CN112381174B - Neural signal classification method for multi-electrode array - Google Patents
Neural signal classification method for multi-electrode array Download PDFInfo
- Publication number
- CN112381174B CN112381174B CN202011380833.7A CN202011380833A CN112381174B CN 112381174 B CN112381174 B CN 112381174B CN 202011380833 A CN202011380833 A CN 202011380833A CN 112381174 B CN112381174 B CN 112381174B
- Authority
- CN
- China
- Prior art keywords
- waveform
- neural
- signal
- signals
- classification
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000001537 neural effect Effects 0.000 title claims abstract description 50
- 238000000034 method Methods 0.000 title claims abstract description 20
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 38
- 238000001514 detection method Methods 0.000 claims abstract description 14
- 238000001914 filtration Methods 0.000 claims abstract description 6
- 238000012847 principal component analysis method Methods 0.000 claims abstract description 5
- 230000008569 process Effects 0.000 claims abstract description 5
- 210000005036 nerve Anatomy 0.000 claims description 10
- 238000000605 extraction Methods 0.000 claims description 3
- 108010076504 Protein Sorting Signals Proteins 0.000 claims description 2
- 239000000284 extract Substances 0.000 abstract description 7
- 230000008901 benefit Effects 0.000 abstract description 5
- 238000004364 calculation method Methods 0.000 abstract description 4
- 239000000203 mixture Substances 0.000 description 4
- 238000003491 array Methods 0.000 description 3
- 238000007635 classification algorithm Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 230000018109 developmental process Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 210000002569 neuron Anatomy 0.000 description 3
- 230000036982 action potential Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011056 performance test Methods 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Other Investigation Or Analysis Of Materials By Electrical Means (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention discloses a neural signal classification method for a multi-electrode array. The invention removes a part of background noise and extracts a signal waveform sequence on a plurality of electrodes through a band-pass filtering and threshold detection algorithm, extracts waveform characteristics by using a principal component analysis method, completes signal waveform classification and overlapped signal waveform detection by using an HDBSCAN clustering method, and completes overlapped signal waveform classification by using template matching. The invention combines the advantages and characteristics of two main algorithms of HDBSCAN clustering and template matching, extracts effective waveform characteristics to the greatest extent and removes redundant data information from the multi-electrode array, realizes complete classification of all waveform overlapping and non-overlapping neural signals, fully utilizes concurrent flows of a computer to accelerate the calculation process, and greatly accelerates the calculation speed on the premise of providing high classification precision.
Description
Technical Field
The invention belongs to the technical field of neural signal classification for multi-electrode arrays, and particularly relates to a neural signal classification method based on machine learning and signal processing.
Background
Acquiring, interpreting and understanding signals transmitted between neurons in the human brain has great significance in both medical research and development in the brain-computer interface field, and thanks to the integrated circuit technology and electrode technology that have been rapidly developed in recent years, multi-electrode arrays that integrate multiple electrodes at intervals of tens of micrometers and even higher have become possible, with which researchers have acquired unprecedented amounts of neural signal data. Before the patterns and rules of the existence of these neural signals are interpreted and understood, however, it is a necessary task to classify them, and neural signal classification methods aim at classifying acquired neuronal action potential waveforms so that they correspond to respective potential emission units. The most primitive classification is done manually, extremely dependent on the experience of the classifier and is labor intensive, which has not been possible to accomplish manually in the face of massive data. Since the beginning of the 21 st century, numerous automated classification algorithms for single electrodes have been developed, but they tend to run slowly and with low accuracy when faced with complex multi-electrode situations, and are not adequate for today's classification tasks for multi-electrode arrays.
