The corrosion electric field is an obvious physical field feature of vessels due to its special characteristics of low frequency, obvious line spectrum features and cannot be eliminated. There are apparent distribution characteristics corresponding to different coating damage regions of a vessel, which are adequate for determining the possible damaged coating region. Therefore, a novel method combining Refined Composite Hierarchical Fluctuation Revise Dispersion Entropy (RCHFRDE) and improved Harris Hawk Optimization-kernel based extreme learning machine (IHHO-KELM) was proposed. We proposed RCHFRDE to extract the feature of electric field signatures. The feature vectors were input into the IHHO-KELM classifier to detect the damage region. The numerical and physical scale experiments were conducted to validate the feasibility and reliability of the proposed method. This damaged region was efficiently predicted, achieving satisfactory accuracy of 94.67% and 89.00% in numerical and measurement data respectively, which provided a complement for non-contact detection methods, especially with less prior environment information.