Method for identifying perceptual signals in information services and related equipment
By introducing attention layer and wavelet transformation into the convolutional neural network, the complexity of perceived signal recognition in information services is solved, efficient identification and research of perceived signals is achieved, and the ability of information perception research is improved.
Patent Information
- Application Number
- CN202210397192.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-04-15
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2042-04-15
AI Technical Summary
The prior art is difficult to effectively identify and study perceived signals in the information perception process, especially in the context of the complex brain nervous system, which leads to difficulties in research on information perception.
The convolutional neural network with attention layer filters, time window selection and wavelet transformation of the collected perceptual signals, extracts the time domain energy entropy, and recognizes the convolutional neural network enhanced by attention mechanism to extract high-level features. To realize the recognition of perceived signals.
Through this method, the perceived signals in information services can be effectively identified and extracted, which improves the characteristic recognition and classification accuracy of the perceived signals, and meets the needs of information perception research.
Smart Images

Figure CN114742107B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of signal recognition, and in particular to a method for recognizing perceptual signals in information services and related equipment. Background Art
[0002] The information perception process is the process of people acquiring and applying knowledge. It is the process of the human brain receiving external information stimulation and performing secondary processing to obtain its own experience method. Information perception is the recognition of the attributes and characteristics of things. It is based on human perception of the connection and relationship between things. It is a series of complex signal processes in the brain. As we all know, the brain is the most complex organizational structure known to humans so far, and the perceptual signals in information services are closely related to the functional state of the brain's nervous system. When there is a stimulus, the perceptual signals in the information service will give a complex response, which also reflects the various links of the information perception processing process. Therefore, through the study of perceptual signals in information services, the process of information perception can be studied more clearly.
[0003] Based on this, a method for identifying perceptual signals in information services is needed to meet the needs of studying information perception. Summary of the invention
[0004] In view of this, the purpose of the present application is to propose a method for identifying perceptual signals in information services and related equipment.
[0005] Based on the above purpose, the present application provides a method for identifying a perceptual signal in an information service, comprising:
[0006] Acquire a first perception signal through a perception signal acquisition channel combination;
[0007] Performing filtering processing and time window selection processing on the first perception signal to obtain a second perception signal;
[0008] performing a wavelet transform on the second perception signal, extracting time domain energy entropy of the second perception signal after the wavelet transform, and using the time domain energy entropy as a third perception signal;
[0009] The third perception signal is recognized by a convolutional neural network with an attention layer to obtain a recognition result of the third perception signal.
[0010] Furthermore, before acquiring the first perception signal through the perception signal acquisition channel combination, the method further includes:
[0011] Acquire a training data set collected through multiple perception signal collection channels, wherein the training data set includes multiple training perception signals;
[0012] Inputting the training data set into a random forest classifier model, an ExtraTree classifier model, and an Adaboost classifier model, and obtaining ranking results of a plurality of the perception signal acquisition channels corresponding to the random forest classifier model, the ExtraTree classifier model, and the Adaboost classifier model, respectively;
[0013] A heat map for measuring the correlation between the plurality of the perception signal acquisition channels is drawn according to the sorting result and the Pearson correlation coefficient, and a preset number of the perception signal acquisition channels are selected according to the heat map to form the perception signal acquisition channel combination.
[0014] Further, performing time window selection processing on the first perception signal includes:
[0015] Using a time window of a preset time length to intercept the first perception signal;
[0016] Marking sampling points of all data in the intercepted first perception signal as first labels, and marking other sampling points as second labels;
[0017] The first label of one of the plurality of continuous data implementation sampling points is retained, and the other data implementation sampling points of the plurality of data implementation sampling points are marked as second labels to obtain the second perception signal.
[0018] Furthermore, the identifying the third perception signal by a convolutional neural network having an attention layer to obtain a recognition result of the third perception signal includes:
[0019] Inputting the third perception signal into the convolution layer of the convolutional neural network, and extracting spatial domain features and temporal domain features of the third perception signal through the convolution layer;
[0020] Extracting high-level features of the third perception signal through the attention layer;
[0021] The spatial domain features, time domain features and high-level features are input into the fully connected layer of the convolutional neural network for classification to obtain the recognition result.
[0022] Further, the training data set is input into a random forest classifier model, an ExtraTree classifier model, and an Adaboost classifier model, and the ranking results of the plurality of perception signal acquisition channels corresponding to the random forest classifier model, the ExtraTree classifier model, and the Adaboost classifier model are obtained, respectively, including:
[0023] Inputting the training data set into a random forest classifier model, wherein the random forest classifier model performs a feature selection operation and a splitting operation on the plurality of perception signals in the training data set to obtain a first weight factor set corresponding to the features of the plurality of perception signals; sorting the plurality of perception signal acquisition channels according to the first weight factor set to obtain a first sorting result corresponding to the plurality of perception signal acquisition channels;
[0024] The training data set is input into the ExtraTree classifier model, and the ExtraTree classifier model performs feature randomization, parameter randomization, model randomization, and split randomization operations on the plurality of perception signals in the training data set to obtain a second weight factor set corresponding to the features of the plurality of perception signals; the plurality of perception signal acquisition channels are sorted according to the second weight factor set to obtain a second sorting result corresponding to the plurality of perception signal acquisition channels;
[0025] The training data set is input into an Adaboost classifier model, and the Adaboost classifier model performs a feature selection operation on the multiple perception signals in the training data set to obtain a third weight factor set corresponding to the features of the multiple perception signals; and the multiple perception signal acquisition channels are sorted according to the third weight factor set to obtain a third sorting result corresponding to the multiple perception signal acquisition channels.
