A Pooling Method Developed for Use in Convolutional Neural Networks

In convolutional neural networks,pooling methods are used to reduce both the size of the data and the number of parameters after the convolution of the models.These methods reduce the computational amount of convolutional neural networks,making the neural network more efficient.Maximum pooling,average pooling,and minimum pooling methods are generally used in convolutional neural networks.However,these pooling methods are not suitable for all datasets used in neural network applications.In this study,a new pooling approach to the literature is proposed to increase the efficiency and success rates of convolutional neural networks.This method,which we call MAM(Maximum Average Minimum)pooling,is more interactive than other traditional maximum pooling,average pooling,and minimum pooling methods and reduces data loss by calculating the more appropriate pixel value.The proposed MAM pooling method increases the performance of the neural network by calculating the optimal value during the training of convolutional neural networks.To determine the success accuracy of the proposed MAM pooling method and compare it with other traditional pooling methods,training was carried out on the LeNet-5 model using CIFAR-10,CIFAR-100,and MNIST datasets.According to the results obtained,the proposed MAM pooling method performed better than the maximum pooling,average pooling,and minimum pooling methods in all pool sizes on three different datasets.

Bayesian partial pooling to reduce uncertainty in overcoring rock stress estimation

The state of in situ stress is a crucial parameter in subsurface engineering,especially for critical projects like nuclear waste repository.As one of the two ISRM suggested methods,the overcoring(OC)method is widely used to estimate the full stress tensors in rocks by independent regression analysis of the data from each OC test.However,such customary independent analysis of individual OC tests,known as no pooling,is liable to yield unreliable test-specific stress estimates due to various uncertainty sources involved in the OC method.To address this problem,a practical and no-cost solution is considered by incorporating into OC data analysis additional information implied within adjacent OC tests,which are usually available in OC measurement campaigns.Hence,this paper presents a Bayesian partial pooling(hierarchical)model for combined analysis of adjacent OC tests.We performed five case studies using OC test data made at a nuclear waste repository research site of Sweden.The results demonstrate that partial pooling of adjacent OC tests indeed allows borrowing of information across adjacent tests,and yields improved stress tensor estimates with reduced uncertainties simultaneously for all individual tests than they are independently analysed as no pooling,particularly for those unreliable no pooling stress estimates.A further model comparison shows that the partial pooling model also gives better predictive performance,and thus confirms that the information borrowed across adjacent OC tests is relevant and effective.

U-Net Based Dual-Pooling Segmentation of Bone Metastases in Thoracic SPECT Bone Scintigrams

In order to enhance the performance of the CNN-based segmentation models for bone metastases, this study proposes a segmentation method that integrates dual-pooling, DAC, and RMP modules. The network consists of distinct feature encoding and decoding stages, with dual-pooling modules employed in encoding stages to maintain the background information needed for bone scintigrams diagnosis. Both the DAC and RMP modules are utilized in the bottleneck layer to address the multi-scale problem of metastatic lesions. Experimental evaluations on 306 clinical SPECT data have demonstrated that the proposed method showcases a substantial improvement in both DSC and Recall scores by 3.28% and 6.55% compared the baseline. Exhaustive case studies illustrate the superiority of the methodology.

Expanding the liver donor pool worldwide with hepatitis C infected livers, is it the time?

Liver transplantation(LT)provides a life-saving option for cirrhotic patients with complications and hepatocellular carcinoma.Despite the increasing number of liver transplants performed each year,the number of LT candidates on the waitlist remains unchanged due to an imbalance between donor organ supply and the demand which increases the waitlist time and mortality.Living donor liver transplant had a great role in increasing the donor pool and shortened waitlist time for LT candidates.Nevertheless,further strategies can be implemented to increase the pool of potential donors in deceased donor LT,such as reducing the rate of organ discards.Utilizing hepatitis C virus(HCV)seropositive liver grafts is one of the expanded donor organ criteria.A yearly increase of hundreds of transplants is anticipated as a result of maximizing the utilization of HCV-positive organs for HCV-negative recipients.Direct-acting antiviral therapy's efficacy has revolutionized the treatment of HCV infection and the use of HCV-seropositive donors in transplantation.The American Society of Transplantation advises against performing transplants from HCV-infected liver donors(D+)into HCV-negative recipient(R-)unless under Institutional Review Board-approved study rules and with full informed consent of the knowledge gaps associated with such transplants.Proper selection of patients to be transplanted with HCV-infected grafts and confirming their access to direct-acting antivirals if needed is im-portant.National and international consensuses are needed to regulate this process to ensure the maximum benefit and the least adverse events.

