Feature extraction based on the expanded triangular episode and its application in pattern recognition

The triangular episode, which is proposed for process trend representation, have not been applied intensively for the representation´s dimension of the signal is not fixed. It brings some difficulties to the succeed pattern matching steps. Based on the triangular episode, a new feature extraction method is proposed. It can extract the signal feature and transfer it to an expression with fixed dimension, which is very convenient for the followed pattern matching process. This method is applied to a benchmark for evaluating feature extraction and classification techniques. Experimental result shows that the method can extract feature effectively and especially have advantage for the case of lack of dataset.