[0026] Furthermore, the step of drawing a heat map for measuring the correlation between the plurality of perception signal acquisition channels according to the ranking result and the Pearson correlation coefficient includes:
[0027] Performing feature screening on the plurality of perception signal acquisition channels according to the first sorting result, the second sorting result, and the third sorting result to obtain a fourth weight factor set for the plurality of perception signal acquisition channels; sorting the plurality of perception signal acquisition channels according to the fourth weight factor set to obtain a fourth sorting result corresponding to the plurality of perception signal acquisition channels;
[0028] A heat map for measuring the correlation between the plurality of perception signal acquisition channels is drawn according to the fourth sorting result and the Pearson correlation coefficient.
[0029] Furthermore, the perception signal is a P300 perception signal.
[0030] Based on the same concept, the present application also provides a device for identifying a perception signal in an information service, comprising:
[0031] A collection module, configured to obtain a first perception signal through a combination of perception signal collection channels;
[0032] A first processing module is configured to perform filtering processing and time window selection processing on the first perception signal to obtain a second perception signal;
[0033] A second processing module is configured to perform a wavelet transform on the second perception signal, extract a time domain energy entropy of the second perception signal after the wavelet transform, and use the time domain energy entropy as a third perception signal;
[0034] The recognition module is configured to recognize the third perception signal through a convolutional neural network with an attention layer to obtain a recognition result of the third perception signal.
[0035] Based on the same concept, the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the methods described above when executing the program.
[0036] Based on the same concept, the present application also provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable the computer to implement any of the methods described above.
[0037] From the above, it can be seen that the method for identifying perceptual signals in information services provided by the present application is based on a convolutional neural network with an attention layer. First, the perceptual signals in the collected information services are filtered and time window selected, and then the perceptual signals in the processed information services are subjected to wavelet transform to extract the time domain energy entropy of the perceptual signals in the information services after the wavelet transform. Finally, the perceptual signals in the information services are identified by a convolutional neural network with an attention layer to extract high-level features of the overall data, and the Attention mechanism is used to identify more fine-grained high-level feature differences, so that the convolutional neural network can extract discriminative features and obtain the recognition results of perceptual signals in the information services. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] In order to more clearly illustrate the technical solutions in the present application or related technologies, the drawings required for use in the embodiments or related technical descriptions are briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0039] Figure 1 A flow chart of a method for identifying a perception signal in an information service according to an embodiment of the present application;
[0040] Figure 2This is a flow chart of a method for selecting a combination of perception signal acquisition channels according to an embodiment of the present application;
[0041] Figure 3 This is a flow chart of the time window selection method of an embodiment of the present application;
[0042] Figure 4 This is a flow chart of a method for identifying a third perception signal according to an embodiment of the present application;
[0043] Figure 5 A schematic diagram of the structure of a device for identifying a perception signal in an information service according to an embodiment of the present application;
[0044] Figure 6 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION
[0045] In order to make the objectives, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.
[0046] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present application should be the usual meanings understood by people with ordinary skills in the field to which the present application belongs. The "first", "second" and similar words used in the embodiments of the present application do not represent any order, quantity or importance, but are only used to distinguish different components. "Including" or "comprising" and similar words mean that the elements or objects appearing in front of the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.
[0047] As described in the background technology section, the perception signal recognition methods in related information services are still difficult to meet the needs of actual research on information perception.
[0048] In the process of implementing this application, the applicant discovered that convolutional neural networks (CNNs) are a type of feedforward neural networks that can construct deep structures and perform convolution operations using feature kernels. They are one of the representative algorithms for deep learning. Convolutional neural networks have strong representational learning capabilities and can perform translation-invariant classification of data according to the dimensions of their hierarchical structure. When dealing with classification tasks, we will use the feature space finally output by several CNN layers as the input of a fully connected layer or a fully connected neural network (FCN), and use the fully connected layer to complete the classification of the target task, that is, to map the input image to the corresponding label set. Of course, the most important task in the whole process is how to iteratively adjust the network weights through training data, which is the backpropagation algorithm. The current mainstream convolutional neural networks (CNNs), such as VGG and ResNet, are all combined by adjusting simple CNNs.
[0049] The Attention mechanism solves the bottleneck problem of information loss caused by the conversion and encoding process from long sequences to fixed-length vectors. It was initially widely used in long-sequence text encoding and decoding work such as machine translation, and then migrated and attached to various deep learning networks, trying to model the influence of different weights of various parts in the problem. Taking the simplest translation problem as an example, in the encoder-decoder structure, the Attention mechanism proposes that the encoder inputs different semantic vectors c at each moment. These vectors are realized by weighted summation of all hidden states of the sequence, and then these semantic vectors are used for decoding, solving the model performance bottleneck caused by the sequence length. The relevant algorithms are as follows:
[0050]
[0051] α t,i =soft max(e t,i )
[0052] e t,i =score(s t-1 ,h i )
[0053] Among them, h i is the result of concatenating the coding layers, α t,i is the tth moment, h i The weight coefficient, s t-1 It is to decode the hidden state of the previous step.
[0054] Evoked EEG signal (P300 brain-computer signal) is a multi-channel time domain signal with a low signal-to-noise ratio and large individual differences. Because the three core advantages of CNN, "local receptive field", "weight sharing" and "pooling operation", can better identify and extract the characteristics of different channels or receptive fields in the image, we can learn from the network advantages of CNN and combine it with specific data processing to process this variable signal into a representation form similar to image data. This specific data processing mainly includes digitization, filtering, evoked EEG signal marking and resampling operations on all collected EEG signals, and further obtains clean, smooth and multi-channel EEG data that meets the input requirements of convolutional neural networks.
[0055] In view of this, one or more embodiments of the present application provide a recognition scheme for perceptual signals in information services. Based on a convolutional neural network with an attention layer, the collected perceptual signals are first filtered and time window selected, and then the processed perceptual signals are subjected to wavelet transform to extract the time domain energy entropy of the perceptual signals after the wavelet transform. Finally, the perceptual signals are recognized through a convolutional neural network with an attention layer to obtain a recognition result of the perceptual signals.
