Implementation of Neural Network-Based Classification Approach on Embedded Platform

摘要

Among a number of challenges present in monitoring systems, an efficient implementation of complex time-consuming algorithms and an identification of relevant features from gathered signals still gain high attention. Compared with the signals captured from human body, the problem of identification and classification of abnormalities in electroencephalography (EEG) and electrocardiography (ECG) signals is correlated to the diagnosis of a number of neurological, neuromuscular, and psychological disorders, such as epilepsy, sleep disorders, and similar. The problem of epileptic seizure detection based on EEG signal is discussed in this contribution. Special emphasis here is given to epileptic seizure detection using real-time signal processing based on Field Programmable Gate Array (FPGA) embedded platforms. Proposed approach involves an implementation of classification algorithm relied on Artificial Neural Networks (ANNs) on FPGA board, whilst the extraction of features from EEG signal is performed offline. Accordingly, real-time implementation of ANN-based approach and its comparison with conventional approaches with respect to accuracy, runtime speedup, and applicability to low-power consumption (wearable) devices is in the main focus. The implementation is based on benchmark data available from public repositories and loopback testing.