In recent years, with the rapid development of machine learning technology and feature engineering, it has become possible to develop a neural signal classification algorithm for multiple electrodes. Many researchers in this field have proposed their own schemes, the Kilosort algorithm proposed by Pachitariu et al obtains the best classification result by optimizing a loss function on waveform characteristics, the herdingspeckes 2 proposed by Hilgen et al and the mountain 4 algorithm proposed by Chung et al develop their own clustering method by combining waveform characteristics and electrode space positions, but none of the above three algorithms can classify signals where multiple waveform overlaps occur. The YASS algorithm proposed by Lee et al uses a strategy of "first excluding and then clustering", first detecting the waveform overlapping signals, then completing the classification of the remaining signals, and finally analyzing and recovering the waveform overlapping signals, which proves to be very effective, but the YASS contains a neural network which requires model training in advance, so that it is not suitable for an unsupervised scene, while the prototype clustering algorithm used by the YASS algorithm assumes that the distribution of samples conforms to a gaussian mixture model, but in fact many studies prove that the waveform sample distribution of the neural signals does not always conform to the gaussian mixture model, such as when electrode drift phenomenon occurs. In the spykingcircle algorithm proposed by Yger et al, the clustering algorithm is used only to generate "templates", and template matching is responsible for completing all classification tasks, which can give more accurate results, but takes too much time in post-processing steps.
The development of a neural signal classification algorithm for a multi-electrode array is still in a bottleneck period, and the problems of how to correctly utilize and process the correlation and redundancy of data on a plurality of electrodes, how to better eliminate the interference of background noise, how to classify signals with time-space overlapping waveforms, how to reduce the time consumption of the algorithm on the premise of ensuring the classification accuracy and the like are still difficulties in the development of the field.
Disclosure of Invention
In order to solve the above-mentioned problems of the prior art, the present invention provides a neural signal classification method for a multi-electrode array.
The invention removes a part of background noise and extracts a signal waveform sequence on a plurality of electrodes through a band-pass filtering and threshold detection algorithm, extracts waveform characteristics by using a principal component analysis method, completes signal waveform classification and overlapped signal waveform detection by using an HDBSCAN clustering method, and completes overlapped signal waveform classification by using template matching.
The technical scheme adopted for solving the technical problems is as follows:
a neural signal classification method for a multi-electrode array, comprising the steps of:
step 1, carrying out 300Hz-3KHz band-pass filtering on multiple paths of original signals acquired from the multi-electrode array, filtering background noise signals, and removing interference.
And 2, executing a threshold detection algorithm on the signals on each electrode channel in parallel, judging the spike positions of the nerve signals according to the threshold, extracting the first 0.75 millisecond and the last 0.75 millisecond of each spike to form complete nerve signal waveforms with the length of 1.5 milliseconds, and simultaneously, only preserving and storing one waveform sequence with the best signal-to-noise ratio on the same nerve signal waveform acquired and extracted on a plurality of electrodes.
And step 3, performing a principal component analysis method on all extracted nerve signal waveforms to extract signal features which can represent waveform differences most, and reducing feature latitudes of waveform sequences.
And 4, using the neural signal waveform after the feature extraction as the input of a clustering algorithm HDBSCAN, and completing the clustering process according to the mutual reachable distance and a single-link algorithm to obtain a classification result of the neural signal and the neural signal with the waveform overlapped with each other. The mutually overlapping neural signals are not discarded and will be identified and classified in a subsequent step.
And 5, calculating the center of each cluster according to the classification result of the fourth step, taking a signal sequence represented by the cluster center as a template of a neural signal waveform, combining two too similar and close templates into one template according to the Euclidean distance between the cluster centers and the spatial arrangement of electrodes, and storing the final template into a standard template library.
And 6, using the template library of the fifth step as a standard reference template of the template matching algorithm, and executing the template matching algorithm to analyze and classify the overlapped signals in the fourth step to obtain a classification result of the overlapped nerve signals.
The invention has the beneficial effects that: the invention combines the advantages and characteristics of two main algorithms of HDBSCAN clustering and template matching, extracts effective waveform characteristics to the greatest extent and removes redundant data information from the multi-electrode array, realizes complete classification of all waveform overlapping and non-overlapping neural signals, fully utilizes concurrent flows of a computer to accelerate the calculation process, and greatly accelerates the calculation speed on the premise of providing high classification precision.
Drawings
Fig. 1 is an overall flow chart of the present invention.
FIG. 2 is a graph comparing the HDBSCAN clustering algorithm used in the present invention with other clustering algorithms.
Detailed Description
The present invention will be further described with reference to the drawings and examples in order to more clearly understand the technical features, objects and effects of the present invention.
