%0 Journal Article
%T 基于深度学习的癫痫识别软硬件协同设计<br>Co-Design of Software and Hardware for EEG Epilepsy Recognition Based on Deep Learning
%A 谈宇轩
%A 涂淑琴
%J Artificial Intelligence and Robotics Research
%P 65-72
%@ 2326-3423
%D 2022
%I Hans Publishing
%R 10.12677/AIRR.2022.111008
%X 针对人工识别脑电图(EEG)的癫痫发作的低效以及现有自动识别算法的高开销等问题，本文提出了一种基于深度学习的癫痫识别的软硬件协同设计。软件部分提出了一种基于深度学习的癫痫识别模型，该模型的正确率在CHB-MIT数据集上能够达到97.08%的同时，尺寸与运算量仅为现有同类方案的20%。硬件部分提出了一种卷积神经网络处理器结构，该结构能够有效提升运算效率，在FPGA平台上完成一次癫痫识别过程仅需要0.6 ms。<br />
To cope with the low efficiency by artificial cognition of epileptic seizure whilst high expense by current automatic recognition algorithm, this work proposes the co-design of software and hardware for EEG epilepsy recognition on basis of deep learning. The software provided a recognition model of EEG epilepsy based on deep learning. The corresponding accuracy reached 97.08% in CHB-MIT data set, yet the size and computation is only 20% of the existing scheme. In the hardware part, an efficient convolutional neural network hardware accelerator structure is proposed, which can effectively improve the computing efficiency. It only takes 0.6 ms to complete one EEG Epilepsy recognition process on the FPGA platform.
%K 深度学习，脑电癫痫识别，软硬件协同设计<br>Deep Learning
%K EEG Epilepsy Recognition
%K Co-Design of Software and Hardware
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=49019