改进YOLOv4的遥感图像目标检测算法

为有效解决遥感图像目标检测算法在复杂背景下的检测效果不佳的问题,提出一种改进YOLOv4的目标检测算法。设计一种跨阶段残差结构,替换原主干网络的简单残差结构,降低模型参数量和计算负担;引入CBAM注意力机制,加强CSP模块间有效特征交互;使用跨阶段分层卷积模块重构特征融合阶段对深层特征图的处理方式,防止网络退化和梯度消失;采用Mish激活函数,增强融合网络对非线性特征的提取能力。在RSOD、DIOR数据集上的实验结果表明,改进YOLOv4算法的测试mAP相比原YOLOv4算法分别高出4.5%、7.3%,其检测速度分别达到48 fps、45 fps,在保证实时性的同时检测精度有较大提升。

Psychological trauma,posttraumatic stress disorder and traumarelated depression:A mini-review

There are various types of traumatic stimuli,such as catastrophic events like wars,natural calamities like earthquakes,and personal trauma from physical and psychological neglect or abuse and sexual abuse.Traumatic events can be divided into type I and type II trauma,and their impacts on individuals depend not only on the severity and duration of the traumas but also on individuals’self-evaluation of the traumatic events.Individual stress reactions to trauma include posttraumatic stress disorder(PTSD),complex PTSD and trauma-related depression.Trauma-related depression is a reactive depression with unclear pathology,and depression occurring due to trauma in the childhood has gained increasing attention,because it has persisted for a long time and does not respond to conventional antidepressants but shows good or partial response to psychotherapy,which is similar to the pattern observed for PTSD.Because trauma-related depression is associated with high risk of suicide and is chronic with a propensity to relapse,it is necessary to explore its pathogenesis and therapeutic strategy.

图采样泛化行人重识别算法

最近的研究表明,度量学习中的深度特征匹配方法,结合大规模、多样化的训练数据,可以显著增强人员再识别的泛化能力。然而,许多现有的方法会产生大量的内存和计算成本,如分类参数或类记忆学习等。为解决上述问题,提出了一种新的基于相关性图采样(correlation graph sampler,CGS)的泛化行人重识别算法,CGS的基本思想是在训练开始时使用局部敏感哈希函数(locality-sensitive Hashing,LSH)和特征度量为所有类构造最近邻关系图。这确保了每一小批训练样本由随机选择的基类和与基类具有相似性的近邻类组成,以提供信息量大且具有挑战性的学习示例,提高行人重识别模型的判别性学习能力。CGS的采样原理会受主干网提取的特征质量影响,因此CGS采样能力会随着主干网的训练而增强,具有可学习性。通过在大规模数据集(包括CUHK03、Market-1501和MSMT17)上交叉评估该方法,广泛的实验结果证实了该方法的有效性,并展示了其在行人重识别应用中的潜力。

边缘优化和注意力融合的遥感地物分割算法

针对遥感地物图像种类众多且目标边缘较复杂的特点,以及现有分割网络中局部卷积的感受野有限,对图像上下文信息利用不足,导致分割目标边缘模糊以及分割精度低等问题,提出一种基于UNet3+网络的遥感地物分割算法。在解码过程中引入相似性感知点关联算子作为上采样方式,通过聚合特征金字塔中的多个建议,改善目标边界细节的分割能力;在编码过程中引入选择性内核模块,优化下采样方式,以实现神经元的自适应感受野大小,充分地获取目标特征的多尺度信息,精准捕捉有用的细节语义信息;在跳跃连接阶段添加双多尺度注意力模块,对不同尺度的特征进行加权融合,使模型更好地关注局部细节和全局上下文信息。在WHDLD、ISPRS Potsdam数据集上的实验表明,改进算法的平均交并比分别达到了64.4%、75.4%,较基线模型分别提升了约2.6个百分点、3.2个百分点,同时验证了改进算法在分割边缘模糊问题上的有效性。

