A type of smart DPT is presented.The input-output non-linearity problem that affects the measurement accuracy of transducer is researched.And BP neural network is used to build the input-output model

and the BP network contains 3 layers.Either input layer or output layer contains only one neuron.The neural network is simulated in MATLAB.With mean square error setting to 10（-6）

the non-linearity error of the trained input-output model can achieve ±0.032 %.The polynomial curve fitting and the least-squares line fitting are also adopted.The comparison of the results of the three methods shows that BP network model can improve the performance of DPT.