Simulation of single bubble dynamic process in pool boiling process under microgravity based on phase field method

We use the phase field method to track the gas-liquid interface based on the gas-liquid two-phase flow in the pool boiling process,and study the bubble nucleation,growth,deformation,departure and other dynamic behaviors on the heating surface under microgravity.By simulating the correlation between liquid undercooling and bubble dynamics,we find that the bubble growth time increases with the increase of liquid undercooling,but the effect of liquid undercooling on bubble height is not significant.Meanwhile,the gas-liquid-solid three-phase contact angle and the gravity level will also have an effect on the bubble growth time and bubble height.With the increase of the contact angle,the bubble growth time and bubble height when the bubble departs also increase.While the effect of gravity level is on the contrary,the smaller the gravity level is,the larger the bubble height and bubble growth time when the bubble separates.

Genotyping ex situ trees of Abies nebrodensis translocated from the original Sicilian population to enrich the gene pool

As a dynamic ex situ conservation strategy,a clonal seed orchard was started in a nursery in Pomaio(POM)in central Italy in 1993 for an assisted migration experiment of Abies nebrodensis(Lojac.)Mattei.Two artifi cial ex situ populations were planted with this gene pool:a seedling arboretum in Pieve Santo Stefano(PSS)and a small dendrological collection in Papiano(PAP),both originating from the Sicilian relict population.Here,using AFLP markers,we estimated the relatedness among the relocated genotypes of the three collections to check whether the three collections had suffi cient genetic variability to be considered as additional sources of variability to the original gene pool for the assisted migration strategy.High individual genetic variability was found in the collections;each plant had a diff erent genotype and was confi rmed to belong to its population of origin.PAP and PSS trees were shown to be only from the original population of A.nebrodensis species and were derived from a limited set of maternal fertile genotypes.Based on the Sicilian fi r population inventory,nursery production in Sicily,and structure clustering analysis,close genetic relationships among POM,PAP and several PSS trees(1–35)were evident.Similarly,the PSS group(36–78)was genetically close to tree 1 of POM and in a lesser proportion to plants 7 and 9 of POM.The sampling of seedlings used to form batches in the nursery might have infl uenced the structure of the resultant plantations.All genotypes will be useful for enriching the original gene pool.

Comparisons of Structured Surface Floors for Pool Boiling Enhancement at Low Heat Fluxes: Hands-On Learning Setup for Heat Transfer Classroom

Various enhanced surfaces have been proposed over the years to improve boiling heat transfer. This paper introduces an experimental setup designed for boiling demonstration in the graduate-level Heat Transfer course. The pool boiling performance of water under atmospheric pressure of 1.025 bar is investigated by using several structured surfaces at heat fluxes of 28 and 35 kW/m2. Surfaces with holes, rectangular grooves, and mushroom fins are manufactured by an NC-controlled vertical milling machine. The heat flux versus excess temperature graph is plotted by using thermocouple measurements of water and base temperatures of the boiling vessel. The separation, rise, and growth of individual vapor bubbles from the surface during boiling were recorded with a digital camera. The results for the plain surface are compared to the Rohsenow correlation. The enhancement of heat transfer coefficient (h) ranged between 15% - 44.5% for all structured surfaces. The highest heat transfer coefficient enhancement is observed between 41% - 56.5% for holed surface-3 (405 holes) compared to the plain surface. The excess temperature dropped around 29% - 34% for holed surface-3 (405 holes) compared to the plain surface. The heat transfer coefficient increases as the spacing between channels or holes decreases. While the bubbles on holed and mushroomed surfaces were spherical, the bubbles on the flat and grooved surfaces were observed as formless. The suggested economical test design could be appropriate to keep students focused and participating in the classroom.

