CN-121996077-A - User intention recognition method and device and terminal equipment

CN121996077ACN 121996077 ACN121996077 ACN 121996077ACN-121996077-A

Abstract

The application provides a user intention recognition method, a device and terminal equipment, which are suitable for the technical field of data processing, wherein the method comprises the steps of performing space-time slicing and projection processing on a plurality of electroencephalogram signals to be processed to generate a plurality of electroencephalogram signal sequence information to be processed; masking the plurality of electroencephalogram signal sequence information to be processed according to a preset masking ratio and a preset dynamic masking processing rule to generate a plurality of electroencephalogram signal masking sequence information, and obtaining a target intention recognition model according to the plurality of electroencephalogram signal masking sequence information and the initial intention recognition model. The method and the system remarkably improve the robustness of the model to complex noise such as vehicle vibration, electromagnetic interference and the like and the generalization capability of the model across users, realize accurate recognition of the motor imagery intention of the driver with low delay and high accuracy, and provide stable and efficient decision support for the non-sensing brain control and man-machine cooperation of the intelligent cabin.

Inventors

GAO FEI
HE JILIANG
ZHAO RUI
WANG ZHIQIANG
GAO ZHENHAI

Assignees

吉林大学

Dates

Publication Date: 20260508
Application Date: 20260410

Claims (10)

