A 360-Degree Panoramic Image Inpainting Network Using a Cube Map

Inpainting has been continuously studied in the field of computer vision. As artificial intelligence technology developed, deep learning technology was introduced in inpainting research, helping to improve performance. Currently, the input target of an inpainting algorithm using deep learning has been studied from a single image to a video. However, deep learning-based inpainting technology for panoramic images has not been actively studied. We propose a 360-degree panoramic image inpainting method using generative adversarial networks (GANs). The proposed network inputs a 360-degree equirectangular format panoramic image converts it into a cube map format, which has relatively little distortion and uses it as a training network. Since the cube map format is used, the correlation of the six sides of the cube map should be considered. Therefore, all faces of the cube map are used as input for the whole discriminative network, and each face of the cube map is used as input for the slice discriminative network to determine the authenticity of the generated image. The proposed network performed qualitatively better than existing single-image inpainting algorithms and baseline algorithms.     

Experimental investigation and numerical modelling of density-driven segregation in an annular shear cell

Granular materials segregate spontaneously due to differences in particle size, shape, density and flow behaviour. In this paper we experimentally investigate density-difference-driven segregation for a range of density ratios and a range of heavy particle concentrations. The experiments are conducted in an annular shear cell with rotating bumpy bottom that yields an exponential shear profile. The cell is initially filled with a layer of light particles and an upper layer of heavier grains and, on top, a load provides confinement. The segregation process is filmed through the transparent side-wall with a camera, and the evolution of particle concentration in space and time is evaluated by means of post-processing image analysis. We also propose a continuum-approach to model density-driven segregation. We use a segregation-diffusion transport equation, constitutive relations for effective viscosity and friction coefficient, and a segregation velocity analogous to the Stokes’ law. The model, which is validated by comparison with experimental findings, can successfully predict density-driven segregation at different density ratios and volumetric fraction.     

基于改进网格运动统计特征的图像匹配算法

为了解决尺度不变特征变换（SIFT）算法在图像匹配中匹配正确率低、耗时长等问题，提出一种基于改进网格运动统计特征RANSAC-GMS的图像匹配算法。首先，利用快速旋转不变性特征（ORB）算法对图像进行预匹配，对预匹配的特征点采用网格运动统计（GMS）来支持估计量以实现正确匹配点与错误匹配点的区分；然后，采用改进的随机抽样一致性（RANSAC）算法通过匹配点间的距离相似性对特征点进行筛选，并采用评价函数对筛选后的新数据集进行重新整理，进而实现对误匹配点的剔除。采用Oxford标准图库和现实中拍摄的图像对图像匹配算法进行测试对比，实验结果表明，所提算法在图像匹配中的平均匹配正确率达到91%以上；与GMS、SIFT、ORB等算法相比，该改进算法的近景匹配正确率和远景匹配正确率分别最少提高了16.15个百分点和3.56个百分点，说明它能有效剔除误匹配点，进一步提高图像匹配精度。     

基于流形学习的光学遥感图像分类

随着光学遥感图像技术的快速发展与广泛应用,对光学遥感图像的准确分类具有深远的研究意义。传统特征提取方式提取的高维特征中夹杂着许多冗余信息,分类过程可能导致过拟合现象,针对传统的线性降维算法不足以保持原始数据的内部结构,容易造成数据失真这一问题,提出基于流形学习的光学遥感图像分类算法。该算法首先提取出图像的SIFT特征,然后将流形学习运用于特征降维,最后结合支持向量机进行训练和识别。实验结果表明,在Satellite、NWPU和UCMerced实验数据中,冰川、建筑群和海滩分类精度得到了有效提高,达到85%左右;针对沙漠、岩石、水域等特殊环境遥感图像,分类精度提高了10%左右。总而言之,基于流形学习的分类算法对通过降维之后的数据能够保持在原高维空间中的拓扑结构,相似特征点能得到有效聚合,预防了"维数灾难",减少了计算量,保证了分类精度。     

Controllable Image‐Based Transfer of Flow Phenomena

Modelling flow phenomena and their related weathering effects is often cumbersome due their dependence on the environment, materials and geometric properties of objects in the scene. Example‐based modelling provides many advantages for reproducing real textures, but little effort has been devoted to reproducing and transferring complex phenomena. In order to produce realistic flow effects, it is possible to take advantage of the widespread availability of flow images on the Internet, which can be used to gather key information about the flow. In this paper, we present a technique that allows the transfer of flow phenomena between photographs, adapting the flow to the target image and giving the user flexibility and control through specifically tailored parameters. This is done through two types of control curves: a fitted theoretical curve to control the mass of deposited material, and an extended colour map for properly adapting to the target appearance. In addition, our method filters and warps the input flow in order to account for the geometric details of the target surface. This leads to a fast and intuitive approach to easily transfer phenomena between images, providing a set of simple and intuitive parameters to control the process.     