Pooled Sputum for Xpert MTB/RIF Testing: A Cost-Saving Strategy in Mwanza, Tanzania

Background: Tuberculosis (TB) diagnostics, such as Xpert MTB/RIF, is still limited by cost. Testing of pooled samples from presumptive TB patients has been thought as a cost-saving strategy to diagnose TB. We assessed the utility and cost-saving of pooled Xpert MTB/RIF testing strategy for the diagnosis of TB in Mwanza, Tanzania. Methods: Sputum samples from Presumptive TB patients were submitted to TB laboratory for routine diagnosis of TB using Xpert MTB/RIF. The TB results from the individual sputum samples were used as the reference standard and were concealed to the investigating laboratory technicians. The remainder of samples were collected serially and were pooled (5 samples per pool) for testing. The agreement of the results between individual sample testing against pooled sample testing and cost-savings was assessed. Results: A total of 250 sputum samples from presumptive TB patients were analyzed and 50 pools were made with each pool containing 5 samples. Of the 50 sputum pools made, Mycobacterium tuberculosis (MTB) was detected in 17 (34.0%) pools. Results from the individual sputum samples MTB/RIF testing were retrieved for all 250 samples and there were 28 (11.2%) samples in which MTB was detected whereas 222 (88.8%) samples had no MTB detected. Following re-analysis of positive pools, all 28 (100%) individual positive MTB samples were detected within the 17 positive pools, with 1 to 3 individual MTB positive samples per pool. The individual sputum samples were correctly identified by pooled sputum on Xpert MTB/RIF testing, with the sensitivity of 100% and specificity of 100%. Using pooling of sputum samples strategy, we saved 46.0% (115/250) of the cartridges. Conclusion: The pooled sputum testing strategy reduced cartridge costs by 46.0% and has the potential to increase the affordability of Xpert MTB/RIF testing in countries with limited resources, such as Tanzania.

表面张力对激光焊接熔池及匙孔的影响

基于FLUENT19.0软件,建立了激光焊接热-流耦合模型,对比分析了不同表面张力温度系数(为负值)对熔池流场的影响.结果表明,随着表面张力温度系数的减小,熔池后方顺时针漩涡的流动趋势逐渐减弱,甚至消失,而且焊接飞溅的数量增多.纵截面熔池长度逐渐增加,纵截面熔池流体最大流动速度逐渐增大,熔池横截面的面积逐渐减小.当表面张力温度系数为-2.5×10^(-4)N/(m·K)时,熔池长度平均值为3.28 mm、熔池流体最大流动速度的平均值为2.89 m/s、熔池横截面面积的平均值为4.52 mm^(2);当表面张力温度系数为-3.5×10^(-4)N/(m·K)时,熔池长度平均值为3.73 mm、熔池流体最大流动速度的平均值为3.53 m/s、熔池横截面面积的平均值为4.03 mm^(2);当表面张力温度系数为-4.9×10^(-4)N/(m·K)时,熔池长度平均值为4.14 mm、熔池流体最大流动速度的平均值为4.09 m/s、熔池横截面面积的平均值为3.28 mm^(2).

基于条带池化与血管增强的眼底图像动静脉分类方法

视网膜血管动静脉管径比是定量分析糖尿病、高血压等慢性疾病的先决条件,是许多心血管疾病的重要风险指标。随着深度学习技术的发展,许多基于卷积神经网络的方法凭借其捕获高级语义的能力,在眼底图像动静脉分类方面取得了较大的进展。然而,这些方法大多是采用叠加局部卷积和池化操作方式,难以很好地应用于条带形状的眼底视网膜血管。在本研究中,为了更有效地提取条带形状的视网膜血管特征,引入条带池化来捕获空间像素远距离依赖关系,同时考虑到动静脉交错复杂的特性,结合空间金字塔池化并提出了一种全新的混合池化技术以扩大神经网络的感受野和学习上下文信息。另一方面,考虑到眼底图像中血管与非血管分布的比例极不平衡,引入了血管增强模块,利用血管分布信息和高斯核函数约束的血管边缘的信息作为权重校正动静脉特征抑制背景特征,进而解决血管与背景分布比例不平衡问题。在分别包含40、22、45张彩色眼底图像的3种国际公开数据集DRIVE、LES和HRF上的实验表明,所提算法的平衡精度(BACC)分别为0.955、0.946、0.967,表明本研究结合条带池化与血管增强的方法能够较好解决眼底图像中动静脉交错复杂和类别不平衡问题,实现对眼底视网膜动静脉的精确分类,具有较高的应用价值。

