In this study,an electron microprobe analyzer(EMPA) was used to map the spatial distribution and the occurrence of invisible gold in pyrite from the Liulincha gold ore belt.EPMA data show that gold mainly occurs as su...In this study,an electron microprobe analyzer(EMPA) was used to map the spatial distribution and the occurrence of invisible gold in pyrite from the Liulincha gold ore belt.EPMA data show that gold mainly occurs as submicroscopic-microscopic inclusions.From the contrast of the major guide elements of pyrite from the Liulincha gold ore belt and those from four hydrothermal-type gold deposits in the Jiaodong region,we can see the pyrites were formed in two stages:the pyrite from wall rock is mainly sedimentogenic,with simple structure;and the pyrite from ore body experienced early sedimentary process to late hydrothermal activity,the pyrite is regular in crystal form and exhibits fractured structure.展开更多
目的车辆多目标跟踪是智能交通领域关键技术,其性能对车辆轨迹分析和异常行为鉴别有显著影响。然而,车辆多目标跟踪常受外部光照、道路环境因素影响,车辆远近尺度变化以及相互遮挡等干扰,导致远处车辆漏检或车辆身份切换(ID switch,IDs...目的车辆多目标跟踪是智能交通领域关键技术,其性能对车辆轨迹分析和异常行为鉴别有显著影响。然而,车辆多目标跟踪常受外部光照、道路环境因素影响,车辆远近尺度变化以及相互遮挡等干扰,导致远处车辆漏检或车辆身份切换(ID switch,IDs)问题。本文提出短时记忆与CenterTrack的车辆多目标跟踪,提升车辆多目标跟踪准确度(multiple object tracking accuracy,MOTA),改善算法的适应性。方法利用小样本扩增增加远处小目标车辆训练样本数;通过增加的样本重新训练CenterTrack确定车辆位置及车辆在相邻帧之间的中心位移量;当待关联轨迹与检测目标匹配失败时通过轨迹运动信息预测将来的位置;利用短时记忆将待关联轨迹按丢失时间长短分级与待匹配检测关联以减少跟踪车辆IDs。结果在交通监控车辆多目标跟踪数据集UA-DETRAC(University at Albany detection and tracking)构建的5个测试序列数据中,本文方法在维持CenterTrack优势的同时,对其表现不佳的场景获得近30%的提升,与YOLOv4-DeepSort(you only look once—simple online and realtime tracking with deep association metric)相比,4种场景均获得近10%的提升,效果显著。Sherbrooke数据集的测试结果,本文方法同样获得了性能提升。结论本文扩增了远处小目标车辆训练样本,缓解了远处小目标与近处大目标存在的样本不均衡,提高了算法对远处小目标车辆的检测能力,同时短时记忆维持关联失败的轨迹运动信息并分级匹配检测目标,降低了算法对跟踪车辆的IDs,综合提高了MOTA。展开更多
文摘In this study,an electron microprobe analyzer(EMPA) was used to map the spatial distribution and the occurrence of invisible gold in pyrite from the Liulincha gold ore belt.EPMA data show that gold mainly occurs as submicroscopic-microscopic inclusions.From the contrast of the major guide elements of pyrite from the Liulincha gold ore belt and those from four hydrothermal-type gold deposits in the Jiaodong region,we can see the pyrites were formed in two stages:the pyrite from wall rock is mainly sedimentogenic,with simple structure;and the pyrite from ore body experienced early sedimentary process to late hydrothermal activity,the pyrite is regular in crystal form and exhibits fractured structure.
文摘目的车辆多目标跟踪是智能交通领域关键技术,其性能对车辆轨迹分析和异常行为鉴别有显著影响。然而,车辆多目标跟踪常受外部光照、道路环境因素影响,车辆远近尺度变化以及相互遮挡等干扰,导致远处车辆漏检或车辆身份切换(ID switch,IDs)问题。本文提出短时记忆与CenterTrack的车辆多目标跟踪,提升车辆多目标跟踪准确度(multiple object tracking accuracy,MOTA),改善算法的适应性。方法利用小样本扩增增加远处小目标车辆训练样本数;通过增加的样本重新训练CenterTrack确定车辆位置及车辆在相邻帧之间的中心位移量;当待关联轨迹与检测目标匹配失败时通过轨迹运动信息预测将来的位置;利用短时记忆将待关联轨迹按丢失时间长短分级与待匹配检测关联以减少跟踪车辆IDs。结果在交通监控车辆多目标跟踪数据集UA-DETRAC(University at Albany detection and tracking)构建的5个测试序列数据中,本文方法在维持CenterTrack优势的同时,对其表现不佳的场景获得近30%的提升,与YOLOv4-DeepSort(you only look once—simple online and realtime tracking with deep association metric)相比,4种场景均获得近10%的提升,效果显著。Sherbrooke数据集的测试结果,本文方法同样获得了性能提升。结论本文扩增了远处小目标车辆训练样本,缓解了远处小目标与近处大目标存在的样本不均衡,提高了算法对远处小目标车辆的检测能力,同时短时记忆维持关联失败的轨迹运动信息并分级匹配检测目标,降低了算法对跟踪车辆的IDs,综合提高了MOTA。