动态查询感知的行人重识别算法

目前无监督的泛化行人重识别算法在某些需要背景信息辅助判断的情况下,可能会忽视图像的局部区域对细粒度特征的关注,导致背景信息被过滤掉,从而降低识别精度。针对上述问题,提出了一种基于稀疏注意力的动态查询感知算法。通过挤压拼接(squeeze and concat,SPC)模块,获取不同通道数的特征图。利用双层路由感知注意力机制,提取不同尺度特征图之间的注意力权重,得到逐级通道注意力向量。对逐级通道注意力向量的权重进行重新校准。将重新标定的权重与相应的特征图进行加权,输出具有更丰富细化特征信息的多尺度特征图。所提模型在大规模公开数据集(Market-1501、DukeMTMC-reID、MSMT17)上进行实验,相较于基线模型Rank-1分别提高了3.2、4.4、15.4个百分点,mAP分别提高了5.5、8.3、16.2个百分点,与现有前沿算法相比,能够实现更好的局部和全局特征通道之间的信息交互,提升模型对图像特征的细节感知能力.

Role of CD36 in central nervous system diseases

CD36 is a highly glycosylated integral membrane protein that belongs to the scavenger receptor class B family and regulates the pathological progress of metabolic diseases.CD36 was recently found to be widely expressed in various cell types in the nervous system,including endothelial cells,pericytes,astrocytes,and microglia.CD36 mediates a number of regulatory processes,such as endothelial dysfunction,oxidative stress,mitochondrial dysfunction,and inflammatory responses,which are involved in many central nervous system diseases,such as stroke,Alzheimer’s disease,Parkinson’s disease,and spinal cord injury.CD36 antagonists can suppress CD36 expression or prevent CD36 binding to its ligand,thereby achieving inhibition of CD36-mediated pathways or functions.Here,we reviewed the mechanisms of action of CD36 antagonists,such as Salvianolic acid B,tanshinone IIA,curcumin,sulfosuccinimidyl oleate,antioxidants,and small-molecule compounds.Moreover,we predicted the structures of binding sites between CD36 and antagonists.These sites can provide targets for more efficient and safer CD36 antagonists for the treatment of central nervous system diseases.

多分支特征映射的遥感图像目标检测算法

由于遥感图像具有背景复杂、目标较小且密集、尺度连续变化大等特点,通用目标检测器难以较好地适应,导致检测效果不佳。针对以上问题,基于YOLOv5s模型,提出一种多分支特征映射的遥感图像目标检测算法。首先,利用结构重参数化技术设计一种结合门控通道转换的RepVGG模块,采用其串联结构替换原主干网络的C3模块,聚合全局上下文信息,增强特征表达和特征提取能力;其次,使用自适应指数加权池化方法以及逆过程重构特征融合网络的采样方式,最大化地保留特征信息,改善较小目标的检测效果;最后,引入EIoU和Focal Loss组合作为模型的损失函数,优化预测框的回归速率以及难易样本的损失权重分配,进一步提高定位精度。在DIOR和NWPU VHR-10数据集上的实验结果表明,提出算法的平均精度均值分别达到92.2%、92.5%,较YOLOv5s分别提高了3.5个百分点、5.6个百分点,达到了更好的检测效果,同时实时性也满足实际场景下的遥感图像目标检测。

基于空间自适应和内容感知的红外小目标检测

由于红外街道图像中小目标像素较少、颜色特征不丰富,容易导致模型漏检、误检以及检测效果不佳等问题,因此提出了一种基于空间自适应和内容感知的红外小目标检测算法。首先,通过堆叠局部注意力与可变形注意力设计一种基于空间自适应的转换器,以增强对长距离依赖特征的建模能力,捕获到更多空间位置信息。其次,采用内容感知特征重组算子进行特征上采样,实现在大感受野内聚合上下文信息以及利用浅层特征信息来自适应地重组特征。最后增加160×160的高分辨率预测头,将输入特征的像素点映射到更细小的检测区域,进一步改善小目标的检测效果。在FILR数据集上的实验结果表明,改进算法的平均精度均值达到85.6%,相较于YOLOX-s算法提高了3.9%,验证了所提算法在红外小目标检测上的优越性。

