The stockwell transform (S transform) is used to conduct analysis and processing of the Acoustic Emission (AE) signal in the welding process. The multi dimension instantaneous frequency information of AE signal in the joint time frequency distribution is obtained to analyze and recognize the welding structure state. The changing paces of window functions under different parameters are analyzed, and the parameters of S transform are determined to realize the optimal adaptive multi resolution and time frequency clustering. The S transform is performed to analyze the time frequency characteristics of the collected AE signal in weldment tensile testing experiment. According to the analysis results of the weldment tensile AE signal, the time frequency distribution and the first-order moment of the AE signal are observed to identify and detect the elastic, plastic deformation, and crack in the welding restrictable experiment. The results indicate that the feature information of the AE signal obtained by S transform can present the welding structural state, which provides an effective method for online detection of welding crack.