A method for epilepsy EEG recognition based on olfactory-hippocampal bionic model

Abstract

The present invention discloses an epilepsy electroencephalogram (EEG) recognition method based on an olfactory-hippocampal bionic model. The method comprises the following steps: obtaining an olfactory-hippocampal bionic model; obtaining a KII model, performing multi-channel input modeling on the anterior olfactory nucleus layer and the piriform cortex layer according to the neural cluster theory, and establishing connections between neurons in the olfactory bulb layer and the entorhinal cortex layer; replacing the obtained multi-channel input model and the connections between the neurons into the obtained olfactory-hippocampal bionic model, and modifying the neuron model in the olfactory-hippocampal bionic model to complete the optimization of the olfactory-hippocampal bionic model; training the optimized model and recognizing an epilepsy EEG dataset; and based on the neural cluster theory, improving the olfactory part of the obtained olfactory-hippocampal bionic model, improving the anterior olfactory nucleus layer and the piriform cortex layer to multi-channel input, and increasing the projection and feedback input between the corresponding neurons, thereby increasing the stability of the model and making it closer to a real neural network.

Description

Technical Field

[0001] The present invention relates to the technical field of neural network bionic models, and in particular to an epileptic EEG recognition method based on an olfactory-hippocampal bionic model. Background Art

[0002] Research on the olfactory nervous system is mainly divided into three aspects: macroscopic, microscopic and functional. The macroscopic level mainly focuses on the study of the olfactory nervous system model, the microscopic level mainly focuses on the study of related mechanisms, and the functional level mainly focuses on the relationship between odor information processed by the sense of smell and higher-level neural activities. The hippocampus is the brain area directly connected to the sense of smell, and its relative position is clear. In the research related to brain functions such as learning and memory, the hippocampus has always been a hot spot in brain research. Therefore, studying the bionic model of the olfactory-hippocampal neural network will help the study of the neural networks of other sensory nervous systems and further explore their working mechanisms.

[0003] However, the research on olfaction and hippocampus faces the following difficulties: (1) In neuroscience experiments, the experimental conditions are demanding, the experimental costs are high, and the operation is difficult, which makes it difficult to promote neurophysiological experiments on a large scale; (2) The research from sensory signals such as olfaction to specific brain regions involves higher-level brain functions, with many influencing factors and complex experimental designs, and there are few studies available for reference. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention provides an epilepsy EEG recognition method based on the olfactory-hippocampal bionic model, performs multi-channel input modeling on the anterior olfactory nucleus layer and the piriform cortex layer in the obtained olfactory-hippocampal bionic model, and re-establishes the connections between neurons in each layer to complete the optimization of the olfactory-hippocampal neural network model.

[0005] To solve the above technical problems, the present invention provides a technical solution: a method for identifying epilepsy EEG based on an olfactory-hippocampal bionic model, characterized by:

[0006] Step 1: Obtain the olfactory-hippocampal bionic model and dynamic equations;

[0007] Step 2: Obtain the KII model. Based on the neural cluster theory, model the multi-channel input of the anterior olfactory nucleus and the piriform cortex, and establish the connections between the neurons in the olfactory bulb and the entorhinal cortex.

[0008] The process of multi-channel input modeling of the anterior olfactory nucleus layer and the piriform cortex layer is as follows: the anterior olfactory nucleus layer and the piriform cortex layer are both represented by multiple KII models connected in parallel, and the number of parallel input channels of the olfactory bulb layer, the anterior olfactory nucleus layer and the piriform cortex layer is the same, and the data output by the olfactory bulb layer is classified using the proximity principle;

[0009] Step 3: replacing the obtained multi-channel input model and the connections between the neurons into the obtained olfactory-hippocampal bionic model, and modifying the neuron model in the olfactory-hippocampal bionic model to complete the optimization of the olfactory-hippocampal bionic model and update the dynamic equation;

[0010] Step 4: After training the optimized olfactory-hippocampal bionic model, the epilepsy EEG dataset is identified.