基于改进Faster R-CNN的热轧带钢表面缺陷检测

热轧带钢是钢铁行业的重要产品,其表面缺陷是影响产品质量的重要因素。针对传统缺陷检测算法存在的过程繁琐、精度不足和效率低下等问题,提出一种基于改进更快速区域卷积神经网络(faster region-based convolutional neural network,Faster R-CNN)的检测算法,实现对热轧带钢表面缺陷的高效、高精度检测。首先,采用特征相加的方法对底层细节特征和高层语义特征进行融合;然后,采用精准的感兴趣区域池化(precise region of interest pooling,Precise ROI Pooling)获取固定大小的特征向量,避免特征出现位置偏差;最后,利用均值偏移聚类算法对带钢数据集进行聚类,获得适用于热轧带钢表面缺陷检测的先验框尺寸。实验结果表明,所提算法在热轧带钢表面缺陷检测数据集上的平均精度均值达到了85.34%,检测速度为23.5帧/s,且鲁棒性良好,满足实际的工业检测需求。

基于自适应图卷积和注意力池化的点云分类与分割

针对现有点云分类与分割方法使用最大池化聚合局部邻域特征,导致最大值以外的重要信息丢失的缺陷,提出一种结合自适应图卷积AdaptConv和注意力池化AP的点云分类与分割网络。首先,采用K近邻算法构建点云局部图结构,根据点的特征生成自适应卷积核,灵活精确地捕获点云的局部邻域特征;其次,为有效提高特征聚合能力,采用注意力池化定义能量函数得到权重值,加权并聚合出更具代表性的点云局部特征;最后,堆叠自适应图卷积和注意力池化逐层提取全局特征,提高网络的分类和分割精度。实验结果表明,相较基准方法,点云分类的平均类别精度提升0.9%,部件分割和语义分割的平均交并比分别提升0.8%和0.3%,证明所提方法可有效提升点云分类与分割的准确率,具有较高的鲁棒性。

冷泉碳酸盐工厂的提出及古气候意义

【意义】碳酸盐工厂的提出及其分类研究对于推动碳酸盐岩沉积学发展具有重要意义,但现行分类方案不足以囊括所有碳酸盐沉积类型,因此亟须从机理和过程层面对碳酸盐工厂类型予以梳理、扩充。【进展】传统碳酸盐工厂的碳源主要为大气—海洋中的无机碳库,而一些特殊海相和陆相碳酸盐工厂的碳则主要源于外源碳库,因而两者具有本质上的区别。后者往往对于追溯深时水体环境和气候事件具有重大意义。【结论与展望】以冷泉碳酸盐岩为例,正式提出“冷泉碳酸盐工厂”概念,综述了其沉积特征、生物组成及生物地球化学过程,并举例阐述其古气候意义。基于外源碳库的碳酸盐工厂类型理应得到重视。

华北局地大暴雨过程中多个β中尺度对流系统发生发展对比分析

利用常规地面高空观测、多普勒雷达、风廓线、VDRAS(Variational Doppler Radar Analysis System)和NCEP再分析资料,对2018年8月5—6日副热带高压(以下简称副高)控制下华北一次局地大暴雨过程中多个β中尺度对流系统触发和发展机制进行了分析。结果表明:这次大暴雨发生在副高控制下,处于高温、高湿气团中,大气层结极不稳定。暴雨由多个相继发展的中尺度对流系统造成,分别是太行山迎风坡上西南—东北向、华北平原地区保定一带南北向、保定至霸州附近西南—东北向和以雄安新区为中心东西向原地生消的准静止MCS-Ⅰ、MCS-Ⅱ、MCS-Ⅲ和MCS-Ⅳ,均属于β中尺度。在相似的环境中,不同中尺度对流系统触发机制有较大差异,太行山迎风坡上的MCS-Ⅰ是由近地层偏东暖湿气流在迎风坡与山风形成的辐合抬升触发;由辐射差异和前期强降水形成的局地冷池受MCS-Ⅰ影响再次加强后,其出流与环境风形成的两条地面辐合线分别触发了MCS-Ⅱ和MCS-Ⅲ,并组织对流沿辐合线呈带状发展;而超低空偏东风增强叠加冷池出流在地形抬升作用下促使沿山暖湿气团进一步抬升,使得原本消亡的MCS-Ⅰ再次重建。MCS-Ⅳ发展最旺盛、持续时间最长,是大暴雨中心的直接制造者,一方面MCS-Ⅱ与MCS-Ⅲ、MCS-Ⅰ与MCS-Ⅳ的两次合并过程,是MCS-Ⅳ增强、持久的重要原因;另一方面边界层偏东风急流为MCS-Ⅳ的发展提供了水汽和不稳定能量等有利条件,同时推动其左前方中尺度涡旋的发展,导致MCS-Ⅳ所在地的气旋性涡度大大增加,加强了以急流轴为中心的垂直次级环流发展,造成MCS-Ⅳ的发展维持,形成华北平原地区以雄安新区为中心的东西向大暴雨带。

