Weld defect detection model based on machine vision

Abstract

Addressing the shortcomings of low accuracy and time-consuming nature in manual weld defect detection, this study applies machine vision to weld defect detection. Based on the SSD target detection network, it firstly introduces the model structure and detection process of the SSD target detection network. Secondly, it describes the relevant evaluation indicators and the preparation of the dataset. Finally, the SSD network is used to train and verify the dataset. The result shows that the average accuracy of the model for these four types of recognition is 66.49%, indicating that the SSD target detection model has good detection ability for weld defects, providing a way for the application of machine vision in recognizing weld defects.