[0056] refer to Figure 1 The method for identifying a perceptual signal in an information service of an embodiment of the present application comprises the following steps:
[0057] Step S101: Acquire a first perception signal through a perception signal acquisition channel combination;
[0058] In this embodiment, there may be multiple perception signal collection channels, and the multiple perception signal collection channels respectively collect different potentials to collect perception signals as detailed as possible. The set of perception signals in the information service collected by the perception signal collection channels in the multiple information services is taken as the first perception signal.
[0059] Step S102: performing filtering processing and time window selection processing on the first perception signal to obtain a second perception signal;
[0060] In this step, the perception signals collected by different perception signal collection channels have different dimensions and contain a lot of interference and random noise, making it difficult to extract features from the first perception signal. Therefore, preprocessing must be performed to reduce noise and other interference and improve the recognition of the features of the first perception signal.
[0061] As a specific example, for the collected P300 EEG signal, there will be interference from different frequency bands such as electromyography and power frequency. The electromyography frequency band is above 30Hz, the power frequency is around 50Hz, and the P300 signal is concentrated in the low-frequency area below 30Hz. It is necessary to separate the original signal from the interference artifacts in the spectrum. Filtering can achieve denoising to a certain extent.
[0062] Step S103: performing a wavelet transform on the second perception signal, extracting the time domain energy entropy of the second perception signal after the wavelet transform, and using the time domain energy entropy as the third perception signal;
[0063] In this step, wavelet transform describes the second perception signal from the time and scale planes. It is an algorithm that decomposes the perception signal into multiple mother wavelets and then superimposes them. Superposition and averaging can enhance the second perception signal. The effectiveness of this method can be verified by mathematical derivation. Let pi(t) represent the second perception signal, that is, the target signal s that is worth studying in the second perception signal is included. i (t) and interference signal q i (t):
[0064] p i (t) = s i (t)+q i (t)
[0065] After N times of superposition and averaging operations, we can get:
[0066]
[0067] Assume that the interference signal has a mean of zero and a variance of δ 2 , the fragments intercepted by different time windows are unrelated to each other, and the signal variance after superposition and averaging operation is:
[0068]
[0069] It can be seen that the variance δ of the interference signal after superposition and averaging is 2 change becomes the original variance δ 2 of The signal-to-noise ratio is significantly enhanced. Therefore, the wavelet transform based on the superposition principle has a strong multi-resolution characteristic and provides a "time-frequency" window that changes with frequency. For any signal f(t), its corresponding continuous wavelet transform is:
[0070]
[0071] The corresponding inverse continuous wavelet transform is:
[0072]
[0073] Parameter a represents the parameter that affects the window shape; parameter b represents the parameter that affects the window position. By influencing the window shape parameter a and the window position parameter b, high frequency resolution and low time resolution of the low-frequency part of the signal can be quickly achieved; as well as low frequency resolution and high time resolution of the high-frequency part of the signal. Therefore, wavelet transform has good time-frequency characteristics.
[0074] In this step, the time domain energy entropy can not only characterize the complexity of the perception signal in the case of time domain changes, but also characterize the characteristics of the perception signal in many frequency domains, and can also reflect the energy distribution characteristics of the perception signal in the time domain and frequency domain. This is because the time domain energy entropy can well reflect the energy distribution information of the collected perception signal in the frequency space.
[0075] In conjunction with wavelet transform, the second perceptual signal after wavelet transform is processed in time domain average, with the purpose of extracting the frequency bands containing P300 and excluding P300. According to the definition of Shannon entropy in information theory, the calculation of time domain energy entropy is as follows:
[0076]
[0077] in,
[0078] E i represents the time domain energy entropy of the i-th signal, P i represents the energy of the i-th segment of the signal, P is the total energy of the signal, and n is the number of segments into which the signal is divided. Energy entropy describes the distribution of signal energy in the time domain: the larger the energy entropy, the greater the proportion of the corresponding time domain segment energy in the total signal energy. As can be seen from the figure, the reconstructed signal containing P300 concentrates most of its energy in the time period of about 300ms, while the reconstructed signal without P300 is more evenly distributed in the entire time domain.
[0079] Step S104: recognizing the third perception signal through a convolutional neural network with an attention layer to obtain a recognition result of the third perception signal.
[0080] In this step, after the first sensory signal is transformed into the third sensory signal through various operations, CNN can compare it to image data and use the combination of convolution units to train the classification model of P300 signal (evoked EEG signal). The attention-enhanced CNN model can use CNN to automatically identify and extract high-level features of the overall data in the convolution operation, and use the Attention mechanism to identify more fine-grained high-level feature differences, so that the model network can extract the most discriminative features. Therefore, the CNN model classifier based on the use of the attention mechanism enhancement is a good EEG signal classification method.
[0081] As a specific example, a four-layer CNN network based on Cecotti and Grase can be used. After the first two convolutional layers extract the spatial and temporal features of the third sensory signal respectively, an attention layer is added. Finally, the extracted features are expanded into one dimension and input into a fully connected layer (MLP) to complete the classification task. In addition, different data sets with induced EEG delay hypotheses and different hyperparameters can be used, and BN layers and Dropout layers can be introduced to improve the model, thereby improving the generalization ability of the model and accelerating the training of the model.
[0082] In some other embodiments, reference Figure 2 , before acquiring the first perception signal through the perception signal acquisition channel combination in the aforementioned embodiment, the method further includes:
[0083] Step S201: acquiring a training data set collected through a plurality of perception signal collection channels, wherein the training data set includes a plurality of training perception signals;
[0084] In this embodiment, the first perception signal obtained by the perception signal acquisition channel combination belongs to multi-channel signal data. Such multi-channel time series data will introduce a large number of interference factors when participating in classification modeling. When the model is calculated based on the channel signal data, it will also cause waste due to the large amount of computing resources, thereby affecting the accuracy.