[0011] Furthermore, in step one, the olfactory-hippocampal bionic model includes the olfactory bulb layer, anterior olfactory nucleus layer, piriform cortex layer, entorhinal cortex layer, dentate gyrus layer, CA3 layer, CA1 layer and subiculum layer simulated by the K series model, with the entorhinal cortex layer as the core and through the connection between neurons in each layer, the bionic simulation between the sense of smell and the hippocampus is realized.

[0012] Furthermore, the K series models are K0, KI, KII, and KIII models constructed by Professor Freeman based on the olfactory nervous system of mammals and on the basis of a large number of neurophysiological experiments.

[0013] Furthermore, in step 2, the connections between the neurons are as follows: the anterior olfactory nucleus layer and the piriform cortex layer receive projections from the olfactory bulb layer, the entorhinal cortex layer receives projections from the piriform cortex layer, and the olfactory bulb layer projects to the entorhinal cortex layer; the anterior olfactory nucleus layer provides feedback to the olfactory bulb layer, the piriform cortex layer provides feedback to the olfactory bulb layer, the entorhinal cortex layer provides delayed feedback to the piriform cortex layer, the piriform cortex provides delayed feedback to the anterior olfactory nucleus layer, and the entorhinal cortex provides feedback input to the olfactory bulb layer.

[0014] Furthermore, in step three, the neuron model in the olfactory-hippocampal bionic model adopts spiking neurons.

[0015] Furthermore, the olfactory-hippocampal bionic model uses the Hebbian learning rule and adaptive learning rule to update the connection weights between neurons.

[0016] Furthermore, in step 3, the updated kinetic equation is:

[0017] (1) Olfactory bulb

[0018]

[0019] (2) Anterior olfactory nucleus

[0020]

[0021] (3) Piriform cortex

[0022]

[0023] (4) Entorhinal cortex

[0024]

[0025] (5) Feedback

[0026]

[0027] In formulas (1)-(5), i=1, 2, ..., n represents the number of parallel input channels; represents the peripheral noise signal introduced by the i-th channel of the R layer. The noise is simulated by a Gaussian distribution with a mean of 0 and a positive mean. N c (t) represents the central noise signal introduced by the AON layer. The noise is simulated by a Gaussian random number with a mean of 0 and a positive mean. R1(t)…R n (t) represents the pulse density variable of the olfactory receptor output; P1(t)…P n (t), periglomerular cells in the PG layer, representing potential state variables; M i1 (t)…M in (t), i = 1, 2, mitral cells in the OB layer, representing the potential state variable; G i1 (t)…G in (t), i = 1, 2, granule cells in the OB layer, representing the potential state variable; E i (t),I i (t), i = 1, 2, excitatory neurons and inhibitory neurons in the AON layer, representing potential state variables; A i (t),B i (t), i = 1, 2, excitatory neurons and inhibitory neurons in the PC layer, representing potential state variables; S i (t),I i (t), i = 1, 2, excitatory neurons and inhibitory neurons in the EC layer, representing potential state variables; DG i (t),i=1,2,D l (t),l=1,2,…,8,D i (t) represents the pulse density variable after different long-delay feedback cycles, l represents the number of delayed feedback units in the bionic model; w (PPL) , …,w (Sub_I) , w represents the connection weight between neurons.

[0028] Furthermore, in step four, the epileptic EEG signals are preprocessed to obtain a data set, the processed data set is divided into a training set and a test set according to an 8:2 ratio, the training set is input into the optimized olfactory-hippocampal bionic model for training, and the trained olfactory-hippocampal bionic model is used to recognize the test set to obtain a recognition result.

[0029] The beneficial effects of the present invention are:

[0030] 1. Based on the neural cluster theory, this application improves the olfactory part of the obtained olfactory-hippocampal bionic model, improves the anterior olfactory nucleus layer and the piriform cortex layer to multi-channel input, that is, distributed modeling, and at the same time increases the projection and feedback input between the corresponding neurons, thereby increasing the stability of the model and making the model closer to a real neural network, providing an effective research object for the study of the nervous system in the brain.

[0031] 2. The neuron model of the olfactory-hippocampal bionic model in this application uses pulse neurons, which can better learn the pattern of data generation. At the same time, the Hebbian learning rule and the adaptive learning rule are used to update the connection weights between neurons, strengthen the desired stimulation pattern or learning pattern, that is, when the pattern is desired, strengthen the connection weights between neurons, and weaken the influence of unnecessary factors in recognition and classification, such as unnecessary stimulation, background noise, etc., to improve model performance.