Fig. 1 is a flow chart of a neural signal classifying method for a multi-electrode array according to the present invention. As shown in the flow chart of fig. 1, first, raw neural signal data acquired on a multi-electrode array (four electrodes are taken as an example in the drawing), and various interference noises including noise from electrode tips, weak signals from neurons in a far-end area and the like exist in an actual acquisition environment, which is collectively called background noise. According to the working frequency band of the signal of the acquisition target area, a band-pass filter of 300Hz-3KHz is arranged to filter out a part of background noise.
And then extracting all single signal waveform sequences from each path of electrode channel data. The neural signal action potential has a significant spike with a magnitude much greater than the remainder, and the threshold detection algorithm determines whether the current data point is a spike based on the set spike threshold, and if so, extracts the first 0.75 ms and the last 0.75 ms of each spike to form a complete neural signal waveform with a length of 1.5 ms. To improve the efficiency of operation, a threshold detection algorithm is executed in parallel on each electrode channel to take full advantage of the concurrency of the computer. In a multi-electrode array, a neural signal excited at a certain moment is captured by a plurality of adjacent electrodes, and for utilizing the advantages of the multiple electrodes and removing redundancy of data information, for the same signal waveform, one of which has the best signal-to-noise ratio is finally selected and stored. If there are k electrodes, k packets containing a waveform sequence are finally obtained, each packet corresponding to an electrode channel having the best signal-to-noise ratio for the waveform.
The features of the waveform sequence are extracted by a principal component analysis method, which selects and retains a plurality of features in which waveform differences are most represented by analyzing variance differences between the features. The necessity of this step is: (1) The information redundancy is further removed, the necessary characteristics of classification are reserved, and the operation of a subsequent algorithm is quickened. (2) Reducing the feature dimension of the waveform sequence, too long dimension can cause the subsequent clustering algorithm to sink into the "dimension curse" phenomenon in machine learning, and the clustering algorithm can perform poorly anyway under the phenomenon.
The clustering algorithm HDBSCAN (Hierarchical Density-Based Spatial Clustering of Applications with Noise) is a clustering algorithm with noise identification and based on sample hierarchy and density, defines a mutual reachable distance between input sample points, firstly converts samples from an original characteristic space to a space based on the mutual reachable distance, has the effects of enabling sparse distribution areas of the samples to be sparse, enabling dense distribution areas of the samples to be dense, then establishing a minimum spanning tree of the space, constructing a cluster hierarchy structure, finally compressing the hierarchy structure into a smaller tree, and extracting an optimal clustering result. In fig. 2 is a cluster effect comparison of HDBSCAN clusters and two other common clustering algorithms kmans and DBSCAN on two standard cluster performance test datasets, different sample point shapes representing different classifications after clustering, where circles represent samples that are algorithmically judged to be noise. In the data set of the bi-crescent profile of the first row kmens, although successful in classifying the samples into two classes, a significant portion of the samples are misclassified, resulting in the fact that kmens assumes that the samples are approximately spherically shaped, and in addition kmens do not support noise detection, and also force classification of some outliers around, it is preferable for them to be detected; the failure of DBSCAN to classify samples into two categories results in the fact that its superparameter is relatively sensitive, can be successfully classified by multiple adjustments to the superparameter, but more desirably the default parameters will work well, and in addition DBSCAN can support at least noise detection, a feature that is important in some applications. The HDBSCAN almost perfectly classifies the two crescent clusters and also supports noise detection, and the result is simply to use default parameters, which benefits from the superior robustness of the HDBSCAN's super parameters, keeping the default parameters to operate in an unsupervised mode. In the three spherically distributed data set with density variation of the second row kmens forces the three spherically clusters into two classes, resulting in that kmens needs to know in advance before working that the samples should be classified into several classes, and typically does not know this information when a clustering algorithm is needed, so kmens may not be called a clustering algorithm but a segmentation algorithm; DBSCAN also fails to classify successfully because its principle of operation results in clusters that do not cope well with sample density variations; HDBSCAN is again nearly perfect in completing the classification task, distinguishing three spherical clusters and detecting noise. Based on the comparison, the HDBSCAN clustering algorithm is adopted by the invention for several reasons: 1. the neural signal classification requires truly unsupervised automatic clustering, the HDBSCAN has reasonable hyper-parameter requirements and is very robust, and the neural signal classification can work in an unsupervised manner by using default parameters. 2. HDBSCAN does not assume any prototype of the sample distribution, it works based on the density and hierarchy of the sample distribution, and the neural signal waveform samples may ideally be distributed in accordance with a gaussian mixture model, but no longer in accordance with a fixed distribution pattern when neuronal bursts and electrode drift occur. 3. The HDBSCAN supports noise detection, and when several neural signal waveforms overlap in time and space, the waveforms will be distorted due to waveform characteristics to represent noise samples different in the characteristic space, so that the HDBSCAN can help to detect overlapping signals. After the waveform sequence subjected to the feature extraction is input into the HDBSCAN, a clustering classification result of the neural signals is obtained, and a waveform sample which is judged to be noise is obtained, wherein the noise is a different sample, the noise can be a combination of a plurality of neural signals with overlapped waveforms or a single neural signal severely interfered by background noise, and the analysis and classification of the neural signals are completed through subsequent steps.
