Online multi-feature space migration identification method for scalp electroencephalogram signals

Abstract

The invention relates to the technical field of medicine, in particular to an online multi-feature space migration recognition method for scalp electroencephalogram signals, which comprises the following steps: step 1, establishing an electroencephalogram signal multi-view source domain data set for model training; step 2, for each view angle, calculating Euclidean distances from samples in all the source domain data sets to a target to be identified, performing clustering analysis by taking the distances as features, and then selecting a cluster where a clustering center closest to the target to be identified is located as a migration source domain; 3, setting model parameters lambda1, lambda2 and lambda3, and performing model training by using the correction samples in the source domain and the target domain selected in the step 2; and step 4, using the trained model to predict unmarked samples in the target domain. From the perspective of multi-feature space representation and induced transfer learning, guarantee is provided for improving the online disease recognition efficiency based on the electroencephalogram signals, and technical support is provided for precise medical services.

Description

technical field [0001] The invention relates to the technical field of medicine, in particular to an online multi-feature space migration identification method of scalp EEG signals. Background technique [0002] As a non-invasive detection method, scalp EEG signals play an important role in the detection of brain diseases such as epilepsy. Among the existing similar patents, Yu Qingshan et al. [Patent No.: CN201410738486.9] invented an online recognition method for multi-class EEG patterns based on the probability output of dual support vector machines. This method establishes a dual support vector machine probability output model, and on this basis, an incremental learning method is introduced to realize the online recognition of multi-type EEG patterns. [0003] Gan Haitao et al. [Patent No.: CN201510922194.5] invented an EEG signal recognition method for self-training learning. The method uses the semi-supervised extreme energy ratio algorithm to extract the feature vec...

AI Technical Summary

Problems solved by technology

[0005] Although some patents have disclosed the intelligent diagnosis of diseases by using scalp EEG signals combined with machine learning, it can be seen from the above technical solutions that there are still some shortcomings: 1) Extracting features from EEG signals Finally, features are still expressed in a single feature space, and EEG signals often contain rich timing, space, and energy information. It is difficult for a single feature representation method to capture enough feature pattern information for machine learning; 2) for a specific For clinical problems, such as epilepsy, transfer learning is often used to solve the problem of insufficient training samples or the cost of manual labeling

However, most of the current migration learning models are trained on the source domain, and do not consider adding a small number of labeled samples in the target domain for model "induced" learning

Images