Yolov7-pose识别运动员姿态时关键点异常的消除方法

本研究首先构建了深度聚合网络（deep layer aggregation,DLA）,用于替换Yolov7-pose算法中原有的特征融合网络,增强了特征的表达;其次,将高效聚合网络（efficient layer aggregation Net, ELAN）和尺度感知注意力网络（scale-aware attention Net, SAAN）融合在一起,引入到Yolov7-pose算法中;然后,引入尺度因子到Yolov7-pose算法的损失函数中,并通过实验确定尺度因子最佳值。实验结果表明,经过本法改进的方法识别篮球比赛视频流中的多名运动员姿态时,不再出现关键点异常现象,姿态识别的精度达到95%以上,运行效率达到24 FPS以上,综合性能显著优于原始Yolov7-pose算法等算法。     

基于DBN特征提取的雷达辐射源个体识别

针对传统特征参数难以表征复杂体制雷达信号个体特征的问题,基于深度置信网络DBN的深层特征提取和高维数据处理能力,提出一种基于DBN特征提取的雷达辐射源个体识别算法。首先建立基于多层受限玻耳兹曼机的DBN模型,然后通过DBN无监督提取脉冲包络前沿特征,再利用标签数据对模型参数进行有监督微调完成训练,最后输入未知辐射源信号脉冲包络前沿特征实现辐射源个体识别。与传统算法相比,该方法能够自适应地提取脉冲深层次细微差异,提取过程减少了对人为经验的依赖。实验结果表明,该算法对脉冲包络特征提取效果明显,有较高的识别精度。     

基于卷积核分解的深度CNN模型结构优化及其在小图像识别中的应用

小图像由于像素少、分辨率低、整幅图像包含信息较少,识别较为困难。目前优秀的深度卷积神经网络模型多为大图像而设计,而用于小图像的模型则存在着层次不够深、难以对特征进行充分抽象的不足。本文基于VGG19模型,依据卷积核分解的原理,设计了一种KDS-DCNN模型,模型深度达到31层,解决了目前超深度模型不能直接用于小图像识别的问题,实验表明该方法不但提升了识别性能,而且还降低了模型的时间复杂度。在CIFAR-10、CIFAR-100和SVHN三个数据集上的验证结果显示,KDS-DCNN模型性能优越,其识别错误率分别降低到29.46%、6.02%和2.17%。     

尿素深施条件下模拟稻田中氮磷的动态特征及其降污潜力分析

采用室外微区模拟试验,于水稻返青生长期,采用尿素造型深施方式,在3、7和11cm等3个不同施肥深度处理(分别表示为t-3、t-7、t-11)条件下,对田面水和土壤中氮、磷、以及悬浮物(SS)等主要污染物的动态特征及其降污潜力进行了研究。结果表明,尿素深施后5～7d内,田面水中氮素呈总体下降趋势,于第7d后达较低的浓度水平;硝氮(NO3--N)各处理间浓度差异较小,但整体上要高于铵氮(NH4+-N),就污染源而言,田面水NO3--N负荷不容忽视。田面水氮素浓度与尿素深施处理没有相关性。与对照处理撒施(表示为t-c)相比,深施处理的田面水中氮素浓度可分别降低NO3--N 29%～47%、NH4+-N 64%～89%、总氮(TN)79%～97%。t-3和t-7的田面水总磷(TP)和可溶性磷(DP)浓度都处于较高水平,与t-7和t-3相比较,t-11可分别降低TP浓度32%～73%、DP浓度92%～99%以及SS浓度50%～80%。因此,尿素造型深施至11cm左右,具有显著的减排降污潜力,是控制农田面源污染的有效措施之一。     

Sea surface temperature records of core ZY2 from the central mud area in the South Yellow Sea during last 6200 years and related effect of the Yellow Sea Warm Current

