Abstract: Titanium alloy materials are widely used in ships, aviation and other fields, and the quality of raw materials and welding is usually evaluated by ultrasonic testing. However, grain noise makes it difficult to recognize signals when detecting large size components and small defects. In this paper, the acoustic properties of typical titanium alloy microstructures were tested by water immersion ultrasound technology. The results showed that the sound velocity and attenuation coefficient of titanium alloy were anisotropic and the key factor affecting the sensitivity of large thickness structures was scattering clutter interference. The wavelet packet energy spectrum was used as the characteristic parameter to study the characteristics of clutter signal and defect signal, and the recognition accuracy of defect signal could reach more than 90% by training and classifying the defect signal with the neural network algorithm.