Based on the Empirical Wavelet Transform (EWT) method, this paper explores the mathematical derivation process of signal modes at different scales. The multi-scale wavelet transform method is proposed and used as the feature extraction method for urban forest landscapes. Then, three theoretical contents related to line spectrum extraction, namely line spectrum frequency band estimation, time-frequency analysis and spectrum refinement, are expounded. With the technical support of the above methods, the collection and extraction of sound sources in urban forest parks are carried out. For the sorted sound source data set, time-domain and frequency-domain analyses and wavelet noise reduction are carried out. Subsequently, taking Guangzhou Baiyun Mountain Forest Park as the research object, the perception frequency, perception intensity and sound energy distribution of urban forest parks were analyzed to construct a natural sound perception model. The results show that among the 9 frequently extracted features, the water source landscape received the most subjective favorability evaluation from tourists and received a score of more than 0.9 in multiple scenic spots. Research indicates that in the process of urban landscape design, more attention should be paid to the maintenance of natural landscapes to assist in the development and construction of its tourism industry.
