Taste pathway calculation model optimization method based on human body taste nerve pathway

Abstract

The invention discloses a taste pathway calculation model optimization method based on a human body taste neural pathway, which comprises the following steps: S1, optimization of a taste pathway calculation model: comparing existing taste pathways according to inquired related information of a human taste pathway, and completing optimization of the taste pathway, a taste model topological structure and a support equation of the model; s2, verifying the performance of the optimized model: verifying some characteristics of the original model, improving the dynamic performance and calculating and verifying some new parameters capable of explaining the bionic property of the model; and S3, preparing a test sample and collecting taste data. According to the method provided by the invention, an existing taste pathway calculation model is effectively improved, auxiliary evaluation of artificial taste is realized, reliable technical support is provided, and the method is used for improving the application level of the taste recognition field and the food field.

Description

Technical Field

[0001] The present invention relates to taste evaluation technology, and in particular to a method for optimizing a taste pathway computational model based on a human taste nerve pathway. Background Art

[0002] At present, the electronic tongue is an advanced instrument that simulates human taste perception. It is designed to identify and evaluate taste substances by detecting and analyzing taste components in liquid samples. However, the original signal of the electronic tongue contains a lot of redundant information. If it is used directly for taste decision-making and evaluation, it will increase the difficulty of decision-making and reduce the evaluation effect. In order to solve this problem, some people use methods such as feature selection and feature dimensionality reduction to process the original data of the electronic tongue, and then use the processed data for further decision-making and evaluation, and obtain better results.

[0003] Although these processing methods can better handle the redundancy of the original data of the electronic tongue and complete some material identification tasks, they do not deeply study the taste mechanism of the electronic tongue. Even various artificial neural networks have not simulated the process of taste information transmission for human taste until the CMTP was proposed.

[0004] CMTP is inspired by Freeman's K series model and the olfactory pathway computational model, and is built in combination with the human taste information pathway. CMTP has good material recognition and taste evaluation capabilities, but the most important thing is its bionic performance. However, compared with the actual taste information transmission process of the human body, CMTP still has some shortcomings, and it cannot fully simulate the entire process of human taste transmission.

[0005] In view of the above problems, there is an urgent need for an optimization method to further improve CMTP and ensure that the model still has good or even better bionic performance while being more in line with the actual taste information transmission process of the human body. Summary of the invention

[0006] The main purpose of the present invention is to provide a method for optimizing the computational model of the taste pathway based on the human taste neural pathway. By studying and summarizing the process of human taste information transmission, the model is optimized. At the same time, combined with the characteristics of artificial networks, a new performance index is provided to illustrate the bionic performance of the network, and then the feasibility of optimization is explained. The optimized taste model is combined with the mature classification model to achieve the evaluation and classification of cigarette taste information.

[0007] The technical solution adopted by the present invention is: a method for optimizing a taste pathway computational model based on a human taste neural pathway, comprising:

[0008] S1. Optimization of the computational model of the taste pathway: Based on the relevant information of the human taste pathway, the existing taste pathways are compared to optimize the taste pathway, the topological structure of the taste model and the supporting equations of the model;

[0009] S2. Verification of the performance of the optimized model: including verification of some characteristics that the original model already possesses, improvement of dynamic performance, and calculation verification of some new parameters that can illustrate the bionics of the model;

[0010] S3. Preparation of test samples and collection of taste data.

[0011] Furthermore, the formula derivation process of the optimized taste pathway calculation model OCMTP is as follows:

[0012] The dynamic characteristic equation of the CN module can be described as shown in Formula 1. Among them, the variable l = 7, 9, 10, respectively represents the three types of neurons in the CN module, and the variable i takes values ​​from 1 to n, where n is the number of channels of the model W. rl and W ee represent the connection parameters between the lth nerve and the external input and the connection parameters between the same type of nerves, and W ee =0.198. i (t) represents the taste stimulus received by the i-th channel. P (t) represents the peripheral noise;

[0013] The dynamic characteristic equations of the IC part are shown in formula (2), formula (3) and formula (4);

[0014] Among them, M D 、M G and M A Represent the neural nodes of the three parts of IC; and They are the connection parameters between the three parts IC and k3 on the left; W Gkr =0.425 and They are the connection parameters between the three parts IC and k3 on the right. DG =1.357, W DA =1.241,W GD =1.357, W GA =0.375, W AD =1.241 and W AG =0.375 respectively represent the connection parameters between the various parts of the IC;

[0015]

[0016] Furthermore, the step S2 comprises:

[0017] Verify the rapid response capability of OCMTP and the chaotic characteristics between key nodes in the model;

[0018] The calculated clustering coefficient of OCMTP is 0.43, which is close to the clustering coefficient of 0.383 of the known primate cortical functional connectivity network;

[0019] Comparison of the 1 / f characteristics and node correlations of OCMTP and CMTP shows that OCMTP is superior.

