Abstract: Aiming at the problems of feature overlap and poor repeatability in the signal detection for different remainder materials of high precision spacecraft, a method for feature identification of remainder materials based on Mel frequency cepstrum coefficient (MFCC) and probabilistic neural network (PNN) was proposed. Inspired by the speech recognition technology, a method for pulse feature extraction of remainder material based on energy weighted MFCC was designed. A model for pulse classification of individual remainder material based on MFCC and optimized PNN was constructed. The pulse classification information of each remainder material was used to build the credibility of the remainder material, and the identification of four kinds of typical materials, including aluminum, solder, plastic and rubber, was realized. The experimental results show that the identification accuracy is above 90% for single remainder material, and is above 80% for two kinds of remainder materials.