Research On Indoor Target Search Of Multi-rotor Aircraft Based On Motion Imagination

The brain-computer interface system(BCI)establishes the communication between brain activity and controlled devices by collecting EEG signals.The common methods include motor imagination,steady-state visual evoked potential and P300.Since motor imagination is independent of external stimuli,it has a higher potential to realize control and communication,and it has a wide range of applications,such as:wheelchair robot drive,aircraft control,etc.Brain-computer interface systems combined with aircraft have become a hot topic in recent years.People usually have to control aircraft with their hands,which makes it impossible for them to perform multiple complex tasks at the same time.In unpredictable practical applications(such as indoor target search),different people’s control level of the aircraft may lead to different performance of the mission.How to design a control mode with better adaptability,stability and intelligence to achieve indoor space target search of multirotor aircraft has become an urgent research problem.In order to solve the above problems,this paper first proposes a brain-computer interface system that integrates navigation and decision-making to realize indoor 3D space target search of low-speed multi-rotor aircraft.The monocular visual navigation subsystem uses scale invariant feature transformation and brute force algorithm to extract and match the key points,and then provides the 3D feasible flight direction for the decision subsystem.The obstacle and its location are estimated by comparing the size change of the key points with the size of the "convex" region of the target of interest.The decision subsystem first collects EEG signals from 15 electrodes for four motor imagination tasks(left/right,foot,and tongue).Then,two five-order Butterworth bandpass filters were used to preprocess the EEG signals.Secondly,the common space mode is used to filter the preprocessed EEG signal.In order to realize the fourclassification motion imagination task,the decision subsystem uses the single convolutional layer convolutional neural network to realize the EEG feature extraction,classification and the final feasible flight direction decision.The experimental results show that the brain-computer interface system has good adaptability and stability.Aiming at the problems of incomplete signal preprocessing and the extraction of spatial features while ignoring the time-frequency features of the proposed braincomputer interface system,improvements and optimization were made.In the pretreatment stage,the independent component analysis method was added to further remove the artifact of the ophthalmic signal.In the feature extraction stage,the wavelet packet decomposition method was added to extract the EEG time-frequency features,and the time-frequency features were fused with the spatial features extracted by the common space pattern algorithm to form a mixed feature set for feature classification.The model structure of convolutional neural network is adjusted and optimized.Finally,the performance of BCI system is further improved through calibration experiment.In this thesis,we propose that the brain-computer interface system based on fourcategory motion imagination task and integrated monocular visual navigation and decision subsystem has good adaptability and control stability.The improved and optimized brain-computer interface system further improves the accuracy and system performance of the four-classification motion imagination task,solves the defects of multi-rotor aircraft used for target search,and provides a new solution for it.Multidomain feature fusion solves the problem of incomplete feature extraction to a certain extent.This thesis is of reference value in theory and practice.