Defect detection on inwall of ferromagnetic pipes through low frequency electromagnetic method has become a research hotspot. However, periodic variation of magnetic flux leakage signal brings inconvenience to extraction of defect information. In addition, dependences between magnetic flux leakage signal and the pipe parameters such as length, diameter and initial thickness severely disturb the detection. To solve the problems, in this paper, we firstly establish a two-dimensional finite element detection model of irregular defects on inwall of the ferromagnetic pipes. Subsequently, the impact of periodic variation of magnetic flux leakage signal on extraction of defect information is eliminated by calculating the ratio of magnetic flux leakage signal to coil current. After that, the impact of dependences between magnetic flux leakage signal and the pipe parameters is greatly reduced through preprocessing of Finally, the defect profile is successfully reconstructed using the regression model between defect profile and preprocessed trained by Gaussian process regression algorithm. Based on the above mentioned method, simulation has been performed, and the results indicate that the RMSEs of reconstructed profiles are all around 0.17mm, which verify that the proposed method can accurately reconstruct profiles of the defects on inwall of ferromagnetic pipes.