Abstract: The acoustic emission signal of welding cracks in wind turbine tower can exhibit nonlinear distortion in the continuous time domain. Using a fixed frequency band to decompose signal components can easily truncate the crack characteristic frequency band across scales, resulting in feature blurring and being easily affected by signal uncertainty, thereby reducing the accuracy of crack detection. Therefore, an online detection method for welding cracks in in-service wind turbine towers under incomplete tree wavelet decomposition is proposed. Firstly, using the binary wavelet transform, the acoustic emission signal of tower welding is discretized in both time and space dimensions to correct the time difference of sensor signals at different positions and avoid nonlinear distortion; Then, the acoustic emission signal of the tower welding area is discretized using incomplete tree wavelet decomposition, and the acoustic emission signal features of the tower welding area are extracted based on sub-band energy. Finally, the uncertainty of the signal is quantified using information entropy, and then K-nearest neighbor algorithm is used to classify the feature attribute value categories according to the information entropy quantification results, classify the crack state, and achieve welding crack detection of in-service wind turbine towers. Through experiments, it has been proven that the method proposed in this article can be applied to detect welding cracks in in-service wind turbine towers. The maximum error between the detection results and the actual CSI values does not exceed 0.05, and the correlation coefficient can be maintained above 0.98. The crack state can be accurately identified, and the online detection results have a high correlation.