[0020] Furthermore, in the step S3, the test sample preparation and taste data collection include:

[0021] Mainstream cigarette smoke was inhaled through a homemade simple smoking machine and filtered through a Cambridge filter to obtain a filter containing cigarette particle information;

[0022] Place the obtained filter disc in a beaker, pour 100 ml of pure water into the beaker and seal it with plastic wrap. Then place the sealed beaker on a heating table and heat it at 37.5°C for 1 hour.

[0023] Shake the heated beaker at a constant speed for 1 minute while keeping it sealed, so that the cigarette substances in the filter are evenly dissolved in the pure water;

[0024] The obtained solution containing the taste information of cigarettes is put into an electronic tongue machine to obtain the taste information.

[0025] Furthermore, in the step S3, the test sample preparation includes the following steps:

[0026] During the non-experimental period, the cigarette samples were placed in sealed bags and the air in the sealed bags was evacuated to ensure that the cigarette samples would not be affected by components such as water vapor in the air;

[0027] When using a smoking machine to extract mainstream cigarette smoke, an electronic nose system can be connected behind the filter to obtain information about the gas phase of the cigarette, which is also convenient for subsequent processing;

[0028] The mainstream smoke of the cigarette sample from the smoking machine was inhaled four times, each time for 6 seconds. After each inhalation, the electronic nose sensor and the air chamber were given enough time to recover.

[0029] Furthermore, in the step S3, the acquisition of taste information includes: the room temperature is maintained at 22±0.5°C and the humidity is 62±1.5%RH during the experiment, and the experiment is performed using the reference solution and other solutions required by the electronic tongue that have been prepared in advance. The heated and soaked solution is placed in the electronic tongue device, and the sampling number is set to 5.

[0030] Advantages of the present invention:

[0031] High bionics: By combining the actual pathways of human taste to further improve the taste model, the present invention can make the existing taste model closer to the actual situation of the human body, and then present analysis results that are more in line with human taste.

[0032] High objectivity and reliability: By introducing the clustering coefficient parameter in the artificial network to describe the compactness of vertices in a graph, the clustering coefficient of the network is obtained using the "coarse details" in the lowest resolution subspace, and compared with the clustering coefficient of the known primate cerebral cortex functional connection network, the bionic results are more objective and reliable.

[0033] The experimental process can be handled flexibly: the cigarette gas filtered by the Cambridge filter can be further transported to the gas phase electronic nose for detection of cigarette gas substances, and then the cigarette particle information measured by the particle phase electronic nose and the cigarette taste information measured by the electronic tongue can be combined to jointly process and analyze the cigarette flavor information.

[0034] Assisting artificial taste evaluation: The optimized taste model combined with the electronic tongue can better handle the redundancy of the original data of the electronic tongue. At the same time, it simulates the process of taste information transmission for human taste, making it have better bionic performance. Therefore, it can effectively assist the traditional artificial cigarette taste evaluation and avoid the errors caused by human taste fatigue.

[0035] Wide application: The perfect model system of this invention can not only be used for the taste evaluation of cigarettes, but can also be extended to other food or traditional Chinese medicine fields to assist in the classification of sample quality, etc., and has broad application prospects.

[0036] The method proposed in the present invention effectively improves the existing taste pathway calculation model, realizes the auxiliary evaluation of artificial taste, and provides reliable technical support for improving the application level in the field of taste recognition and food.

[0037] In addition to the above-described purposes, features and advantages, the present invention has other purposes, features and advantages. The present invention will be further described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The drawings constituting a part of this application are used to provide a further understanding of the present invention. The illustrative embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0039] Figure 1 This is the improved transmission pathway of human taste information of the present invention, which is demonstrated by taking a unilateral pathway as an example;

[0040] Figure 2It is the taste potential lateralization problem pathway structure of the present invention;

[0041] Figure 3 is the topological structure of the optimized taste pathway computational model of the present invention;

[0042] Figure 4 is a node response diagram of the optimized taste pathway computational model of the present invention under no stimulation;

[0043] Figure 5 It is a node response diagram of the optimized taste pathway computational model of the present invention under stimulation;

[0044] Figure 6 is an output phase diagram of key nodes of the optimized taste pathway computational model of the present invention;

[0045] Figure 7 is a box plot of the slope of the function fitted by the frequency characteristics of the key nodes in the optimized taste pathway computational model of the present invention;