1. A method for identifying user intention, comprising: acquiring a plurality of user brain electrical signals; Generating user intention recognition information according to the plurality of user electroencephalogram signals and the target intention recognition model; The target intention recognition model is obtained through the following steps: Acquiring a plurality of electroencephalogram signals to be processed; performing space-time slicing and projection processing on the plurality of electroencephalogram signals to be processed to generate a plurality of electroencephalogram signal sequence information to be processed; Masking the plurality of electroencephalogram signal sequence information to be processed according to a plurality of preset mask ratios and preset dynamic mask processing rules to generate a plurality of electroencephalogram signal mask sequence information; And obtaining a target intention recognition model according to the plurality of electroencephalogram signal mask sequence information and the initial intention recognition model.
2. The method for recognizing user intention according to claim 1, wherein the step of performing space-time slicing and projection processing on the plurality of to-be-processed electroencephalogram signals to generate a plurality of to-be-processed electroencephalogram signal sequence information specifically comprises: performing space-time slicing processing on the plurality of electroencephalogram signals to be processed according to preset electroencephalogram signal space-time slicing size information to obtain a plurality of electroencephalogram signal space-time block information to be processed; generating a plurality of electroencephalogram signal linear projection information according to the plurality of electroencephalogram signal space-time block information to be processed and a preset electroencephalogram signal linear projection model; And generating a plurality of electroencephalogram sequence information to be processed according to the plurality of electroencephalogram linear projection information and preset electroencephalogram position coding information.
3. The user intent recognition method of claim 1, wherein, The preset mask ratio includes a preset first mask ratio and a preset second mask ratio; Wherein the preset first mask ratio is smaller than the preset second mask ratio; the plurality of electroencephalogram signal sequence information to be processed comprises a plurality of first electroencephalogram signal sequence information to be processed and a plurality of second electroencephalogram signal sequence information to be processed; The plurality of electroencephalogram signal mask sequence information comprises a plurality of first electroencephalogram signal mask sequence information and a plurality of second electroencephalogram signal mask sequence information; the step of performing mask processing on the plurality of electroencephalogram signal sequence information to be processed according to a plurality of preset mask ratios and preset dynamic mask processing rules to generate a plurality of electroencephalogram signal mask sequence information specifically comprises the following steps: generating first mask matrix information according to a preset first mask ratio and a preset dynamic mask processing rule; performing dot multiplication calculation according to the first mask matrix information and a plurality of first electroencephalogram signal sequence information to be processed, and generating a plurality of first electroencephalogram signal mask sequence information; generating second mask matrix information according to a preset second mask ratio and a preset dynamic mask processing rule; And performing dot multiplication calculation according to the second mask matrix information and the plurality of second electroencephalogram signal sequence information to be processed, and generating a plurality of second electroencephalogram signal mask sequence information.
4. The user intent recognition method of claim 1, wherein, The initial intention recognition model comprises an initial electroencephalogram signal coding feature generation sub-model, an initial intention recognition inference distribution generation sub-model, an initial intention recognition reference distribution generation sub-model and an initial intention recognition electroencephalogram signal reconstruction sub-model; The step of obtaining a target intention recognition model according to the plurality of electroencephalogram signal mask sequence information and the initial intention recognition model specifically comprises the following steps: Generating a sub-model according to the plurality of electroencephalogram signal mask sequence information and the initial electroencephalogram signal coding characteristics, and generating a plurality of electroencephalogram signal mask sequence characteristic information; generating a plurality of electroencephalogram mask sequence inferred distribution information and a plurality of electroencephalogram sequence reference distribution information according to the plurality of electroencephalogram mask sequence feature information, the plurality of electroencephalogram sequence information to be processed, the initial intention recognition inferred distribution generation sub-model and the initial intention recognition reference distribution generation sub-model; Deducing distribution information according to the plurality of electroencephalogram signal mask sequences and carrying out alignment processing on the plurality of electroencephalogram signal sequence reference distribution information to generate a plurality of aligned electroencephalogram signal mask sequence distribution information; Generating a plurality of electroencephalogram signal mask sequence reconstruction information according to the plurality of aligned electroencephalogram signal mask sequence distribution information and the initial intention recognition electroencephalogram signal reconstruction sub-model; according to the reconstruction information of the plurality of electroencephalogram signal mask sequences and the plurality of electroencephalogram signals to be processed, calculating to obtain reconstruction loss information of the electroencephalogram signals; And obtaining a target intention recognition model according to the initial intention recognition model, the electroencephalogram signal reconstruction loss information and a preset electroencephalogram signal reconstruction loss threshold value.
5. The method for recognizing user intention according to claim 4, wherein the step of generating the plurality of electroencephalogram mask sequence inferred distribution information and the plurality of electroencephalogram sequence reference distribution information from the plurality of electroencephalogram mask sequence feature information, the plurality of electroencephalogram sequence information to be processed, the initial intention recognition inferred distribution generation sub-model, and the initial intention recognition reference distribution generation sub-model specifically comprises: generating a submodel according to the characteristic information of the plurality of electroencephalogram signal mask sequences and the initial intention recognition inference distribution, and generating a plurality of electroencephalogram signal mask sequence inference distribution information; generating a sub-model according to the plurality of electroencephalogram signal sequence information to be processed and the initial intention recognition reference distribution, and generating a plurality of electroencephalogram signal sequence reference distribution information.
6. The method for recognizing user intention according to claim 1, wherein the step of generating the user intention recognition information based on the plurality of the user electroencephalogram signals and the target intention recognition model specifically comprises: performing spatial filtering processing according to the plurality of user brain electrical signals to obtain a plurality of filtered user brain electrical signals; Performing fine adjustment processing on the target intention recognition model according to the plurality of filtered user electroencephalograms to generate a fine-adjusted target intention recognition model; And generating user intention recognition information according to the plurality of user electroencephalogram signals and the target intention recognition model after fine adjustment.
7. The user intent recognition method of claim 6, wherein, The target intention recognition model comprises a target electroencephalogram signal coding feature generation sub-model and a target intention recognition information generation sub-model; the step of performing fine adjustment processing on the target intention recognition model according to the plurality of filtered user electroencephalograms to generate a fine-adjusted target intention recognition model specifically comprises the following steps: Generating a sub-model according to the plurality of filtered user electroencephalogram signals and target electroencephalogram signal coding characteristics, and generating a plurality of user electroencephalogram signal coding characteristic information; generating a sub-model according to the coding characteristic information of the plurality of the user electroencephalogram signals and the target intention identification information, and generating the user electroencephalogram signal intention identification information; and performing fine adjustment processing on the target intention recognition model according to the user electroencephalogram intention recognition information and preset user intention recognition label information to generate a fine-adjusted target intention recognition model.
8. A user intention recognition apparatus, comprising: The electroencephalogram signal acquisition module of the user, the method comprises the steps of acquiring a plurality of user brain electrical signals; And the user intention identification information generation module is used for generating user intention identification information according to the plurality of user electroencephalogram signals and the target intention identification model.
9. A terminal device, characterized in that it comprises a memory, a processor, on which a computer program is stored which is executable on the processor, the processor executing the computer program to carry out the steps of the method according to any one of claims 1 to 7.
10. A computer readable storage medium storing a computer program, characterized in that the computer program when executed by a processor implements the steps of the method according to any one of claims 1 to 7.

