Complex Defects Detection of 3-D-Printed Lattice Structures: Accuracy and Scale Improvement in YOLO V7

Various complex and minor defects exist inside 3D-printed lattice structures that are difficult to detect using traditional detection methods. This study focuses on enhancing the accuracy of YOLO V7 in identifying these intricate defects. Initially, a Convolutional Block Attention Module (CBAM) was integrated into the YOLO V7 network to emphasize key image details and enhance detection precision. Then, Adaptively Spatial Feature Fusion (ASFF) was used to enable the model to select and fuse features of different scales, as needed, to improve the detection performance of the model. These improvements address the issue of inconsistent attention to features at different scales in the classic Feature Pyramid Network (FPN) of the YOLO V7. To assess the performance of the proposed model, this study utilized two datasets—the Face-Centered Cubic (FCC) dataset and the combined Body-Centered Cubic (BCC) and FCC datasets. The improved YOLO V7 demonstrated an average accuracy of 96.9%, a 2.4% enhancement compared to the unmodified model.