1 桂林电子科技大学,a.电子工程与自动化学院
2 桂林电子科技大学,b.计算机与信息安全学院, 广西 桂林 541000
针对红外无人机目标识别过程中特征信息较少、特征丢失严重、识别准确率低等问题, 提出一种基于YOLOv7的红外无人机目标检测方法。通过引入注意力机制来增强目标区域的特征表达能力, 提升图像的空间信息含量。采用改进的串行连接方式将通道注意力模块与空间注意力模块连接, 在结合了目标通道特征信息和空间特征信息的同时, 改进结构降低了通道注意力对红外图像识别的负面影响, 可以更好地实现对红外目标的特征加强作用。选择基于角度向量回归的SIoU损失函数作为边框损失函数, 进一步提升了模型的收敛性和检测精度。实验结果表明, 改进的算法模型推理速度达到了43帧/s, 准确率为95.4%, 召回率为87.3%, mAP为96.1%,在红外无人机检测任务中取得了更好的检测效果。
无人机 红外图像 注意力机制 损失函数 UAV infrared image YOLOv7 YOLOv7 attention mechanism loss function
1 中国科学院红外探测与成像技术实验室，上海 200083
2 中国科学院上海技术物理研究所，上海 200083
3 中国科学院空间应用工程与技术中心太空应用重点实验室，北京 100094
4 中国科学院大学，北京 100049
傅里叶变换高光谱仪器在定量化遥感领域展现出极大的优势，非线性校正是保证在轨辐射定标精度不可缺少的过程。针对搭载于风云四号静止轨道干涉式红外探测仪（FY-4/GIIRS）运行轨道受日晒分布不均匀、仪器环境温度日变化剧烈的特点，推导出一种基于仪器光谱响应率修正的非线性校正算法，通过测量一组标准参考辐射源光谱量化值和光谱响应率，拟合得到光谱响应率一次项修正系数。在仪器环境温度变化后，利用一次项修正系数、黑体观测光谱值和黑体观测光谱响应率重新计算光谱响应率常数项修正系数，就可以得到任意仪器环境温度下的仪器非线性校正系数。经仪器发射前地面热真空（TVAC）定标试验数据验证，该算法简单有效，在各试验环境温度工况下，对180~320 K观测范围内的辐射定标精度均有明显的提高。
红外探测 辐射定标 非线性校正 傅里叶变换光谱仪 高光谱 infrared detection radiometric calibration nonlinearity correction FTIR hyper spectrum
1 华南理工大学 材料科学与工程学院，广东 广州 510640
2 华南理工大学 海洋科学与工程学院，广东 广州 510640
3 华南理工大学 华南软物质科学与技术高等研究院，广东 广州 510640
4 中山大学 中山眼科中心 眼科学国家重点实验室，广东省眼科视觉科学重点实验室，广东 广州 510060
高分子薄膜通常具有结晶、相分离等微尺度结构，同时在制备和加工过程中也会形成表面缺陷、气泡等不均匀结构，这些都会严重影响其性能。然而，表征这些结构目前还没有很好的方法。本文探究了数字全息显微镜（Digital Holographic Microscopy，DHM）这一原位、无损、高精度的光学三维显微技术在高分子薄膜表面及内部结构、缺陷表征中的应用。利用自建的DHM和三维连续定位算法对100 nm聚苯乙烯纳米颗粒在液相薄层中的分布、晶圆涂覆聚氨酯后表面的形貌、聚甲基丙烯酸酯薄膜与水的相分离过程、聚乙二醇薄膜干燥过程中的边缘结晶、聚乙烯醇薄膜内的微纳气泡进行了表征，证明DHM具有很高的成像分辨率，能够对高分子薄膜的表面和内部同时实现高精度的原位三维缺陷监测与结构表征，具有独特的优势和应用前景。
高分子薄膜 数字全息 三维成像 polymer films digital holography three-dimensional imaging
实现电学性能优良的高Al组分AlGaN外延层是制备深紫外光电器件最重要的环节之一。本工作利用分子束外延(MBE)技术, 基于周期热脱附的生长方式, 通过改变Al源供应量调控Al组分, 并用Si进行n型掺杂, 在AlN/蓝宝石衬底上得到了系列高Al组分的Si-AlxGa1-xN外延层(x＞0.60)。对外延层相关物理性质进行了表征测试, 结果表明, 外延层Al组分与生长过程中Al束流大小呈现线性关系, 这为制备精确Al组分的AlGaN外延层奠定了基础。AFM结果表明, 高Al组分AlGaN外延层的表面形貌强烈依赖于Ga的供应量, 在生长过程中提高Ga束流可以显著降低外延层的粗糙度。基于范德堡法测量Si-AlGaN外延层电学性能, 证实其载流子特性良好, 其中 Al组分为0.93的样品室温下自由电子浓度、电子迁移率和电阻率分别达到了8.9×1018 cm-3和3.8 cm2·V-1·s-1和0.18 Ω·cm。
高Al组分AlGaN 分子束外延 Si掺杂 载流子特性 周期热脱附 high Al-content AlGaN molecular beam epitaxy Si doping carrier property periodic thermal desorption
Author Affiliations
Abstract
1 Key Laboratory on High Power Laser and Physics, Shanghai Institute of Optics and Fine Mechanics, Chinese Academy of Sciences, Shanghai, China
2 Center of Materials Science and Optoelectronics Engineering, University of Chinese Academy of Sciences, Beijing, China
3 Shanghai Institute of Ceramics, Chinese Academy of Sciences, Shanghai, China
