基于各向异性测度的路面三维图像裂缝识别

为准确而完整地识别路面裂缝,提出了基于1 mm/像素的路面三维图像裂缝自动识别算法．该算法主要包括各向异性测度计算与自适应优化阈值分割、深度验证和多分辨率去噪处理3个部分．首先,针对路面图像像素特征,基于0°、45°、90°和135°四个方向的线性邻域的均值和标准差计算每个像素的各向异性测度（表征方向性的强弱）,并应用最大类间方差法确定最优阈值,将路面图像分为强方向性和弱方向性像素两类;其次,根据半径为d的正方形邻域深度均值设定阈值,用方向性强且深度低于或等于该阈值的像素形成初步的裂缝图像;最后,将裂缝图像划分为多个子块,设计去噪模板对裂缝图像进行滑动窗口去噪处理,获得最终裂缝图像．基于166幅含有各类裂缝的三维路面图像（2048×2048像素）进行测试分析,结果显示,本文算法获得了较高的准确率（均值91．57％）和召回率（均值81．29％）,最终以84．26％的F1均值优于种子识别算法（F1均值69．19％）、Canny边缘检测（F1均值8．15％）和OTSU分割（F1均值5．11％）．     

基于航拍图像的海上目标定位系统

针对海事闭路电视(CCTV)系统未能直接通过视频图像获得海上目标位置信息的现状,提出一种基于航拍图像的海上目标定位系统.该系统基于Visual C++开发平台,以模块化设计思路,实现基于航拍图像的海上目标定位功能.介绍系统组成、主要功能、概位算法以及该定位系统在海事监管、海上搜救中的应用.该定位系统可弥补海事CCTV监管的不足,提升海事监管、海上搜救的效率.     

The motion analysis of fire video images based on moment features and flicker frequency

In this paper, motion analysis methods based on the moment features and flicker frequency features for early fire flame from ordinary CCD video camera were proposed, and in order to describe the changing of flame and disturbance of non-flame phenomena further more, the average changing pixel number of the first-order moments of consecutive flames has been defined in the moment analysis as well. The first-order moments of all kinds of flames used in our experiments present irregularly flickering, and their average changing pixel numbers of first-order moments are greater than fire-like disturbances. For the analysis of flicker frequency of flame, which is extracted and calculated in spatial domain, and therefore it is computational simple and fast. The method of extracting flicker frequency from video images is not affected by the catalogues of combustion material and distance. In experiments, we adopted two kinds of flames, i.e. , fixed flame and movable flame. Many comparing and disturbing experiments were done and verified that the methods can be used as criteria for early fire detection.     

基于光流和水平集模型的运动目标自动跟踪方法

在分析直升机在天空飞行序列图像特点的基础上,提出了基于光流和水平集模型的运动目标自动跟踪方法。该方法通过对图像进行高通滤波预处理,获得运动目标的特征图像,用特征图像来求解运动目标的光流,用光流对运动区域进行分割,获得运动区域的中心和半径,以该中心和半径的圆作为初始水平集曲线,以光流速度作为曲线演化的外力修改Chan and Vese模型,用改进后的模型对图像进行分割。实验证明,用本文方法能够快速、准确地自动跟踪运动目标。     