[0046] Figure 8 is a mutual information entropy heat map between key nodes in the optimized taste pathway computational model of the present invention;

[0047] Fig. 9 It is a heat map of mutual information entropy between key nodes in the original model of the present invention, and a difference map of mutual information entropy between key nodes in the taste pathway calculation model before and after optimization;

[0048] Fig.10 It is a technical roadmap of the present invention for using the optimized taste pathway calculation model to perform cigarette taste identification. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0050] refer to Figures 1 to 10 and Tables 1 to 3, the main purpose of the present invention is to provide a method for optimizing the taste pathway calculation model. Through this method, a more complete taste pathway model is obtained to achieve taste recognition and evaluation of cigarettes, thereby assisting traditional manual taste evaluation and improving the objectivity and reliability of the evaluation.

[0051] The present invention provides a method for optimizing a taste pathway computational model based on a human taste neural pathway, comprising:

[0052] S1. Optimization of the computational model of the taste pathway: Based on the relevant information of the human taste pathway, the existing taste pathways are compared to optimize the taste pathway, the topological structure of the taste model and the supporting equations of the model;

[0053] S2. Verification of the performance of the optimized model: including verification of some characteristics that the original model already possesses, improvement of dynamic performance, and calculation verification of some new parameters that can illustrate the bionics of the model;

[0054] S3. Preparation of test samples and collection of taste data.

[0055] The formula derivation process of the optimized taste pathway computational model OCMTP (Optimized Computational Model of Taste Pathways) is as follows:

[0056] The dynamic characteristic equation of the CN module can be described as shown in Formula 1. Among them, the variable l = 7, 9, 10, respectively represents the three types of neurons in the CN module, and the variable i takes values ​​from 1 to n, where n is the number of channels of the model W. rl and W ee represent the connection parameters between the lth nerve and the external input and the connection parameters between the same type of nerves, and W ee =0.198. i (t) represents the taste stimulus received by the i-th channel. P (t) represents the peripheral noise;

[0057] The dynamic characteristic equations of the IC part are shown in formula (2), formula (3) and formula (4) (taking the left side as an example). D 、M G and M A Represent the neural nodes of the three parts of IC; and They are the connection parameters between the three parts IC and k3 on the left; W Gkr =0.425 and They are the connection parameters between the three parts IC and k3 on the right. DG =1.357, W DA =1.241,W GD =1.357, W GA =0.375, W AD =1.241 and W AG =0.375 respectively represent the connection parameters between the various parts of the IC;

[0058]

[0059] The step S2 comprises:

[0060] Verify the rapid response capability of OCMTP and the chaotic characteristics between key nodes in the model;

[0061] The calculated clustering coefficient of OCMTP is 0.43, which is close to the clustering coefficient of 0.383 of the known primate cortical functional connectivity network;

[0062] Comparison of the 1 / f characteristics and node correlations of OCMTP and CMTP shows that OCMTP is superior.

[0063] In the step S3, the test sample preparation and taste data collection include:

[0064] Mainstream cigarette smoke was inhaled through a homemade simple smoking machine and filtered through a Cambridge filter to obtain a filter containing cigarette particle information;

[0065] Place the obtained filter disc in a beaker, pour 100 ml of pure water into the beaker and seal it with plastic wrap. Then place the sealed beaker on a heating table and heat it at 37.5°C for 1 hour.

[0066] Shake the heated beaker at a constant speed for 1 minute while keeping it sealed, so that the cigarette substances in the filter are evenly dissolved in the pure water;

[0067] The obtained solution containing the taste information of cigarettes was placed into the SA-402B electronic tongue machine from Insent Company to obtain the taste information.

[0068] In the step S3, the test sample preparation includes the following steps:

[0069] During the non-experimental period, the cigarette samples were placed in sealed bags and the air in the sealed bags was evacuated to ensure that the cigarette samples would not be affected by components such as water vapor in the air;

[0070] When using a smoking machine to extract mainstream cigarette smoke, an electronic nose system can be connected behind the filter to obtain information about the gas phase of the cigarette, which is also convenient for subsequent processing;

[0071] Due to the addition of the electronic nose, the cigarette samples of the smoking machine need to inhale the mainstream smoke four times, each time for 6 seconds. After each inhalation, the electronic nose sensor and air chamber are given enough time to recover.

[0072] In the step S3, the acquisition of taste information includes: the room temperature is maintained at 22±0.5°C and the humidity is 62±1.5%RH during the experiment, and the electronic tongue required solution and the reference solution prepared in advance are used for the experiment. The heated and soaked solution is placed in the electronic tongue device, and the sampling number is set to 5 to prevent the oxidation of the cigarette taste solution caused by the long sampling time of a single sample.

