Abstract: Noise in ultrasonic TOFD testing signal often leads to difficulties in identifying the defect accurately.Wavelet packet decomposition technique was adopted to analyze the characteristics of defect's signals in time and frequency domain.The denoising of the ultrasonic TOFD detection signal is carried out by the wavelet packet sqtwolog,the noise reduction results of the soft and hard threshold function were compared.The results indicate that the soft and hard threshold is applied to the length of 10 mm,the depth of 5 mm crack defect signal denoising,the signal-to-noise ratio is increased from the original 22.88 dB to 186.66 dB and 176.65 dB respectively,the soft and hard threshold is applied to length of 28 mm,the depth of 8 mm inclusion defect signal denoising,the signal-to-noise ratio is increased from the original 16.62 dB to 33.74 dB and 28.16 dB respectively.Based on the wavelet packet soft and hard threshold denoising,the image reconstruction can effectively suppress the interference fringes and improve the resolution of the defective feature image,and the soft threshold method is used to completely remove the noise fringes in the original ultrasonic TOFD detection image.