In order to effectively monitor the insulation condition of metal oxide arresters, a leakage current denoising method based on improved empirical mode decomposition is proposed. Firstly, the end point is extended to suppress the end effect by considering both waveform and amplitude similarity comprehensively, and then the leakage current signal without noise is reconstructed according to the comprehensive index that incorporates the smoothness and correlation of the intrinsic mode function. Subsequently, the weight parameters were dynamically adjusted and updated through the coordinate descent method to ensure the rationality and accuracy of the reconstructed signals. Finally, through analysis of simulated and measured data, it is confirmed that the proposed method can effectively eliminate the noise interference in the leakage current signal, and identify abnormal monitoring values, meeting the practical requirements of engineering.