Autism functional magnetic resonance signal classification method based on multiple time scales

The invention relates to an autism functional magnetic resonance signal classification method based on multiple time scales. The method comprises the following steps: preprocessing an acquired functional magnetic resonance image; constructing a brain connection network for each piece of ROI functional magnetic resonance data of each tested brain region after preprocessing by using LASSO, and performing feature recursion elimination on a connection matrix to obtain classification features; for each ROI sequence, an ASD patient, a healthy control group and all subjects are used to jointly train an HMM model, the likelihood value of each brain region is solved, and the likelihood values calculated under the three models are combined to serve as features. For each brain region sequence, all tested training auto-encoders are used, each brain region encoding result is used as a feature value to classify ASD and healthy tested objects, and a brain region feature with the highest classification accuracy is selected;