Newborn electroencephalogram signal convulsion discharge detection system based on multi-scale feature fusion network

The invention discloses a newborn electroencephalogram signal convulsion discharge detection system based on a multi-scale feature fusion network. According to the method, anomaly detection is carried out on clinical electroencephalogram signal data of newborns on the basis of multi-scale fusion features; a highly parallel multi-branch one-dimensional convolution model is established to perform feature extraction on the marked newborn electroencephalogram signal data, and the multi-branch one-dimensional convolution model can reduce the operation time as much as possible while extracting the multi-scale features of the electroencephalogram signals; an integrated learning model based on a deep network is used, a plurality of weak supervision models are fused into a strong supervision model, multi-scale fusion features are utilized to the maximum extent, interference of random noise is reduced, and it is guaranteed that the model has high anti-interference performance. The method is a deep network modeling meth