Flight interaction system driven by non-invasive electroencephalogram signals

Abstract

The invention discloses a flight interaction system driven by non-invasive electroencephalogram signals. The flight interaction system comprises an electroencephalogram collection module, a topology modeling module, an intention recognition module, an instruction mapping module, a flight control module, a multi-mode feedback module and an individual self-adaption module. According to the system, electroencephalogram signals are collected through non-intrusive equipment, topological features are extracted through a persistent coherent network, user intentions are recognized in combination with a depth model, recognition results are mapped into flight control instructions, and unmanned aerial vehicle control is achieved. The system constructs a closed-loop interaction process, improves the control stability through a feedback mechanism, dynamically optimizes an identification model based on a historical deviation sample, and enhances the adaptability and generalization ability of the system.

Description

Technical Field

[0001] The present invention relates to the fields of brain-computer interface technology and flight control and human-computer interaction system technology, and in particular to a non-invasive EEG signal-driven flight interaction system. Background Art

[0002] Brain-computer interfaces (BCIs), a key research direction at the intersection of neuroengineering and artificial intelligence, aim to transform information between EEG signals and external devices, providing a new interaction paradigm for applications such as disability rehabilitation, intelligent control, and augmented reality. Non-invasive EEG-based control systems, due to their convenience and safety, have become the mainstream path for the promotion of BCI applications. Enabling users to naturally control the flight platform through their thoughts is a key technical challenge, particularly in highly dynamic scenarios such as aircraft operation, remote navigation, and virtual flight simulation.

[0003] Most existing brain-controlled flight interaction systems use pattern recognition methods based on traditional EEG features. A typical processing flow is as follows: multi-channel EEG signals are first acquired using a non-invasive EEG acquisition device. Preprocessing (such as filtering, artifact removal, and normalization) is then performed, and features are extracted using classic methods such as power spectrum analysis, wavelet transform, event-related potentials, and regular spatial patterns. Classification models such as support vector machines, linear discriminant analysis, and convolutional neural networks are then used to identify user intent. Ultimately, the recognition results are converted into flight control commands to drive the drone or simulated flight platform.

[0004] Although the above methods have been initially applied in static control tasks (such as single command triggering, direction selection, etc.), there are still significant bottlenecks in complex and continuous flight missions. First, as a weak electrophysiological signal, EEG signals are highly volatile and subject to significant noise interference. Traditional time domain, frequency domain or linear spatial filtering methods are difficult to capture their high-dimensional nonlinear structural characteristics. This shallow feature description method is prone to problems such as decreased recognition accuracy and unclear category distinction in multi-task and multi-state scenarios, resulting in unstable intention output, which seriously affects the continuity and safety of flight missions.

[0005] Secondly, most current EEG control systems use a one-way decoding structure, a linear process of "acquisition-recognition-execution," lacking effective feedback mechanisms and closed-loop user perception channels. In flight control scenarios, due to the large spatial displacement, sensitive response delays, and continuous execution paths of tasks, it is difficult for users to judge whether the system response is consistent with their expectations based on their subjective intentions. Once the system recognizes deviations or execution lags, there is a lack of sensory feedback channels such as vision and touch to correct the current operating state, resulting in reduced control accuracy and a broken interactive experience, increasing the risk of flight errors.

[0006] Thirdly, existing EEG intention recognition systems are seriously deficient in their ability to adapt to individual differences. EEG signals are highly personalized, and the same task exhibits significant differences in neural responses between different users. Existing systems often rely on static model training with fixed parameters, making it difficult to adapt to the dynamic neural changes of different users. Some systems attempt to improve generalization capabilities through transfer learning or fine-tuning, but these systems have low adaptation efficiency, high sample requirements, and a lack of automatic feedback mechanisms during the training process, making continuous adaptive updates difficult. This severely limits the cross-population deployment and long-term operation capabilities of such systems.

[0007] In terms of flight control, traditional brain-controlled systems use mostly discrete state-transition commands, making it difficult to precisely express the user's intended continuous information, such as speed, attitude, and heading. This results in coarse platform execution granularity and delayed response. Furthermore, the mapping relationship between flight control commands and EEG intent often lacks structural hierarchy and parameter associations, relying solely on intent labels to drive fixed action templates. This results in a lack of flexibility and adjustability in the execution path, making it unable to meet the actual needs of high-precision operations and complex path planning.

