Multi-task deep network-based sleep epilepsy electrical persistence state quantification method

The invention discloses a sleep epilepsy electrical persistence state quantification method based on a multi-task deep network. Firstly, a signal is preprocessed. And then inputting the preprocessed signal into a shared feature extraction network for feature extraction to obtain a task shared feature. And then, respectively inputting the task sharing features into a task special signal channel attention module to obtain a task-independent self-adaptive weighted feature map. And inputting the feature map of the epilepsy sample activity segmentation task into an epilepsy sample activity decoder to obtain a total discharge time length and a ratchet slow wave discharge index of a single sample. And inputting the feature map of the sleep staging task into a sleep staging classifier to obtain a sleep stage of each sample. And finally, the two are combined to realize identification and quantification of the epileptic electrical continuous activity in the sleep of the children. According to the method, the epilepsy-l