脑电癫痫波的自动检测与分类是具有重要临床意义的课题。现存的算法大都着重于对棘、尖波形的检测 ,而忽略了慢波所包含的有用信息。为满足临床要求 ,论文提出了一种改进的脑电癫痫波自动分析系统。系统采用“分层次、多方法”的检测策略 ,兼顾了各种癫痫病理波形 ;整个处理过程综合应用了自适应预测、小波变换、人工神经网络、启发式规则等多种信号处理方法。经临床数据测试 ,该系统对癫痫波的总检测率达 83.6 % ,误检率为 1.1%。通过分层次处理 ,运用多方法的结合 ,可以提高检测敏感度和特异度 ,减少计算量 ,适合对长程脑电数据进行分析 EEG epilepsy wave automatic detection and classification is an issue of great clinical significance. The existing algorithms mostly focus on the detection of spine and spike waveform, while ignoring the useful information contained in the slow wave. In order to meet the clinical requirements, the paper presents an improved EEG automatic analysis system. The system adopts the “multi-level and multi-method” detection strategy, taking into account all kinds of epilepsy pathological waveforms; a variety of signal processing methods such as adaptive prediction, wavelet transform, artificial neural networks and heuristic rules are applied in the whole process. The clinical data test, the system of epileptic wave of the total detection rate of 83.6%, false detection rate was 1.1%. Through hierarchical processing, using a combination of methods, can improve the detection sensitivity and specificity, reduce the amount of computation, suitable for long-range EEG data analysis