We design a discrete seizure-specific wavelet (Seizlet) to model the seizure signals. Four patterns are designed by Seizlet cone of influence map and modulus maximas lines. Features are defined by mapping CIM series from patterns and fitted conic ellipse. Features are tuned by Honeybee Hive optimization with LVRA and Elman neural network. 7-channel seizure signals are detected by AdaBoost classifiers in a cascade structure.