[0032] In order to make the above and other objects, features and advantages of the present invention more clearly understood, preferred embodiments are given below with reference to the accompanying drawings for detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only four of the drawings of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0034] Figure 1 It is a flowchart of the present invention;

[0035] Figure 2 This is the structural diagram of the obtained olfactory-hippocampal model;

[0036] Figure 3 This is the structural diagram of the optimized olfactory-hippocampal model;

[0037] Figure 4 The following is a line graph showing the recognition results of different models on the epilepsy EEG dataset. DETAILED DESCRIPTION

[0038] Embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Although certain embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the drawings and embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0039] Example

[0040] like Figure 1 As shown, a method for identifying epilepsy EEG based on an olfactory-hippocampal bionic model is provided, which is characterized by:

[0041] Step S1, obtaining an olfactory-hippocampal bionic model and a dynamic equation;

[0042] Step S2: Obtain the KII model, perform multi-channel input modeling on the anterior olfactory nucleus and the piriform cortex according to the neural cluster theory, and establish connections between neurons in the olfactory bulb and the entorhinal cortex;

[0043] The process of multi-channel input modeling of the anterior olfactory nucleus layer and the piriform cortex layer is as follows: the anterior olfactory nucleus layer and the piriform cortex layer are both represented by multiple KII models connected in parallel, and the number of parallel input channels of the olfactory bulb layer, the anterior olfactory nucleus layer and the piriform cortex layer is the same, and the data output by the olfactory bulb layer is classified using the proximity principle;

[0044] Step S3, replacing the obtained multi-channel input model and the connections between the neurons into the obtained olfactory-hippocampal bionic model, and modifying the neuron model in the olfactory-hippocampal bionic model to complete the optimization of the olfactory-hippocampal bionic model and update the dynamic equation;

[0045] Step S4: After training the optimized olfactory-hippocampal bionic model, the epilepsy EEG dataset is recognized.

[0046] In step S1, Figure 2 As shown, the olfactory-hippocampal bionic model includes the periocular layer (PG layer), olfactory bulb layer (OB layer), anterior olfactory nucleus layer (AON layer), piriform cortex layer (PC layer), entorhinal cortex layer (EC layer), dentate gyrus layer (DG layer), CA3 layer, CA1 layer and subiculum layer (Sub layer) simulated by the K series model. With the entorhinal cortex layer as the core and through the connection between neurons in each layer, the bionic simulation between the olfactory sense and the hippocampus is realized.

[0047] The K series models are K0, KI, KII, and KIII models constructed based on Professor Freeman's work on the mammalian olfactory nervous system and a large number of neurophysiological experiments. Among them, the R unit is simulated by multiple parallel-connected K0 models, the periocular layer is simulated by multiple parallel-connected KI models, the olfactory bulb is simulated by multiple parallel-connected KII models, the anterior olfactory nucleus is simulated by a KII model, the piriform cortex is simulated by a KII model, the dentate gyrus is simulated by a single KI model, the CA3 layer is simulated by a single KII model, the CA1 layer is simulated by a single KII model, the subiculum is simulated by a single KII model, and the entorhinal cortex is simulated by a single KII model. The piriform cortex and entorhinal cortex are connected in a one-to-one manner, and the dentate gyrus, CA3 layer, CA1 layer, and subiculum are connected in a one-to-one manner.

[0048] R units are olfactory receptors that transmit odor information, P units are periglomerular cells that preprocess odor information, M and G correspond to mitral cells and granule cells, respectively, which together complete the spatial transformation of odor information, E and I correspond to excitatory neurons and inhibitory neurons in the anterior olfactory nucleus, respectively; S and I correspond to stellate cells in the entorhinal cortex, corresponding to interneurons; CA3_E, CA1_E, and Sub_E represent excitatory neurons, and CA3_I, CA1_I, and Sub_I represent inhibitory neurons.

