针对当前基于视觉的跌倒检测方法无法较好平衡检测精度与响应速度、所设实验环境单一的问题,提出了一种基于YOLOv11n改进的轻量级跌倒检测算法MWA-YOLO。将轻量级的通道注意力机制MLCA(mixed local channel attention)引入C3K2模块,增...针对当前基于视觉的跌倒检测方法无法较好平衡检测精度与响应速度、所设实验环境单一的问题,提出了一种基于YOLOv11n改进的轻量级跌倒检测算法MWA-YOLO。将轻量级的通道注意力机制MLCA(mixed local channel attention)引入C3K2模块,增强通道特征的表现力以减少通道分割导致的信息丢失。引入WIoU替换原模型中的CIoU作为新的边界框损失函数,提升模型在复杂环境中边界框的定位精度。将网络结构中的部分标准卷积Conv层替换为改进后的下采样ADown模块,增强模型对有效信息捕捉能力的同时保持轻量化。实验结果表明,MWA-YOLO跌倒检测算法较YOLOv11n在mAP@0.5指标上从92.3%提升到了93.5%,参数量和计算量仅有2.2×10^(6)和6.0×10^(9),可部署在计算资源有限的硬件设备中,用于日常生活场景中的跌倒检测。展开更多
对道路交通参与者中的行人、骑行者以及车辆进行检测是实现自动驾驶的核心任务之一。在光照不均、遮挡、密集目标和远距离小目标等复杂场景中往往会存在误检及漏检情况;基于此,提出了一种改进YOLOv8模型的复杂交通场景目标检测算法。基...对道路交通参与者中的行人、骑行者以及车辆进行检测是实现自动驾驶的核心任务之一。在光照不均、遮挡、密集目标和远距离小目标等复杂场景中往往会存在误检及漏检情况;基于此,提出了一种改进YOLOv8模型的复杂交通场景目标检测算法。基于GhostNet轻量化网络结构,对原始YOLOv8模型的主干和颈部网络进行改进,利用幻影卷积(ghost convolution,GhostConv)来替换标准卷积(convolution,Conv),并将幻影瓶颈(ghost bottleneck,G-bneck)结合C3模块代替C2f模块,这样就有效抑制了冗余检测,提升了检测效率;应用混合局部通道注意力机制(mixed local channel attention,MLCA)对多元化信息进行整合,以增强模型的特征提取能力;添加小目标检测层,可保留更多细节特征信息,提高对远距离小目标的检测能力;采用WIoU(wise intersection over union loss)损失函数加速了网络收敛并增强了在复杂工况下的鲁棒性。研究结果表明:改进YOLOv8模型在所构建的复杂交通场景数据集RCCW-Dataset中的平均精度均值为0.872,较原始模型提高了2.1%;模型参数量和大小分别降低了41%和37%,能有效完成对实时复杂交通场景目标任务的检测。展开更多
为解决现有道路裂缝检测模型精度低、易漏检误检等问题,提出了一种基于混合局部通道注意力机制的YOLOv10n道路裂缝检测模型。针对道路裂缝细长且不规则的形状特点,首先在模型特征提取网络引入混合局部通道注意力机制(mixed local channe...为解决现有道路裂缝检测模型精度低、易漏检误检等问题,提出了一种基于混合局部通道注意力机制的YOLOv10n道路裂缝检测模型。针对道路裂缝细长且不规则的形状特点,首先在模型特征提取网络引入混合局部通道注意力机制(mixed local channel attention,MLCA),增强模型对道路裂缝特征的提取能力;其次,采用重参数化泛化特征金字塔网络(reparameterized generalized feature pyramid network,RepGFPN)优化原始颈部网络,充分融合多尺度下的裂缝特征信息;最后使用Focaler-IoU替换CIoU损失函数,调整模型训练不同裂缝样本的权重,加快收敛速度。在RDD2022_China数据集上的实验结果表明,改进后的模型相较于原始YOLOv10n模型检测准确率提升4.4%,平均精度均值(mean average precision,mAP)提高2.9%。与其他主流目标检测模型相比,改进后的模型在准确率、召回率和计算成本等方面均展现出最佳性能,验证了本文方法在道路裂缝检测任务中的有效性和优越性。展开更多
Multilayer ceramic actuator(MLCA)has been widely employed in actuators due to the large cumulative displacement under the low driving voltage.In this work,the MLCA devices consisting of a lead-free MnCO_(3-)and CuO-do...Multilayer ceramic actuator(MLCA)has been widely employed in actuators due to the large cumulative displacement under the low driving voltage.In this work,the MLCA devices consisting of a lead-free MnCO_(3-)and CuO-doped 0.96(K_(0.48)Na_(0.52))(Nb_(0.96)Ta_(0.04))O_(3)-0.04CaZrO_(3) piezoelectric ceramics and a base nickel(Ni)metal inner electrode were well co-fired by the two-step sintering process in a reducing atmosphere.The ceramic layer/electrode interface is well-integrated and clearly continuous without distinct interdiffusion and chemical reaction,which is beneficial to the electrical reliability of the MLCA.As a result,the MLCA laminated with nine active ceramic layers obtains an ultrahigh piezoelectric coefficient d_(33) of 3157 pC/N,about 9 times than bulk ceramics.The 0.5 mm-thick MLCA composed of a series of~50μm-thick ceramic layers and~3μm-thick Ni electrodes reaches a high 1.8μm displacement under the low applied voltage of 200 V(the same displacement requires a voltage as high as 3700 V for~1 mm-thick bulk ceramics).The excellent electrical performance and low-cost base electrode reveal that the(K,Na)NbO_(3)(KNN)-based MLCAs are promising lead-free candidate for actuator application.展开更多