Optical redox imaging of ANT1-de-cient muscles

Adenine nucleotide translocator(ANT)is a mitochondrial protein involved in the exchange of ADP and ATP across the mitochondrial inner membrane.It plays a crucial role in cellular energy metabolism by facilitating the transport of ATP synthesized within the mitochondria to the cytoplasm.The isoform ANT1 predominately expresses in cardiac and skeletal muscles.Mutations or dysregulation in ANT1 have been implicated in various mitochondrial disorders and neuromuscular diseases.We aimed to examine whether ANT1 deletion may affect mitochondrial redox state in our established ANT1-de-cient mice.Hearts and quadriceps resected from age-matched wild type(WT)and ANT1-de-cient mice were snap-frozen in liquid nitrogen.The Chance redox scanner was utilized to perform 3D optical redox imaging.Each sample underwent scanning across 3–5 sections.Global averaging analysis showed no signi-cant differences in the redox indices(NADH,flavin adenine dinucleotide containing-flavoproteins Fp,and the redox ratio Fp/(NADH+Fp)between WT and ANT1-de-cient groups.However,quadriceps had higher Fp than hearts in both groups(p¼0:0004 and 0.01,respectively).Furthermore,the quadriceps were also more oxidized(a higher redox ratio)than hearts in WT group(p¼0:004).NADH levels were similar in all cases.Our data suggest that under non-stressful physical condition,the ANT1-de-cient muscle cells were in the same mitochondrial state as WT ones and that the signi-cant difference in the mitochondrial redox state between quadriceps and hearts found in WT might be diminished in ANT1-de-cient ones.Redox imaging of muscles under physical stress can be conducted in future.

Human immunodeficiency virus-associated dementia complex with positive 14-3-3 protein in cerebrospinal fluid:A case report

BACKGROUND Human immunodeficiency virus(HIV)-associated dementia(HAD)is a subcortical form of dementia characterized by memory deficits and psychomotor slowing.However,HAD often presents with symptoms similar to those of Creutzfeldt-Jakob disease(CJD),particularly in patients with acquired immune deficiency syndrome(AIDS).CASE SUMMARY We report the case of a 54-year-old male who exhibited cognitive dysfunction and secondary behavioral changes following HIV infection and suspected prion exposure.The patient was diagnosed with HIV during hospitalization and his cerebrospinal fluid tested positive for 14-3-3 proteins.His electroencephalogram showed a borderline-abnormal periodic triphasic wave pattern.Contrast-enhanced magnetic resonance imaging revealed moderate encephalatrophy and demyelination.Initially,symptomatic treatment and administration of amantadine were pursued for presumed CJD,but the patient’s condition continued to deteriorate.By contrast,the patient’s condition improved following anti-HIV therapy.This individual is also the only patient with this prognosis to have survived over 4 years.Thus,the diagnosis was revised to HAD.CONCLUSION In the diagnostic process of rapidly progressive dementia,it is crucial to rule out as many potential causes as possible and to consider an autopsy to diminish diagnostic uncertainty.The 14-3-3 protein should not be regarded as the definitive marker for CJD.Comprehensive laboratory screening for infectious diseases is essential to enhance diagnostic precision,especially in AIDS patients with potential CJD.Ultimately,a trial of diagnostic treatment may be considered when additional testing is not feasible.

伏诺拉生用于幽门螺杆菌根除治疗的临床疗效评价

目的优化幽门螺杆菌(HP)根除治疗的用药方案。方法选取2020年1月至2022年12月在医院体检中心或门诊发现感染HP且胃镜检查无溃疡及其他胃黏膜病变的患者312例,根据根除方案用药的不同分为质子泵抑制剂(PPI)四联组(124例)、伏诺拉生(VPZ)四联组(84例)、VPZ二联组(97例)。PPI四联组予艾司奥美拉唑肠溶胶囊[每次20 mg、每日2次(bid)]+阿莫西林胶囊(每次1 g、bid)+克拉霉素缓释胶囊(每次500 mg、bid)+胶体果胶铋胶囊[每次200 mg、每日3次(tid)];VPZ四联组予富马酸伏诺拉生片(每次20 mg、bid)+阿莫西林胶囊(每次1 g、bid)+克拉霉素缓释胶囊(每次500 mg、bid)+胶体果胶铋胶囊(每次200 mg、tid);VPZ二联组予富马酸伏诺拉生片(每次20 mg、bid)+阿莫西林胶囊(每次1 g、tid)。3组均为口服给药,疗程均为14 d。结果PPI四联组、VPZ四联组及VPZ二联组HP根除成功率分别为85.48%,90.48%,88.66%,组间比较无显著差异(P>0.05),且仅VPZ四联组根除效果满意(根除率>90%);不良事件发生率分别为15.32%,16.67%,7.72%,组间比较有显著差异(P<0.05)。结论在无溃疡及胃黏膜病变人群的HP根除治疗中,含VPZ四联疗法的根除成功率及不良事件发生率均稍高于传统的含PPI四联疗法;VZP(联合高剂量阿莫西林)二联疗法HP根除成功率虽略高于PPI四联疗法,但仍欠满意。未来还需更多临床研究优化,以获得更理想的根除效果。

