Amplification method of micro defect data on workpiece surface based on super-resolution feature fusion

Abstract

Aiming at the difficulty of detecting surface micro defects in defect detection, a data amplification model based on super-resolution feature fusion was proposed. The model consisted of three layers: Data (Data) layer, Super Resolution-image Repair(SR-Re) layer and data Merge-Amplification (M-A) layer. Data layer completed the sample division, and the sample with defect feature pixel ratio less than 0. 333% was used as the output of micro defect data; SR-Re layer used a dual-channel structure to process the input data in parallel for completing the super-resolution feature extraction of the input image data and sample repair; M-A layer achieved sample amplification by Poisson fusion of super-resolution features and defect-free samples. The difficulty problems such as the identification of micro defects on the surface of workpiece, the construction of detection model and the industrial inspection due to the inconspicuous image features were solved by the proposed model. The accuracy of defect detection models was improved through the amplification of micro defect samples. Five types of aluminum profile samples were used to complete the experiment, which verified the effectiveness and feasibility of the proposed model.