Based on a simple deconvolution model of multi-layer interfaces, the reasons of wave number variation of the interfacial echo signal were analyzed to explore a method for feature recognition of the superficial soft tissue interfaces. The interfacial echo signal data were decomposed and reconstructed by Mallat multisolution analysis, with the number of the reconstructed interface signal as the feature. The results showed that the deconvolution model was effective for extracting the interface echo signal features in the superficial soft tissue and allowed identification and location of tissue defects.