基于非下采样轮廓波变换的遥感地物分割算法

针对遥感地物图像具有背景复杂且种类众多的特点,利用传统算法进行分割会导致边缘模糊、信息丢失及分割精度低的问题,提出了一种基于改进DeepLabV3+网络的语义分割算法。首先,在主干网络中引入改进后的特征提取网络CHRNet;其次,使用非下采样轮廓波变换(NSCT)算法重构空洞空间金字塔池化(ASPP)模块中的全局池化操作;最后,在模型编码和解码阶段添加无参数的注意力机制SimAM,加强模块间的特征传递,提高特征利用率。实验表明,在PASCAL VOC2012和WHDLD数据集上,改进算法的平均交并比(MIoU)分别达到了81.56%和64.2%,较原有算法分别提升了约4.61和2.8个百分点,改进算法在保证分割速率的同时,提升了分割精度。

改进YOLOv5的SAR图像近海岸舰船目标检测算法研究

SAR图像舰船目标检测时,因近海岸港口存在着复杂背景的问题,以至于重叠舰船目标无法被准确提取特征信息,造成近海岸的舰船目标出现漏检、误检的情况.针对以上问题,提出一种复杂场景下的SAR图像舰船检测算法,该算法基于YOLOv5进行改进,采用SPPF结构加强提取特征信息,并融合原YOLOv5的SPP结构提取的特征信息,这种多级金字塔模块并列融合的方式能有效的检测多尺度舰船目标,使特征信息更好的表达;然后将原模型中的GIOU改进为CIOU,使其可以准确的回归出预测框的位置;最终为了更合理的筛选高于阈值的预测框,改进NMS(Non-Maximum-Suppression),采用Soft-NMS方法去惩罚衰减高于阈值的边框得分,合理的去除预测框.试验结果表明,该文改进的模型相比于原模型在SSDD、SAR-Ship-Dataset数据集上的mAP(mean Average Precision)提高了5.15%和5.06%,改进模型能有效检测近海岸中复杂背景下的SAR图像舰船目标.

非局部稀疏关注的YOLOv4优化算法

传统目标检测网络如Fast R-CNN、ReseNet等在下采样提取图像特征的过程中,损失了大量的空间位置信息表征,存在对较小目标检测效果差的问题。在保留空间位置信息的基础上,提出了一种非局部稀疏关注的级联残差高分辨率网络(cascaded residual high resolution network)。该网络架构从一个高分辨率的子网络开始,逐步增加从高到低分辨率的子网络,形成更多的阶段,将多个分辨率的子网络并行连接,使用级联残差模块(cascaded residual module,CrModule)进行同分辨率特征流间的特征提取;利用多尺度特征图融合,使得每个从高到低分辨率的表示反复地从其他并行表示接收信息,产生丰富语义表征和空间位置表征的高分辨率表示;引入NLSA(nonlocal sparse attention)算法实现深层网络特征块超分重构,挖掘不同尺度相同物体间的结构关联,提高较小物体的特征表示,使之与大物体特征类似,提升较小目标的特征可学习性。在VOC2007数据集的广泛评估表明,将CrHRnet作为YOLOv4的主干特征提取网络,能有效提高目标检测的准确率;CrHRnet-YOLOv4测试mAP(mean average precision)比YOLOv4、YOLOv5_s、YOLOv5_m分别高出1.8、9.5、3.4个百分点,在相同的设备下检测单张图片的FPs较YOLOv4网络提升了30%。