In the result of the HDBSCAN clustering, calculating the average value of the classified waveform sequences to obtain a clustering center of a class of waveforms to be used as a template of the class of waveforms. Grouping k sets of waveform sequences each produces a plurality of templates belonging thereto, but all templates cannot be stored directly in a standard template library, and a template merging step is required. The necessity of template merging is that, although each waveform is retained only in the electrode channel group whose signal-to-noise ratio is optimal after the threshold detection step, if a certain neuron is located equidistantly in the middle of two electrodes, the result is that the signal waveform it produces will remain in the corresponding group of two electrodes, and the templates derived from them should be merged. Template merging is based on two criteria: (1) The two electrodes that produce the template are indeed neighboring electrodes in spatial position distribution. (2) The euclidean distance between cluster centers as templates should be small enough. After the template merging step is completed, all templates are stored in a standard template library.
The final template matching step aims at analyzing and classifying the neural signals with overlapped waveforms according to the standard template library, the algorithm matches the current waveform to be analyzed each time and subtracts an optimal template waveform until the residual waveform does not meet the threshold value of threshold detection, which indicates that the residual waveform does not exist any effective neural signal waveform any more, and the result is the composition category of the neural signals with overlapped waveforms, and successfully analyzes and classifies the overlapped neural signal waveforms. The whole flow outputs classification results of all waveform samples.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, and alternatives falling within the spirit and principles of the invention.
Claims (4)
1. A nerve signal classification method for a multi-electrode array is characterized by comprising the following steps:
step 1, carrying out 300Hz-3KHz band-pass filtering on multiple paths of original signals acquired from a multi-electrode array, filtering background noise signals and removing interference;
step 2, executing a threshold detection algorithm on the signals on each electrode channel in parallel, judging the spike positions of the nerve signals according to the threshold, extracting the first 0.75 millisecond and the last 0.75 millisecond of each spike to form a complete nerve signal waveform with the length of 1.5 milliseconds, and storing the complete nerve signal waveform;
step 3, performing a principal component analysis method on all extracted nerve signal waveforms to extract signal features which can represent waveform differences most, and reducing feature latitudes of waveform sequences;
step 4, using the neural signal waveform after the feature extraction as the input of a clustering algorithm HDBSCAN, completing the clustering process according to the mutual reachable distance and a single-link algorithm, and obtaining a classification result of the neural signal and the neural signal with the waveform overlapped with each other;
step 5, calculating the center of each cluster according to the classification result in the step 4, and storing the signal sequence represented by the cluster center into a template library as a template of a type of neural signal waveform;
and 6, using the template library in the step 5 as a standard reference template of a template matching algorithm, and executing the template matching algorithm to analyze and classify the neural signals with mutually overlapped waveforms in the fourth step to obtain the classification result of the overlapped neural signals.
2. A method of classifying neural signals for a multi-electrode array according to claim 1, wherein: in the step 2, the same neural signal waveform acquired and extracted on a plurality of electrodes simultaneously only retains and stores one waveform sequence in which the signal-to-noise ratio is optimal.
3. A method of classifying neural signals for a multi-electrode array according to claim 1, wherein: in step 4, the mutually overlapping neural signals are also successfully classified in step 6, but not discarded.