[0085] In order to obtain the optimal combination of sensory signal acquisition channels, it is necessary to perform feature extraction on the original channel data, remove irrelevant or redundant features, sort the importance of sensory signal acquisition channels, and filter out feature combinations that are easy to classify and robust. The starting points of feature selection usually include two points:
[0086] The first point is whether the feature is divergent. Information has information entropy, which represents the complexity and amount of information. It is generally believed that the greater the information entropy of feature data, the better the divergence, and the easier it is to participate in subsequent tasks.
[0087] The second point is whether there is correlation between features. Features with strong correlation will change closely with each other, and appropriately discarding strongly correlated features will help save computing resources.
[0088] Therefore, in this embodiment, a multi-model integrated embedding method is used to solve the optimal perception signal acquisition channel combination. The multi-model integrated embedding method is a feature sorting method based on a machine learning model. The focus of this method is to select an appropriate machine learning model algorithm, train the machine learning model on the processed full feature dimension data, and establish a corresponding model for each dimension feature and the corresponding known variable label.
[0089] Step S202: input the training data set into a random forest classifier model, an ExtraTree classifier model, and an Adaboost classifier model, and obtain sorting results of a plurality of the perception signal acquisition channels corresponding to the random forest classifier model, the ExtraTree classifier model, and the Adaboost classifier model, respectively;
[0090] In some embodiments, step S202 may specifically include:
[0091] Step 1: inputting the training data set into a random forest classifier model, wherein the random forest classifier model performs a feature selection operation and a splitting operation on the plurality of perception signals in the training data set to obtain a first weight factor set corresponding to the features of the plurality of perception signals; and sorting the plurality of perception signal acquisition channels according to the first weight factor set to obtain a first sorting result corresponding to the plurality of perception signal acquisition channels;
[0092] In this step, random forest is a kind of bagging ensemble algorithm, which refers to integrating multiple classifiers in parallel. CART decision tree is the core of random forest algorithm and the base learner of random forest. Then, the model is trained for T rounds by random sampling and random feature selection, and then T weak learners are added together to get the final learner.
[0093] Random forest performs self-sampling on samples, that is, sampling with replacement. The number of samples taken in each round of training is equal to the total number of samples. Because of the replacement sampling, some samples may be sampled repeatedly, and some samples may not be sampled. Feature selection also randomly selects a fixed number of features to train the model. This ensures that the obtained weak learners are not completely independent of each other, but have a small correlation, which can increase the overall generalization ability of the model. The process of the random forest model is as follows;
[0094] (1) First, the input is a sample set D = (x1, y1), (x2, Y2), ..... (x m ,y m );
[0095] (2) Randomly select training data sets and sample features for T rounds of training, for t = 1, 2, ... T;
[0096] a) Perform random sampling on the training set for the tth time, a total of m times, to obtain a sampling set D containing m samples t
[0097] b) Using sampling set D t Train the tth decision tree model G t(x) When training the nodes of the decision tree model, a part of the sample features are selected from all the sample features on the node, and an optimal feature is selected from these randomly selected sample features to divide the decision tree into left and right subtrees.
[0098] (3) Output the final strong learner F(x);
[0099] (4) If it is a classification algorithm prediction, the category or one of the categories that the T weak learners cast the most votes is the final category. If it is a regression algorithm, the arithmetic average of the regression results obtained by the T weak learners is the final model output.
[0100] Step 2: input the training data set into the ExtraTree classifier model, and the ExtraTree classifier model performs feature randomization, parameter randomization, model randomization, and split randomization operations on the plurality of perception signals in the training data set to obtain a second weight factor set corresponding to the features of the plurality of perception signals; sort the plurality of perception signal acquisition channels according to the second weight factor set to obtain a second sorting result corresponding to the plurality of perception signal acquisition channels;
[0101] In this step, the ET or Extra-Trees (Extremely randomized trees) algorithm is very similar to the random forest algorithm, both of which are composed of many decision trees. The main difference between the extreme tree and the random forest is that the randomForest uses the bagging model, and the extraTree uses all samples, but the features are randomly selected. Because the split is random, it is better than the random forest in some ways. The random forest obtains the best bifurcation attribute in a random subset, while the ET obtains the bifurcation value completely randomly, thereby bifurcation of the decision tree.
[0102] Step 3: Input the training data set into an Adaboost classifier model, wherein the Adaboost classifier model performs a feature selection operation on the multiple perception signals in the training data set to obtain a third weight factor set corresponding to the features of the multiple perception signals; and sort the multiple perception signal acquisition channels according to the third weight factor set to obtain a third sorting result corresponding to the multiple perception signal acquisition channels.
[0103] In this step, the Boosting method is a type of algorithm in ensemble learning, and the core idea is to promote weak learners to strong learners. The so-called weak learner is a learner whose generalization performance is slightly better than random guessing. Adaptive Boosting means adaptive improvement, which means that the algorithm can adapt to the training error rate of each weak classifier. The two most core points of this algorithm are as follows: Pay more attention to the misclassified samples: increase their sample weights in the next round of training, so each successive classifier is forced to focus on the training data that was misclassified in the previous step. Voting synthesis (the last step in the algorithm): The weight of the weak classifier with a small classification error rate is large, and the weight of the weak classifier with a large classification error rate is small.