A Preliminary Study on the Breeding of Late-maturing Daylily Cultivar Jinlei No.1

In order to stagger the early harvest stage of daylily and critical busy summer farming season in Qingyang City, late-maturing daylily cultivar Jinlei No. 1 was systematically bred from natural variant offspring of Malian daylily. According to the results, late-maturing daylily eultivar Jinlei No. 1 exhibited good agro- nomic traits and strong resistance, The first day of harvest of Jinlei No. 1 was more than 10 d later compared with Malian daylily and Xian daylily ; harvest duration of Jinlei No. 1 was 4 d longer than Malian daylily and 5 d than Xian daylily; the yield of Jinlei No. 1 was improved by 3.41% and 5.16% compared with Malian daylily and Xian daylily, respectively. The cultivation of late-maturing daylily cuhivar Jinlei No. 1 can effectively avoid heavy economic losses caused by overlap period of early harvest of daylily and critical busy summer farming in Qingyang City.

基于拓扑优化的锁紧块优化设计

现有的锁紧块结构单一,且对孔轴的配合精度要求高,这对使用和加工带来诸多不便。通过对现有常规锁紧结构进行优化,以达到宽裕的配合精度和轻量化产品。首先通过拓扑优化,然后根据优化结果和设计经验进行结构的形状优化,再进行尺寸优化,得到最优尺寸参数,最后通过疲劳分析进行校核。结果显示:优化后的模块的重量减轻38%,使用同样锁紧力作用下,检测点P点的变形量是原变形量的2.24倍,达到轻量化设计和宽裕的配合精度,同时5年工况环境可以满足使用。说明使用拓扑优化可以实现产品初期设计和产品的轻量化设计,仿真可完成具体的尺寸参数优化和疲劳强度校核。

Interlayer and intralayer co-modified flexible V_(2)CT_(X) MXene@SWCNT films for high-power Li-ion capacitors

As an emerging member of the two-dimensional(2D)material family,V_(2)CT_(X)MXene shows great potential in the application of lithium-ion capacitors(LICs)due to its unique structure and excellent electrical conductivity.However,severe nanosheets stacking and intra-layer transport barriers have limited the further development of V_(2)CT_(X)MXene-based materials.Herein,we prepared Kions and–O functional group co-modified V_(2)CT_(X)MXene(VCT-K)and further incorporated it with single-walled carbon nanotube(SWCNT),obtaining freestanding V_(2)CT_(X)composite films(VCT-K@C)with the 3D conductive network.Significantly,K+ions were introduced into V_(2)CT_(X)MXene to stabilize the interlayer structure and prevent the aggregation of nanosheets,the terminal group of–O was controllably modified on the surface of MXene to improve the Li+ions storage reversible capacities and the SWCNT acted as the bridge between MXene nanosheets to opens up the channels for ion/electron transportation in the longitudinal direction.Benefited from the synergistic effect of VCT-K and SWCNT,the VCT-K@C exhibits superior reversible specific capacities of 671.8 mA h g^(-1)at 0.1 A g^(-1)and 318 mA h g^(-1)at 1.0 A g^(-1).Furthermore,the assembled LICs with VCT-K@C anode coupling activated carbon(AC)cathode deliver an outstanding power density of 19.0 kW kg^(-1)at 67.4 Wh kg^(-1),a high energy density of 140.5 Wh kg^(-1)at 94.8 W kg^(-1)and a stable capacitance retention of 86%after 6000 cycles at 10 A g^(-1).Such unique structures and excellent electrochemical properties are expected to pave the way for the large-scale application in LICs of MXene-based materials.