[0049] The connections between neurons in the obtained olfactory-hippocampal biomimetic model are as follows: the R unit represents an olfactory receptor, projecting to an olfactory glomerulus, either directly or through P cells to the OB layer. The AON and PC layers receive projections from the OB layer, with M1 cells projecting to E1 and A1 cells, respectively. The EC layer receives projections from the PC layer, with B1 cells projecting to S1 cells. The AON layer provides feedback to the OB layer, with E1 cells projecting to M1 neurons and G1 cells, respectively. The PC layer provides feedback to the OB and AON layers, with B1 cells projecting to G1 cells and A1 cells projecting to I1 cells. The PC layer provides delayed feedback to the AON layer, and the EC layer provides delayed feedback to the PC layer, with S1 cells projecting to B1 cells. The entorhinal cortex (EC) can project directly to the DG and CA1 layers or through S1 cells. The CA1 and Sub layers receive projections from the DG, with CA3_E1 cells projecting to CA1_E1 and Sub_E1 cells, respectively. The EC layer receives projections from the CA1 and Sub layers, with CA1_I1 and Sub_I1 cells projecting to I1 cells. The CA1 layer provides delayed feedback to the CA3 layer, with CA1_E1 cells projecting to CA3_E1 cells. The Sub layer provides delayed feedback to both the CA3 and CA1 layers, with Sub_E1 cells projecting to CA1_I1 cells and Sub_I1 cells projecting to CA3_I1 cells.

[0050] The dynamic equation of the obtained olfactory-hippocampal bionic model is:

[0051] The neural dynamics equations of the olfactory-hippocampal bionic model are solved using the forward Euler method.

[0052] (1) Input layer

[0053]

[0054] (2)PG layer

[0055]

[0056] (3) OB layer

[0057]

[0058] (4) AON layer

[0059]

[0060] (5) PC layer

[0061]

[0062] (6)EC layer

[0063]

[0064] (7) DG layer

[0065]

[0066] (8) CA3 layer

[0067]

[0068] (9) CA1 layer

[0069]

[0070] (10) Sublayer

[0071]

[0072] (11) Feedback

[0073]

[0074] In formulas (3)-(13), i=1, 2, ..., n represents the number of parallel input channels; represents the peripheral noise signal introduced by the i-th channel of the R layer. The noise is simulated by a Gaussian distribution with a mean of 0 and a positive mean. N c (t) represents the central noise signal introduced by the AON layer. The noise is simulated by a Gaussian random number with a mean of 0 and a positive mean. R1(t)…R n (t) represents the pulse density variable of the olfactory receptor output; P1(t)…P n (t), periglomerular cells in the PG layer, representing potential state variables; M i1 (t)…M in (t), i = 1, 2, mitral cells in the OB layer, representing the potential state variable; G i1 (t)…G in (t), i = 1, 2, granule cells in the OB layer, representing the potential state variable; E i (t),I i (t), i = 1, 2, excitatory neurons and inhibitory neurons in the AON layer, representing potential state variables; A i (t),B i (t), i = 1, 2, excitatory neurons and inhibitory neurons in the PC layer, representing potential state variables; S i (t),I i (t), i = 1, 2, excitatory neurons and inhibitory neurons in the EC layer, representing potential state variables; DG i(t), i = 1, 2, excitatory neurons and inhibitory neurons in the DG layer, representing potential state variables; CA3_E i (t),CA3_I i (t), i = 1, 2, excitatory neurons and inhibitory neurons in the CA3 layer, representing potential state variables; CA1_E i (t),CA1_I i (t), i = 1, 2, excitatory neurons and inhibitory neurons in the CA1 layer, representing potential state variables; Sub_E i (t),Sub_I i (t), i = 1, 2, excitatory neurons and inhibitory neurons in the Sub layer, representing potential state variables; D l (t),l=1,2,…,8,D i (t) represents the pulse density variable after different long-delay feedback cycles, l represents the number of delayed feedback units in the bionic model; w (PPL) , …,w (Sub_I) , w represents the connection weight between neurons.

[0075] In step S2, the neural cluster theory refers to that cell clusters composed of similar neurons have similar functions and consistent characteristics and can serve as building blocks of the entire nervous system.

