This paper presents an enhanced method for classifying transformer winding deformation faults, aiming to achieve accurate and swift diagnoses, which holds great importance for power suppliers. Firstly, the polar plot is plotted by using the measured frequency response data (including amplitude and phase information). Secondly, the Digital image processing (DIP) technology is employed to extract texture features, such as Histogram of Oriented Gradient (HOG), Local Binary Pattern (LBP), and Gray-level Co-occurrence Matrix (GLCM) features, from polar plots.. Finally, three enhanced Support Vector Machines (SVMs) are trained independently using extracted texture features. These models are then combined to create a robust classifier for classifying transformer winding deformation faults. In addition, the paper utilizes an enhanced genetic algorithm (IGA) with an Emperor-Selective (EMS) mating scheme to optimize SVM parameters. The proposed method's feasibility and accuracy are demonstrated using experimental data from a model transformer. Comparisons with traditional methods reveal the proposed approach's superior performance in classifying transformer winding deformation faults.