Abstract: The traditional iterative compressed sensing reconstruction algorithm is difficult to play a role in actual wearable devices because of its high computational complexity and poor real-time data processing. In this paper, a non-iterative compressed sensing real-time reconstruction algorithm suitable for wearable health monitoring is proposed by combining one-dimensional dilated convolution and residual network in deep learning. The proposed method trains a network model for compressed sensing reconstruction based on a large number of physiological signal data, and the trained neural network model can accurately reconstruct physiological signals at a very fast speed. Experiments on 2 open physiological signal data sets show that the proposed method has higher reconstruction accuracy than the existing reconstruction algorithms based on deep learning. The proposed method can reconstruct a 2 s signal frame in only about 0.7 ms on the computer used in this paper. This is about 2～3 orders of magnitude faster than the traditional iterative compressed sensing reconstruction algorithm. Therefore, the method proposed in this paper has excellent real-time performance.