gas identification processor to execute the process of identifying the gas type of the target gas through an asynchronous logic control circuit.

5. The method of claim 4, wherein the asynchronous logic control circuit comprises a signal control unit and a signal flip-flop; the gas identification processor comprises a plurality of different functional modules, and each functional module is connected with a signal control unit and a signal trigger;

the process of controlling the gas identification processor to identify the gas type of the target gas through the asynchronous logic control circuit comprises the following steps:

converting the starting signal of the current functional module into a control signal through the signal control unit;

and inputting the control signal into the current functional module through the signal trigger to indicate the current functional module to start executing a corresponding function.

6. The method according to any one of claims 1 to 3, wherein the generating of the network parameters of the plurality of gases comprises:

acquiring pulse signals of a plurality of sample gases;

7. The method of claim 6, wherein the network parameters include weight information of neuronal synapses between excitable and suppressive layers in the biomimetic olfactory neural network, and pulse emission times of each of the sample gases at the excitable layer;

updating the network parameters of the bionic olfactory neural network through each pulse signal until the bionic olfactory neural network converges, including:

for each sample gas, acquiring a first pulse emission time at the excitation layer and a second pulse emission time at the inhibition layer of a pulse signal of the sample gas;

and performing iteration for multiple times, and determining the convergence of the bionic olfactory neural network if the weight variation is zero.

8. The method of claim 7, further comprising:

and controlling the bionic olfactory neural network to execute a process of updating the network parameters of the bionic olfactory neural network through each pulse signal until the bionic olfactory neural network converges by an asynchronous logic control circuit.

9. The method of claim 8, wherein the asynchronous logic control circuit comprises a signal control unit and an enable control unit, and the determining that the bionic olfactory neural network converges if the weight variation is zero comprises:

and if the weight variation is zero, sending an enabling signal to a signal control unit connected with the excitation layer through the enabling control unit, and indicating the convergence of the bionic olfactory neural network.

10. The method of claim 6, further comprising:

and after the bionic olfactory nerve network converges, storing the network parameters of the multiple sample gases.

11. A gas identification device, the device comprising:

the gas identification processor is used for identifying the gas type of the collected target gas based on the prestored network parameters of various sample gases to obtain a gas type identification result of the target gas;

12. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor realizes the steps of the method of any one of claims 1 to 10 when executing the computer program.

13. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 10.

14. A computer program product comprising a computer program, characterized in that the computer program realizes the steps of the method of any one of claims 1 to 10 when executed by a processor.