[0104] The Adaboost algorithm can be briefly described as three steps:
[0105] (1) First, the weight distribution D1 of the training data is initialized. Assuming there are N training sample data, each training sample is initially assigned the same weight:
[0106] w i =1 / N
[0107] (2) Iterate m=1, 2, .....M
[0108] a) Using the initial weight w i Select a weak classifier G with the minimum current error m (x) conducting training;
[0109] b) And calculate the error of the weak classifier on distribution D1:
[0110]
[0111] c) Calculate the weak classifier G m (x) The weight of the final classifier:
[0112] α m =log((1-err m ) / err m )
[0113] d) Update the weight distribution of training samples:
[0114] w i ←w i ·exp[α m ·I(y i ≠G m (x i ))], i = 1, 2, ..., N
[0115] (3) Finally, the weak classifiers are combined according to their weights, and a strong classifier is obtained through the action of the sign function:
[0116]
[0117] After the training process of each weak classifier is completed, the weight of the weak classifier with a small classification error rate is increased so that it plays a greater decisive role in the final classification function, while the weight of the weak classifier with a large classification error rate is reduced so that it plays a smaller decisive role in the final classification function. In other words, the weak classifier with a low error rate has a larger weight in the final classifier, otherwise it is smaller.
[0118] Step S203: Draw a heat map for measuring the correlation between the plurality of the perception signal acquisition channels according to the sorting result and the Pearson correlation coefficient, and select a preset number of the perception signal acquisition channels according to the heat map to form the perception signal acquisition channel combination.
[0119] In some embodiments, step S203 may specifically include:
[0120] Performing feature screening on the plurality of perception signal acquisition channels according to the first sorting result, the second sorting result, and the third sorting result to obtain a fourth weight factor set for the plurality of perception signal acquisition channels; sorting the plurality of perception signal acquisition channels according to the fourth weight factor set to obtain a fourth sorting result corresponding to the plurality of perception signal acquisition channels;
[0121] A heat map for measuring the correlation between the plurality of perception signal acquisition channels is drawn according to the fourth sorting result and the Pearson correlation coefficient.
[0122] In this embodiment, model ensemble learning using a model ensemble embedding method is used to correctly combine multiple weak models to obtain a more accurate and robust model.
[0123] Therefore, in the present embodiment, the three models of random forest, ExtraTree and Adaboost are integrated, and the integrated model uses the feature selection technology of supervised data. When the relationship between the feature dimension and the response variable is nonlinear, the sample features are determined by tree-based methods (decision tree, random forest), etc. The tree-based method is relatively easy to use, and does not require too much debugging when modeling nonlinear relationships. By scoring the features through this training, the importance score of the features can be obtained, thereby obtaining the optimal channel combination for each subject. When using random forest, ExtraTree, Adaboost and integrated models to solve this problem, it is possible to take into account the differences in the eigenvalues of different subjects, weigh the degree of association and contribution, and obtain a better feature sorting result.
[0124] In this embodiment, based on the signal channel ranking results generated by the above three single models, the three-model integrated embedding method is used to perform feature screening on the signal channel of each subject again, and the weight factors of each feature are obtained, and then the perceptual signal channels are sorted according to the weight factors. The Pearson correlation coefficient is then used to measure the correlation between multiple channels, and a heat map is drawn to eliminate strongly correlated features. The Pearson correlation coefficient measures the degree of correlation between different channels.
[0125] In other embodiments, when multiple subjects collect corresponding first perception signals respectively, the integrated model can be used again to sort the feature weights according to the optimal EEG channel combination of each subject, taking into account the channel contribution of each subject, to obtain a set of optimal channel combinations suitable for all subjects.
[0126] In some other embodiments, reference Figure 3 , the step of performing time window selection processing on the first perception signal described in the foregoing embodiment includes:
[0127] Step S301: intercepting the first perception signal using a time window of a preset time length;
[0128] Step S302: Mark all data sampling points in the intercepted first perception signal as first labels, and mark other sampling points as second labels;
[0129] Step S303: retain the first label of one of the plurality of continuous data implementation sampling points, and mark the other data implementation sampling points in the plurality of data implementation sampling points as second labels to obtain the second perception signal.
[0130] As a specific example, the P300 EEG signal has a positive peak about 300 milliseconds after a low-probability stimulus occurs. In order to locate the P300 signal, a time window of 100-600 milliseconds is usually used to capture the original signal. Sampling point marking is implemented for the captured data, and multiple sampling points that receive the P300 stimulus signal are marked as "1", otherwise they are marked as "0" to facilitate the subsequent binary classification of the signal. Furthermore, the sampling points that are continuously marked as "1" are compressed and identified as the initial stimulus point "1", and the rest are not retained, and the stimulus signal is defaulted to "0".
[0131] In some other embodiments, reference Figure 4 , for the aforementioned embodiment, identifying the third perception signal by using a convolutional neural network with an attention layer, obtaining a recognition result of the third perception signal includes:
[0132] Step S401: input the third perception signal into the convolution layer of the convolutional neural network, and extract spatial domain features and temporal domain features of the third perception signal through the convolution layer;
[0133] Step S402: extracting high-level features of the third perception signal through the attention layer;
[0134] Step S403: input the spatial domain features, temporal domain features and high-level features into the fully connected layer of the convolutional neural network for classification to obtain the recognition result.
[0135] A specific implementation scenario of the present application is given below. The goal is to analyze the P300 brain-computer interface experimental data of 5 healthy adult subjects (S1-S5), extract the P300 potential characteristics of the subjects' gaze at the "target characters", obtain the corresponding classification and recognition method, and then find out the 10 target characters to be recognized in the subject test set.
[0136] A total of 5 subjects participated in five rounds of experiments. In each round of the experiment, a gray mark will be used in the upper left corner to prompt the subject to look at the "target character", where the "target character" is located in a character matrix consisting of 36 characters. The subjects are required to identify the "target character" in a row or column of the character matrix that flashes in a random order. The interval time between flashing rows and columns is 160 milliseconds, and the potential sampling frequency is 250Hz. When the subject recognizes the "target character", a P300 potential will appear in the EEG signal. In order to collect the EEG signals of the subjects, the experiment adopts the international 10-20 system electrode channel distribution, which is characterized by odd numbers on the left and even numbers on the right; smaller numbers are used close to the midline and larger numbers are used away from the midline; the electrode name includes the first letter of the head partition where the electrode is located. The experiment selected 20 of the potentials as the perceptual signal acquisition channels in the information service. The collected training data has the following characteristics, see Table 1;
[0137] (1) Each subject collected 20 potential data for each of the 12 known target characters "B", "D", "G", "L", "O", "Q", "S", "V", "Z", "4", "7", and "9", and obtained 12 sampling data tables, with a total of 186,702 samples;
[0138] (2) Each subject took 40 samples during the interval between flickering periods. One round of the experiment took 1920 milliseconds, and the total number of potential samples was at least 480.
