To assess a novel cell manipulation technique of tissue engineering with respect to its ability to augment superparamagnetic iron oxide particles （SPIO） labeled mesenchymal stem cells （MSCs） density at a localized...To assess a novel cell manipulation technique of tissue engineering with respect to its ability to augment superparamagnetic iron oxide particles （SPIO） labeled mesenchymal stem cells （MSCs） density at a localized cartilage defect site in an in vitro phantom by applying magnetic force. Meanwhile, non-invasive imaging techniques were use to track SPIO-labeled MSCs by magnetic resonance imaging （MRI）. Human bone marrow MSCs were cultured and labeled with SPIO. Fresh degenerated human osteochondral fragments were obtained during total knee arthroplasty and a cartilage defect was created at the center. Then, the osteochondral fragments were attached to the sidewalls of culture flasks filled with phosphate-buffered saline （PBS） to mimic the human joint cavity. The SPIO-labeled MSCs were injected into the culture flasks in the presence of a 0.57 Tesla （T） magnetic force. Before and 90 min after cell targeting, the specimens underwent T2-weighted turbo spin-echo （SET2WI） sequence of 3.0 T MRI. MRI results were compared with histological findings. Macroscopic observation showed that SPIO-labeled MSCs were steered to the target region of cartilage defect. MRI revealed significant changes in signal intensity （P0.01）. HE staining exibited that a great number of MSCs formed a three-dimensional （3D） cell ＂sheet＂ structure at the chondral defect site. It was concluded that 0.57 T magnetic force permits spatial delivery of magnetically labeled MSCs to the target region in vitro. High-field MRI can serve as an very sensitive non-invasive technique for the visualization of SPIO-labeled MSCs.展开更多
针对金属表面缺陷检测任务中检测目标小、尺度变化大、背景复杂等问题,提出了一种基于RTDETR(real-time detection Transformer)的改进模型——HAS-DETR(high accurancy for small object-DETR)。HAS-DETR通过在骨干网络中引入复合差分...针对金属表面缺陷检测任务中检测目标小、尺度变化大、背景复杂等问题,提出了一种基于RTDETR(real-time detection Transformer)的改进模型——HAS-DETR(high accurancy for small object-DETR)。HAS-DETR通过在骨干网络中引入复合差分卷积,增强对小目标的特征提取能力;构建双重多尺度特征融合模块,有效捕获全局语义信息与细节特征,解决目标尺度变化大的问题;设计全局多尺度注意力机制,替代AIFI(attentionbased intra-scale feature interaction)模块中的多头注意力机制,提高模型在复杂背景和多尺度目标场景中的鲁棒性和精确度。在金属表面缺陷数据集上,HAS-DETR在mAP50和mAP50-95上分别较RT-DETR提升了6.5%和4.5%;在公开ADPPP数据集上,mAP50提升了2%,mAP50-95提升了1.3%。实验结果表明:HAS-DETR在保持较高检测效率的同时,有效提升了在复杂背景中对小目标的检测精度,具有良好的实际应用前景。展开更多
针对印刷电路板(Printed Circuit Board,PCB)表面缺陷检测任务中模型体积和参数量较大的问题,提出了一种基于通道剪枝的轻量级YOLOv8n网络PCB缺陷检测算法。为有效提升对PCB小目标缺陷的特征提取能力,采用RepViT作为特征提取网络;为提...针对印刷电路板(Printed Circuit Board,PCB)表面缺陷检测任务中模型体积和参数量较大的问题,提出了一种基于通道剪枝的轻量级YOLOv8n网络PCB缺陷检测算法。为有效提升对PCB小目标缺陷的特征提取能力,采用RepViT作为特征提取网络;为提升网络对小目标的关注度,减少神经网络推理过程中的梯度信息重复,将颈部网络的卷积模块替换为Rep-Net with Cross-Stage Partial CSP and ELAN(RepNCSPELAN4);为降低缺陷重叠时检测框失真现象,在预测部分使用Focaler-MPDIoU替换完全交并比(Complete Intersection over Union,CIoU);利用层自适应幅度分数剪枝(Layer Adaptive Magnitude based Pruning,LAMP)方法对融合改进方法的模型进行修剪,去除模型中冗余的梯度信息和权重,减少参数量和浮点运算量,压缩模型体积。实验结果表明,在PCB公开数据集中,经过LAMP之后,该算法相较于YOLOv8n,参数量下降60.8%,模型体积减小50.8%,计算量下降48.8%,平均精度均值(mean Average Precision,mAP)提高3.8%。在提高精度的同时,计算量、参数量和模型体积都低于原模型,满足在低配置设备下的使用需求。展开更多