As optical parametric chirped pulse amplification has been widely adopted for the generation of extreme intensity laser sources, nonlinear crystals of large aperture are demanded for high-energy amplifiers. Yttrium calcium oxyborate (YCa4O(BO3)3, YCOB) is capable of being grown with apertures exceeding 100 mm, which makes it possible for application in systems of petawatt scale. In this paper, we experimentally demonstrated for the first time to our knowledge, an ultra-broadband non-collinear optical parametric amplifier with YCOB for petawatt-scale compressed pulse generation at 800 nm. Based on the SG-II 5 PW facility, amplified signal energy of approximately 40 J was achieved and pump-to-signal conversion efficiency was up to 42.3%. A gain bandwidth of 87 nm was realized and supported a compressed pulse duration of 22.3 fs. The near-field and wavefront aberration represented excellent characteristics, which were comparable with those achieved in lithium triborate-based amplifiers. These results verified the great potential for YCOB utilization in the future.
optical parametric amplification petawatt ultra-short pulse yttrium calcium oxyborate High Power Laser Science and Engineering
2023, 11(1): 010000e2
1 武汉科技大学信息科学与工程学院，湖北 武汉 430081
2 武汉科技大学冶金自动化与检测技术教育部工程研究中心，湖北 武汉 430081
现有的光场图像显著性检测算法不能有效地衡量聚焦度信息，从而影响了检测目标的完整性，造成信息的冗余和边缘模糊。考虑到焦堆栈不同的图像及全聚焦图像对于显著性预测发挥着不同的作用，提出有效通道注意力（ECA）网络和卷积长短期记忆模型（ConvLSTM）网络组成特征融合模块，在不降低维度的情况下自适应地融合焦堆栈图像和全聚焦图像的特征；然后由交互特征模块（CFM）组成的反馈网络细化信息，消除特征融合之后产生的冗余信息；最后利用ECA网络加权高层特征，更好地突出显著性区域，从而获得更加精确的显著图。所提网络在最新的数据集中，F-measure和平均绝对误差（MAE）分别为0.871和0.049，表现均优于现有的红、绿、蓝（RGB）图像、红、绿、蓝和深度（RGB-D）图像以及光场图像的显著性检测算法。实验结果表明，提出网络可以有效分离焦堆栈图像的前景区域和背景区域，获得较为准确的显著图。
图像处理 显著性检测 深度学习 光场图像 卷积神经网络 激光与光电子学进展
2022, 59(22): 2210006
Author Affiliations
Abstract
1 National Engineering Research Center of Electromagnetic Radiation Control Materials, University of Electronic Science and Technology of China, Chengdu 610054, China
2 State Key Laboratory of Electronic Thin-Films and Integrated Devices, University of Electronic Science and Technology of China, Chengdu 610054, China
3 Institute of Fundamental and Frontier Sciences, University of Electronic Science and Technology of China, Chengdu 610054, China
4 College of Optoelectronic Engineering, Chengdu University of Information Technology, Chengdu 610225, China
5 State Key Laboratory of Vanadium and Titanium Resources Comprehensive Utilization, Panzhihua 617000, China
6 Microsystem & Terahertz Research Center, China Academy of Engineering Physics (CAEP), Chengdu 610200, China
7 Institute of Electronic Engineering, China Academy of Engineering Physics (CAEP), Mianyang 621900, China
Active metasurfaces whose optical properties can be tuned by an external stimulus have attracted great research interest recently. Introduction of phase change material in all-dielectric metasurfaces has been demonstrated to modulate the resonance wavelength and amplitude in the visible to near-infrared wavelength range. In this study, we report a mid-infrared active metasurface based on hybrid meta-atoms. By incorporating thin films in different locations of all-dielectric