[0139] (3) A positive peak appeared about 300 milliseconds after each subject recognized the “target character”, indicating that there was a signal delay. The total number of samples in five rounds was greater than 2400.
[0140] Table 1. Statistics of training data characteristics
[0141]
[0142]
[0143] According to the control variable criterion, except for the “target character” not being provided in the test data, the perceptual signal collection channels and the test subjects in the remaining information services remained consistent.
[0144] In order to judge the 10 characters to be recognized in the test set of 5 subjects, it is necessary to filter and extract features from the training data, classify the P300 potential patterns of different subjects fixating on the "target characters", and realize target discrimination through the recognition method of perceptual signals in the information service of the embodiment of the present application, and give specific characters to be recognized.
[0145] (1) Data. Assuming that the P300 EEG signal delay is 300ms\400ms\600ms respectively, the EEG data is labeled, and three training sets are collected. The distributed input model is trained to obtain the best classification accuracy of the test set. (2) Hyperparameters. The manually set hyperparameters of the neural network are set according to the specific business scenario. The quality of the hyperparameters often affects the convergence speed and model accuracy of the model. In this experiment, the specific hyperparameter selection is shown in Table 2 below. The dimensions of the two convolution kernels are (10, 1) and (1, 20) respectively; the dropout value of the two fully connected layers is 0.5, the batch_size is 36, the epoch is 30, the optimizer compares Adam and SGD, the learning rate is 1e-5, and the data set is split into 28, and the experiment is conducted using 10-fold cross validation.
[0146] Table 2 Hyperparameter setting reference table
[0147]
[0148] According to the recognition method of perceptual signals in the information service of the embodiment of the present application and several related methods such as manual character recognition, it is known that the remaining characters char13-char22 of subject S1 are M, 4, 5, 2, Y, P, J, V, I and U respectively. This result is obtained by first identifying the P300 signal and retrieving the stimulated row and column flashes using the convolutional neural network P300 signal classification algorithm enhanced by the attention mechanism, and then identifying the target test characters. Similarly, the target characters to be recognized of subjects S2, S3, S4 and S5 are obtained, and the result Table 3 is prepared as follows.
[0149] Table 3 Recognition character table of 5 subjects in the test set (test_data)
[0150]
[0151]
[0152] The P300 event-related potential is a positive peak relative to the baseline that appears about 300 milliseconds after the target character flickering stimulus occurs. Due to individual differences, the occurrence time of P300 is also different, and the P300 waveform will also appear about 450 milliseconds after the stimulus occurs. As an endogenous component, the P300 potential is not affected by the physical characteristics of the stimulus, but is related to perception or perceptual psychological activities, and is closely related to processing processes such as attention, memory, and intelligence.
[0153] In some embodiments, the training set is manually annotated from train_data and train_event of subjects S1-S5. The primary task of the annotation is to confirm the delay time of the P300 signal. Assuming that the P300 signal is delayed by 300ms, 400ms and 800ms, three training sets are input into the classification model. According to the prediction accuracy achieved by the model, the Percision score, Recall score and F1-Score score on the test set, the training set of the P300 signal is more accurate under the assumed delay of 300ms. The upper limit of the model depends on the data quality, and this part of the operation optimizes the performance of the subsequent model.
[0154] It can be seen that in this embodiment, through the recognition method of perceptual signals in the information service of the embodiment of the present application, the processed data is classified and learned, and the built Attention-enhanced CNN model can use CNN to automatically identify and extract high-level features of the overall data in the convolution operation, and use the Attention mechanism to identify more fine-grained high-level feature differences, so that the model network extracts the most discriminative features and effectively recognizes the P300 signal. Then predict the characters to be recognized in the test data set of the subject.
[0155] It should be noted that the method of the embodiment of the present application can be performed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario and completed by multiple devices cooperating with each other. In the case of such a distributed scenario, one of the multiple devices can only perform one or more steps in the method of the embodiment of the present application, and the multiple devices will interact with each other to complete the described method.
[0156] It should be noted that the above describes some embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the above embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0157] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present application also provides a device for identifying perceptual signals in information services.
[0158] refer to Figure 5 , the recognition device of the perception signal in the information service comprises:
[0159] The acquisition module 501 is configured to acquire a first perception signal through a perception signal acquisition channel combination;
[0160] A first processing module 502 is configured to perform filtering processing and time window selection processing on the first perception signal to obtain a second perception signal;
[0161] The second processing module 503 is configured to perform a wavelet transform on the second perception signal, extract a time domain energy entropy of the second perception signal after the wavelet transform, and use the time domain energy entropy as a third perception signal;
[0162] The recognition module 504 is configured to recognize the third perception signal through a convolutional neural network with an attention layer to obtain a recognition result of the third perception signal.
[0163] In some other embodiments, the apparatus for identifying a perceptual signal in an information service further includes a selection module 500, and the selection module 500 is configured to:
[0164] Acquire a training data set collected through multiple perception signal collection channels, wherein the training data set includes multiple training perception signals;
[0165] Inputting the training data set into a random forest classifier model, an ExtraTree classifier model, and an Adaboost classifier model, and obtaining ranking results of a plurality of the perception signal acquisition channels corresponding to the random forest classifier model, the ExtraTree classifier model, and the Adaboost classifier model, respectively;
[0166] A heat map for measuring the correlation between the plurality of the perception signal acquisition channels is drawn according to the sorting result and the Pearson correlation coefficient, and a preset number of the perception signal acquisition channels are selected according to the heat map to form the perception signal acquisition channel combination.
