Neonatal convulsion electroencephalogram signal classification system of time-frequency-space domain CNN-LSTM introducing attention mechanism

The invention discloses a neonatal convulsion electroencephalogram signal classification system of time-frequency-space domain CNN-LSTM introducing an attention mechanism. After the collected data are sliced, feature extraction is carried out on information contained in the time domain and the frequency domain of each slice, and feature vectors are obtained through splicing; time and frequency characteristics of the multi-lead electroencephalogram signals are mapped into a characteristic matrix according to an electrode space relation, and electrode space position information is fully reserved; an attention mechanism is introduced in feature processing, time-frequency-space domain features significantly related to convulsion symptoms are enhanced, and interference of irrelevant signals is reduced; feature vector sequences obtained by different slices are sent to two-layer bidirectional LSTM coding, and then a self-attention mechanism is utilized to strengthen the slice effect related to abnormal significance