Sea surface temperature (SST) records in the South Yellow Sea during the last 6200 years are reconstructed by the unsaturation index of long-chain alkenones (K 37 U ’) in sediment core ZY2 from the central mud area.The SST records varied between 14.1 and 16.5°C (15.6°C on average),with 3 phases:(1) A high SST phase at 6.2-5.9 cal ka BP;(2) A low and intensely fluctuating SST phase at 5.9-2.3 cal ka BP;and (3) A high and stable SST phase since 2.3 cal ka BP.Variation of the SST records is similar to intensity of the Kuroshio Current (KC),and corresponds well in time to global cold climate events.However,the amplitude of the SST response to cooling events was significantly different in different phases.The SST response to global cooling event was weak while the KC was strong;and the SST response was strong while the KC was weak.The difference in amplitude of the SST response is possibly caused by the modulation effect of the Yellow Sea Warm Current which acts as a shelf branch of the KC and a compensating current induced by the East Asia winter monsoon.The warm waters brought by the Yellow Sea Warm Current cushion the SST decrease induced by climate cooling,and both the Kuroshio and East Asian winter monsoon play important roles in the modulation mechanism.The SST records display a periodicity of 1482 years.The same period was found in the KC records,indicating that variation of the SST records in the central South Yellow Sea is strongly affected by KC intensity.The same period was also found in Greenland ice cores and North Atlantic and Arabian Sea sediment cores,showing a regional response of marine environmental variability in the East China Seas to that in the global oceans.     

Molybdenum isotopic composition as a tracer for low-medium temperature hydrothermal ore-forming systems: A case study on the Dajiangping pyrite deposit,western Guangdong Province,China

Little is known about the range or controls on the molybdenum isotopic composition of low-medium temperature hydrothermal ore-forming systems. We present molybdenum isotope data from 12 hydrothermal syndepositional silicalite and carbonaceous slate samples from the Dajiangping pyrite deposit in western Guangdong Province, South China. The δ ^97/95 Mo values from Orebody III range from −0.02‰ to 0.29‰, with an average of 0.18‰. In contrast, the composition values from Orebody IV display a larger variation from −0.70‰ to 0.62‰. However, the five samples from the main ore bed all show strong negative values. This indicates that the significant variation in molybdenum isotopic composition supports different hydrothermal ore-forming metallogenesis and a metallogenic environment between the two orebodies. Orebody III is likely to have been deposited from submarine exhalative hydrothermal fluids under a relatively strong reducing environment and Orebody IV may have also been influenced by hydrothermal superimposition in a more oxidized disequilibrium condition. In addition, the δ ^97/95 Mo values of Orebody IV are clearly negative, together with the values increasing stratigraphically upward in the ore beds, suggesting that the metallogenic environment of Orebody IV is different from the open oceanic systems. There might be dynamic fractionation in this restricted environment. Therefore, the molybdenum isotope can be used as an effective tracer for the ore-forming fluid and metallogenic environment.     

脑深部神经刺激微电极的表面改性研究

为了提高脑深部刺激电极的生物相容性,降低脑组织的免疫反应,减少或抑制电极周围包裹物的形成,对电极植入端聚氨酯材料进行了表面改性处理.首先用N₂/H₂进行等离子体处理,在电极表面生成活性基团——氨基,再利用电极表面的氨基基团与YIGSR多肽分子发生聚合反应,从而实现在电极表面修饰上对神经细胞具有促进生长作用的多肽分子.改性电极的大鼠植入实验结果发现,改性电极能够在一定程度上减少胶质包裹物的形成,有利于电极与神经细胞之间接触,增加有效刺激体积,证明了电极聚氨酯材料表面接枝聚合生物大分子的可行性,对今后电极表面改性的深入研究具有一定的参考意义.     

工业废渣对海泥烧结砖泛霜程度影响的试验

研究两种不同特性工业废渣铁尾矿、粉煤灰对海泥烧结多孔砖样品泛霜程度的影响及机理.结果表明:铁尾矿减轻海泥烧结砖泛霜程度的有利因素多,粉煤灰提高海泥烧结砖泛霜程度的不利因素多.此外,掺加外加剂可以在某种程度上降低海泥烧结砖的泛霜等级;当掺量达到某一范围时,能够较大幅度减少砖坯成型用水量,烧成中产生液相的非瘠性工业废渣-铁...     

贵州天柱坑头金矿构造变形特征探讨

天柱坑头金矿床主要受层间构造控制,金矿体主要富集在层间滑动构造带.在纵弯褶皱作用过程中,由于顺层差异性滑动,形成一系列的层间滑动构造.通过对坑头金矿控矿构造中变形岩石及石英脉体的宏观变形特征分析,初步得出研究区控制金矿的层间滑动变形带以脆韧性变形为主.     

深基坑出碴方案比选

深基坑出碴在一些情况下会成为制约工期的关键因素,尤其是复杂地形条件下,在首先满足工期的前提下,还要综合考虑现有机械情况、技术条件和经济效益以及其他作业干扰,这使得深基坑出碴变得更加复杂。文章通过准朔铁路黄河特大桥工程特点,对几种深基坑出碴方案进行了分析对比,综合考虑后选择了履带吊出碴方案。