铣削加工竹制品表面缺陷分类及数据集的构建OA北大核心

Accurate detection of surface defects in bamboo product processing is critical for improving quality,and existing detection models are however constrained by scarcity of high-quality datasets to achieve the complete detection,particularly for defects on the surface of milled bamboo materials.Focusing on the surface defects,this study uses a high-resolution industrial camera to capture 8 types of defects,including burrs,cracks,and mildew,on semi-finished bamboo toothbrush handles,and applies a data augmentation method to expand minority-class samples,so as to construct the bamboo milling surface defect(BMSD)dataset comprising 4 162 images.Twelve architectures,including single-stage,two-stage and Transformer-based DETR architectures are trained and evaluated to validate the effectiveness of BMSD dataset in the detection.Furthermore,gradient heat-maps characterizing defect density are generated to reveal distribution patterns and formation mechanisms of the defects,and their correlations with environmental and processing factors.The BMSD dataset supports defect detection model training and provides data support for research on milling process optimization and quality control.