The spiral drum of the coal shearer generates instantaneous alternating impact loads when cutting coal and rock layers, causing the drum to vibrate in all directions. The vibration state varies with different cutting conditions. It is difficult to effectively perceive the current cutting state by extracting only a single vibration feature. This paper proposes a coal-rock cutting state perception system based on D-S fusion. Both the cutting force and the two front-end vibration signals are collected as features, and the corresponding transformed feature vectors are input into the RBF neural network. First, the recognition result of a single signal is obtained, and then the D-S fusion based on the evidence correlation coefficient is carried out at the decision-making level. This scheme can increase the average recognition accuracy rate to 96.5%.