Domain adaptive tinnitus classification method and classification system based on adversarial training

Abstract

The invention relates to the technical field of medical treatment, in particular to a domain self-adaptive tinnitus classification method and system based on adversarial training. An experiment time sequence signal passes through a pre-trained time sequence Transform model, long-term dependency and a complex mode in the signal are captured by utilizing a self-attention mechanism, an original time sequence signal is mapped into high-dimensional feature representation, general feature information of data is extracted, and the extracted features are transmitted to a task classifier for learning a specific task; in the training process, the gradient inversion layer inverts the gradient in back propagation, so that the feature extractor cannot distinguish the features of the source domain and the target domain, and the feature extractor is promoted to learn universal features which are not distinguishable to the domain, so that the universal features adapt to the features of the target task and the target domain. According to the method, the requirement for large-scale annotation data can be effectively reduced, and the diagnosis and treatment capacity of a medical intelligent system is improved.

Images