瓷砖表面非常规尺寸瑕疵的检测

与普通目标检测任务不同, 瓷砖表面瑕疵检测的困难之处在于检测小尺寸和大长宽比等非常规尺寸的目标. 为了解决这两个问题, 本文提出了一种基于改进 Cascade R-CNN 的新型瓷砖表面瑕疵检测算法. 为了提高模型对小瑕疵的检测能力, 本文模型利用侧向连接结构进行上下层语义信息的融合, 使用可切换空洞率的空洞卷积来增加模型的感受野; 为了提高模型对于大长宽比瑕疵的检测能力, 本文模型在标准卷积上引入偏移域以更好提取目标特征信息. 此外, 本文模型调整 Cascade R-CNN 框架中预选锚框的大小和长宽比例. 实验结果表明, 在从瓷砖工厂收集的数据集上, 本文所提出算法的平均精度均值 (mean average precision, mAP) 达到了 73.5%, 比改进前的 Cascade R-CNN 模型提高了9.7%. 本文实验代码可从以下链接获取: https://github.com/mashibin/Ceramic-tile-defect-detection.     

一种基于证据推理的视频运动目标融合检测算法

为了解决复杂场景下传统的运动目标检测问题, 利用证据推理—–谨慎有序加权平均方法(COWA-ER), 提出一种综合使用混合高斯、均值滤波和码本的多方法融合的检测方法. 该融合检测算法以上述3 种检测方法为准则建立一个多准则决策框架, 通过双阈值检测法来表征检测过程中的不确定性, 最终利用COWA-ER 方法进行决策级融合, 实现多种方法的优势互补. 实验表明, 所提出的融合检测算法具有更理想的目标检测效果, 能有效应对诸如阴影及光照突变等问题对检测性能的影响.

     

Improved magnetic particle tracking technique in dense gas fluidized beds

Noninvasive monitoring of multiphase flow is rapidly gaining increased interest. More specifically noninvasive particle tracking techniques have received a lot of attention in recent years to study dense granular flow. However, these techniques are usually quite expensive and require strict safety measures. An improved magnetic particle tracking (MPT) technique for dense granular flow will be presented in this article. The improvements of the analysis technique for MPT will be demonstrated and rigorously tested with a three‐dimensional system and two‐dimensional sensor system. The strengths and limitations of the MPT technique will also be reported. Finally, the results of the MPT are compared with data obtained from a combined particle image velocimetry and digital image analysis technique. © 2014 American Institute of Chemical Engineers AIChE J, 60: 3133–3142, 2014     

Distribution of large biomass particles in a sand‐biomass fluidized bed: Experiments and modeling

The axial distribution of large biomass particles in bubbling fluidized beds comprised of sand and biomass is investigated in this study. The global and local pressure drop profiles are analyzed in mixtures fluidized at superficial gas velocities ranging from 0.2 to 1 m/s. In addition, the radioactive particle tracking technique is used to track the trajectory of a tracer mimicking the behavior of biomass particles in systems consisting of 2, 8, and 16% of biomass mass ratio. The effects of superficial gas velocity and the mixture composition on the mixing/segregation of the bed components are explored by analyzing the circulatory motion of the active tracer. Contrary to low fluidization velocity (U = 0.36 m/s), biomass circulation and distribution are enhanced at U = 0.64 m/s with increasing the load of biomass particles. The axial profile of volume fraction of biomass along the bed is modeled on the basis of the experimental findings. © 2014 American Institute of Chemical Engineers AIChE J, 60: 869–880, 2014     

轮式移动机器人滑移轨迹跟踪控制策略研究

针对轮式移动机器人在滑移条件下的轨迹跟踪问题,提出一种自适应模糊控制策略。利用模糊状态观测器对未知的复杂系统模型进行速度估计和补偿,通过设计输出向量将位置约束转换为输出约束,并构造Barrier Lyapunov函数来保证机器人的运动约束,在此基础上设计更简单的纵向滑移模型,利用李亚普诺夫定理分析闭环系统的稳定性,证明了闭环系统中所有信号都是有界的。最后通过仿真和实验验证了自适应模糊控制策略在轮式机器人纵向滑移情况下的有效性和实用性。     

