Manufacturing, Radiographic Testing
MBCF-YOLO: Enabling Automated Defect Detection and Area Quantification
in Explosive Columns with an Adapted YOLO Framework
ABSTRACT
Explosive columns are solid energetic charges used in munitions and industrial blasting that may contain internal holes of various shapes and sizes. In X-ray computed tomography (XCT) images, these internal defects can compromise product quality and safety. However, accurate hole identification remains challenging, and unreliable area estimation further limits quantitative inspection. To address this issue, this paper proposes a multiscale bidirectional context fusion model (MBCF-YOLO), based on YOLOv11, for hole defect instance segmentation in explosive columns. The proposed model enhances multiscale representation and feature fusion by introducing a context fusion block (CFB-C3K2), a multiscale bidirectional fusion module (MSBF-SPPF), and a CARAFE-based upsampling strategy. The model directly outputs instance masks for each hole, while area quantification is performed as a separate post-processing step by applying the Shoelace theorem to the extracted hole contours.
KEYWORDS: explosive column holes, MBCF-YOLO, defect detection, instance segmentation, area quantification
https://doi.org/10.32548/2026.me-04571
