The liquid level of different types of oil in an oil tank can be judged on the basis of the amplitude of time-domain signals, but the characteristic information obtained by this signal is relatively simple. In addition, the liquid level intelligent detection for oil tanks based on machine learning techniques, such decision tree algorithm, artificial neural network algorithm, and support vector machine, has stimulated considerable research interest. To obtain more characteristic information and improve the liquid level detection rate of different medium signals in an oil tank, a liquid level intelligent detection method based on empirical mode decomposition (EMD) and deep belief network (DBN) for a steel oil tank of 5 mm thickness is provided in this paper. Firstly, the ipsilateral phase detection method of the air-coupled ultrasonic Lamb wave was adopted to detect the oil tank with the help of the A0 model. Then, the intrinsic mode function (IMF) of each order was obtained by analyzing the signals of different media in the oil tank by EMD, and the correlation characteristics of the time/frequency domain signals of each order IMF component were analyzed. Finally, the time/frequency domain signals of the IMF component served as the input signals of the DBN model. The liquid level is divided into 15 sections as the output of DBN. The experimental results of the combination of EMD and DBN show that the liquid levels of different media in the oil tank can be accurately identified and further classified within the range of 10 mm, the detection rate can reach 99%, and the detection range meets the actual testing requirements of the oil tank.