基于改进YOLOv5的雾霾环境下船舶红外图像检测算法

从监控图像中准确检测船舶对于港区水域船舶交通智能监管具有重要意义。为解决雾霾条件下传统YOLOv5目标检测算法对船舶红外图像检测准确率低、小目标特征提取能力弱等问题,提出了基于Swin Transformer的改进YOLOv5船舶红外图像检测算法。为扩大原始数据集的多样性,综合考虑船舶红外图像轮廓特征模糊、对比度低、抗云雾干扰能力强等特点,改进算法提出基于大气散射模型的数据集增强方法;为增强特征提取过程中全局特征的关注能力,改进算法的主干网络采用Swin Transformer提取船舶红外图像特征,并通过滑动窗口多头自注意力机制扩大窗口视野范围;为增强网络对密集小目标空间特征提取能力,通过改进多尺度特征融合网络（PANet）,引入底层特征采样模块和坐标注意力机制（CA）,在注意力中捕捉小目标船舶的位置、方向和跨通道信息,实现小目标的精确定位;为降低漏检率和误检率,采用完全交并比损失函数（CIoU）计算原始边界框的坐标预测损失,结合非极大抑制算法（NMS）判断并筛选候选框多次循环结构,提高目标检测结果的可靠性。实验结果表明：在一定浓度的雾霾环境下,改进算法的平均识别精度为93.73%,平均召回率为98.10%,平均检测速率为每秒38.6帧;与RetinaNet、Faster R-CNN、YOLOv3 SPP、YOLOv4、YOLOv5和YOLOv6-N算法相比,其平均识别精度分别提升了13.90%、11.53%、8.41%、7.21%、6.20%和3.44%,平均召回率分别提升了11.81%、9.67%、6.29%、5.53%、4.87%和2.39%。综上,所提的Swin-YOLOv5s改进算法对不同大小的船舶目标识别均具备较强的泛化能力,并具有较高的检测精度,有助于提升港区水域船舶的监管能力。     

基于亮度特性的高空间分辨率遥感影像阴影处理研究

在高密度的城市环境中,建筑物产生的阴影遮挡了许多重要信息,导致车辆信息提取不准确甚至无法提取,因此对阴影的处理显得尤为重要.利用遥感影像阴影的几何特征和光谱特征,采用基于亮度的阴影检测方法,通过双峰曲线选取合适的亮度阈值把阴影和光照区进行分割;并用灰度线性映射的阴影补偿方法,通过线性映射对阴影区域进行信息补偿,对高空间分辨率遥感影像的样本数据中的阴影进行试验处理和结果分析,结果证明这一方法是行之有效的.     

局部紧度量空间的闭映象

借助弱紧k网络,获得了局部紧度量空间闭映象内部特征一个新的刻画,即空间X是局部紧度量空间的闭映象当且仅当X是具有点可数的弱紧k网络的Fr(e)chet空间.     

Elements of automated survey of pavements and a 3D methodology

Sound transportation infrastructure is critical for economic development and sustainability.Pavement condition is a primary concern among agencies of the roadway infrastructure.Automation has become possible in recent years on collecting data and producing results for certain aspects of pavement performance,while challenges remain in several other categories,such as automated cracking survey.This paper reviews the technological advances on automated survey of pavements,and discusses the most recent breakthr...     

大型编组站视频监控系统方案研究

研究目的：铁路大型编组站规模庞大、场区布局分散、环境复杂、部门众多，给维护管理带来极大的不便。为保证编组站货运安全、减少事故发生，针对编组站视频监控系统构成进行研究，提出合理方案，为编组站设计提供借鉴。研究结论：以新丰镇编组站视频监控系统构成为例，阐述了编组站监控系统的构成方案。通过分析比较，采用模拟采集＋数字处理监控技术，可充分利用既有监控设备和骨干传输平台，满足了各监控中心对画面控制和图像清晰度的要求。采用H．264标准图像压缩编码方式，码流结构网络适应性强，增加了差错恢复能力，能更好地适应IP和无线网络的应用。为保证视频监控系统的正常运行，应采取防雷措施。     

Multi-Model Ensemble Deep Learning Method to Diagnose COVID-19 Using Chest Computed Tomography Images

Deep learning based analyses of computed tomography (CT) images contribute to automated diagnosis of COVID-19,and ensemble learning may commonly provide a better solution.Here,we proposed an ensemble learning method that integrates several component neural networks to jointly diagnose COVID-19.Two ensemble strategies are considered:the output scores of all component models that are combined with the weights adjusted adaptively by cost function back propagation;voting strategy.A database containing 8347 CT slices of COVID-19,common pneumonia and normal subjects was used as training and testing sets.Results show that the novel method can reach a high accuracy of 99.37％ (recall:0.9981;precision:0.9893),with an increase of about 7％in comparison to single-component models.And the average test accuracy is 95.62％ (recall:0.9587;precision:0.9559),with a corresponding increase of 5.2％.Compared with several latest deep learning models on the identical test set,our method made an accuracy improvement up to 10.88％.The proposed method may be a promising solution for the diagnosis of COVID-19.