基于机器学习方法的 ITO薄膜划痕临界载荷识别研究

The critical scratch load of an Indium Tin Oxide(ITO) film is a key indicator for evaluating its mechanical stability, However, traditional manual assessment methods are subjective and inefficient. To address the challenges of limited image data and severe class imbalance in scratch test analysis, an automated recognition method based on deep learning was proposed. The effects of three data augmentation strategies on the performance of a VGG19 model, i. e., basic geometric transformations, feature-space interpolation using SMOTE, and image-space generation using a Generative Adversarial Network(GAN), were systematically compared. The research results show that the GAN generative enhancement strategy can most effectively improve the classification performance of the model. Compared with the traditional geometric enhancement, its harmonic mean F₁ value (precision and recall) increases by 2.6%, and compared with SMOTE, the precision and F₁ increase by 4.88% and 2.69% respectively. The optimized model has strong robustness and generalization ability when dealing with samples under different process parameters (power, print point spacing, etc.), proving that the proposed method not only ensures the objectivity of the evaluation of film mechanical properties but also improves the efficiency of the evaluation.