[0073] Embodiments of the present invention:

[0074] S1. Optimization of taste pathway and topological structure

[0075] The present invention relates to the actual transmission of human taste information. Studies have shown that the transmission of human taste information from the tongue to the brain is completed bilaterally. Figure 1 Shown is the process of taste information transmission using the left side as an example.

[0076] Mammals, including rodents and primates, have generally the same taste information transmission process, except that the information transmission of primates does not pass through the parabrachial nucleus (Pbn). Specifically, the taste information transmission of primates can be summarized as follows: the taste receptors (taste buds, etc.) in the mouth receive taste stimuli from the outside world and generate certain stimulation information. These taste stimulation information is transmitted to the NST via the facial nerve, glossopharyngeal nerve, and vagus nerve. As the location of the second-order neurons in the taste pathway, the NST is responsible for further processing the taste stimulation signals from the taste receptors and transmitting the processed information to the ipsilateral VPMpc. As the third-order neurons in the taste nerve pathway, the VPMpc neurons are important inputs for taste production. They transmit the processed taste signals to the insular cortex to produce taste sensations corresponding to the stimulation. The insular cortex can be divided into three parts, namely the granular (gIC), the dysgranular (dIC) and the agranular (aIC). In most cases, the areas that process taste are gIC and dIC, and in a few cases, aIC. Among gIC and dIC, gIC is more taste selective. Therefore, gIC and dIC generally receive information from VPMpc and integrate taste information in IC. At the same time, gIC outputs feedback signals to VPMpc and NST to achieve feedback regulation of taste stimulus information. The feedback connection between IC and VPMpc has been confirmed by horseradish peroxidase tracing, and the feedback connection between IC and NST has also been confirmed from the anatomical structure of mammals.

[0077] Human taste information is transmitted bilaterally. The main pathways on both sides are roughly the same. After the taste receptors in the mouth receive the corresponding stimulation, they are transmitted to the Nucleus of the Solitary Tract (NST) via the facial nerve (CNVII), glossopharyngeal nerve (CNIX) and vagus nerve (CNX). Then the NST transmits the processed information to the parvicellular portion of the Ventroposteromedial Nucleus (VPMpc) and the insular cortex (IC). The IC can be subdivided into three interconnected areas, namely the granular (gIC), the dysgranular (dIC) and the agranular (aIC). In most cases, the gIC and dIC are the areas that process taste, and in a few cases, the aIC is used. At the same time, there is a bidirectional transmission between the VPMpc and the IC, and the IC also feeds back the corresponding taste information to the NST. The connection between the two sides of taste is completed by the transmission from the left VPMpc to the right IC.

[0078] In the step S1, the existing taste pathway is optimized and improved according to the relevant information of human taste information transmission.

[0079] Figure 2 In the past, at least five different bilateral structures of taste have been proposed for the bilateral pathway model of taste. However, two technologies, magnetoencephalography (MEG) and functional magnetic resonance imaging (fMRI), have proven that the two bilateral structures of taste are the same. Figure 2 The correctness of the concept shown in this figure is compared with other concepts of bilateral taste perception. It can be roughly described as follows: For bilateral taste perception, oral stimulation information is transmitted to NST via the three major CN nerves, and then transmitted to VPMpc by NST. VPMpc then transmits information to IC via nerve fibers, and IC also feeds back the generated taste information to VPMpc and NST. The left and right neural pathways are mainly connected through the transmission from VPMpc in the right pathway to IC in the left pathway.

[0080] Figure 3It is the topological structure of the computational model of the taste pathway after optimization according to the optimization scheme. It is worth noting that compared with the topological structure of the original model: first, the structure of bilateral transmission of taste is modified, that is, the unilateral pathway of the original model topological structure is expanded to a bilateral pathway, and the actual connection between the bilateral pathways is guaranteed; secondly, it is believed that after the taste receptors such as taste buds on the surface of the human tongue receive taste stimulation, when CNVII, CNIX and CNX transmit it to NST, each nerve that transmits taste bud information eventually terminates in NST together, so in OCMTP, various nerves are "interconnected"; thirdly, since there is no clear evidence that there is a relationship between the glossopharyngeal nerve and the vagus nerve, the connection pathway between the two is removed in the optimized topological structure, and similarly, the pathway from NST to IC is also removed; finally, according to the subdivision at IC, the IC in the original topological structure is expanded into three parts.

