场景统计类红外图像非均匀性校正算法研究

场景统计类非均匀性校正算法对场景分布进行假定,获得探测单元接收红外能量的一、二阶矩,进而估计探测单元的响应参数,校正非均匀性。分析比较了现有的场景统计类非均匀性校正算法,并应用交互多模(IMM)算法校正非均匀性,实验结果表明以连续图像序列作为观测数据时,能够有效地校正非均匀性,扩展了卡尔曼滤波法的适用范围。对各种方法进行仿真,表明删算法具有较好的收敛特性。     

感光材料分辨力测定实验研究

使用ＰＩ－Ｃ感光材料分辨力测定仪,测定了感光材料分辨力。测定试验不仅证明感光材料分辨力测定仪器操作稳定可靠、重复性好,而且测定实验加深了对感光材料的感光影象特性的研究。实验表明,最佳焦面实际集中在各种胶片乳剂层表层。     

A spherical code and stress tensor inversion

A spherical code is a set of many points arranged at more or less uniform intervals on a hypersphere. Here, we present a spherical code comprising 60,000 points in five-dimensional space and the list of the uniformly distributed reduced stress tensors corresponding to the points. This code benefits tectonic studies. By the way of example, it is demonstrated how the code improves the resolution of stress tensor inversion.     

Colour names and focal colours on electronic displays

This study identified the colours that were perceived as being the best examples (focal colours) of 10 common colour names when viewed on an electronic display. A total of 204 colours derived from a reflective surface colour measurement system (Munsell) were converted to the Commission Internationale de l'Eclairage (CIE) 1976 system and presented on a cathode ray tube (CRT). Forty subjects chose what was perceived as the best examples for each of the following colour terms: purple, blue, aqua, green, yellow, orange, red, pink, grey, and white. Results show that focal colours on a CRT do not always correspond to the typical colours used or to the best examples for reflective surface colours. The display designer should consider these differences when selecting colours for a display.     

国外红外焦平面探测器组件可靠性研究综述

介绍了国外焦平面探测器组件可靠性研究的主要特点和内容.焦平面探测器组件的主要制造商,建立了较完善的可靠性保证体系,可靠性研究融入设计与制造全过程,是基于过程的可靠性研究.在研制阶段进行失效模式的充分暴露与加速试验研究;批生产阶段进行工艺优化,并降低系统成本;建立数据库,进行可靠性评估方法研究.     

改进SSD的输电铁塔鸟窝检测

输电铁塔作为架空输电线的重要组成部分,其安全状况将影响整个电力系统的运行,鸟窝的搭建作为影响输电线路正常运行的重要因素之一,需要对此进行监控.而现有的监控手段不仅效率低,还需要耗费大量的人力物力.本文针对这一现象提出一种基于SSD算法的实时检测方法,并在SSD的网络结构基础上将前置网络VGGNet替换为ResNet-101,提高其特征提取能力,并将Softmax loss用Focal loss代替以改善SSD算法中样本不平衡的问题,并利用数据增广提高数据多样性,以提高模型的鲁棒性.实验结果表明本文所提出的方法检测精度对比原SSD算法在准确度和召回率上分别有3.17%和6.35%的提升.     

改进残差神经网络实现心音分类

心音的变化可以较早表征心脏疾病体征,基于深度学习实现心音信号分类具有心脏疾病临床辅助无创诊断作用。秉承特征获取简易和深度学习模型简单理念,设计了融合CBAM注意力机制、Focal Loss损失函数和多尺度特征的ResNet152网络模型对PhysioNet/CinC 2016心音数据集进行深度学习;详细介绍了CBAM注意力机制以及在网络瓶颈结构中的融合方式、Focal Loss损失函数原理、多尺度特征获取方式,并设计了5组对比消融实验和横向对比实验。实验结果表明,CBAM注意力机制、Focal Loss损失函数和多尺度特征提高了ResNet152基线网络模型分类准确率。     

基于阶段聚焦损失和并行增广策略的 遥感图像场景分类

随着深度学习的不断普及,卷积神经网络已成为遥感图像场景分类的主要手段,然而当前的研究主要集中于多网络主干的信息融合以及注意力机制等领域,在提高分类精度的同时也带来更高的计算复杂度。针对上述问题,分别从改进卷积神经网络损失函数和设计新的样本训练策略两个角度出发,在不增加计算复杂度的前提下,提升卷积神经网络的分类性能。首先,在对传统交叉熵和Focal loss损失函数进行分析的基础上,提出一种阶段聚焦损失函数,该损失函数可以在训练阶段对卷积网络进行有侧重的性能挖掘。其次,设计了一种并行样本训练策略,将采用Gridmask算法增广后的样本图像和原始样本图像,分为两路输入卷积网络进行并行训练,进一步提升卷积网络的分类性能。实验结果表明,所提出的算法分别在AID和NWPU-RESISC45两个大规模数据库上取得了96.72%和93.95%的检测精度,可以显著提升遥感图像场景分类的性能。     

应用四杆靶图像测量热成像系统焦距

采用实验室中四杆靶红外目标源,使其成像在热成像系统自身的焦面探测器上,采用放大倍率法测量与计算出系统的焦距.由于热成像系统的探测器像元尺寸一般较CCD像元大,为了降低像元尺寸对测量精度的影响,根据误差理论,合理选择四杆靶目标,使其能够达到满意的测试精度.这种焦距测量方法,原理简单直接,测量范围大,对于热成像系统的检测与装调,有着很好的实际应用价值.文中详细分析了影响焦距测量不确定度的因素,焦距的测量扩展不确定度可达到U=0.4%.     

Software for slip-tendency analysis in 3D: A plug-in for Coulomb

Slip-tendency analysis is a valuable tool in fault reactivation evaluation and seismic hazard assessment as it provides a means of quantifying the slip potential on mapped or suspected faults in a known or inferred stress field. We developed an interactive graphic tool to perform slip-tendency analysis. The application is written in MATLAB in the form of plug-ins for COULOMB, a graphic-rich deformation and stress change open-source software. In addition to identifying the faults most prone to reactivation, we compute and plot synthetic focal mechanisms from the direction and sense of likely slip. This allows compatibility between focal mechanisms and geological structures to be verified. Both individual faults and fault networks can be considered in three dimensions. The potential for slip depends on the prevailing stress field, the fault surface orientation and the coefficient of friction. These parameters are given interactively in a Windows environment. The application thus offers an easy-to-use graphical interface with the possibility of fast parameter modification and 3D visualization of the results.