Description

User intention recognition method and device and terminal equipment Technical Field The application belongs to the technical field of data processing, and particularly relates to a user intention recognition method, a user intention recognition device and terminal equipment. Background With the rapid development of intelligent network-connected automobile and intelligent cabin technology, cabin man-machine interaction is being upgraded from physical buttons and touch control voice to multi-mode intention perception directions, and brain-machine interface technology based on brain-computer signals becomes a leading direction for realizing cabin noninductive control because the brain-machine interface technology can directly read motor imagination intention of a driver. The vehicle-mounted environment belongs to a typical strong noise and non-stable scene, the electroencephalogram signal of a driver is easily influenced by vehicle vibration, electromagnetic interference and self psychological state, how to extract high signal-to-noise ratio and high separability features becomes the key of the vehicle-mounted brain-computer interface technology landing, and the current motor imagery brain electrolysis code method mainly comprises three types of methods of traditional machine learning, conventional deep learning and a model generation. The traditional machine learning method adopts a co-space mode, a variant of the co-space mode and other artificial design feature extraction algorithms, and is matched with a support vector machine to finish classification, the traditional deep learning method utilizes a convolutional neural network to extract inter-electrode space correlation and a cyclic neural network to capture time sequence dependence, end-to-end feature learning and classification are realized, the method based on a generated model is used for enhancing electroencephalogram data and extracting features by reconstructing potential distribution of learning data through signals, and the method is used for directly processing electroencephalogram features in Euclidean space and finishing decoding classification. The traditional machine learning method is sensitive to non-stationary signals, has poor generalization capability and is easy to generate aliasing, is difficult to capture a global topological structure of a whole brain channel, lacks a self-adaptive recovery mechanism of signal loss and strong interference, has drastically reduced performance after being interfered, cannot capture a low-dimensional manifold geometric structure of an electroencephalogram signal based on a method for generating a model, has poor characteristic space decoupling effect, is easy to alias with different movement intention characteristics, has insufficient model generalization capability, and is difficult to adapt to a vehicle-mounted complex dynamic noise environment. Disclosure of Invention In view of the above, the embodiment of the application provides a method, a device and a terminal device for identifying user intention, which aim to solve the problems of difficult signal analysis, weak anti-interference capability, poor characteristic decoupling effect, insufficient generalization capability across users, easy aliasing of intention characteristic space and difficult accurate identification existing in a vehicle-mounted complex environment in the prior art. A first aspect of an embodiment of the present application provides a user intention recognition method, including: acquiring a plurality of user brain electrical signals; and generating user intention recognition information according to the plurality of user electroencephalogram signals and the target intention recognition model. One aspect of the embodiments of the present application provides a generating step of a target intention recognition model, including: Acquiring a plurality of electroencephalogram signals to be processed; performing space-time slicing and projection processing on the plurality of electroencephalogram signals to be processed to generate a plurality of electroencephalogram signal sequence information to be processed; Masking the plurality of electroencephalogram signal sequence information to be processed according to a plurality of preset mask ratios and preset dynamic mask processing rules to generate a plurality of electroencephalogram signal mask sequence information; And obtaining a target intention recognition model according to the plurality of electroencephalogram signal mask sequence information and the initial intention recognition model. A second aspect of an embodiment of the present application provides a user intention recognition apparatus, including: The electroencephalogram signal acquisition module of the user, the method comprises the steps of acquiring a plurality of user brain electrical signals; And the user intention identification information generation module is used for generating user intention identification