[0076] The process of multi-channel parallel modeling of the anterior olfactory nucleus layer and the piriform cortex layer is as follows: the anterior olfactory nucleus layer and the piriform cortex layer are both represented by parallel connections of multiple KII models, and the number of parallel input channels of the olfactory bulb layer, the anterior olfactory nucleus layer and the piriform cortex layer is the same, and the data output by the olfactory bulb layer is classified using the proximity principle.

[0077] Because the Olfactory-Hippocampal Biomimetic Model primarily learns patterns in sample data through the olfactory bulb layer for pattern recognition, for example, when training sample data is input, the olfactory bulb layer of the biomimetic model learns the pattern and then stores the training sample data. The data output by the olfactory bulb layer can, to some extent, be considered the cluster center of that pattern. During recognition and classification, test sample data is input into the pattern. Based on the cluster centers of the previous pattern, the Euclidean distance formula is used to calculate the closest cluster center to the olfactory bulb layer output. The data is then classified into the category represented by that cluster center, ensuring the success rate and efficiency of automatic and accurate pattern recognition.

[0078] Among them, the connections between each neuron are as follows: the anterior olfactory nucleus layer and the piriform cortex layer receive projections from the olfactory bulb layer, the entorhinal cortex layer receives projections from the piriform cortex layer, and the olfactory bulb layer projects to the entorhinal cortex layer; the anterior olfactory nucleus layer provides feedback to the olfactory bulb layer, the piriform cortex layer provides feedback to the olfactory bulb layer, the entorhinal cortex provides delayed feedback to the piriform cortex, the piriform cortex provides delayed feedback to the anterior olfactory nucleus layer, and the entorhinal cortex provides feedback input to the olfactory bulb layer.

[0079] Among them, the neuron model in the olfactory-hippocampal bionic model uses spiking neurons, namely the Hodgkin–Huxley (HH) model. This type of neuron has biological characteristics, such as spiking characteristics, which can better learn the patterns of data generation.

[0080] The olfactory-hippocampal biomimetic model uses both the Hebbian learning rule and the adaptive learning rule to update the connection weights between neurons, primarily updating the connection weights of neurons in the olfactory bulb. The Hebbian learning rule primarily reinforces the desired stimulation pattern or learning pattern, strengthening the connection weights between neurons when the pattern is desired. The adaptive learning rule primarily mitigates the influence of unnecessary factors in recognition and classification, such as unnecessary stimulation and background noise.

[0081] After the optimization of the olfactory-hippocampal bionic model, the neuronal dynamic equations corresponding to its internal structure also changed. This is mainly reflected in the improvements of the peripheral bulb (PG layer), olfactory bulb (OB layer), anterior olfactory nucleus (AON layer), piriform cortex (PC layer), and entorhinal cortex (EC layer) structures. The dentate gyrus (DG layer), CA3 layer, CA1 layer, and subiculum (Sub layer) were not improved. The improved parts were then replaced with the obtained olfactory-hippocampal bionic model, and the optimized olfactory-hippocampal bionic model was formed together with the unimproved parts. The neuronal dynamic equations of the olfactory bulb, anterior olfactory nucleus, piriform cortex, and entorhinal cortex structures in the optimized olfactory-hippocampal bionic model are as follows:

[0082] (1) Olfactory bulb

[0083]

[0084] (2) Anterior olfactory nucleus

[0085]

[0086] (3) Piriform cortex

[0087]

[0088] (4) Entorhinal cortex

[0089]

[0090] (5) Feedback

[0091]

[0092] In step S4, the epileptic EEG signal is preprocessed to obtain a data set, the processed data set is divided into a training set and a test set according to an 8:2 ratio, the training set is input into the optimized olfactory-hippocampal bionic model for training, and the trained olfactory-hippocampal bionic model is used to recognize the test set to obtain a recognition result.

[0093] The optimized olfactory-hippocampus bionic model (O-OHB) of the present application, the obtained olfactory-hippocampus bionic model (OHB) and the KIII model were compared in the following test.

[0094] This experiment was conducted in MATLAB R2018b, where the model algorithm was implemented and comparative experiments were conducted. The operating system was Windows 10 Professional, the processor was an AMD Ryzen 5 2600X Six-Core Processor (3.60GHz), the memory was 16GB DDR4, and the hard drive was 3TB. The neurodynamic equations were solved using the Forward Euler method.