nanodisks, we demonstrate different modulation amplitude of the electric or magnetic resonance scattering cross sections, leading to drastically different transmission spectrum upon insulator to metal phase transition. The physical mechanism is originated from the field profiles of the resonance modes, which interact with differently depending on its locations. Based on this mechanism, we experimentally demonstrated a large modulation of the transmittance from 82% to 28% at the 4.6 μm wavelength. Our work demonstrates a promising potential of -based active all-dielectric metasurface for mid-infrared photonic applications such as infrared camouflage, chemical/biomedical sensing, optical neuromorphic computing, and multispectral imaging.
Photonics Research
2022, 10(2): 02000373
Author Affiliations
Abstract
1 School of Physics University of Electronic Science and Technology of China Chengdu, Sichuan 610054, P. R. China
2 Department of Biomedical Engineering Southern University of Science and Technology Shenzhen, Guangdong 518055, P. R. China
3 School of Electrical Science and Engineering University of Electronic Science and Technology of China Chengdu, Sichuan 610054, P. R. China
Microwave-induced thermoacoustic imaging (MI-TAI) remains one of the focus of attention among biomedical imaging modalities over the last decade. However, the transmission and distribution of microwave inside bio-tissues are complicated, thus result in severe artifacts. In this study, to reveal the underlying mechanisms of artifacts, we deeply investigate the distribution of specific absorption rate (SAR) inside tissue-mimicking phantoms with varied morphological features using both mathematical simulations and corresponding experiments. Our simulated results, which are confirmed by the associated experimental results, show that the SAR distribution highly depends on the geometries of the imaging targets and the polarizing features of the microwave. In addition, we propose the potential mechanisms including Mie-scattering, Fabry- Perot-feature, small curvature effect to interpret the diffraction effect in different scenarios, which may provide basic guidance to predict and distinguish the artifacts for TAI in both fundamental and clinical studies.
Imaging microwave thermoacoustic imaging artifacts specific absorption rate Journal of Innovative Optical Health Sciences
2021, 14(5): 2150011
中北大学仪器科学与动态测试教育部重点实验室, 山西 太原 030051
容器液位检测是工业生产及化工原料储存、运输过程中的重要环节,针对现有液位检测技术中传感器布置容易受空间限制,在高温高压、灰尘、潮湿等特殊环境下传感器寿命短等问题,提出了一种基于深度学习的红外目标成像液位检测方法。通过对容器红外图像标注数据集进行优化训练,得到可以准确识别容器内液体百分比含量的模型。首先,构建储罐液位标准数据集,并搭建基于Pytorch的深度学习目标检测框架。然后,在输入端对图像进行数据增强,调整模型的宽度和深度,优化训练检测模型。最后,采用特征金字塔网络和路径聚合网络结构融合不同尺寸特征图的特征信息,用联合交并比计算边界框的回归损失,并在后处理过程中引入加权非极大值抑制。实验结果表明,该模型具有较好的鲁棒性和识别效果,在交并比为0.5时的平均精度均值可达到0.804。
图像处理 液位检测 红外成像 深度学习 目标检测 光学学报
2021, 41(21): 2110001
