Abstract
A fuzzy radial basis inference network with grouped signal feature embedding (GFE-FINN) classification model is proposed for multi-source time-varying signal fusion analysis and feature knowledge embedding, which multi-channel signals are divided into several groups according to the sources, attribute, features and sensitivity of signals. Each pattern class of the grouped signal sample set is divided into several pattern subclasses which are more similar features according to the grouping index, and typical feature samples are extracted to implicitly express the category features knowledge of the grouped signal. A fuzzy radial basis process neuron (FRBPN) is defined, which is used as parametric membership functions, and the typical feature signal samples of the grouped pattern subclass are used as the kernel centers of FRBPN to realize the embedding of the diverse feature knowledge. Through the kernel transformation in FRBPN, the input signals of each group are fuzzified respectively. Fuzzy multiplication operation is used to realize the information synthesis based on the membership degree of grouped pattern subclasses and establish fuzzy reasoning and classification rules. The proposed method can realize the feature fusion based on fuzzy membership degree and the semantic representation based on fuzzy rules hierarchically. Through the learning of the sample set, fuzzy membership function, reasoning and classification rules are established adaptively. A comprehensive learning algorithm was given. An experiment was conducted using 4-groups 12-lead long ECG signals in diagnosis of difficult heart disease. The correct recognition rate reaches 87.95%, and the performance evaluation index and generalization ability are significantly improved.
Highlights
I N the field of practical engineering and scientific research, there are many problems of nonlinear dynamic systems state evaluation and pattern classification based on multisensor monitoring signals [1], [2]
At the same time, based on the L2-norm distance, there may be a situation that the intraclass distance is large and the inter-class distance is small between the multi-channel signal samples of different pattern classes, which makes it difficult to achieve accurate similarity
If in the construction of multi-source time-varying signal classification model and algorithm, the multi-channel signals can be divided into several groups according to the signal source, attribute, feature, sensitivity and other factors, and the rule-based reasoning and knowledge representation ability of fuzzy system are combined with the adaptive learning and classification mechanism of neural network, the discovery and extraction of diversity typical signal features of various pattern classes are realized, the embedding and utilization mechanism of prior signal feature knowledge is established, and the memory of the diversity signal characteristics of pattern classes and their combination patterns are strengthened
Summary
I N the field of practical engineering and scientific research, there are many problems of nonlinear dynamic systems state evaluation and pattern classification based on multisensor monitoring signals [1], [2]. (1) Aiming at the problem of multi-source time-varying signal classification and feature knowledge embedding, a fuzzy radial basis inference neural network with grouped signal feature embedding (GFE-FINN) integrated analysis model is proposed. It can embed the feature knowledge of grouped signal, hierarchically realize the information processing and feature fusion in groups, and strengthen the role of signal typical category features in classification. The multi-source signal feature information fusion based on fuzzy set membership degree and grouped signal feature knowledge is realized It is suitable for the fast establishment of multi-source signal classification model in the case of unknown or low cognitive level, as well as small and unbalanced dataset, and it has good identification ability of signal feature and robustness
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call