Research Of Pigeon Turning Motion Decoding Based On ReliefF-PLS

The neural signal decoding of movement intention has always been a hot issue in brain-computer interface research.It can help us to understand the animal's movement state from neural signals and deepen the understanding of the specific functions and working mechanisms of the brain.It has very important theoretical and practical significance.The detected neural signal contains a large amount of noise and redundant information,which affects the accuracy of the decoding.Therefore,it is especially important to extract features from the signal before decoding the neural signal.To solve this problem,we use the local field potential(LFP)signal of the pigeon nidopallium caudolaterale of the dove to study the best feature extraction and decoding problems.In this thesis,we designed the plus maze goal-directed experiment firstly to collect the LFP signal in the NCL during pigeon turning,and extracted the frequency band carrying the turning motion information through wavelet transform.Then analyzing the correlation of the signals,it was found a large amount of information redundancy between channels,and some channels also contain a lot of noise,which will affect the accuracy of the decoding.Aiming at the characteristics of large signal noise,high redundancy,and large feature dimension of LFP,the method of combining ReliefF algorithm and partial least squares(PLS)was applied to feature extraction of neural signals.The ReliefF feature selection method was used to remove the interference features,and each feature was assigned a corresponding weight value.According to the weight threshold,an appropriate subset of feature features was selected.Then PLS was used to extract the principal component features with the greatest degree of information relevance to the turning category.The extraction effect of ReliefF-PLS features was evaluated experimentally in terms of dimensionality reduction,feature dispersion,and feature classification accuracy,and compared with ReliefF and PLS.The results showed the ReliefF-PLS algorithm significantly reduced feature dimensions,effectively removed interference and redundant information,increased the dispersion of features,and improves the reparability of features.Experiments and analysis verified the effectiveness of the ReliefF-PLS algorithm.For LFP signal features extracted by ReliefF-PLS,linear discriminant analysis,support vector machine,and k nearest neighbors were used for decoding.The results showed that the correctness and stability of k nearest neighbor decoding are better than support vector machine and linear discriminant analysis.It had better noise immunity and was more suitable for the decoding of features extracted by ReliefF-PLS.