1 华中光电技术研究所—武汉光电国家研究中心,湖北武汉430223
2 武汉设计工程学院信息工程学院,湖北武汉430225
红外小目标检测通常受制于较远的成像距离,使得提取目标特征成为了一种困难,如何增强目标的特征表达是近些年的主要方向之一。而过于复杂的特征表达会损失推理速度,这对于有实时性要求的红外小目标检测任务是不利的。通过使用重参数化技术结合领域中常用的残差网络作为特征提取网络,再使用额外注意力与通道注意力作为特征增强模块与特征融合模块,在数据集上取得了较好的结果。提出的模型在 SIRST 与 IRSTD-1K 数据集上分别取得了 0.734 与 0.638 的 mIoU 值,同时参数量和计算复杂度只有 0.306 M 与1.114 G FLOPs。该模型能够在推理阶段保持较少参数的同时拥有和其他领先的方法相近甚至领先的性能,在串行运行的环境上有着明显的优势。
红外小目标检测 深度学习 卷积神经网络 模型压缩 注意力机制 infrared small targets detection deep learning convolutional neural networks model compression attention mechanism
Author Affiliations
Abstract
1 Peking University, National Engineering Research Center of Visual Technology, Beijing, China
2 Hangzhou Dianzi University, School of Automation, Hangzhou, China
3 Medical School of Nanjing University, Nanjing, China
4 Hangzhou Dianzi University, School of Communication Engineering, Hangzhou, China
5 Lishui Institute of Hangzhou Dianzi University, Lishui, China
Light-field fluorescence microscopy (LFM) is a powerful elegant compact method for long-term high-speed imaging of complex biological systems, such as neuron activities and rapid movements of organelles. LFM experiments typically generate terabytes of image data and require a substantial amount of storage space. Some lossy compression algorithms have been proposed recently with good compression performance. However, since the specimen usually only tolerates low-power density illumination for long-term imaging with low phototoxicity, the image signal-to-noise ratio (SNR) is relatively low, which will cause the loss of some efficient position or intensity information using such lossy compression algorithms. Here, we propose a phase-space continuity-enhanced bzip2 (PC-bzip2) lossless compression method for LFM data as a high-efficiency and open-source tool that combines graphics processing unit-based fast entropy judgment and multicore-CPU-based high-speed lossless compression. Our proposed method achieves almost 10% compression ratio improvement while keeping the capability of high-speed compression, compared with the original bzip2. We evaluated our method on fluorescence beads data and fluorescence staining cells data with different SNRs. Moreover, by introducing temporal continuity, our method shows the superior compression ratio on time series data of zebrafish blood vessels.
light-field microscopy lossless compression phase space entropy judgment Advanced Photonics Nexus
2024, 3(3): 036005
提出并演示了一个光子辅助的集成雷达和通信系统,该系统利用光子辅助拍频在W波段产生毫米波信号。通过将正交相移键控(QPSK)信号编码到线性调频连续波(LFMCW)雷达信号上,实现了传感和通信波形的集成。一体化波形可以通过去啁啾分离通信信号与雷达感知信号,并通过脉冲压缩实现高分辨率感知。实验结果表明,在91 GHz频段内可实现单目标和双目标检测,感知精度约为2.0 cm。此外,成功实现了W波段下2 m、10 m和50 m传输距离的20 Gbit/s高质量无线通信。该系统还适用于各种成分的一体化波形,为高速通信和高分辨率雷达感知融合提供了有效参考。
光通信 通信与雷达感知一体化系统 毫米波通信 脉冲压缩 一体化波形
Author Affiliations
Abstract
1 SwissFEL, Paul Scherrer Institute, Villigen PSI, Switzerland
2 Photonics Institute, Technische Universität Wien, Vienna, Austria
3 Institute of Applied Physics, University of Bern, Bern, Switzerland
4 Institute for Quantum Electronics, Physics Department, ETH Zurich, Zurich, Switzerland
We demonstrate the generation, spectral broadening and post-compression of second harmonic pulses using a thin beta barium borate (BBO) crystal on a fused-silica substrate as the nonlinear interaction medium. By combining second harmonic generation in the BBO crystal with self-phase modulation in the fused-silica substrate, we efficiently generate millijoule-level broadband violet pulses from a single optical component. The second harmonic spectrum covers a range from long wave ultraviolet (down to 310 nm) to visible (up to 550 nm) with a bandwidth of 65 nm. Subsequently, we compress the second harmonic beam to a duration of 4.8 fs with a pulse energy of 0.64 mJ (5 fs with a pulse energy of 1.05 mJ) using chirped mirrors. The all-solid free-space apparatus is compact, robust and pulse energy scalable, making it highly advantageous for generating intense second harmonic pulses from near-infrared femtosecond lasers in the sub-5 fs regime.