4. A method of classifying neural signals for a multi-electrode array according to claim 1, wherein: in step 5, two too similar and close templates are combined into one template according to the Euclidean distance between the cluster centers and the spatial arrangement of the electrodes.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011380833.7A CN112381174B (en) | 2020-12-01 | 2020-12-01 | Neural signal classification method for multi-electrode array |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011380833.7A CN112381174B (en) | 2020-12-01 | 2020-12-01 | Neural signal classification method for multi-electrode array |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112381174A CN112381174A (en) | 2021-02-19 |
| CN112381174B true CN112381174B (en) | 2024-02-02 |
Family
ID=74590087
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011380833.7A Active CN112381174B (en) | 2020-12-01 | 2020-12-01 | Neural signal classification method for multi-electrode array |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112381174B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113057656B (en) * | 2021-03-29 | 2022-02-01 | 浙江大学 | Method, device and system for detecting brain nerve spike potential signal on line based on self-adaptive threshold |
| CN115081471B (en) * | 2022-05-13 | 2023-05-30 | 浙江大学 | Peak potential detection and classification method based on multi-band composite waveform and application |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2007058950A2 (en) * | 2005-11-10 | 2007-05-24 | Cyberkinetics Neurotechnology Systems, Inc. | Biological interface system with neural signal classification systems and methods |
| CN109171707A (en) * | 2018-10-24 | 2019-01-11 | 杭州电子科技大学 | A kind of intelligent cardiac figure classification method |
-
2020
- 2020-12-01 CN CN202011380833.7A patent/CN112381174B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2007058950A2 (en) * | 2005-11-10 | 2007-05-24 | Cyberkinetics Neurotechnology Systems, Inc. | Biological interface system with neural signal classification systems and methods |
| CN109171707A (en) * | 2018-10-24 | 2019-01-11 | 杭州电子科技大学 | A kind of intelligent cardiac figure classification method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112381174A (en) | 2021-02-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107909117B (en) | Classification device for early and late mild cognitive impairment based on brain function network characteristics | |
| Zhao et al. | Cloud shape classification system based on multi-channel cnn and improved fdm | |
| CN108491077B (en) | Surface electromyographic signal gesture recognition method based on multi-stream divide-and-conquer convolutional neural network | |
| Lewicki | A review of methods for spike sorting: the detection and classification of neural action potentials | |
| CN112381174B (en) | Neural signal classification method for multi-electrode array | |
| Funke et al. | Efficient automatic 3D-reconstruction of branching neurons from EM data | |
| CN111150393B (en) | Electroencephalogram epilepsy spike discharge joint detection method based on LSTM multichannel | |
| CN109645990A (en) | A kind of CRT technology method of epileptic's EEG signals | |
| CN111539152A (en) | Rolling bearing fault self-learning method based on two-stage twin convolutional neural network | |
| US11622727B2 (en) | Process and hardware implementation of adaptive real-time neural spike sorting | |
| CN110399846A (en) | A kind of gesture identification method based on multichannel electromyography signal correlation | |
| CN108846346B (en) | Neuron peak potential on-line classification method based on density peak clustering algorithm | |
| CN113486752B (en) | Emotion recognition method and system based on electrocardiosignal | |
| CN112330609A (en) | Intelligent fault diagnosis method for magnetic suspension rotor system | |
| CN113627391B (en) | Cross-mode electroencephalogram signal identification method considering individual difference | |
| CN100412884C (en) | Human face quick detection method based on local description | |
| CN109662710A (en) | A kind of EMG Feature Extraction based on convolutional neural networks | |
| CN112101319B (en) | Vein image classification method and device based on topographic point classification | |
| KR102298709B1 (en) | Device and method for learning connectivity | |
| Iqtidar et al. | Image pattern analysis towards classification of skin cancer through dermoscopic images | |
| US20210158154A1 (en) | Apparatus and method for distinguishing neural waveforms | |
| CN113057654B (en) | Memory load detection and extraction system and method based on frequency coupling neural network model | |
| CN112932431B (en) | Heart rate identification method based on 1DCNN + Inception Net + GRU fusion network | |
| Keshtkaran et al. | Unsupervised spike sorting based on discriminative subspace learning | |
| Yang et al. | On-chip spike clustering & classification using self organizing map for neural recording implants |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |