基于深度学习的高速公路团雾检测方法

随着我国高速公路总里程的迅猛增加,由团雾引发的交通事故也不断增多。团雾形成突然、范围小、分布随机性高,难以被气象预报所检测,且其浓度通常较高,严重影响了道路安全。传统的团雾检测方法多依托无线传感器、激光器等建立监测站,存在技术繁琐、实施困难、经济性差等缺点。为解决上述问题,论文提出了一种基于深度学习的高速公路团雾检测方法,以高速公路监控为依托,快速检测团雾等级,可以大大降低团雾检测成本。该检测方法以道路监控图像为输入,首先通过车道线分割网络得到二值化的车道线特征,并设计了浓雾区分割网络分支提取浓雾特征;然后综合可见车道线特征、浓雾区域特征、路面梯度特征输入至特征融合网络训练得到出路面水平可见距离,进而预测出路面团雾等级。实验结果表明,论文提出的快速团雾检测方法能够通过路面监控图像预测团雾等级,在复杂的团雾道路环境下具有较高的准确性和鲁棒性。

基于图像灰度熵的团块目标检测方法

针对面向自动目标识别(ATR)的图像数据压缩方法,感兴趣区域(ROI)检测器的设计是需要解决的一个关键问题。图像灰度熵可以有效地反映图像中像素灰度变化的剧烈程度,并且与图像均值有关。提出了一种基于图像灰度熵的团块目标检测方法,实现了对可见光和红外图像中坦克、飞机、舰船等团块目标的有效检测。实验表明该方法具有较好的稳健性和适应性,优于基于分水岭的检测方法,满足面向ATR图像数据压缩方法ROI检测器的设计要求。

基于视频图像对比度的团雾检测算法

提出了一种基于视频图像对比度的团雾检测算法。首先,利用二维多点各向异性高斯滤波器检测视频图像中边缘线的位置和方向。然后,利用本文提出的彩色图像对比度模型计算每幅视频图像边缘线区域的对比度值。最后,利用支撑向量机区分有雾视频图像所具有的不同对比度值,实现团雾浓度的检测。实验结果表明,该算法对复杂场景下的团雾检测效果优于传统的方法,是一种有效的团雾检测方法。

基于图像暗通道的高速公路团雾检测应用算法

针对高速公路上团雾预报的需要,提出一种基于图像暗通道理论的团雾检测预报算法。该算法从暗通道理论出发,通过对同一高速公路情景上的有雾及无图像雾参照对比,并进行代数逻辑运算,求出单一角度处的团雾扩散深度。经实验测试,该方法计算出的团雾雾厚距离与实际雾厚距离在数值上基本吻合,误差在±1. 5 m范围内,满足实际应用需求误差小于2 m的范围,具有较高的可信性及实际应用性。

基于视频局部特征分析的团雾检测方法研究

团雾检测在交通等领域应用需求强烈,文章介绍了基于视频局部特征分析技术的检测方法,能克服传统物理检测诸多因素的约束.还分析了进行团雾有效检测的具体实现方法与其技术特点,随着5G通信的日益发展,视频分析技术的应用将会更好地服务于智慧交通等领域的团雾实时检测预警.

基于视频的带电作业中组合间隙的智能检测

在带电作业实训项目中,作业人员进出电场需要保持正确的姿态,以保证自身的组合间隙小于最小组合间隙,避免引起安全事故,传统的带电作业培训中主要依靠现场教练的主观经验判断作业人员的实时组合间隙,判断的标准具有较强的主观性,从而导致一定的安全隐患。文中针对现有问题,创新性的提出了一种基于视频的智能检测算法,运用了自适应混合高斯背景建模、团块检测、前景融合等模式识别算法。该算法以500 k V输电线塔架为实验对象,实验表明该算法能够准确检测出作业人员的实时组合间隙,并在发生违规操作时,可及时准确的给予警示,有助于规范作业人员的操作,可有效提高带电作业的安全系数。

基于时间序列的团雾监测与预警系统

为解决团雾监测系统研发和建设中存在的硬件成本过高、监测数据分散及道路条件限制等问题,基于时间序列图像处理技术提出一种团雾监测与预警系统。首先提取海量现有视频监控数据,以10 s间隔分割视频并利用高斯混合模型提取背景序列图像,进行图像预处理。在此基础上,以图像对比度、灰度值和边缘特征分析突发团雾时背景像元变化规律,拟合特征值与能见度的关系,建立基于背景像元序列的能见度模型;通过相邻摄像头数据的对比分析,判断团雾;判定结果通过云端进行数据处理传输与信息发布,实现团雾的监测与预警。根据模型验证结果可知,该系统检测准确率相对于单幅图像检测准确率更高,系统运行成本低,可实现不同路段和道路全程的团雾监测。

基于自顶向下分裂聚类的图像匹配算法研究

为了提高图像匹配的效果,提出一种自顶向下分裂聚类的图像匹配算法,该算法可以获得多个目标级别的对应关系的聚类,进而找到两幅图像共存的多个目标。在互k近邻图表示模型的基础上,通过团检测方法来获得图中的团,主要是利用分裂聚类的思想,并定义了一个团密度函数,根据此函数来确定分裂终止条件。根据团检测技术获得的团恢复出团内的对应关系,从而达到图像匹配的目的。实验结果表明:该算法有较好的性能,可以应用到很多图像匹配问题中。

基于机器视觉的新生仔猪目标识别方法研究与实现

[目的]新生仔猪目标检测是母猪分娩监测的关键环节。[方法]通过自制图像采集器采集母猪分娩视频图像,机器视觉系统获取分娩图像信息,选取Canny算子对图像进行边缘检测,采用Otsu算法对图像进行二值变换,应用滑动平均算法和形态学开运算对二值图像滤波消噪,提取图像最大连通域,利用团序列检测算法对母猪目标进行分割,对分割后区域进行仔猪目标识别。[结果]试验结果表明,团序列检测算法能够准确分割出母猪目标,检测仔猪目标的正确率达到95.5%。[结论]提出一种能够有效识别新生仔猪目标的方法,为仔猪的出生预警提供了技术支撑。

高速公路团雾影响分析与检测预警系统开发

为解决团雾对高速公路交通运输造成的危害,保障团雾天气下高速公路的交通安全性和通畅性,通过总结团雾的特征与发生规律,团雾天气下高速公路发生交通事故的特点,分析团雾对高速公路交通安全的影响。结果表明：团雾对高速公路交通安全的影响主要在于降低能见度、削弱路面附着系数和增加驾驶人身心负担三个方面。针对雾气的消光性原理,提出高速公路团雾实时检测与预警系统。从模块的构建到设计原理的阐述,以及系统实施方式都具有较高的可行性、实用型和经济性,该系统的应用将能大大降低团雾对交通的危害,提高公路运行的科学管理水平和公路的运营效益。

市域高速公路恶劣天气预警智慧管控系统

以高速公路易发浓雾和团雾路段展开研究,利用高速公路路侧先进的气象检测设备对浓雾能见度进行识别,并利用其他交通特征参数检测设备,结合分析路段运行情况做出分析,最后结合路侧预警系统或者其他方式对道路使用者发出预警信息或道路运行状况信息。由实际运行效果可见,在易发生浓雾、团雾的高速路段,加装团雾检测和预警系统后,该路段因团雾和浓雾发生事故的概率大幅降低,团雾预警系统相较于其他而言具有更多的检测方式与数据分析模型,较大提升了团雾检测的准确性。

Palladium-Carbon Composite Nanoparticle Films with Enhanced Electrocatalytic Activity for Hydrogen Peroxide Sensors

A nanocomposite electrocatalyst was prepared with the method of cluster beam deposition of palladium nanoparticle thin lms on carbon nanoparticle supporting layers and used as sensitive nonenzyme hydrogen peroxide sensors. An enhancement on the electrocatalytic activity of the palladium nanoparticles toward H2O2 reduction was observed, which was related to the coverage of the carbon nanoparticles. With one monolayer of carbon nanoparticles, the H2O2 detection sensitivity reached the maximum, which was more than twice of that of the pure Pd nanoparticles.

Preliminary Data on Legionella Detection in Water Distribution Systems in Cameroon

After the declaration of the first case of Legionnaire＇s disease in Cameroon in 2007, the Centre Pasteur of Cameroon implemented the detection method for Legionella. The introduction of this new method was put in places in order to investigate Legionella spp. colonization of water distribution systems （WDS） of large buildings including hospitals, hotels and Off Shore Exploitations Sites （OSES） in an attempt to identify risk factors for Legionella spp. Water systems of 6 hotels, 6 hospitals and 6 ships were investigated for the presence of Legionella spp.. A total of 130 samples were collected, 77 from hotels, 27 from hospitals and 26 from ships. 51 Legionella spp. were isolated from 41 （31.54%） water samples. Of a total of 51 positive isolates, 40/51 （78.4%） were L. pneumophila with 21 （52.5%） Legionellapneumophila serogroup （sg） 1, 16 （40%） L. pneumophila sg 5, 2 （5%） L. pneumophila sg 6, 1 （2.5%） L. pneumophila sg 7 and 11/51 Legionella spp. with 10 （90%） L. anisa, 1 （10%） Legionella dumoffii. 5 L. pneumophila sg 1 were associated with 5 L. pneumophila sg 5 and 4 L. pneumophila sg 1 were associated with 4 L. anisa. These results showed that WDS of hospitals, hotels and ships can be heavily colonized by Legionella spp. and may present a risk of Legionnaires＇ disease. Based on these preliminary results, we have just put in place a Legionella survey protocol in Cameroon.

A Way Finding the Classical Ramsey Number and r（4,6）=36

Two definitions are given that Definitionl： an induced subgraph by a vertex vie G and its neighbors in G is defined as a vertex adjacent closed subgraph, and denoted by Qi （=G[V（Nvi）]）, with the vertex vi called the hub; and Definition2： A r（k,I）-I vertices graph is called the （k,l）-Ramsey graph, denoted by RG（k,1）, if RG（k,1） only contains cliques Kk.1 and the intersect QiNQj of any two nonadjacent vertices vi and vj of RG（k,I） contains only Kk-2. Meanwhile, the RG（k,l）＇s complement RG（I,k） contains only cliques Kl.l, and the intersect QiNQj of any two nonadjacent vertices vi and vj of RG（I,k） contains only Ki.2. On the basis of those definitions, two theorems are put forward and proved in this paper. They are Theoreml： the biggest clique in G is contained in some Qi of G, and Theorem 2： r（k,1） = [V（RG（k,I））I ＋ 1. With those definitions and theorems as well as analysis of chord property, a method for quick inspection and building RG（k,1） is proposed. Accordingly, RG（4,6） is built, it is a strongly 14-regular graph on order 35. We have tested RG（4,6） and its complement, as a result, they meet the defintion2, so we proclaim that r（4,6）=36.

Meso-damage cracking characteristics analysis for rock and soil aggregate with CT test

The aim of this paper is to investigate the damage cracking characteristics of rock and soil aggregate(RSA)by X-ray computed tomography(CT)under uniaxial compressive loading.The mean CT value for the region of interest(ROI)is used to analyze the cracking characteristics.Also,the mathematical morphology method based on the image threshold segmentation is used to obtain characteristic parameters of cracks to describe the cracking evolution of RSA.Results show that the elastic mismatch between rock blocks and soil matrix is the primary reason for RSA cracking.The mean CT value for the RSA specimen,rock block inclusions,and their adjacent soil regions decreases with the increasing stress level.However,it is more sensitive for block inclusions than soil regions.Using the image segmentation method,length,area and mean width of cracks obey to power function distribution.Crack statistical characteristics are closely related to the rock block’s distribution and morphology.These results may be useful to reveal the mesoscopic cracking mechanism,establish meso-damage evolution equation,and constitutive relation for RSA.