文摘针对当前基于视觉的跌倒检测方法无法较好平衡检测精度与响应速度、所设实验环境单一的问题,提出了一种基于YOLOv11n改进的轻量级跌倒检测算法MWA-YOLO。将轻量级的通道注意力机制MLCA(mixed local channel attention)引入C3K2模块,增强通道特征的表现力以减少通道分割导致的信息丢失。引入WIoU替换原模型中的CIoU作为新的边界框损失函数,提升模型在复杂环境中边界框的定位精度。将网络结构中的部分标准卷积Conv层替换为改进后的下采样ADown模块,增强模型对有效信息捕捉能力的同时保持轻量化。实验结果表明,MWA-YOLO跌倒检测算法较YOLOv11n在mAP@0.5指标上从92.3%提升到了93.5%,参数量和计算量仅有2.2×10^(6)和6.0×10^(9),可部署在计算资源有限的硬件设备中,用于日常生活场景中的跌倒检测。
文摘对道路交通参与者中的行人、骑行者以及车辆进行检测是实现自动驾驶的核心任务之一。在光照不均、遮挡、密集目标和远距离小目标等复杂场景中往往会存在误检及漏检情况;基于此,提出了一种改进YOLOv8模型的复杂交通场景目标检测算法。基于GhostNet轻量化网络结构,对原始YOLOv8模型的主干和颈部网络进行改进,利用幻影卷积(ghost convolution,GhostConv)来替换标准卷积(convolution,Conv),并将幻影瓶颈(ghost bottleneck,G-bneck)结合C3模块代替C2f模块,这样就有效抑制了冗余检测,提升了检测效率;应用混合局部通道注意力机制(mixed local channel attention,MLCA)对多元化信息进行整合,以增强模型的特征提取能力;添加小目标检测层,可保留更多细节特征信息,提高对远距离小目标的检测能力;采用WIoU(wise intersection over union loss)损失函数加速了网络收敛并增强了在复杂工况下的鲁棒性。研究结果表明:改进YOLOv8模型在所构建的复杂交通场景数据集RCCW-Dataset中的平均精度均值为0.872,较原始模型提高了2.1%;模型参数量和大小分别降低了41%和37%,能有效完成对实时复杂交通场景目标任务的检测。
文摘为解决现有道路裂缝检测模型精度低、易漏检误检等问题,提出了一种基于混合局部通道注意力机制的YOLOv10n道路裂缝检测模型。针对道路裂缝细长且不规则的形状特点,首先在模型特征提取网络引入混合局部通道注意力机制(mixed local channel attention,MLCA),增强模型对道路裂缝特征的提取能力;其次,采用重参数化泛化特征金字塔网络(reparameterized generalized feature pyramid network,RepGFPN)优化原始颈部网络,充分融合多尺度下的裂缝特征信息;最后使用Focaler-IoU替换CIoU损失函数,调整模型训练不同裂缝样本的权重,加快收敛速度。在RDD2022_China数据集上的实验结果表明,改进后的模型相较于原始YOLOv10n模型检测准确率提升4.4%,平均精度均值(mean average precision,mAP)提高2.9%。与其他主流目标检测模型相比,改进后的模型在准确率、召回率和计算成本等方面均展现出最佳性能,验证了本文方法在道路裂缝检测任务中的有效性和优越性。
基金supported by the National Natural Science Foundation of China(GrantNos.52072150 and 51972146)Shandong Province Key Fundamental Research Program(Grant No.ZR2022ZD39)+1 种基金State Key Laboratory of New Ceramics and Fine Processing,Tsinghua University(Grant No.KF202002)Open Foundation of Guangdong Key Laboratory of Electronic Functional Materials andDevices(Grant No.EFMD2021002Z).
文摘Multilayer ceramic actuator(MLCA)has been widely employed in actuators due to the large cumulative displacement under the low driving voltage.In this work,the MLCA devices consisting of a lead-free MnCO_(3-)and CuO-doped 0.96(K_(0.48)Na_(0.52))(Nb_(0.96)Ta_(0.04))O_(3)-0.04CaZrO_(3) piezoelectric ceramics and a base nickel(Ni)metal inner electrode were well co-fired by the two-step sintering process in a reducing atmosphere.The ceramic layer/electrode interface is well-integrated and clearly continuous without distinct interdiffusion and chemical reaction,which is beneficial to the electrical reliability of the MLCA.As a result,the MLCA laminated with nine active ceramic layers obtains an ultrahigh piezoelectric coefficient d_(33) of 3157 pC/N,about 9 times than bulk ceramics.The 0.5 mm-thick MLCA composed of a series of~50μm-thick ceramic layers and~3μm-thick Ni electrodes reaches a high 1.8μm displacement under the low applied voltage of 200 V(the same displacement requires a voltage as high as 3700 V for~1 mm-thick bulk ceramics).The excellent electrical performance and low-cost base electrode reveal that the(K,Na)NbO_(3)(KNN)-based MLCAs are promising lead-free candidate for actuator application.