[0081] S2. Improvement of the dynamic characteristic equation of the taste pathway computational model

[0082] In step S2, the rapid response capability of OCMTP and the chaotic characteristics between key nodes in the model are first verified. Both of these characteristics are possessed by CMTP, and both are consistent with some characteristics of the human brain to a certain extent, which is consistent with the characteristics of bionics. Secondly, the clustering coefficient of OCMTP is calculated to be 0.43, which is close to the clustering coefficient of the known primate cerebral cortex functional connection network of 0.383, which further proves that this network is highly bionic. Finally, the comparison of the 1 / f characteristics and node correlation of OCMTP and CMTP shows that OCMTP is better.

[0083] Compared with the topological structure of the original model, the topological structure of the optimized taste pathway computational model not only adds bilateral structures and refines the IC structure, but also enriches the afferent structures of the three major nerves. We believe that after the taste receptors such as taste buds on the surface of the human tongue receive taste stimuli, they are transmitted to the NST by CNVII, CNIX and CNX. Each nerve that transmits taste bud information eventually terminates in the NST. Therefore, in OCMTP, the various nerves are "interconnected" and the information transmission from NST to IC in the original model is removed.

[0084] The dynamic characteristic equation structure of CMTP can be summarized as formula (5), which is a second-order differential equation. Where a and b are constants, and a = 0.220, b = 1.000. Unless otherwise specified, the meanings of a and b in all equations below are the same. x and y represent the nodes of CMTP. x″, x′ and x represent the derivatives and initial values ​​of the potential state of the x node, respectively. W xyrepresents the connection parameter between nodes x and y. t represents the running time. Q(y(t)) represents the output of node y connected to node x. I(t) represents the external input of node x at time t. W xI represents the connection parameter between x and I(t). However, since the topological structure of the taste model has been optimized, some of its dynamic characteristic equations also need to be modified accordingly.

[0085]

[0086] First, in OCMTP, we “interconnect” various types of afferent nerves, so the dynamic characteristic equation of the CN module can be described as shown in formula (1). Among them, the variable l = 7, 9, 10, respectively represents the three types of nerves in the CN module, the variable i takes values ​​from 1 to n, n is the number of channels in the model, and the meanings of i and n in the equations of all modules are the same. ki W represents the i-th and l-th neural nodes. rl and W ee represent the connection parameters between the lth nerve and the external input and the connection parameters between the same type of nerves, and W ee =0.198. i (t) represents the taste stimulus received by the i-th channel. P (t) represents peripheral noise, which is simulated using a Gaussian distribution random number with a positive mean, whose mean can be set to 0.10 and variance can be set to 0.70. In addition, there is central noise in the entire model, which is a Gaussian distribution with a mean of 0 and a variance of 0.07.

[0087]

[0088] Secondly, we further subdivide the IC, so the dynamic characteristic equations of the IC part are shown in formula (2), formula (3) and formula (4) (taking the left side as an example). D 、M G and M A Represent the neural nodes of the three parts of IC respectively. and These are the connection parameters between the three parts of IC and k3 on the left. W Gkr =0.425 and They are the connection parameters between the three parts IC and k3 on the right. DG =1.357, W DA =1.241,W GD =1.357, W GA =0.375, W AD =1.241 and W AG =0.375 represents the connection parameters between various parts of the IC.

[0089]

[0090] Since the optimization process takes into account the bilaterality of the taste pathway, in addition to improving some dynamic characteristic equations, some parts need to be added, and the dynamic characteristic equation at the IC also needs to be further improved according to the bilaterality theory. For example, the IC part on the right does not have and

[0091] S3. Validation of the optimized taste pathway computational model

[0092] First, verify the fast response capability of the optimized model. Because the two neural structures, NST and VPMpc, play an important role in the formation of human taste and the transmission and feedback of taste signals in the brain, this section simulates and verifies the two key parts of the model, NST and VMPpc. During the simulation, the total time period is selected as 0-1100ms, the time step is set to 1ms, and the model is given pulse square wave stimulation in the time period of 201-800ms. The rest of the time period is the rest time or recovery time of the model. Figure 4 is the J of the first channel of the left NST when there is no external stimulation. 1l , K 1l , J of the right VPMpc 3r and K 3r The response curves of the six nodes, gM nodes of the left and right ICs. It can be seen that when there is no external stimulation, the response value of each node fluctuates around zero, that is, each node can be regarded as being in a resting state. When stimulated by a certain pulse, the response curve of the node changes. A simple description is: in 0-200ms, due to the absence of external stimulation, each node is still in a resting state; Figure 5 , when the stimulus is received at 201ms, each node responds quickly, and the change time is extremely short. This phenomenon is the fast response characteristic of the node, which is an existing characteristic of the original model and is consistent with the rapid response phenomenon of the cerebral cortex after receiving sudden stimulation.

