A novel approach to peak detection using sequential learning algorithm

Peak detection is a facile wing of signal processing. Conventional peak detection algorithms detect peaks when the entire signal is made available to them. In contrast, we propose a method that is based on recognizing the fundamental shapes of a signal, and the overall method intuitive in nature. Towards this, we use a feedforward neural network that is trained using the online sequential learning algorithm which provides better convergence performance relative to the back propagation algorithm. Moreover, the training avoids complex pre-processing tasks and feature extraction. Most importantly, the entire signal is not required for the algorithm to detect the peaks.