In ICU (Intensive Care Unit), the trachea of patients with ventilator should be supervised all time to avoid the sputum depositing. The sputum situation identification by using traditional lung auscultation is time-consuming and related skill is difficult to acquire. Therefore it needs the medical staff to have a good training and experience. In this paper, an automatic sputum situation detection method is proposed. A system which is used to acquire respiratory sound was also developed. 46 features were extracted from the respiratory sounds based upon Empirical Mode Decomposition (EMD). And then Random Forest classifier is used as the classifier for recognition of sputum situation. In the experiment, 803 respiratory sound samples were collected from 14 patients, with each sample corresponding to one respiratory cycles. The classification results shows that this algorithm can achieve the accuracy of 92.02%.