TopPixelLoss:类别不均衡的遥感影像语义分割损失函数

针对遥感影像中类别不均衡的小目标分割效果不理想的问题，提出了一种类别不均衡小目标二分类分割的损失函数——TopPixelLoss损失函数。首先计算出每个像素的交叉熵，然后将所有像素的交叉熵按从大到小进行排序，随后确定一个K值作为阈值，筛选出前K个交叉熵最大的像素，最后对于筛选出的K个像素交叉熵取平均，做为损失值。在ISPRS 提供的 Vaihingen 数据集上，使用PSPNet网络与普通交叉熵、FocalLoss、TopPixelLoss三种损失函数分别对车辆进行二分类分割试验。结果表明，不同的K值，使用TopPixelLoss损失函数的平均交并比（MIoU）、F1-score、准确度（ACC）都最高;当K值为5×104时效果最佳，MIoU、F1-score、ACC分别比FocalLoss提高了3.0%、5.0%、0.1%。TopPixelLoss损失函数是一种针对类别不均衡分割非常有效的损失函数     

A novel satellite image encryption algorithm based on hyperchaotic systems and Josephus problem

With the rapid growth of the number of Earth observation satellite (EOS) supporting critical applications, it is required to improve the security techniques to protect the sensitive data and images during the transmission between the satellites and the ground stations. This paper introduces a new satellite image encryption algorithm based on the Linear Feedback Shift Register (LFSR) generator, SHA 512 hash function, hyperchaotic systems, and Josephus problem. LFSR generates a matrix that is used to construct the 512-bits value of the hash function. These bits are used to set the initial values and parameters of the proposed encryption algorithm. Firstly, the six dimensions (6-D) hyperchaotic system is divided into three parts, where every two equations are considered as one part. Secondly, the 1-D hyperchaotic logistic-tent system is considered as the controller to select one part. The selected part is used to generate a matrix that is XORed with the original image. Thirdly, the scrambling operation by Josephus sequences is applied to the output of the previous step by scrambling the rows and the columns according to the selected part to produce the pre-encrypted image. Finally, if the number of iterations is less than the required number which is considered as a parameter of the secret key, the previous operations will be repeated in the pre-encrypted image; otherwise, the pre-encrypted image is considered as the final cipher image. Experimental and analyses results show that the proposed algorithm has good performance in terms of high level of security, large enough key-space, tolerance to Single Event Upsets (SEU) as well as low time complexity.     

Near-wall topological patterns and flow structures over a simplified Danaus plexippus model

Studies on the high-lift mechanisms of butterfly gliding flights shed light on the design of the micro air vehicles. The flow field around a simplified Danaus plexippus model is investigated using the hydrogen bubble visualization and the Particle Image Velocimetry (PIV) techniques. There are three near-wall topological patterns with different Angles of Attack (AoAs): the separation bubble, the Leading-Edge Vortex (LEV) and the high AoAs flow. For the separation bubble pattern, two saddles and two foci form in the middle of the model. The features of the LEV pattern are the leading-edge separation lines. The topological characteristics of the separation lines are changed by the interaction between the LEV and the Wing-Tip Vortex (WTV). For the high AoAs flow pattern, four unstable foci are found at the forewing and the hindwing respectively. The angle between the trajectory of the WTV and the model increases with increasing AoA even though the slope of the WTV angle versus AoA curve declines at the moderate AoAs.     

我国饮料广告中女性形象的符号学解析

从符号学角度出发,以我国饮料广告中女性形象为分析对象,进行解析,提出女性形象的新类型,并从符号学角度对其所指、能指进行了解读。     

基于多尺度 3D-2D卷积神经网络的高光谱图像分类

设计提出了 1种针对高光谱图像分类任务的 3D-MSCNN模型。在 PCA降维的基础上,利用 3D空谱特征提 取网络和 2D多尺度特征提取网络实现高光谱图像特征提取,充分发挥高光谱图像空谱信息价值,增强对不同尺度 地表覆盖的表达能力。最后,利用 Softmax分类损失函数实现高光谱图像分类任务。实验结果表明,本文算法在 In. dian Pines和 Pavia University数据集上都取得了较好的分类效果。与 CD-CNN、3D-CNN、SS-Net和 HybirdSN等方法相 比,本文算法能够有效提升总体精度、平均精度和 Kappa系数等客观评价指标。     

光波纳米推进技术中的表面等离激元增益光镊作用机理

为研究天线不同结构对表面等离激元（SPP）增益光镊作用的影响,建立天线-基底的电磁波传输耦合激元电场触发光梯度力的数值模型,并利用数值模型分析天线不同结构对SPP电场的影响规律,同时,这种电场变化规律对光镊增益有直接的数学关系,基于上述物理机制,获得不同天线结构对光梯度力产生的优化策略。为验证上述物理机制与优化策略的有效性,开展纳米颗粒的粒子图像测速（PIV）试验,天线材料为银,基底为二氧化硅,纳米颗粒为银,试验能够完好地观测到纳米颗粒在天线通道的运动情况。结果表明,光梯度力的增益机制在于激元电场强度和梯度两方面因素,前者随天线不对称性增强而先增大后减小,后者呈现一直增大趋势;纳米颗粒推动作用力的计算误差约为5.5%~13.8%,且试验值与计算值的趋势相符,一方面验证本文研究机理及优化策略的有效性,另一方面证明PFP技术的原理可行。     

一种新的能量谱熵图像聚焦评价函数

针对计算机视觉中图像聚焦评价函数的选择与设计问题,首先对图像聚焦评价函数的基本概念和特征进行了讨论,然后对目前几种主要的图像聚焦评价函数（ 灰度变化函数、梯度函数、图像灰度熵函数、频域类函数） 进行了分析．在此基础之上提出了一种新的能量谱熵函数评价方法,并且给出了相应的计算公式．该函数的理论依据较之目前常用的评价函数更为充分,最后给出了与其它几种主要评价函数的实验对比结果,证明了能量谱熵函数比其它评价函数在评价图像聚焦质量方面更为准确、稳定和可靠．     

贝叶斯景象匹配技术

景象匹配技术以其极高的末制导精度在导航系统中起着关键性的作用。经典的景象匹配定位方法只能提供象素级的定位精度,为了获得更高的精度可采用ＦＦＴ匹配算法、最小二乘匹配算法等。该文首次提出了一种新的贝叶斯景象匹配算法。其基本思想在于：匹配定位位置可看作为随机量,先利用匹配图像间的信息求其条件概率分布,再利用图像中的大量样本来求解其贝叶斯估计值。同时研究了基于贝叶斯估计的景象匹配技术的原理及算法,并将其与     

基于表面再现 Marching cubes算法改进与实现

从医学体数据中得到对象的三维可视图像有广泛的应用领域,例如医学诊断、法医学、古人类学等.本文讨论并分析了一种基于表面再现的 Marching cubes算法的优点及存在的不足,给出了算法在时空开销、不确定性等方面的改进和解决方案,改进后的算法在实际系统中取得了良好的效果.     

机载红外搜索跟踪系统关键技术分析

概述了红外搜索跟踪系统的组成及主要技术特点,对红外成像系统中的红外望远光学系统、红外探测器组件等进行了分析;探讨了稳定与瞄准系统实现方案;对图像跟踪器中的跟踪算法及工程实现中采取的措施进行了论述。