Aiming at the problem of low accuracy of steam detection in the process of automation of "steam detection and steaming" of Chinese spirits. This paper proposes a steam detection method based on support vector machine (SVM) to improve the accuracy of automatic steam detection. First, the infrared thermal imager is used to collect the infrared grayscale image of the surface of the wine cellar in the barrel and perform histogram processing, then extract histogram’s multiple features. Combined with the engineering experience to add tags to form the training set, and through the support vector machine training to obtain the exploration model in the process of the captain. Through testing, the method has a high detection accuracy of 96%, which can meet the requirements of Chinese spirits production process.