Abstract: To address the issue of noise processing in ultrasonic echo signals for nondestructive testing, starting from the characteristics of ultrasonic echo signal itself, this paper proposes an ultrasound echo signal processing method based on an improved sparse representation algorithm. The emitted signal is used as the basic unit atom and incorporated into the training samples, which are sequentially combined to construct a sparse dictionary. This approach ensures high efficiency and matching accuracy in the dictionary construction. The matching pursuit (MP) algorithm, based on this dictionary, exhibits predictable patterns when selecting the optimal atom. Accordingly, a proposal is made to improve the MP algorithm by adopting a skip search approach. The results show that the proposed algorithm outperforms wavelet transform and sparse representation algorithms using discrete cosine dictionaries in both denoising performance and computational efficiency. Furthermore, for the same dictionary, the improved MP algorithm shows a 30%-60% increase in efficiency compared to the original MP algorithm. The experimental results demonstrate the effectiveness and efficiency of the proposed method in handling ultrasonic echo noise signals.