基金supported by a grant from the National Natural Sciences Foundation of China (No. 30870639)
文摘To assess a novel cell manipulation technique of tissue engineering with respect to its ability to augment superparamagnetic iron oxide particles （SPIO） labeled mesenchymal stem cells （MSCs） density at a localized cartilage defect site in an in vitro phantom by applying magnetic force. Meanwhile, non-invasive imaging techniques were use to track SPIO-labeled MSCs by magnetic resonance imaging （MRI）. Human bone marrow MSCs were cultured and labeled with SPIO. Fresh degenerated human osteochondral fragments were obtained during total knee arthroplasty and a cartilage defect was created at the center. Then, the osteochondral fragments were attached to the sidewalls of culture flasks filled with phosphate-buffered saline （PBS） to mimic the human joint cavity. The SPIO-labeled MSCs were injected into the culture flasks in the presence of a 0.57 Tesla （T） magnetic force. Before and 90 min after cell targeting, the specimens underwent T2-weighted turbo spin-echo （SET2WI） sequence of 3.0 T MRI. MRI results were compared with histological findings. Macroscopic observation showed that SPIO-labeled MSCs were steered to the target region of cartilage defect. MRI revealed significant changes in signal intensity （P0.01）. HE staining exibited that a great number of MSCs formed a three-dimensional （3D） cell ＂sheet＂ structure at the chondral defect site. It was concluded that 0.57 T magnetic force permits spatial delivery of magnetically labeled MSCs to the target region in vitro. High-field MRI can serve as an very sensitive non-invasive technique for the visualization of SPIO-labeled MSCs.
文摘针对金属表面缺陷检测任务中检测目标小、尺度变化大、背景复杂等问题,提出了一种基于RTDETR(real-time detection Transformer)的改进模型——HAS-DETR(high accurancy for small object-DETR)。HAS-DETR通过在骨干网络中引入复合差分卷积,增强对小目标的特征提取能力;构建双重多尺度特征融合模块,有效捕获全局语义信息与细节特征,解决目标尺度变化大的问题;设计全局多尺度注意力机制,替代AIFI(attentionbased intra-scale feature interaction)模块中的多头注意力机制,提高模型在复杂背景和多尺度目标场景中的鲁棒性和精确度。在金属表面缺陷数据集上,HAS-DETR在mAP50和mAP50-95上分别较RT-DETR提升了6.5%和4.5%;在公开ADPPP数据集上,mAP50提升了2%,mAP50-95提升了1.3%。实验结果表明:HAS-DETR在保持较高检测效率的同时,有效提升了在复杂背景中对小目标的检测精度,具有良好的实际应用前景。
文摘针对印刷电路板(Printed Circuit Board,PCB)表面缺陷检测任务中模型体积和参数量较大的问题,提出了一种基于通道剪枝的轻量级YOLOv8n网络PCB缺陷检测算法。为有效提升对PCB小目标缺陷的特征提取能力,采用RepViT作为特征提取网络;为提升网络对小目标的关注度,减少神经网络推理过程中的梯度信息重复,将颈部网络的卷积模块替换为Rep-Net with Cross-Stage Partial CSP and ELAN(RepNCSPELAN4);为降低缺陷重叠时检测框失真现象,在预测部分使用Focaler-MPDIoU替换完全交并比(Complete Intersection over Union,CIoU);利用层自适应幅度分数剪枝(Layer Adaptive Magnitude based Pruning,LAMP)方法对融合改进方法的模型进行修剪,去除模型中冗余的梯度信息和权重,减少参数量和浮点运算量,压缩模型体积。实验结果表明,在PCB公开数据集中,经过LAMP之后,该算法相较于YOLOv8n,参数量下降60.8%,模型体积减小50.8%,计算量下降48.8%,平均精度均值(mean Average Precision,mAP)提高3.8%。在提高精度的同时,计算量、参数量和模型体积都低于原模型,满足在低配置设备下的使用需求。