[0093] Secondly, it shows that the phase diagram of the model node after optimization still has chaotic characteristics. Various parameters remain unchanged, and a 2V pulse square wave is input from 201 to 800ms. The phase diagrams between the left J1 and J2, K1 and K2, J1 and K1, J3 nodes and the right J1 and K1, K1 and K3, K1 and K2, K3 outputs are shown in the figure below. Figure 6 As shown in the figure, the phase diagram between each node shows a regular limit cycle. Even the phase diagram between three nodes conforms to this rule, which shows that under the stimulation of the pulse square wave, the two modules NST and VPMpc have chaotic stability, and this property is a common feature of human brain activity.

[0094] Afterwards, the high biomimetic property of the optimized model is further illustrated by calculating the clustering coefficient of the optimized model. The clustering coefficient is an indicator used in graph theory to measure the concentration of vertices in a graph. It is particularly used in network analysis to describe the closeness of vertices in a graph. The steps for calculating the clustering coefficient of the model can be described as follows: First, simplify the OCMTP network into a directed unweighted network and obtain the proximity matrix of this network. The matrix acquisition rule is: if there is a directional connection from node i to node j, then in the proximity matrix a ij Take 1, otherwise take 0. Subsequently, the obtained proximity matrix is ​​subjected to multi-scale wavelet decomposition, and then the "coarse and fine details" in the lowest resolution subspace after decomposition can be used to characterize the clustering coefficient of the network, that is, the difference between the smallest diagonal element and the largest off-diagonal element in the subspace is used to characterize the clustering coefficient. Finally, the proximity matrix is ​​decomposed into 4 layers, and it is calculated that when the number of channels of OCMTP is 10, that is, the number of nodes is 173, its clustering coefficient is 0.43, which is close to 0.383 of the functional connection network of the primate cerebral cortex. The closeness of the clustering coefficient of the optimized model to the clustering coefficient of the known brain network just illustrates the bionic nature of the model.

[0095] Then, the 1 / f characteristics of the model before and after optimization are compared to illustrate the feasibility of the present invention. The power spectral density (PSD) and frequency of 1 / f noise satisfy a proportional relationship under double logarithmic coordinates. Therefore, under double logarithmic coordinates, the obtained PSD and frequency curves can be linearly fitted. The closer the absolute value of the slope of the curve fitting is to 1, the better the release effect. The specific fitting calculation process can be described as follows: first, only a 2V pulse stimulation is input at 201-800ms, and the response values ​​of the J1 and K1 nodes in the bilateral NST are obtained. Then, the PSD characteristics of the node output under stimulation are described by the Welch method of the periodogram, and the specific parameters are set as follows: the sampling frequency is 58Hz, a Hamming window is used, and its width is 64, the number of overlapping points of adjacent intercepted segments is 32, and the number of sampling points of the fast Fourier transform is 128. Finally, the method of constructing a fitting function is used to determine the PSD characteristics of the node output. The PSD output by the node and f are fitted in the logarithmic coordinate system. The constructed fitting function is in the form of y=k*x+b, and y is 10*log 10 PSD, k is the slope, x is the frequency, and b is the intercept. The calculation results of the absolute value of k at each node on the left side of the optimized model are as follows: Figure 7As shown in the figure, 20 calculation results are selected to eliminate the error. It can be seen from the figure that the absolute value of k of the nodes of the optimized model is close to 1. At the same time, the absolute value of k of each node of the original model is calculated and subtracted from 1 to obtain Table 1. The results show that whether it is the J1 node or the K1 node, the difference between the absolute value of k of the optimized model and 1 is smaller, that is, the 1 / f characteristic of the optimized model is more obvious and the bionic performance is stronger.

[0096] Finally, by comparing the mutual information entropy between the nodes of the model before and after optimization, it can be shown that the correlation between the model nodes has been improved. When the value of the mutual information entropy is 0, it means that the two variables are independent of each other; when the value is greater than 0, it means that there is a connection between the two variables; and the maximum value of the mutual information entropy will not exceed 1, so the closer its value is to 1, the stronger the dependence of the two variables. The specific steps can be described as follows: First, within 201-800ms, a 2V pulse square wave is input into the optimized model, and the node outputs of J1, J2, K1, K2, J3, K3 and gM on both sides of the model are obtained. The size of each node output is (1, 600), where 600 is the length of the node output within 201-800ms. Then the mutual information entropy between each point is calculated based on the output information obtained, such as Figure 8 As shown in the figure, the mutual information entropy between the output nodes of the optimized model ranges from (0.8564, 0.8804) to (0.8537, 0.8817), indicating that the nodes of the optimized model have high synchronization. The mutual information entropy between the corresponding nodes of the original model is calculated in the same way, and the difference is made with the optimized one, as shown in Fig. 9 As shown, after comparison, the mutual information entropy range of the optimized model is closer to 1, and each difference is greater than 0, indicating that the synchronization of the optimized model has been improved, indicating the feasibility of this invention.

