Abstract: The coarse columnar crystal structure of the austenitic stainless steel weld leads to the low signal-to-noise ratio of the conventional TOFD detection signal, and it is difficult to determine the defect diffraction wave in the detection signal. In order to improve the signal-to-noise ratio of the detection signal, based on the difference between the defect diffraction wave signal and the weld structure noise, a deep resampling and superposition algorithm is proposed and a test is carried out.The research results show that, compared with the phase distribution of the defect diffraction wave signal, the phase distribution of the structure noise is random; the deep resampling superposition technology can effectively suppress the structure noise of the tissue and improve the signal-to-noise ratio of the detection signal, playing an important role in improving the TOFD inspection ability of austenitic stainless steel welds.