奥氏体不锈钢薄板激光焊表面张力驱动熔池形成的数值模拟分析

为探究奥氏体不锈钢薄板激光焊接过程中表面张力对熔池形状、熔池流速、熔池静压的影响,以期优化焊接工艺并提高焊接质量。基于数值模拟分析方法,建立了移动旋转高斯曲面体热源模型,通过改变Marangoni对流系数来模拟不同表面张力条件,分析其对熔池形成过程、熔池形状、流速及熔池内静压力的影响。模拟结果表明,表面张力是影响熔池形态和内部静压力的关键因素,表面张力指向熔池内部,驱动熔池向中心汇聚,从而促进熔池的形成;表面张力的大小与熔池峰值温度、熔深、深宽比呈负相关,与峰值流速、熔宽、熔池内部静压力呈正相关;铜衬底的加入显著增大了过冷度,加速了熔池的冷却,有助于避开奥氏体不锈钢的敏化温度区间。

基于自适应融合的实时车辆检测

针对传统的车辆检测技术检测速度慢和精度低的问题,提出了一种融合注意力的自适应金字塔网络的交通目标检测算法(fusion attentiont adaptive pyramid network,FAAP-Net),可以显著降低交通事故的发生率。为了降低计算复杂度,设计了一种轻量级的互补池化结构(CPS),该结构在宽度和高度上采用了两组不同的池化组合,在保持高精度的同时,显著降低了网络的浮点运算数(GFLOPs)和参数量。为了解决智能交通系统特征图生成过程中的信息损失问题,通过将自适应注意力模块(AAM)和特征增强模块(FEM)引入自适应融合特征金字塔网络(AF-FPN),以融入车辆检测的形状特征。针对车辆细节特征表征弱的问题,引入了一种按通道维度分组的注意力(SA)机制,以增强主干网络对不同车辆检测细节特征的关注,有效提取车辆细节的显著特征。在BDD100K数据集上的实验结果表明,FAAP-Net算法相比于传统算法,平均精度从30.3%提升到43.7%。

三次台风登陆后雨带列车效应特征对比

为探讨台风雨带列车效应导致极端降水的可能形成方式和流型配置,利用多源观测资料,对2015年第13号台风苏迪罗(过程1)、2013年第23号台风菲特(过程2)和2005年第9号台风麦莎(过程3)登陆减弱后的对流雨带列车效应现象进行对比分析。结果表明:三次过程大暴雨区都发生在浙江东部丘陵东侧迎风坡,且具有雨带走向与环境背景气流基本一致及水汽通量辐合集中在850 hPa以下的共性,但其环境背景明显不同。过程1和过程2均由台风外围对流雨带引起,其中过程1雨带位于台风低压与副热带高压之间,垂直风切变和对流有效位能较大,水汽来自低纬度海面且湿层深厚;过程2发生在陆上台风残涡与海上另一个台风之间的鞍型场内,垂直风切变和对流有效位能较小,水汽来自同纬度海面,湿层仅位于对流层中下部。过程3由台风内核区雨带引起,垂直风切变强,对流有效位能最小。进一步分析发现在不同环境垂直风廓线和湿度廓线下三次过程雨带的结构和组织方式呈现出明显差异。在过程1中,对流发展较高,冷池与偏东气流间形成较深厚边界层辐合,所构成边界的两侧斜压结构促进来自海面暖湿空气在力管项作用下在边界暖湿一侧抬升,雨带两侧正负散度配置加强了对流的组织化程度,使对流单体长时间沿雨带长轴移动;在过程2中,偏东气流与弱冷池之间风向相反,风速相当,辐合较浅薄,暖云降水起主要作用,新单体不断在雨带东边界生成、西边界消亡,雨带停滞少动引发持续强降水;在过程3中,内核区雨带较大程度受台风涡旋动力学影响,对流发展高度低,为典型的热带海洋降水型,结构随高度略向外侧倾斜,对流单体在台风本体东侧低空东南风急流脉动产生的风速辐合作用下反复被激发并向下游移动,引发极端强降水。以上事实说明,导致极端降水的台风雨带列车效应形成方式多样,其短时临近预报面临重大挑战。