[0008] From a signal modeling perspective, traditional methods ignore the global topological structure information that may be contained in EEG signals. In recent years, topological data analysis has been widely used in biosignal modeling, demonstrating its advantages in processing non-Euclidean spaces and extracting persistent homology features. Persistent homology networks can map high-dimensional EEG signals into point cloud topologies and extract stable characteristic parameters such as connected branches and loop persistence. These parameters, such as Betti numbers and persistent bar graphs, are used to represent structural stability and pattern transitions in neural dynamics. However, currently, no mature system has applied topological data analysis to brain-controlled flight systems, failing to fully realize its potential in disturbance rejection, structural perception, and pattern discrimination.

[0009] Therefore, there is an urgent need to build a closed-loop brain-controlled flight system that integrates topological modeling, depth recognition, control-level mapping and adaptive feedback to meet the actual needs of high robustness, high precision and high interactive experience in complex flight missions. Summary of the Invention

[0010] One purpose of the present invention is to propose a non-invasive EEG signal-driven flight interaction system. The present invention integrates a closed-loop system architecture of topological feature modeling, deep intent recognition, parametric control mapping and multimodal interactive feedback, which can significantly improve the stability, accuracy and interactive fluency of brain-controlled flight missions.

[0011] A non-invasive EEG signal-driven flight interaction system according to an embodiment of the present invention includes the following modules:

[0012] Non-invasive EEG signal acquisition module, used to collect the user's EEG signals;

[0013] A persistent coherence network module, used for topologically modeling the EEG signal and extracting characteristic parameters with global topological stability;

[0014] an EEG signal intention recognition module, configured to recognize the user's EEG control intention based on the characteristic parameters;

[0015] A flight control module, configured to map the EEG control intentions into control instructions for the flight platform;

[0016] A flight execution module, configured to control the flight operation of the flight platform according to the control instructions;

[0017] The feedback module is used to convert the status information of the flight platform into visual or tactile feedback signals and feed them back to the user.

[0018] A non-invasive EEG signal-driven flight interaction method according to an embodiment of the present invention includes the following steps:

[0019] S1. Collecting the user's EEG signals and obtaining EEG data through a non-invasive EEG signal acquisition device;

[0020] S2. performing topological modeling on the EEG signal, extracting global topological features of the EEG signal using a topological data analysis method of a persistent coherent network, and obtaining characteristic parameters with global topological stability;

[0021] S3. Inputting the characteristic parameters into an EEG signal intention recognition module to identify the user's EEG control intention and perform feature mapping;

[0022] S4, mapping the EEG control intention into a control instruction of the flight platform, and realizing a closed-loop mapping between the intention and the control instruction of the flight platform through the flight control module;

[0023] S5. According to the control instructions, the flight execution module executes the flight operation of the flight platform and generates flight status feedback;

[0024] S6. Through the feedback module, the status information of the flight platform is converted into visual or tactile feedback signals and fed back to the user.

[0025] Optionally, the S1 specifically includes:

[0026] S11. Place the wearable non-invasive EEG acquisition device on the user's scalp surface in the preset brain area, and use a dry electrode array to record multi-channel EEG signals with a sampling frequency of f. s Hz, the acquisition time window length is T w seconds, generate the original EEG time series data matrix X∈R C×N , where C is the number of channels, N = f s ×Tw is the number of sampling points per channel;

[0027] S12, bandpass filtering is performed on the original EEG signal matrix X, and the filtering range is [f l ,f h ], where f l is the lower limit frequency, f h As the upper limit frequency, the signal matrix X limited by the frequency band is obtained f ;

[0028] S13, the filtered signal X f Sliding window slicing is performed on each channel, and the sliding window size is set to L, and the overlap rate is γ∈[0,1), to obtain multiple time domain segment vectors Where K is the number of sliding windows x i ∈R L represents the i-th signal segment;

[0029] S14, all the fragments {x i}Map to high-dimensional point cloud space and construct a sample point cloud set Where d is the embedding dimension, forming the original neural temporal point cloud for topological modeling;

[0030] S15. Construct an ∈-adjacency graph on the point cloud set P, and define any two points p i 、p j The edge connections between them satisfy:

[0031] ||p i -p j ||2<∈;

[0032] Where ||·||2 represents the Euclidean norm, ∈ is the neighborhood threshold;

[0033] S16. Generate Rips complex based on ∈-adjacency graph and calculate Betti number sequence {β k}, further construct a persistent bar chart D = {(b i ,d i )}, where b i d i They represent the birth time and death time of the i-th topological feature, respectively, and serve as the structural feature input for subsequent intent recognition.