Multiple therapies relieve long-term tardive dyskinesia in a patient with chronic schizophrenia:A case report

BACKGROUND Tardive dyskinesia(TD)is a serious and disabling movement disorder;it impairs social function and quality of life and increases the mortality rate.TD is usually induced by the use of antipsychotic drugs;however,the underlying mechanism remains unclear.Pharmacotherapy of TD includes cholinergic drugs,benzodiazepines,ginkgo biloba extract(GBE),antioxidants,amantadine,propanolol,botulinum toxin,valbenazine,and deutetrabenazine,whereas the non-pharmacotherapy approach includes modified electroconvulsive therapy(MECT)and deep brain stimulation.We successfully treated a chronic schizophrenia patient with comorbid long-term severe TD using deutetrabenazine,clozapine,and MECT.CASE SUMMARY A 69-year-old woman who was diagnosed as having schizophrenia 16 years ago developed severe TD after 6-mo prescription of risperidone oral solution.Her TD symptoms did not resolve despite various treatments,such as GBE,vitamin E,trihexyphenidyl,promethazine,benzodiazepines,and switching to quetiapine and olanzapine.After admission,she was given deutetrabenazine 6 mg bid.Her buccal tremor was slightly resolved 3 d later;however,her tongue remained protruded and could not be retracted.Quetiapine was switched to clozapine on day 4,and the buccal tremor remarkably resolved,and the tongue could be retracted into the mouth from day 6 onward.After three sessions of MECT,the buccal tremor resolved further.Since then,she has been able to take a semifluid diet,and her quality of life improved remarkably during 6 mo of follow-up.CONCLUSION TD is a serious condition which could be caused by antipsychotic medications;however,the best strategy against TD is prevention and monitoring during using antipsychotics.For patients with TD caused by antipsychotic medication use,multiple measures should be considered like switching to clozapine,adjunction with deutetrabenazine,or even MECT.

肺炎患儿血清G-CSF、sTREM-1、sICAM-1水平的变化及意义

目的探讨肺炎患儿血清粒细胞集落刺激因子(G-CSF)、可溶性髓系细胞触发受体-1(sTREM-1)、可溶性细胞间黏附分子-1(sICAM-1)水平的变化及对肺炎的诊断价值。方法选择该院2015年6月至2016年6月收治的128例肺炎患儿,根据患儿临床表现及影像学诊断结果分为重症肺炎组(78例)及轻症肺炎组(50例),另选取同期进行健康检查的50例健康体检儿童为对照组,应用酶联免疫吸附试验测定各组血清G-CSF、sTREM-1、sICAM-1水平,并应用受试者工作特征(ROC)曲线分析血清G-CSF、sTREM-1、sICAM-1在肺炎中的诊断价值。结果重症肺炎组、轻症肺炎组患儿血清G-CSF、sTREM-1、sICAM-1水平显著高于对照组(P<0.05),且重症肺炎组血清G-CSF、sTREM-1、sICAM-1水平高于轻症肺炎组(P<0.05)。重症肺炎存活组患者随着病情好转,血清G-CSF、sTREM-1、sICAM-1水平显著下降,至出院时恢复正常水平(P<0.05);死亡组患者随着时间推移,血清G-CSF、sTREM-1、sICAM-1水平不断升高,并在死亡时达到最大值(P<0.05)。3项指标诊断重症肺炎的ROC曲线下面积(AUC)分别为0.712、0.789、0.756,3项指标联合诊断的AUC为0.802。经Pearson相关分析可知,G-CSF水平与sTREM-1、sICAM-1呈正相关(r=0.356、0.337,P=0.000),而sTREM-1水平与sICAM-1呈正相关(r=0.322,P=0.000)。结论血清G-CSF、sTREM-1、sICAM-1可作为肺炎患儿有价值的诊断指标,通过监测3项指标的动态变化趋势能够反映肺炎患儿病情进展及预后。

接触网支柱号牌定位与字符分割方法

针对高速巡检车成像质量较差的情况,提出一种基于形态学变换的定位算法,在此基础上,采用投影分割法提取字符。使用形态学变换得到号牌候选区域,利用其结构特征与HOG&SVM(梯度直方图与支持向量机)进行筛选,通过仿射变换与两次垂直投影提取号牌上下边界,利用投影矩阵的极值点得到候选分割点,分析其距离的变异系数得到合理的字符分割点。在普速铁路上的实验结果表明,号牌定位与字符分割正确率分别达到96.9%与95%,相对于传统垂直投影分割法与自适应投影分割法,字符分割正确率分别提升了19.9%与6.5%。