[0167] In some other embodiments, the first processing module 502 is further configured to:
[0168] Using a time window of a preset time length to intercept the first perception signal;
[0169] Marking sampling points of all data in the intercepted first perception signal as first labels, and marking other sampling points as second labels;
[0170] The first label of one of the plurality of continuous data implementation sampling points is retained, and the other data implementation sampling points of the plurality of data implementation sampling points are marked as second labels to obtain the second perception signal.
[0171] In some other embodiments, the identification module 504 is further configured to:
[0172] Inputting the third perception signal into the convolution layer of the convolutional neural network, and extracting spatial domain features and temporal domain features of the third perception signal through the convolution layer;
[0173] Extracting high-level features of the third perception signal through the attention layer;
[0174] The spatial domain features, time domain features and high-level features are input into the fully connected layer of the convolutional neural network for classification to obtain the recognition result.
[0175] In some other embodiments, the selection module 500 is further configured to:
[0176] Inputting the training data set into a random forest classifier model, wherein the random forest classifier model performs a feature selection operation and a splitting operation on the plurality of perception signals in the training data set to obtain a first weight factor set corresponding to the features of the plurality of perception signals; sorting the plurality of perception signal acquisition channels according to the first weight factor set to obtain a first sorting result corresponding to the plurality of perception signal acquisition channels;
[0177] The training data set is input into the ExtraTree classifier model, and the ExtraTree classifier model performs feature randomization, parameter randomization, model randomization, and split randomization operations on the plurality of perception signals in the training data set to obtain a second weight factor set corresponding to the features of the plurality of perception signals; the plurality of perception signal acquisition channels are sorted according to the second weight factor set to obtain a second sorting result corresponding to the plurality of perception signal acquisition channels;
[0178] The training data set is input into an Adaboost classifier model, and the Adaboost classifier model performs a feature selection operation on the multiple perception signals in the training data set to obtain a third weight factor set corresponding to the features of the multiple perception signals; and the multiple perception signal acquisition channels are sorted according to the third weight factor set to obtain a third sorting result corresponding to the multiple perception signal acquisition channels.
[0179] In some other embodiments, the selection module 500 is further configured to:
[0180] Performing feature screening on the plurality of perception signal acquisition channels according to the first sorting result, the second sorting result, and the third sorting result to obtain a fourth weight factor set for the plurality of perception signal acquisition channels; sorting the plurality of perception signal acquisition channels according to the fourth weight factor set to obtain a fourth sorting result corresponding to the plurality of perception signal acquisition channels;
[0181] A heat map for measuring the correlation between the plurality of perception signal acquisition channels is drawn according to the fourth sorting result and the Pearson correlation coefficient.
[0182] For the convenience of description, the above device is described in terms of functions divided into various modules. Of course, when implementing the present application, the functions of each module can be implemented in the same or multiple software and / or hardware.
[0183] The device of the above embodiment is used to implement the method for identifying the perception signal in the corresponding information service in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here.
[0184] Based on the same inventive concept, corresponding to any of the above-mentioned embodiments and methods, the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for identifying perceptual signals in the information service described in any of the above embodiments is implemented.
[0185] Figure 6A more specific schematic diagram of the hardware structure of an electronic device provided in this embodiment is shown, and the device may include: a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040 are connected to each other through the bus 1050 in the device.
[0186] The processor 1010 can be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.
[0187] The memory 1020 may be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 may store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program codes are stored in the memory 1020 and are called and executed by the processor 1010.
[0188] The input / output interface 1030 is used to connect the input / output module to realize information input and output. The input / output module can be configured in the device as a component (not shown in the figure), or it can be externally connected to the device to provide corresponding functions. The input device may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device may include a display, a speaker, a vibrator, an indicator light, etc.
[0189] The communication interface 1040 is used to connect a communication module (not shown) to realize communication interaction between the device and other devices. The communication module can realize communication through a wired mode (such as USB, network cable, etc.) or a wireless mode (such as mobile network, WIFI, Bluetooth, etc.).
[0190] The bus 1050 includes a path that transmits information between the various components of the device (eg, the processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040).
[0191] It should be noted that, although the above device only shows the processor 1010, the memory 1020, the input / output interface 1030, the communication interface 1040 and the bus 1050, in the specific implementation process, the device may also include other components necessary for normal operation. In addition, it can be understood by those skilled in the art that the above device may also only include the components necessary for implementing the embodiments of the present specification, and does not necessarily include all the components shown in the figure.
[0192] The electronic device of the above embodiment is used to implement the method for identifying the perception signal in the corresponding information service in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here.
[0193] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present application also provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the method for identifying perceptual signals in information services as described in any of the above embodiments.
[0194] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.
[0195] The computer instructions stored in the storage medium of the above embodiment are used to enable the computer to execute the method for identifying perceptual signals in information services as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0196] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present application (including the claims) is limited to these examples. In line with the concept of the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.
[0197] In addition, to simplify the description and discussion, and in order not to make the embodiments of the present application difficult to understand, the known power supply / ground connection with the integrated circuit (IC) chip and other components may or may not be shown in the provided drawings. In addition, the device can be shown in the form of a block diagram to avoid making the embodiments of the present application difficult to understand, and this also takes into account the fact that the details of the implementation of these block diagram devices are highly dependent on the platform to be implemented in the embodiments of the present application (that is, these details should be fully within the scope of understanding of those skilled in the art). In the case of elaborating specific details (e.g., circuits) to describe exemplary embodiments of the present application, it is obvious to those skilled in the art that the embodiments of the present application can be implemented without these specific details or when these specific details are changed. Therefore, these descriptions should be considered to be illustrative rather than restrictive.