基于深度学习的单目视觉水下机器人目标跟踪方法研究

介绍了一种基于深度学习的单目视觉水下机器人目标跟踪新方法,采用该方法能够使自主水下机器人(AUV)在水下复杂环境中实现图像增强处理与目标识别跟踪,对于视频图像中的每个图像,使用预先经过训练的深度卷积神经网络计算图像传输图,图像传输图提供了图像深度信息的相关估计,该方法能够识别水下目标区域,并标明水下目标运动及跟踪方向。实验表明:该方法能够更精确、更稳定地获取水下环境中的定位数据。     

Multiple lines of evidence to identify the sources of fecal pollution at a freshwater beach in Hamilton Harbour, Lake Ontario

Multiple microbial source-tracking methods were investigated to determine the source of elevated Escherichia coli levels at Bayfront Park Beach in Hamilton Harbour, Lake Ontario. E. coli concentrations were highest in wet foreshore sand (114,000 CFU/g dry sand) and ankle-depth water (177,000 CFU/100mL), declining rapidly in deeper waters. Many gull and geese droppings were enumerated each week on the foreshore sand within 2m of the waterline. Both antimicrobial resistance analysis and rep-PCR DNA fingerprinting of E. coli collected at the beach and nearby fecal pollution sources indicated that E. coli in sand and water samples were predominantly from bird droppings rather than from pet droppings or municipal wastewater. Both methods indicated a trend of decreasing bird contamination, and increasing wastewater contamination, moving offshore from the beach. When foreshore sand was treated as a reservoir and secondary source of E. coli, waterborne E. coli were found to be more similar to sand isolates than bird or wastewater isolates out to 150 m offshore. Multiple lines of evidence indicated the importance of bird droppings and foreshore sand as primary and secondary sources of E. coli contamination in beach water at Bayfront Park.     

Microbial source tracking using host specific FAME profiles of fecal coliforms

The objective of this study was to investigate the host-specific differences in fatty acid methyl ester (FAME) profiles of fecal coliforms (FC). A known-source library was constructed with 314 FC isolates cultured from 6 possible sources of fecal pollution; 99 isolates from sewage; 29 from bovine; 29 from poultry; 50 from swine; 46 from waterfowl; and 61 from deer. It was found that the hydroxy FAMEs 12:0 2 OH, 12:03 OH, and 14:02 OH were exclusively associated with isolates of human origin. On the other hand, 3 saturated FAMEs, 10:0, 15:0, and 18:0 were found only in isolates from non-human sources, 15:0 being associated with livestock samples only. In addition to the presence of these signature FAMEs, the mean relative masses of 16:1 omega7c and 16:1 ISO/14:03 OH were significantly different between the isolates of human and non-human origins. A linear discriminant function differentiated FC isolates of human origin from those of livestock and wildlife origin at 99% accuracy. These results strongly suggest that the FAME profiles of FC show statistically significant host specificity and may have the potential to be used as a phenotypic microbial source tracking tool.     

Comparison of the efficacy of an existing versus a locally developed metabolic fingerprint database to identify non-point sources of faecal contamination in a coastal lake

A comparison of the efficacy of an existing large metabolic fingerprint database of enterococci and Escherichia coli with a locally developed database was undertaken to identify the sources of faecal contamination in a coastal lake, in southeast Qld., Australia. The local database comprised of 776 enterococci and 780 E. coli isolates from six host groups. In all, 189 enterococci and 245 E. coli biochemical phenotypes (BPTs) were found, of which 118 and 137 BPTs were unique (UQ) to host groups. The existing database comprised of 295 enterococci UQ-BPTs and 273 E. coli UQ-BPTs from 10 host groups. The representativeness and the stability of the existing database were assessed by comparing with isolates that were external to the database. In all, 197 enterococci BPTs and 179 E. coli BPTs were found in water samples. The existing database was able to identify 62.4% of enterococci BPTs and 64.8% of E. coli BPTs as human and animal sources. The results indicated that a representative database developed from a catchment can be used to predict the sources of faecal contamination in another catchment with similar landuse features within the same geographical area. However, the representativeness and the stability of the database should be evaluated prior to its application in such investigation.     

三高测量技术探测岩溶

阐述了三高测量技术的内涵和特点，以珠江水泥厂岩溶探测为例，阐述了三高技术对溶岩探测的应用及效果，经处理直观，形象地反映了岩溶的发育分布情况及形态。