[0034] Optionally, the S2 specifically includes:

[0035] S21. Vectorize and encode the persistent bar graph data to obtain a topological feature vector of fixed dimension, where each dimension in the vector reflects the stability characteristics of different topological structures;

[0036] S22. Input the topological feature vector into the deep intent recognition network, which is composed of a multi-layer neural structure and extracts intent discriminant features through a nonlinear activation function;

[0037] S23. Output the probability distribution of flight intention and determine the category corresponding to the current user intention based on the maximum probability strategy;

[0038] S24. Extracting a matching flight control parameter template from the preset control template based on the recognition result, and performing parameter-level fusion in combination with the current topological feature vector;

[0039] S25. Output the fused control vector and transmit it to the flight control module to drive the platform to execute the next flight instruction.

[0040] Optionally, the S3 specifically includes:

[0041] S31. Using the identified intent tag to index the corresponding flight control template and extract the corresponding control parameter set;

[0042] S32, fusing the current topology feature vector with the template parameters to obtain a personalized flight control vector;

[0043] S33. Perform normalization processing on the fused control vector so that the values ​​of each dimension fall within the safe control range of the flight platform;

[0044] S34: Transmit the processed control vector as the final flight control instruction to the control module;

[0045] S35. Synchronously record the intent labels, topological features, control instructions, and flight response information in this interaction for subsequent optimization and learning.

[0046] Optionally, the S4 specifically includes:

[0047] S41, splitting the flight control command vector into three types of parameters: attitude adjustment, speed adjustment, and direction control;

[0048] S42. Generate a three-dimensional attitude control matrix of the flight platform according to the attitude adjustment parameters to control its pitch, roll, and yaw angles in space;

[0049] S43. Setting the propulsion speed of the flight platform in three axes according to the speed parameters, and dynamically adjusting it through the speed controller;

[0050] S44, using the direction control parameters to correct the current flight heading, and calculating the heading deviation angle to guide the real-time correction of the flight path;

[0051] S45. Integrate the attitude, speed and direction control quantities to generate a complete control signal set, and transmit it to the flight execution interface to complete the flight adjustment.

[0052] Optionally, the S5 specifically includes:

[0053] S51, the flight execution module receives the control signal set and starts the flight platform to perform attitude adjustment and path control;

[0054] S52, collecting the spatial position, velocity state and attitude information of the flight platform in real time to form a current state vector;

[0055] S53, comparing the current state with the intended target state, and calculating the state deviation;

[0056] S54. When the deviation exceeds the set threshold, the feedback mechanism is automatically triggered, a visual or tactile feedback channel is selected according to the deviation type, and a feedback signal is constructed;

[0057] S55. Feedback information is conveyed to the user in real time through the human-computer interaction terminal to prompt the user's adjustment intention, thereby building a complete closed-loop interaction process.

[0058] Optionally, the S6 specifically includes:

[0059] S61. After each interaction, record the user's EEG features, intention labels, control instructions, and flight results to construct an individual interaction sample set;

[0060] S62, filtering the interaction records with large deviations in the sample set and marking them as error sample sets;

[0061] S63. Build an individualized feature distribution model based on deviation samples to reflect the user's unique EEG control mode;

[0062] S64: Using the distribution model to update the parameter distribution of the intent recognition module to improve the model's adaptability to the user;

[0063] S65. Write error samples into the individual buffer and regularly trigger the model fine-tuning process, so that the system has good generalization and evolution capabilities when facing different users and different scenarios.

[0064] The beneficial effects of the present invention are:

[0065] First, by introducing persistent coherence modeling methods from topological data analysis, this invention elevates traditional EEG signal representation from time or frequency domain to structural modeling. This method utilizes topological invariants such as Betti numbers and persistent bar graphs to extract high-dimensional stable features from neural signals, enhancing the system's robustness to EEG noise and physiological perturbations. Compared to existing feature extraction methods based on power spectral density or spatial filtering, this method can capture the intrinsic topological structure of EEG time-series point clouds and identify the "cognitive state transition patterns" that occur during user control intent, significantly improving the accuracy and consistency of intent recognition.