[0198] Although the present application has been described in conjunction with specific embodiments of the present application, many replacements, modifications and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may use the embodiments discussed.
[0199] The embodiments of the present application are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present application should be included in the scope of protection of the present application.
Claims
1. A method for identifying perceptual signals in information services, characterized in that: include: Acquire a first perception signal through a perception signal acquisition channel combination; Performing filtering processing and time window selection processing on the first perception signal to obtain a second perception signal; performing a wavelet transform on the second perception signal, extracting time domain energy entropy of the second perception signal after the wavelet transform, and using the time domain energy entropy as a third perception signal; The third perception signal is identified by a convolutional neural network having an attention layer to obtain a recognition result of the third perception signal; wherein, before acquiring the first perception signal by the perception signal acquisition channel combination, the method further includes: Acquire a training data set collected through multiple perception signal collection channels, wherein the training data set includes multiple training perception signals; Inputting the training data set into a random forest classifier model, an ExtraTree classifier model, and an Adaboost classifier model, and obtaining ranking results of a plurality of the perception signal acquisition channels corresponding to the random forest classifier model, the ExtraTree classifier model, and the Adaboost classifier model, respectively; A heat map for measuring the correlation between the plurality of the perception signal acquisition channels is drawn according to the sorting result and the Pearson correlation coefficient, and a preset number of the perception signal acquisition channels are selected according to the heat map to form the perception signal acquisition channel combination.
2. The method according to claim 1, characterized in that The performing time window selection processing on the first perception signal includes: Using a time window of a preset time length to intercept the first perception signal; Marking sampling points of all data in the intercepted first perception signal as first labels, and marking other sampling points as second labels; The first label of one of the plurality of continuous data implementation sampling points is retained, and the other data implementation sampling points of the plurality of data implementation sampling points are marked as second labels to obtain the second perception signal.
3. The method according to claim 1, characterized in that The identifying the third perception signal by a convolutional neural network having an attention layer to obtain the identification result of the third perception signal includes: Inputting the third perception signal into the convolution layer of the convolutional neural network, and extracting spatial domain features and temporal domain features of the third perception signal through the convolution layer; Extracting high-level features of the third perception signal through the attention layer; The spatial domain features, time domain features and high-level features are input into the fully connected layer of the convolutional neural network for classification to obtain the recognition result.
4. The method according to claim 1, characterized in that: The step of inputting the training data set into a random forest classifier model, an ExtraTree classifier model, and an Adaboost classifier model, and obtaining sorting results of a plurality of the perception signal acquisition channels corresponding to the random forest classifier model, the ExtraTree classifier model, and the Adaboost classifier model, respectively, comprises: Inputting the training data set into a random forest classifier model, wherein the random forest classifier model performs a feature selection operation and a splitting operation on the plurality of perception signals in the training data set to obtain a first weight factor set corresponding to the features of the plurality of perception signals; sorting the plurality of perception signal acquisition channels according to the first weight factor set to obtain a first sorting result corresponding to the plurality of perception signal acquisition channels; The training data set is input into the ExtraTree classifier model, and the ExtraTree classifier model performs feature randomization, parameter randomization, model randomization, and split randomization operations on the plurality of perception signals in the training data set to obtain a second weight factor set corresponding to the features of the plurality of perception signals; the plurality of perception signal acquisition channels are sorted according to the second weight factor set to obtain a second sorting result corresponding to the plurality of perception signal acquisition channels; The training data set is input into an Adaboost classifier model, and the Adaboost classifier model performs a feature selection operation on the multiple perception signals in the training data set to obtain a third weight factor set corresponding to the features of the multiple perception signals; and the multiple perception signal acquisition channels are sorted according to the third weight factor set to obtain a third sorting result corresponding to the multiple perception signal acquisition channels.
5. The method according to claim 4, characterized in that Drawing a heat map for measuring the correlation between the plurality of sensory signal acquisition channels according to the sorting result and the Pearson correlation coefficient includes: Performing feature screening on the plurality of perception signal acquisition channels according to the first sorting result, the second sorting result, and the third sorting result to obtain a fourth weight factor set for the plurality of perception signal acquisition channels; sorting the plurality of perception signal acquisition channels according to the fourth weight factor set to obtain a fourth sorting result corresponding to the plurality of perception signal acquisition channels; A heat map for measuring the correlation between the plurality of perception signal acquisition channels is drawn according to the fourth sorting result and the Pearson correlation coefficient.
6. The method according to claim 1, characterized in that The perception signal is a P300 perception signal.
7. A device for identifying perceptual signals in information services, characterized in that: include: A collection module, configured to obtain a first perception signal through a combination of perception signal collection channels; A first processing module is configured to perform filtering processing and time window selection processing on the first perception signal to obtain a second perception signal; A second processing module is configured to perform a wavelet transform on the second perception signal, extract a time domain energy entropy of the second perception signal after the wavelet transform, and use the time domain energy entropy as a third perception signal; A recognition module is configured to recognize the third perception signal through a convolutional neural network with an attention layer to obtain a recognition result of the third perception signal; Wherein, the acquisition module is further configured to: acquire a training data set acquired through multiple perception signal acquisition channels, wherein the training data set includes multiple training perception signals; Inputting the training data set into a random forest classifier model, an ExtraTree classifier model, and an Adaboost classifier model, and obtaining ranking results of a plurality of the perception signal acquisition channels corresponding to the random forest classifier model, the ExtraTree classifier model, and the Adaboost classifier model, respectively; A heat map for measuring the correlation between the plurality of the perception signal acquisition channels is drawn according to the sorting result and the Pearson correlation coefficient, and a preset number of the perception signal acquisition channels are selected according to the heat map to form the perception signal acquisition channel combination.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable by the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 6 is implemented.
9. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to make a computer execute the method according to any one of claims 1 to 6.
