为解决传统纸张表面缺陷检测人工依赖强、原始YOLOv5(You Only Look Once version 5)模型轻量化不足、多尺度缺陷检测精度低等问题,提出融合分组可分离卷积(Group Shuffle Convolution, GSconv)与双向加权特征金字塔网络(Bi-directional...为解决传统纸张表面缺陷检测人工依赖强、原始YOLOv5(You Only Look Once version 5)模型轻量化不足、多尺度缺陷检测精度低等问题,提出融合分组可分离卷积(Group Shuffle Convolution, GSconv)与双向加权特征金字塔网络(Bi-directional Feature Pyramid Network, BIFPN)的改进YOLOv5检测模型,构建纸张表面缺陷在线检测系统。模型通过GSconv替代骨干网络中的传统卷积实现轻量化,引入BIFPN强化多尺度特征融合,同时结合锚框优化与卷积块注意力模块提升小目标检测能力。实验结果显示,与原YOLOv5相比,改进模型骨干网络的参数减少28.3%,计算量降低31.5%;小目标缺陷检测精度提升12.3%,m AP达94.7%,较原始YOLOv5提升7.6%;推理速度达65 FPS,较原始模型提升32.1%。该系统满足了工业在线检测的实时性与精度需求,为纸张质量控制提供了新的技术支撑。展开更多
Microspheres containing nerve growth factor for sustained release were prepared by a compound method, and implanted into chitosan conduits to repair 10-mm defects on the right buccal branches of the facial nerve in ra...Microspheres containing nerve growth factor for sustained release were prepared by a compound method, and implanted into chitosan conduits to repair 10-mm defects on the right buccal branches of the facial nerve in rabbits. In addition, chitosan conduits combined with nerve growth factor or normal saline, as well as autologous nerve, were used as controls. At 90 days post-surgery, the muscular atrophy on the right upper lip was more evident in the nerve growth factor and normal sa- line groups than in the nerve growth factor-microspheres and autologous nerve groups. Electro- physiological analysis revealed that the nerve conduction velocity and amplitude were significantly higher in the nerve growth factor-microspheres and autologous nerve groups than in the nerve growth factor and normal saline groups. Moreover, histological observation illustrated that the di- ameter, number, alignment and myelin sheath thickness of myelinated nerves derived from rabbits were higher in the nerve growth factor-microspheres and autologous nerve groups than in the nerve growth factor and normal saline groups. These findings indicate that chitosan nerve conduits com- bined with microspheres for sustained release of nerve growth factor can significantly improve facial nerve defect repair in rabbits.展开更多
文摘为解决传统纸张表面缺陷检测人工依赖强、原始YOLOv5(You Only Look Once version 5)模型轻量化不足、多尺度缺陷检测精度低等问题,提出融合分组可分离卷积(Group Shuffle Convolution, GSconv)与双向加权特征金字塔网络(Bi-directional Feature Pyramid Network, BIFPN)的改进YOLOv5检测模型,构建纸张表面缺陷在线检测系统。模型通过GSconv替代骨干网络中的传统卷积实现轻量化,引入BIFPN强化多尺度特征融合,同时结合锚框优化与卷积块注意力模块提升小目标检测能力。实验结果显示,与原YOLOv5相比,改进模型骨干网络的参数减少28.3%,计算量降低31.5%;小目标缺陷检测精度提升12.3%,m AP达94.7%,较原始YOLOv5提升7.6%;推理速度达65 FPS,较原始模型提升32.1%。该系统满足了工业在线检测的实时性与精度需求,为纸张质量控制提供了新的技术支撑。
基金supported by the National Natural Science Foundation of China,No.30872898,81371116the Natural Science Foundation of Beijing,No.7132173
文摘Microspheres containing nerve growth factor for sustained release were prepared by a compound method, and implanted into chitosan conduits to repair 10-mm defects on the right buccal branches of the facial nerve in rabbits. In addition, chitosan conduits combined with nerve growth factor or normal saline, as well as autologous nerve, were used as controls. At 90 days post-surgery, the muscular atrophy on the right upper lip was more evident in the nerve growth factor and normal sa- line groups than in the nerve growth factor-microspheres and autologous nerve groups. Electro- physiological analysis revealed that the nerve conduction velocity and amplitude were significantly higher in the nerve growth factor-microspheres and autologous nerve groups than in the nerve growth factor and normal saline groups. Moreover, histological observation illustrated that the di- ameter, number, alignment and myelin sheath thickness of myelinated nerves derived from rabbits were higher in the nerve growth factor-microspheres and autologous nerve groups than in the nerve growth factor and normal saline groups. These findings indicate that chitosan nerve conduits com- bined with microspheres for sustained release of nerve growth factor can significantly improve facial nerve defect repair in rabbits.