[0066] Secondly, the present invention breaks through the control mechanism of the traditional brain-computer interface system based on "label trigger-fixed action" and establishes a parameter cascade mapping between intention recognition results and flight control instructions for the first time. By fusing the EEG intention category with the structured flight control parameters to form multi-dimensional continuous control vectors such as speed, attitude and direction, the system can achieve a more refined and natural flight control response. This structured mapping mechanism not only improves the operational granularity of flight control, but also provides underlying control capabilities for the brain control system to expand complex flight tasks (such as fixed-point hovering, smooth path transition, and dynamic obstacle avoidance).

[0067] Furthermore, this invention proposes a closed-loop flight feedback mechanism. By converting the flight platform's status information into visual or tactile feedback signals and transmitting them to the user, this complete loop design of "perception-recognition-control-feedback" effectively bridges the "human-machine perception gap" in the existing system interaction chain. This feedback mechanism enhances the user's subjective perception of the system's response results, enabling rapid cognitive correction when flight intent deviates or system execution mismatches occur, thereby enhancing the continuity and safety of interaction.

[0068] Furthermore, the present invention introduces an individual adaptive mechanism, dynamically reconstructing the user feature distribution model based on historical deviation samples. By optimizing the initial state of parameters in the neural network through kernel density estimation and continuous learning, the system can dynamically adjust the intent recognition model based on the EEG characteristics of different users, effectively improving adaptability and generalization across populations and scenarios. Compared to existing technologies that require repeated training or manual parameter adjustment, the present system can automatically adjust and self-evolve during long-term use, demonstrating excellent deployment scalability and engineering application value.

[0069] Finally, the present invention not only has a breakthrough innovation in the EEG signal modeling method, but also realizes the transformation from one-way recognition to closed-loop control in the system structure, and significantly enhances the user interaction experience through structured control instructions and multimodal feedback mechanism. At the same time, through individual adaptive optimization, the scalability and actual deployment capabilities of the system are further improved, which has significant engineering value and application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0071] Figure 1 This is a schematic diagram of the structure of a non-invasive EEG signal-driven flight interaction system proposed by the present invention;

[0072] Figure 2 This is the overall flow chart of a non-invasive EEG signal-driven flight interaction method proposed in the present invention. DETAILED DESCRIPTION

[0073] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0074] refer to Figure 1 , a non-invasive EEG signal-driven flight interaction system, including the following modules:

[0075] Non-invasive EEG signal acquisition module, used to collect the user's EEG signals;

[0076] A persistent coherence network module, used for topologically modeling the EEG signal and extracting characteristic parameters with global topological stability;

[0077] an EEG signal intention recognition module, configured to recognize the user's EEG control intention based on the characteristic parameters;

[0078] A flight control module, configured to map the EEG control intentions into control instructions for the flight platform;

[0079] A flight execution module, configured to control the flight operation of the flight platform according to the control instructions;

[0080] The feedback module is used to convert the status information of the flight platform into visual or tactile feedback signals and feed them back to the user.

[0081] The non-invasive EEG signal-driven flight interaction system provided by the present invention has an overall architecture that integrates topological data analysis and deep neural recognition mechanisms. For the first time, it integrates the high-dimensional point cloud structure, individualized neural features, and flight control parameter mapping in the EEG signal into a closed-loop human-computer interaction paradigm, and realizes the full-process interaction chain of "acquisition-modeling-recognition-control-feedback-optimization" in a modular manner. The system includes multiple core modules such as EEG acquisition, topological modeling, intent recognition, command mapping, flight execution, multimodal feedback, and individual adaptation. The modules operate in coordination, retaining the global topological stability characteristics of the EEG signal while also having real-time adjustable and user-adaptive control response capabilities. The system can effectively improve the robustness of intent recognition to complex cognitive states and environmental interference, significantly enhance the precision, continuity, and human-computer interaction experience of flight control, and is suitable for intelligent brain-controlled flight applications in multi-task, multi-user scenarios such as drone mind control, virtual flight training, and rehabilitation-assisted flight.

