The invention relates to a DAS seismic data noise reduction method based on convolutional neural network, and belongs to the noise reduction method of two-dimensional DAS seismic data. On the basis ofthe convolutional neural network, a network structure is modified and parameters are set up to adapt to a noise reduction network model of DAS seismic data noise reduction processing, the convolutional neural network automatically extracts characteristics of effective signals and noise and intelligent noise reduction of the DAS seismic data is further realized according to characteristic differences. According to the DAS seismic data noise reduction method, noise in the DAS seismic data can be effectively removed, reflection coaxial information is well protected, while the signal to the noiseratio of the DAS seismic data is increased, resolution and fidelity of the data are unaffected, powerful guarantee is provided for subsequent accurate acquisition of reflection amplitude, velocity and frequency information, and the DAS seismic data noise reduction method can be widely used in noise suppression of the DAS seismic data.