post-compression second harmonic generation self-phase modulation supercontinuum generation High Power Laser Science and Engineering
2024, 12(2): 02000e16
强激光与粒子束
2024, 36(2): 025016
南京航空航天大学电子信息工程学院,江苏 南京 211106
随着传感器技术的不断发展,三维点云被广泛应用于自动驾驶、机器人、遥感、文物修复、增强现实、虚拟现实等领域的视觉任务中。然而,直接应用收集到的海量原始点云数据得到的效果不佳,因此,基于深度学习的点云处理方法受到了越来越多的关注和研究。本文综述了近6年来基于深度学习的三维点云处理方法的研究进展。首先给出了三维点云的基本概念和获取方式,简述了4种点云处理任务;然后针对点云去噪和滤波、点云压缩、点云超分辨率以及点云修复-补全-重建任务,重点阐述了相应的深度学习方法的原理,并分析了其优缺点;随后介绍了22种点云数据集和4类评价指标,同时给出了性能对比结果;最后探讨了点云处理方法目前存在的问题,并对未来的研究趋势进行了展望。
深度学习 三维点云处理 点云去噪 点云压缩 点云修复
1 电子科技大学计算机科学与工程学院,四川 成都 611731
2 北京空间机电研究所,北京 100094
3 清华大学电子工程系,北京 100084
光学感前计算是一种在光电传感器前端的光学域进行信息计算处理的技术,包括编码压缩、全光智能推理等计算范式,具有传输即计算、结构即功能等显著特点,在卫星光学遥感领域有着广泛的应用前景。首先对用于感前计算的光场调制器件进行介绍,包括数字微镜器件(DMD)、液晶空间光调制器(LC-SLM)、衍射光学元件(DOE)及超表面等。然后分别梳理了感前编码压缩及全光智能推理的相关技术发展,在此基础上,着重讨论光学感前计算在卫星遥感领域的应用途径和未来发展趋势。
光学感前计算 编码压缩 全光智能推理 卫星遥感 激光与光电子学进展
2024, 61(2): 0211030
中国工程物理研究院激光聚变研究中心等离子体物理科学与技术实验室,四川 绵阳 621900
在基于光参量啁啾脉冲放大的拍瓦级超短超强飞秒激光装置中,光参量相位是阻碍脉冲时域压缩的关键因素。对中国工程物理研究院的数拍瓦全光参量啁啾脉冲放大装置(SILEX-II)的光参量相位演化进行了详细研究。研究结果表明,通过光参量放大过程累积的群延迟色散高达532 fs2,三阶色散高达5782 fs3,在未补偿光参量相位的情况下,压缩脉冲的时域峰值强度仅为傅里叶变换极限脉冲的43%。通过调节压缩器光栅间距,补偿了光参量相位的群延迟色散,将压缩脉冲的时域峰值强度增加至傅里叶变换极限脉冲的94%。研究结果为SILEX-II激光装置的脉冲时域压缩提供了有效指导,同时也为未来基于全光参量啁啾脉冲放大技术的10~100 PW高峰值功率激光器的设计提供了依据。
激光器 光参量啁啾脉冲放大 光参量相位 时域压缩 傅里叶变换极限脉冲