[0082] refer to Figure 2 , a non-invasive EEG signal-driven flight interaction method, comprising the following steps:

[0083] S1. Collecting the user's EEG signals and obtaining EEG data through a non-invasive EEG signal acquisition device;

[0084] S2. performing topological modeling on the EEG signal, extracting global topological features of the EEG signal using a topological data analysis method of a persistent coherent network, and obtaining characteristic parameters with global topological stability;

[0085] S3. Inputting the characteristic parameters into an EEG signal intention recognition module to identify the user's EEG control intention and perform feature mapping;

[0086] S4, mapping the EEG control intention into a control instruction of the flight platform, and realizing a closed-loop mapping between the intention and the control instruction of the flight platform through the flight control module;

[0087] S5. According to the control instructions, the flight execution module executes the flight operation of the flight platform and generates flight status feedback;

[0088] S6. Through the feedback module, the status information of the flight platform is converted into visual or tactile feedback signals and fed back to the user.

[0089] The method of the present invention defines the full-process processing and flight mission execution method of EEG control driven by topological modeling and deep recognition between various functional modules, ensuring the formation of a clear data flow and control closed loop from EEG acquisition, topological feature extraction, intention recognition to control instruction generation, flight execution and multimodal feedback. This method not only retains the high-dimensional topological structure characteristics of the user's EEG signal, but also establishes a dynamic mapping relationship from intention recognition to flight control, supporting the system's stable analysis and individual adaptation to nonlinear EEG signals. By structuring and parameterizing the EEG intention conversion process, the method achieves the standardization of the control path and the continuity of the response mechanism, providing the flight platform with a highly consistent and highly generalized control logic in a multi-task, multi-user environment. This method improves the interactive accuracy and operational robustness of the overall system, and helps to build a human-machine fusion control platform and brain-controlled flight interaction system for intelligent unmanned systems.

[0090] In this embodiment, S1 specifically includes:

[0091] S11. Place the wearable non-invasive EEG acquisition device on the user's scalp surface in the preset brain area, and use a dry electrode array to record multi-channel EEG signals with a sampling frequency of f. s Hz, the acquisition time window length is T w seconds, generate the original EEG time series data matrix X∈R C×N , where C is the number of channels, N = f s ×T w is the number of sampling points per channel;

[0092] S12, bandpass filtering is performed on the original EEG signal matrix X, and the filtering range is [f l ,f h ], where f l is the lower limit frequency, f h As the upper limit frequency, the signal matrix X limited by the frequency band is obtained f ;

[0093] S13, the filtered signal X f Sliding window slicing is performed on each channel, and the sliding window size is set to L, and the overlap rate is γ∈[0,1), to obtain multiple time domain segment vectors Where K is the number of sliding windows x i ∈R L represents the i-th signal segment;

[0094] S14, all the fragments {x i}Map to high-dimensional point cloud space and construct a sample point cloud set Where d is the embedding dimension, forming the original neural temporal point cloud for topological modeling;

[0095] S15. Construct an ∈-adjacency graph on the point cloud set P, and define any two points p i 、p j The edge connections between them satisfy:

[0096] ||p i -p j ||2<∈;

[0097] Where ||·||2 represents the Euclidean norm, ∈ is the neighborhood threshold;

[0098] S16. Generate Rips complex based on ∈-adjacency graph and calculate Betti number sequence {β k}, further construct a persistent bar chart D = {(b i ,d i )}, where b i d i They represent the birth time and death time of the i-th topological feature, respectively, and serve as the structural feature input for subsequent intent recognition.

[0099] In this embodiment, S2 specifically includes:

[0100] S21. Vectorize and encode the persistent bar graph data to obtain a topological feature vector of fixed dimension, where each dimension in the vector reflects the stability characteristics of different topological structures;

[0101] S22. Input the topological feature vector into the deep intent recognition network, which is composed of a multi-layer neural structure and extracts intent discriminant features through a nonlinear activation function;

[0102] S23. Output the probability distribution of flight intention and determine the category corresponding to the current user intention based on the maximum probability strategy;

[0103] S24. Extracting a matching flight control parameter template from the preset control template based on the recognition result, and performing parameter-level fusion in combination with the current topological feature vector;

[0104] S25. Output the fused control vector and transmit it to the flight control module to drive the platform to execute the next flight instruction.