[0097] S4. Comparison of cigarette taste recognition using the taste models before and after optimization:

[0098] like Fig.10As shown in the figure, it is the entire technical flow chart of cigarette taste recognition. First, a solution containing cigarette taste information needs to be obtained. The details of the selected cigarettes are shown in Table 2. The process of obtaining the solution containing cigarette information can be described as follows: First, obtain the Cambridge filter containing cigarette particle information. Before and after the experiment, the selected cigarette samples need to be placed in a sealed bag and the air in the bag needs to be extracted to prevent the cigarette from getting damp. In each experiment, a cigarette is taken out and put into the smoking machine. A new unused Cambridge filter is placed behind the smoking machine to filter the particle information in the mainstream cigarette smoke. At the same time, an air item electronic nose can also be connected to the air chamber of the smoking machine to further utilize the smoke information generated in the experiment. A cigarette can be inhaled in 4 puffs, and a filter containing cigarette particle information is obtained together. After that, the filter is placed in a beaker with pure water and the beaker is sealed. After that, the beaker is heated to a constant temperature of about 37.5℃ to simulate the temperature of the human mouth for one hour. Then, shake the sealed beaker for one minute to evenly blend the substance in the filter into the solution. Finally, cool the shaken cigarette solution to room temperature.

[0099] After that, the taste information of cigarettes can be collected. The SA-402B electronic tongue designed by Insent was used for collection, and the room temperature was maintained at 22±0.5℃ and the humidity was 62±1.5%RH during the collection process. Before the experiment, the corresponding solutions required for the electronic tongue, such as positive and negative electrode immersion solutions, reference solutions, etc., need to be configured. When collecting, the number of collections can be set to 4 times. This operation is to prevent the solution containing the taste information of cigarettes from being exposed to the air for too long, which may cause errors in the collected taste information. The collected data set contains five different brands and types of cigarettes. Each cigarette was tested 40 times, resulting in a total of 200 groups of samples.

[0100] Finally, the obtained data set and the optimized taste model and classifier can be used to realize the taste recognition and classification of cigarettes. The process can be described as follows: First, the obtained data set is input into the CN module of the optimized model at 201-800ms. Because the data set is (200, 12), 200 represents the number of samples, and 12 represents the taste information obtained by the electronic tongue, which is sour, sweet, bitter, salty, fresh, astringent and the aftertaste information of these six flavors, so the channel of the taste model is selected as 12. After that, the model is run. Because NST is the secondary neuron in the taste conduction pathway and is connected to other taste structures, the output of the J1 node in the NST module can be used to identify the substance of the taste model, and the output data size is (200, 600), where 600 is the output length of each channel from 201-800ms. Then, the output of the J1 node of each channel is analyzed by the standard deviation (SD) method, and the formula can be expressed as:

[0101]

[0102] in It represents the average SD of the output of the J1 node of the i-th channel, that is, the average SD is obtained for the number of channels. S represents the number of segments into which the 201-800ms time period is evenly divided. In this verification process, S can be 60, that is, 201-800ms is evenly divided into 60 segments. SD ij It represents the SD of the jth period of time under the i-th channel.

[0103] The data of size (200, 600) obtained from the taste model are processed according to the above SD method to obtain a new data set of size (200, 12), where 12 represents the total number of average SDs output by the 12-channel J1 node. Then, the 200 sets of data are divided into 140 training sets and 60 test sets by the KS algorithm, and the training data set is used for model training of the PSO-SVM model, while the test data set is used for substance recognition by the trained PSO-SVM, where the number of particles and the number of iterations in the PSO-SVM are set to 20 and 200 respectively, and the search ranges of c and g are set to (0.1, 100) and (0.01, 1000) respectively.

[0104] Finally, according to the above steps, the same cigarette taste data is transmitted to the original taste model, and the recognition and classification are completed. The cigarette taste recognition results of the taste model before and after optimization are shown in Table 3. It can be seen that the cigarette recognition results of the optimized taste model are better. At the same time, in order to illustrate that the high accuracy after optimization is not simply caused by the large number of features, the "left side" data of the model is selected for classification and comparison with the original model. The recognition process is consistent with the above, and the obtained cigarette taste recognition results are shown in Table 3. The results also show that the optimized model has a better classification effect.