水泥工厂视频监控智能预警平台研究及规划

视频监控系统作为智能化工厂体系架构的子模块，在水泥工厂得到了广泛的部署和应用，为安全、生产流程等管理需求提供了远程可视化的技术支撑手段。为进一步提升工厂智能化水平，提高管理效率，本文提出了基于视频图像分析技术的智能监控预警平台的路线规划，实施步骤和方法，并对该平台的初步应用效果进行了量化评估，为后续平台建设提供科学决策依据。     

卷积神经网络用于关节角度识别与姿势评估

为了快速准确地输出各种工作姿势风险评估结果，提出采用Kinect v2与卷积神经网络识别人体各关节角度，并输出标准姿势风险的评估得分。首先使用亚像素角点提取的棋盘标定算法标定Kinect两个摄像头，其次使用改进后的双边滤波对深度图像去噪，使用卷积神经网络识别人体关节二维位置，结合深度信息获取实际三维坐标并计算人体关节角度，最后输出姿势风险评估得分。通过两种实验分别验证了提出的Kinect角度识别与姿势评估的准确性，表明该方法关节角度识别与姿势风险评估的准确率均较高，是一种低成本、高可靠性的姿势评价方法，具有一定的科学意义和工程应用价值。     

变分自编码器模型综述

变分自编码器（VAE）作为深度隐空间生成模型的一种，近年来其表现性能取得了极大的成功，尤其是在图像生成方面。变分自编码器模型作为无监督式特征学习的重要工具之一，可以通过学习隐编码空间与数据生成空间的特征映射，进而在输出端重构生成输入数据。梳理了传统变分自编码器模型及其衍生变体模型的发展与研究现状，并就此做了总结和对比，最后分析了变分自编码器模型存在的问题与挑战，并就可能的发展趋势做了展望。     

结合MNF变换和Canny算子的遥感影像变化检测

针对大面积区域的多时相遥感影像变化检测的需求，提出了一种基于最小噪声分离（MNF）的Canny边缘检测提取影像变化信息的检测方法。对多时相影像采用多种变换组合成具有多维波段信息的影像，采用最小噪声分离法分离噪声并得到单波段差异图，通过Canny边缘检测法计算梯度幅值，采用高低双阈值法细化边缘，从而提取差异图变化边缘，有效突出了变化信息。以1995年和2003年加扎勒河的两期遥感影像为例，利用两时相影像进行土地覆被变化检测。实验结果表明，该方法适用于监测大面积区域内地物的突变情况。在数据基础上进行最小噪声分离可以有效解决传统Canny边缘检测提取边缘时造成的伪边缘现象，同时采用高低双阈值法有效去除伪边缘点，从而获得更加精确、直观的变化检测效果，在自然地理变化监测、地理国情灾害监测等有很好的应用价值。     

结合深度学习与相关反馈的遥感图像检索

目的 针对基于内容的图像检索存在低层视觉特征与用户对图像理解的高层语义不一致、图像检索的精度较低以及传统的分类方法准确度低等问题，提出一种基于卷积神经网络和相关反馈支持向量机的遥感图像检索方法。方法 通过对比度受限直方图均衡化算法对遥感图像进行预处理，限制遥感图像噪声的放大，采用自学习能力良好的卷积神经网络对遥感图像进行多层神经网络的监督学习提取丰富的图像特征，并将支持向量机作为基分类器，根据测试样本数据到分类超平面的距离进行排序得到检索结果，最后采用相关反馈策略对检索结果进行重新调整。结果 在UC Merced Land-Use遥感图像数据集上进行图像检索实验，在mAP（mean average precision）精度指标上，当检索返回图像数为100时，本文方法比LSH（locality sensitive Hashing）方法提高了29.4%，比DSH（density sensitive Hashing）方法提高了37.2%，比EMR（efficient manifold ranking）方法提高了68.8%，比未添加反馈和训练集筛选的SVM（support vector machine）方法提高了3.5%，对于平均检索速度，本文方法比对比方法中mAP精度最高的方法提高了4倍，针对复杂的遥感图像数据，本文方法的检索效果较其他方法表现出色。结论 本文提出了一种以距离评价标准为核心的反馈策略，以提高检索精度，并采用多距离结合的Top-k排序方法合理筛选训练集，以提高检索速度，本文方法可以广泛应用于人脸识别和目标跟踪等领域，对提升检索性能具有重要意义。