[0105] In this embodiment, S3 specifically includes:

[0106] S31. Using the identified intent tag to index the corresponding flight control template and extract the corresponding control parameter set;

[0107] S32, fusing the current topology feature vector with the template parameters to obtain a personalized flight control vector;

[0108] S33. Perform normalization processing on the fused control vector so that the values ​​of each dimension fall within the safe control range of the flight platform;

[0109] S34: Transmit the processed control vector as the final flight control instruction to the control module;

[0110] S35. Synchronously record the intent labels, topological features, control instructions, and flight response information in this interaction for subsequent optimization and learning.

[0111] In this embodiment, the S4 specifically includes:

[0112] S41, splitting the flight control command vector into three types of parameters: attitude adjustment, speed adjustment, and direction control;

[0113] S42. Generate a three-dimensional attitude control matrix of the flight platform according to the attitude adjustment parameters to control its pitch, roll, and yaw angles in space;

[0114] S43. Setting the propulsion speed of the flight platform in three axes according to the speed parameters, and dynamically adjusting it through the speed controller;

[0115] S44, using the direction control parameters to correct the current flight heading, and calculating the heading deviation angle to guide the real-time correction of the flight path;

[0116] S45. Integrate the attitude, speed and direction control quantities to generate a complete control signal set, and transmit it to the flight execution interface to complete the flight adjustment.

[0117] In this embodiment, the S5 specifically includes:

[0118] S51, the flight execution module receives the control signal set and starts the flight platform to perform attitude adjustment and path control;

[0119] S52, collecting the spatial position, velocity state and attitude information of the flight platform in real time to form a current state vector;

[0120] S53, comparing the current state with the intended target state, and calculating the state deviation;

[0121] S54. When the deviation exceeds the set threshold, the feedback mechanism is automatically triggered, a visual or tactile feedback channel is selected according to the deviation type, and a feedback signal is constructed;

[0122] S55. Feedback information is conveyed to the user in real time through the human-computer interaction terminal to prompt the user's adjustment intention, thereby building a complete closed-loop interaction process.

[0123] In this embodiment, S6 specifically includes:

[0124] S61. After each interaction, record the user's EEG features, intention labels, control instructions, and flight results to construct an individual interaction sample set;

[0125] S62, filtering the interaction records with large deviations in the sample set and marking them as error sample sets;

[0126] S63. Build an individualized feature distribution model based on deviation samples to reflect the user's unique EEG control mode;

[0127] S64: Using the distribution model to update the parameter distribution of the intent recognition module to improve the model's adaptability to the user;

[0128] S65. Write error samples into the individual buffer and regularly trigger the model fine-tuning process, so that the system has good generalization and evolution capabilities when facing different users and different scenarios.

[0129] Example 1:

[0130] To verify the feasibility of this invention, the joint experimental center of the Neuroengineering Laboratory and Flight Control at a research institute was used to conduct multiple rounds of mind-controlled drone operation experiments with ten subjects in a standard indoor flight test environment. The experimental scenario simulated a complete flight mission process: indoor fixed-point takeoff and landing, mid-altitude hovering, target guidance, path correction, and automatic landing. The experimental space was approximately 15 meters long, 10 meters wide, and 5 meters high. A high-precision positioning system was deployed on the ceiling and side walls to assess flight accuracy. Indoor interference sources were kept below -60dB to ensure the clarity and consistency of EEG signal acquisition.

[0131] Before the experiment began, each subject wore the system's non-invasive EEG acquisition device (a 16-channel dry electrode headband) and completed a 2-minute resting signal baseline acquisition in a quiet state to initialize the individual feature distribution model. The system automatically extracted resting EEG topological features, constructed user-adaptive initialization parameters, and conducted five rounds of pre-training intent recognition (such as "takeoff," "hover," "fly left," and "landing," among other standard control intentions). The system automatically completed the topological modeling, feature embedding, and intent output mapping processes. Each training round took an average of approximately 12 seconds, and the recognition accuracy reached over 85%.

[0132] During the actual flight phase, the user triggers the drone's takeoff through an intention-triggered system. After hovering stably at approximately 3 meters, the drone, guided by visual cues, completes control tasks such as a 2-meter left deviation, a 3-meter forward movement, an obstacle avoidance maneuver, and an automatic return to the center point. During these tasks, the system continuously monitors EEG topological characteristics and updates control commands in real time. The flight platform's speed regulation and attitude correction are driven by the fused control vectors in the execution module.