[0105] Table 1 Details of cigarette samples

[0106]

[0107] Table 2 Comparison of the difference between the absolute value of k and 1 before and after optimization

[0108]

[0109] Table 3 Cigarette taste recognition results

[0110]

[0111] Table 1 is a table showing the difference between the slope of the function fitted by the frequency characteristics of the key nodes in the optimized taste pathway calculation model of the present invention and the corresponding slope of the original taste pathway calculation model.

[0112] Table 2 is a detailed list of cigarette samples used in the cigarette taste recognition experiment in the optimized taste pathway calculation model of the present invention.

[0113] Table 3 is a comparison table of the cigarette taste recognition results in the optimized taste pathway calculation model of the present invention and the cigarette taste recognition results in the original taste pathway calculation model.

[0114] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for optimizing a taste pathway computational model based on a human taste neural pathway, characterized in that: include: S1. Optimization of the computational model of the taste pathway: Based on the relevant information of the human taste pathway, the existing taste pathways are compared to optimize the taste pathway, the topological structure of the taste model and the supporting equations of the model; S2. Verification of the performance of the optimized model: including verification of some characteristics that the original model already possesses, improvement of dynamic performance, and calculation verification of some new parameters that can illustrate the bionics of the model; S3. Preparation of test samples and collection of taste data.

2. The method for optimizing the taste pathway computational model based on the human taste neural pathway according to claim 1, characterized in that: The formula derivation process of the optimized taste pathway calculation model OCMTP is as follows: The dynamic characteristic equation of the CN module can be described as follows: Formula (1); where variable l = 7, 9, 10, representing the three types of neurons in the CN module, variable i takes values ​​from 1 to n, and n is the number of channels in the model W. rl and W ee represent the connection parameters between the lth nerve and the external input and the connection parameters between similar nerves, respectively, and W ee =0.198; R i (t) represents the taste stimulation received by the i-th channel; N P (t) represents the peripheral noise; The dynamic characteristic equations of the IC part are as follows: Formula (2), Formula (3) and Formula (4); where M D 、M G and M A Represent the neural nodes of the three parts of IC; and They are the connection parameters between the three parts IC and k3 on the left; W Gkr =0.425 and They are the connection parameters between the three parts IC and k3 on the right; W DG =1.357, W DA =1.241,W GD =1.357, W GA =0.375, W AD =1.241 and W AG =0.375 respectively represent the connection parameters between the various parts of the IC; 3. The method for optimizing the taste pathway computational model based on the human taste neural pathway according to claim 1, characterized in that: The step S2 comprises: Verify the rapid response capability of OCMTP and the chaotic characteristics between key nodes in the model; The calculated clustering coefficient of OCMTP is 0.43, which is close to the clustering coefficient of 0.383 of the known primate cortical functional connectivity network; Comparison of the 1 / f characteristics and node correlations of OCMTP and CMTP shows that OCMTP is superior.

4. The method for optimizing the taste pathway computational model based on the human taste neural pathway according to claim 1, characterized in that: In the step S3, the test sample preparation and taste data collection include: Mainstream cigarette smoke was inhaled through a homemade simple smoking machine and filtered through a Cambridge filter to obtain a filter containing cigarette particle information; The obtained filter disc was placed in a beaker, and 100 ml of purified water was poured into the beaker and sealed with plastic wrap. The sealed beaker was then placed on a heating table for heating at 37.5°C for 1 hour. Shake the heated beaker at a constant speed for 1 minute while keeping it sealed, so that the cigarette substances in the filter are evenly dissolved in the pure water; The obtained solution containing the taste information of cigarettes is put into an electronic tongue machine to obtain the taste information.

5. The method for optimizing the taste pathway computational model based on the human taste neural pathway according to claim 4, characterized in that: In the step S3, the test sample preparation includes the following steps: During the non-experimental period, the cigarette samples were placed in sealed bags and the air in the sealed bags was removed; When using a smoking machine to extract the mainstream smoke of a cigarette, an electronic nose system is connected behind the filter to obtain the gas phase information of the cigarette; The mainstream smoke of the cigarette sample from the smoking machine was inhaled four times, each time for 6 seconds. After each inhalation, the electronic nose sensor and the air chamber were given enough time to recover.

6. The method for optimizing the taste pathway computational model based on the human taste neural pathway according to claim 4, characterized in that: In the step S3, the acquisition of taste information includes: the room temperature is maintained at 22±0.5°C and the humidity is 62±1.5%RH during the experiment, the experiment is carried out using a pre-configured reference solution and other solutions required for the electronic tongue, the heated and soaked solution is placed in the electronic tongue device, and the sampling number is set to 5.

Images