[0095] The experimental data used epilepsy EEG data collected by the Epilepsy Research Laboratory at the University of Bonn in Germany. This epilepsy EEG dataset consists of five subsets, labeled Z, O, N, F, and S. Each subset contains 100 single-channel EEG signals with a length of 4097, a duration of 23.6 seconds, and a sampling frequency of 173.61 Hz. Subsets Z and O were collected from a control group of five healthy individuals.

[0096] (1) Data processing: To increase the data volume, the epileptic EEG signals were preprocessed. The length of the epileptic EEG signals was divided into 30, 40, 50, and 60 segments, and the corresponding subsample lengths were 136, 102, 81, and 68, respectively. The number of channels of the epileptic EEG signals was divided into 2, 5, 10, and 50 segments, and the corresponding subsample numbers were 50, 20, 10, and 5. 16 different sample data sizes were obtained. Therefore, 16 different sample data sizes can be combined.

[0097] (2) Training and Classification: The processed dataset was divided into a training set and a test set with an 8:2 ratio. The training set and test set were input into the olfactory-hippocampus bionic model (OHB), the optimized olfactory-hippocampus bionic model (O-OHB), and the KIII model to obtain recognition results. The experiment was repeated three times with different sample data, and the average result of the three tests was taken.

[0098] In this experiment, accuracy, a common method used in pattern recognition and classification, was selected as the evaluation metric. Furthermore, time was incorporated into the experimental results. Table 1 shows the recognition and classification results of different models on the dataset. As can be seen from Table 1, on the epilepsy EEG dataset, the O-OHB model had the highest average recognition rate, while the KIII model had the lowest average computation time. Furthermore, on the epilepsy EEG dataset, the average recognition rate of the model gradually increased with the number of channels. When the sample size was 50*81, the model achieved the highest recognition rate of 99.00%.

[0099] Table 1 Recognition results of three models on epilepsy EEG dataset

[0100]

[0101] In order to intuitively show the experimental results of classification and recognition of O-OHB model, OHB model and KIII model on different data sets, the experimental results in Table 1 are shown in the form of a line graph, as shown in Figure 4 As shown. Among them, Figure 4 The horizontal axis represents the sample data and data feature dimensions, and the vertical axis represents the recognition results of different models. Figure 4 It can be seen from the figure that: overall, as the data feature dimension increases, the recognition results of different models will fluctuate, but the overall situation is constantly improving. The three models have good recognition results on different data sets, but the O-OHB model has the best recognition result, and the KIII model has the worst recognition result; therefore, the bionic model has stronger characteristics after optimization.

[0102] The above description of the various embodiments tends to emphasize the differences between the various embodiments. The same or similar aspects can be referenced to each other. For the sake of brevity, this application will not go into details.

[0103] Note that the above are only preferred embodiments of the present invention and the technical principles employed. Those skilled in the art will appreciate that the present invention is not limited to the specific embodiments herein, and that various obvious changes, readjustments, and substitutions are possible for those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments and may include many other equivalent embodiments without departing from the scope of the present invention. The scope of the present invention is determined by the scope of the appended claims.

Claims

1. A method for identifying epilepsy EEG based on an olfactory-hippocampal bionic model, characterized by: Step 1: Obtain the olfactory-hippocampal bionic model and dynamic equations; Step 2: Obtain the KII model. Based on the neural cluster theory, model the multi-channel input of the anterior olfactory nucleus and the piriform cortex, and establish the connections between the neurons in the olfactory bulb and the entorhinal cortex. The process of multi-channel input modeling of the anterior olfactory nucleus layer and the piriform cortex layer is as follows: the anterior olfactory nucleus layer and the piriform cortex layer are both represented by multiple KII models connected in parallel, and the number of parallel input channels of the olfactory bulb layer, the anterior olfactory nucleus layer and the piriform cortex layer is the same, and the data output by the olfactory bulb layer is classified using the proximity principle; Step 3: replacing the obtained multi-channel input model and the connections between the neurons into the obtained olfactory-hippocampal bionic model, and modifying the neuron model in the olfactory-hippocampal bionic model to complete the optimization of the olfactory-hippocampal bionic model and update the dynamic equation; Step 4: After training the optimized olfactory-hippocampal bionic model, the epilepsy EEG dataset is identified.