[0133] Unlike traditional brain control methods, the present invention demonstrates significant advantages in "path control" and "attitude stability." Taking "attitude deviation angle less than 3 degrees during flight" as the standard for stable flight, the system achieved an average achievement rate of 92% in 50 flights by 10 users, a significant improvement over the 72% achieved by the traditional spectrum recognition method in the control group. The average intention-response delay was 420 milliseconds, approximately 270 milliseconds less than the control system, and 90% of the user's subjective questionnaire feedback rated "natural and smooth response." Furthermore, through a multimodal feedback mechanism (vibration + visual illustration), the user's self-correction rate when control deviations occur is increased to 88%, effectively forming a closed-loop interaction.

[0134] After the flight is complete, the system automatically collects and evaluates execution data, including EEG intention recognition accuracy, flight path deviation, task completion time, and number of status feedback. Ultimately, it calculates the system's overall performance and stability. In actual applications, it was observed that when the user operated for more than three consecutive rounds, the system further optimized the individual recognition model based on an adaptive learning mechanism, improving recognition accuracy by an average of 4.7% over the next three rounds, demonstrating excellent online evolution capabilities.

[0135] To illustrate the technical advantages of the present invention in complex interaction scenarios, some key performance indicators are summarized as follows to form a data summary table.

[0136] Table 1 Comparison of non-invasive brain-controlled flight performance based on topological modeling and feedback mechanism

[0137]

[0138] It can be seen from the above embodiments and data tables that the present invention demonstrates high recognition stability, flight accuracy, and response sensitivity in scenarios with multiple users, complex task processes, and multi-round interactions. It effectively solves key problems such as large recognition deviation, lack of closed-loop feedback, and lack of adaptive capabilities in existing non-invasive EEG flight control systems, verifying the engineering feasibility and technological advancement of this system in the field of intelligent control and human-computer interaction.

Claims

1. A non-invasive EEG signal-driven flight interaction system, characterized in that: Includes the following modules: Non-invasive EEG signal acquisition module, used to collect the user's EEG signals; A persistent coherence network module, used for topologically modeling the EEG signal and extracting characteristic parameters with global topological stability; an EEG signal intention recognition module, configured to recognize the user's EEG control intention based on the characteristic parameters; A flight control module, used to map the EEG control intention into a control instruction for the flight platform; A flight execution module, configured to control the flight operation of the flight platform according to the control instructions; The feedback module is used to convert the status information of the flight platform into visual or tactile feedback signals and feed them back to the user.

2. The non-invasive EEG signal-driven flight interaction system according to claim 1, characterized in that: The modules are implemented as follows: S1. Collecting the user's EEG signals and obtaining EEG data through a non-invasive EEG signal acquisition device; S2. performing topological modeling on the EEG signal, extracting global topological features of the EEG signal using a topological data analysis method of a persistent coherent network, and obtaining characteristic parameters with global topological stability; S3. Inputting the characteristic parameters into an EEG signal intention recognition module to identify the user's EEG control intention and perform feature mapping; S4, mapping the EEG control intention into a control instruction of the flight platform, and realizing a closed-loop mapping between the intention and the control instruction of the flight platform through the flight control module; S5. According to the control instructions, the flight execution module executes the flight operation of the flight platform and generates flight status feedback; S6. Through the feedback module, the status information of the flight platform is converted into visual or tactile feedback signals and fed back to the user.

3. The non-invasive EEG signal-driven flight interaction system according to claim 2, characterized in that: Said S1 specifically includes: S11. Place the wearable non-invasive EEG acquisition device on the user's scalp surface in the preset brain area, and use a dry electrode array to record multi-channel EEG signals with a sampling frequency of f. s Hz, the acquisition time window length is T w seconds, generate the original EEG time series data matrix X∈R C×N , where C is the number of channels, N = f s ×T w is the number of sampling points per channel; S12, bandpass filtering is performed on the original EEG signal matrix X, and the filtering range is [f l ,f h ], where f l is the lower limit frequency, f h As the upper limit frequency, the signal matrix X limited by the frequency band is obtained f ; S13, the filtered signal X f Sliding window slicing is performed on each channel, and the sliding window size is set to L, and the overlap rate is γ∈[0,1), to obtain multiple time domain segment vectors Where K is the number of sliding windows x i ∈R L represents the i-th signal segment; S14, all the fragments {x i }Map to high-dimensional point cloud space and construct a sample point cloud set Where d is the embedding dimension, forming the original neural temporal point cloud for topological modeling; S15. Construct an ∈-adjacency graph on the point cloud set P, and define any two points p i 、p j The edge connections between them satisfy: ||p i -p j ||2<∈; Where ||·||2 represents the Euclidean norm, ∈ is the neighborhood threshold; S16. Generate Rips complex based on ∈-adjacency graph and calculate Betti number sequence {β k }, further construct a persistent bar chart D = {(b i ,d i )}, where b i d i They represent the birth time and death time of the i-th topological feature, respectively, and serve as the structural feature input for subsequent intent recognition.

