基于脑-机接口技术获取的脑电信号（EEG）已广泛应用于人机交互领域，并取得了灵活、自然的交互效果。但由于EEG信号微弱，采集过程中容易混入噪声，给解码带来巨大困难。目前，脑-机接口技术一般可记录锋电位信号（Spike）和局部场电位信号（Local Field Potential, LFP），Spike信号有较好的解码效果，但峰电位的记录效果会随着时间的延长而质量下降，最终影响信号质量。局部场电位信号可长久稳定记录，并且噪声信号相对较少，可以弥补单独记录锋电位信号的不足。本文基于脑-机接口技术采集了自由活动下大鼠脑电信号。针对信号中的LFP信号，提出了精细回归树算法的大鼠运动步态解码方法，通过低通滤波器对LFP进行滤波处理，同时基于不同的频段对LFP信号进行划分并提取信号特征，讨论了不同频率下LFP信号的解码效果。发现基于短时傅里叶变换（STFT）和重采样的连续小波变换（CWT）方法对信号进行特征提取更为有效，提取后的脑电信号特征经过精细回归树方法可以更好地与步态信号建立联系。

Electroencephalographic (EEG) signals acquired based on brain-machine interface technology have been widely used in the field of human-computer interaction, and have achieved flexible and natural interaction effects. However, due to the weakness of the EEG signal, it is easy to mix in noise during the acquisition process, which poses great difficulties in decoding. At present, brain-machine interface technology can generally record Spike and Local Field Potential (LFP) signals. Spike signals have a good decoding effect, but the quality of peak potential recording will deteriorate over time, which eventually affects the signal quality. The local field potential signal can be recorded consistently for a long time, and the noise signal is relatively low, which can compensate for the lack of recording the fronto-potential signal alone. In this paper, the EEG signals of rats under free movement were acquired based on brain-computer interface is filtered by a low-pass filter, and the LFP signal is divided into different frequency bands and the signal features are extracted, and the decoding effect of the LFP signal at different frequencies is discussed. The short-time Fourier transform (STFT) and resampled continuous wavelet transform (CWT) methods were found to be more effective in extracting the features, and the extracted EEG features could be better related to the gait signal by the fine regression tree method.