Industrial Defect Comprehensive Evaluation Based on Gated Multilayer Perceptron

This paper proposes a comprehensive evaluation method for industrial defect characteristics based on gated multilayer perceptron (gMLP), aiming to address the challenges of multi-feature fusion and comprehensive diagnosis in industrial defect evaluation. The method builds upon the traditional MLP by introducing the Sigmoid-weighted linear unit (SiLU) and one-dimensional depth-wise separable convolutional layer (DWConv) to efficiently aggregate features, significantly enhancing feature representation capabilities while reducing model parameter size and computational complexity. Additionally, root mean square normalization (RMSNorm) is adopted to replace traditional regularization methods, further optimizing the model training process and improving convergence speed and stability. By integrating multiple defect features, the proposed method outputs a comprehensive evaluation result for defects, offering higher detection accuracy and robustness, making it suitable for defect diagnosis and evaluation in complex industrial scenarios.