2. The epilepsy EEG recognition method based on the olfactory-hippocampal bionic model according to claim 1, characterized in that: In step 1, the olfactory-hippocampal bionic model includes the olfactory bulb layer, anterior olfactory nucleus layer, piriform cortex layer, entorhinal cortex layer, dentate gyrus layer, CA3 layer, CA1 layer and subiculum layer simulated by the K series model, with the entorhinal cortex layer as the core and through the connection between neurons in each layer, the bionic simulation between the sense of smell and the hippocampus is realized.

3. The epilepsy EEG recognition method based on the olfactory-hippocampal bionic model according to claim 2, characterized in that: The K series models are K0, KI, KII, and KIII models constructed by Professor Freeman based on the mammalian olfactory nervous system and a large number of neurophysiological experiments.

4. The epilepsy EEG recognition method based on the olfactory-hippocampal bionic model according to claim 1, characterized in that: In step 2, the connections between the neurons are as follows: the anterior olfactory nucleus layer and the piriform cortex layer receive projections from the olfactory bulb layer, the entorhinal cortex layer receives projections from the piriform cortex layer, and the olfactory bulb layer projects to the entorhinal cortex layer; the anterior olfactory nucleus layer provides feedback to the olfactory bulb layer, the piriform cortex layer provides feedback to the olfactory bulb layer, the entorhinal cortex provides delayed feedback to the piriform cortex, the piriform cortex provides delayed feedback to the anterior olfactory nucleus layer, and the entorhinal cortex provides feedback input to the olfactory bulb layer.

5. The epilepsy EEG recognition method based on the olfactory-hippocampal bionic model according to claim 1, characterized in that: In the step 3, the neuron model in the olfactory-hippocampal bionic model adopts a spiking neuron.

6. The epilepsy EEG recognition method based on the olfactory-hippocampal bionic model according to claim 5, characterized in that: The olfactory-hippocampal bionic model utilizes the Hebbian learning rule and the adaptive learning rule to update the connection weights between neurons.

7. The epilepsy EEG recognition method based on the olfactory-hippocampal bionic model according to claim 1, characterized in that: In step 3, the updated kinetic equation is: (1) Olfactory bulb (2) Anterior olfactory nucleus (3) Piriform cortex (4) Entorhinal cortex (5) Feedback In formulas (1)-(5), i=1, 2, ..., n represents the number of parallel input channels; represents the peripheral noise signal introduced by the i-th channel of the R layer. The noise is simulated by a Gaussian distribution with a mean of 0 and a positive mean. N c (t) represents the central noise signal introduced by the AON layer. The noise is simulated by a Gaussian random number with a mean of 0 and a positive mean. R1(t)…R n (t) represents the pulse density variable of the olfactory receptor output; P1(t)…P n (t), periglomerular cells in the PG layer, representing potential state variables; M i1 (t)…M in (t), i = 1, 2, mitral cells in the OB layer, representing the potential state variable; G i1 (t)…G in (t), i = 1, 2, granule cells in the OB layer, representing the potential state variable; E i (t),I i (t), i = 1, 2, excitatory neurons and inhibitory neurons in the AON layer, representing potential state variables; A i (t),B i (t), i = 1, 2, excitatory neurons and inhibitory neurons in the PC layer, representing potential state variables; S i (t),I i (t), i = 1, 2, excitatory neurons and inhibitory neurons in the EC layer, representing potential state variables; DG i (t),i=1,2,D l (t),l=1,2,…,8,D i (t) represents the pulse density variable after different long-delay feedback cycles, l represents the number of delayed feedback units in the bionic model; w represents the connection weight between neurons.

8. The epilepsy EEG recognition method based on the olfactory-hippocampal bionic model according to claim 1, characterized in that: In step 4, the epileptic EEG signals are preprocessed to obtain a data set, the processed data set is divided into a training set and a test set according to an 8:2 ratio, the training set is input into the optimized olfactory-hippocampal bionic model for training, and the trained olfactory-hippocampal bionic model is used to recognize the test set to obtain a recognition result.

Images