“门诊共济”改革背景下南京市职工医保统筹基金的可持续性研究

以人口老龄化的南京市为例,通过构建保险精算模型的方法,预测并分析门诊共济保障机制改革对南京市2021—2035年期间职工医保统筹基金收支、结余的影响,模拟延迟退休和扩大筹资、支付方式改革等政策对提升医保基金可持续性的影响,基于此提出政策建议。研究发现:按现行门诊共济方案,到2032年,南京市职工医保统筹基金将发生穿底风险;与现行政策相比,实施男女同龄退休将使得累计结余亏空时间点推迟一年;实行扩大筹资政策,在职和退休职工个人均向统筹基金缴费后,统筹基金在中期内将消除亏损风险;实施支付方式改革政策,也能使统筹基金在中期内可持续发展。结论:在人口老龄化背景下实施的门诊共济改革可能导致统筹基金出现支付压力,长期来看需要组合多种配套政策,才能实现统筹基金的可持续发展。

基于改进YOLOv5s的道路裂缝检测算法

为了解决道路巡检系统光学传感器采集的裂缝图像中颜色特征不明显且尺寸不规则造成检测精度不高、泛化能力不足的问题,提出改进YOLOv5s的裂缝检测算法。将结合深度可分离卷积(Depthwise Separable Convolution, DSC)的全局注意力(Global Attention Mechanism, GAM)引入主干特征提取网络,在降低注意力复杂度的同时获得丰富的跨维度特征,增强了裂缝的识别能力;采用空间金字塔软池化网络(Spatial Pyramid Softpool, SPSF),通过Softpool池化保留多维语义以减少信息弥散,提高了边界框回归的准确性;在颈部特征增强网络,运用空洞深度可分离卷积(Atrous DSC)进行下采样,通过扩大感受野加强深层和浅层信息的聚合能力,提高裂缝识别的泛化性。经过在自制道路裂缝数据集上的实验,相较于YOLOv5s,改进算法的mAP提高2.2%,有效提升了道路裂缝检测的准确性和对不同背景下裂缝识别的泛化能力。

基于改进YOLOv8的煤矿输送带异物检测

现有基于深度学习的输送带异物检测模型较大,难以在边缘设备部署,且对不同尺寸异物和小目标异物存在错检、漏检情况。针对上述问题,提出一种基于改进YOLOv8的煤矿输送带异物检测方法。采用深度可分离卷积、压缩和激励(SE)网络将YOLOv8主干网络中C2f模块的Bottleneck重新构建为DSBlock,在保持模型轻量化的同时提升检测性能;为增强对不同尺寸目标物体信息的获取能力,引入高效通道注意力(ECA)机制,并对ECA的输入层进行自适应平均池化和自适应最大池化操作,得到跨通道交互MECA模块,以增强模块的全局视觉信息,进一步提升异物识别精度;将YOLOv8的3个检测头修改为4个轻量化小目标检测头,以增强对小目标的敏感性,有效降低小目标异物的漏检率和错检率。实验结果表明:改进YOLOv8的精确度达91.69%,mAP@50达92.27%,较YOLOv8分别提升了3.09%和4.07%;改进YOLOv8的检测速度达73.92帧/s,可充分满足煤矿输送带异物实时检测的需求;改进YOLOv8的精确度、mAP@50、参数量、权重大小和每秒浮点运算数均优于SSD,Faster-RCNN,YOLOv5,YOLOv7-tiny等主流目标检测算法。