Femtosecond laser processing is an important machining method for micro-optical components such as Fresnel zone plate(FZP).However,the low processing efficiency of the femtosecond laser restricts its application.Here,...Femtosecond laser processing is an important machining method for micro-optical components such as Fresnel zone plate(FZP).However,the low processing efficiency of the femtosecond laser restricts its application.Here,a femtosecond laser Bessel beam is proposed to process micro-FZP,which is modulated from a Gaussian beam to a Bessel annular beam.The processing time for FZP with an outer diameter of 60μm is reduced from 30 min to 1.5 min on an important semiconductor material gallium arsenide(GaAs),which significantly improves the processing efficiency.In the modulation process,a central ablation hole that has an adverse effect on the diffraction performance is produced,and the adverse effect is eliminated by superimposing the blazed grating hologram.Meanwhile,the FZP machined by spatial light modulator(SLM)has good morphology and higher diffraction efficiency,which provides a strong guarantee for the application of micro-FZP in computed tomography and solar photovoltaic cells.展开更多
针对可见光通信信号在传输中易受信道环境和背景噪声干扰等因素影响调制格式识别精度的问题,提出一种用于可见光通信信号调制格式识别的改进YOLOv5s(You Only Look Once)算法。首先,通过YOLOv5s算法网络输入端引入Mixup数据增强方式,将...针对可见光通信信号在传输中易受信道环境和背景噪声干扰等因素影响调制格式识别精度的问题,提出一种用于可见光通信信号调制格式识别的改进YOLOv5s(You Only Look Once)算法。首先,通过YOLOv5s算法网络输入端引入Mixup数据增强方式,将其与原网络中的Mosaic数据增强方式相结合,提升网络的鲁棒性,并增强算法在不同调制格式信号间的泛化能力;其次,将自适应空间特征融合(ASFF)引入到Neck网络中,充分提取不同层次的特征,提高检测精度。实验结果表明,在混合信噪比条件下,所提改进算法的平均精度均值(mAP)达到了0.903,比原始YOLOv5s算法提升了0.7%,且在信噪比为20 dB时mAP高达0.993。展开更多
基金Projects(51875584,51875585,51975590)supported by the National Natural Science Foundation of China。
文摘Femtosecond laser processing is an important machining method for micro-optical components such as Fresnel zone plate(FZP).However,the low processing efficiency of the femtosecond laser restricts its application.Here,a femtosecond laser Bessel beam is proposed to process micro-FZP,which is modulated from a Gaussian beam to a Bessel annular beam.The processing time for FZP with an outer diameter of 60μm is reduced from 30 min to 1.5 min on an important semiconductor material gallium arsenide(GaAs),which significantly improves the processing efficiency.In the modulation process,a central ablation hole that has an adverse effect on the diffraction performance is produced,and the adverse effect is eliminated by superimposing the blazed grating hologram.Meanwhile,the FZP machined by spatial light modulator(SLM)has good morphology and higher diffraction efficiency,which provides a strong guarantee for the application of micro-FZP in computed tomography and solar photovoltaic cells.
文摘由于低照度图像具有对比度低、细节丢失严重、噪声大等缺点,现有的目标检测算法对低照度图像的检测效果不理想.为此,本文提出一种结合空间感知注意力机制和多尺度特征融合(Spatial-aware Attention Mechanism and Multi-Scale Feature Fusion,SAM-MSFF)的低照度目标检测方法 .该方法首先通过多尺度交互内存金字塔融合多尺度特征,增强低照度图像特征中的有效信息,并设置内存向量存储样本的特征,捕获样本之间的潜在关联性;然后,引入空间感知注意力机制获取特征在空间域的长距离上下文信息和局部信息,从而增强低照度图像中的目标特征,抑制背景信息和噪声的干扰;最后,利用多感受野增强模块扩张特征的感受野,对具有不同感受野的特征进行分组重加权计算,使检测网络根据输入的多尺度信息自适应地调整感受野的大小.在ExDark数据集上进行实验,本文方法的平均精度(mean Average Precision,mAP)达到77.04%,比现有的主流目标检测方法提高2.6%~14.34%.
文摘针对可见光通信信号在传输中易受信道环境和背景噪声干扰等因素影响调制格式识别精度的问题,提出一种用于可见光通信信号调制格式识别的改进YOLOv5s(You Only Look Once)算法。首先,通过YOLOv5s算法网络输入端引入Mixup数据增强方式,将其与原网络中的Mosaic数据增强方式相结合,提升网络的鲁棒性,并增强算法在不同调制格式信号间的泛化能力;其次,将自适应空间特征融合(ASFF)引入到Neck网络中,充分提取不同层次的特征,提高检测精度。实验结果表明,在混合信噪比条件下,所提改进算法的平均精度均值(mAP)达到了0.903,比原始YOLOv5s算法提升了0.7%,且在信噪比为20 dB时mAP高达0.993。
