Application of improved YOLOv7 based on Swin Transformer in defect detection of 3D printed lattice structures

Abstract

This paper utilizes an enhanced YOLOv7 network model, incorporating the Swin Transformer as the backbone network, to enable automated identification of internal defects within 3D printed lattice structures. By harnessing the robust adaptability and contextual capturing capabilities of the Swin Transformer, it effectively mitigates the limitations of YOLOv7 in handling diverse image sizes and detecting small objects. Through validation using CT slice images of the 3D printed lattice structure, the results indicate the recognition accuracy of 96.2%, surpassing the conventional YOLOv7 approach by 1.7%. The effectiveness and superiority of the methods suggested in this study are supported by these findings.