Abstract: Aiming at the problems of inconspicuous auscultatory features of pulmonary hypertension associated with congenital heart disease and the low accuracy of existing machine-assisted diagnostic algorithms, a feature extraction method based on phase space reconstruction (PSR) is proposed, which preserves the kinetic properties of the nonlinear heart sound system. First, the cardiac cycle is extracted by an adaptive dual-threshold segmentation method. Then, the main subbands of the main energy are extracted by tunable Q-factor wavelet transform (TQWT), which is used as the reference variable of PSR for feature extraction. Finally, a ConvNeXt network incorporating multi-scale convolution, spatial group-wise enhancement (SGE) attention mechanism and Efficient channel attention (ECA) improvement is designed as a classification model. In the three classification tasks, the accuracy, precision, sensitivity, specificity and F1 scores reached 91.16%, 91.19%, 91.16%, 95.58% and 91.14%, respectively. The results show that the algorithm proposed in this study has a significant effect in the auxiliary diagnosis of pulmonary hypertension associated with congenital heart disease.