4. The non-invasive EEG signal-driven flight interaction system according to claim 2, characterized in that: The S2 specifically includes: S21. Vectorize and encode the persistent bar graph data to obtain a topological feature vector of fixed dimension, where each dimension in the vector reflects the stability characteristics of different topological structures; S22. Input the topological feature vector into the deep intent recognition network, which is composed of a multi-layer neural structure and extracts intent discriminant features through a nonlinear activation function; S23. Output the probability distribution of flight intention and determine the category corresponding to the current user intention based on the maximum probability strategy; S24. Extracting a matching flight control parameter template from the preset control template based on the recognition result, and performing parameter-level fusion in combination with the current topological feature vector; S25. Output the fused control vector and transmit it to the flight control module to drive the platform to execute the next flight instruction.

5. The non-invasive EEG signal-driven flight interaction system according to claim 2, characterized in that: The S3 specifically includes: S31. Using the identified intent tag to index the corresponding flight control template and extract the corresponding control parameter set; S32, fusing the current topology feature vector with the template parameters to obtain a personalized flight control vector; S33. Perform normalization processing on the fused control vector so that the values ​​of each dimension fall within the safe control range of the flight platform; S34: Transmitting the processed control vector as the final flight control instruction to the control module; S35. Synchronously record the intent labels, topological features, control instructions, and flight response information in this interaction for subsequent optimization and learning.

6. The non-invasive EEG signal-driven flight interaction system according to claim 2, characterized in that: The S4 specifically includes: S41, splitting the flight control command vector into three types of parameters: attitude adjustment, speed adjustment, and direction control; S42. Generate a three-dimensional attitude control matrix of the flight platform according to the attitude adjustment parameters to control its pitch, roll, and yaw angles in space; S43. Setting the propulsion speed of the flight platform in three axes according to the speed parameters, and dynamically adjusting it through the speed controller; S44, using the direction control parameters to correct the current flight heading, and calculating the heading deviation angle to guide the real-time correction of the flight path; S45. Integrate the attitude, speed and direction control quantities to generate a complete control signal set, and transmit it to the flight execution interface to complete the flight adjustment.

7. The non-invasive EEG signal-driven flight interaction system according to claim 2, characterized in that: The S5 specifically includes: S51, the flight execution module receives the control signal set and starts the flight platform to perform attitude adjustment and path control; S52, collecting the spatial position, velocity state and attitude information of the flight platform in real time to form a current state vector; S53, comparing the current state with the intended target state, and calculating the state deviation; S54. When the deviation exceeds the set threshold, the feedback mechanism is automatically triggered, a visual or tactile feedback channel is selected according to the deviation type, and a feedback signal is constructed; S55. Feedback information is conveyed to the user in real time through the human-computer interaction terminal to prompt the user's adjustment intention, thereby building a complete closed-loop interaction process.

8. The non-invasive EEG signal-driven flight interaction system according to claim 2, characterized in that: The S6 specifically includes: S61. After each interaction, record the user's EEG features, intention labels, control instructions, and flight results to construct an individual interaction sample set; S62, filtering the interaction records with large deviations in the sample set and marking them as error sample sets; S63. Build an individualized feature distribution model based on deviation samples to reflect the user's unique EEG control mode; S64: Using the distribution model to update the parameter distribution of the intent recognition module to improve the model's adaptability to the user; S65. Write error samples into the individual buffer and regularly trigger the model fine-tuning process, so that the system has good generalization and evolution capabilities when facing different users and different scenarios.

Images