Surface roughness grade evaluation of milling workpiece based on deep transfer learning

Abstract

ABSTRACT At present, when using machine vision to detect surface roughness, the statistical analysis of image gray value ignores the advantages of multidimensionality of color images and subjective judgment of visual system. Traditional deep learning is computationally intensive, time-consuming, and heavily dependent on a large number of samples. Therefore, a surface roughness level recognition model Deep CORAL AlexNet based on deep transfer learning color image is proposed. Using the difference of the definition of the virtual image of the red and green blocks formed on the milling surface with different roughness and the correlation between the definition and the surface roughness, transfer learning automatically extracts more general roughness-related features to develop the model, reducing the amount of data required for the model, and reducing the data distribution difference between the source domain (training set) and the target domain (test set). The experimental results show that compared with the gray level co-occurrence matrix and artificial neural network detection methods, Deep CORAL AlexNet has a cross-domain recognition accuracy of 99.39 % in various light environments, and has good robustness to complex light environments. The method is accurate, convenient, and has a wide measurement range.