Psychological quality directly affects the playing level of a piano player’s ability. In this paper, we use the advantage of multi-source data information fusion to extract multiple physiological signal features of the piano player, combine with RBF kernel function to optimize the SVM emotion recognition, and form the DBN-SVM-based performance emotion state recognition algorithm. The initial feature set of physiological signals is established, the feature selection results for various classification tasks are examined, and the statistical data on effective features are obtained. The recognition of three emotional states of calmness, fear, and pleasure is done, and relevant strategies are proposed to cultivate good psychological qualities in players. The accuracy rate of emotional classification of the subjects is analyzed, and the accuracy rate of different subjects is more than 77.50%, indicating that the classification effect is evident. The psychological characteristics of the performers were clarified to facilitate the implementation of corresponding cultivation strategies.