基于子空间投影和边缘增强的低剂量CT去噪

低剂量计算机断层扫描（CT）是一种相对安全的疾病筛查手段,但低剂量CT图像往往包含较多噪声和伪影,严重影响医生的诊断。针对该问题,本文提出一种基于子空间投影和边缘增强网络（SPEENet）。SPEENet为自编码器结构,包含双流编码器和解码器两个主要模块。双流编码器可以被分为噪声图像编码流及边缘信息编码流两部分,噪声图像编码流对低剂量CT图像进行特征提取,利用图像特征去除低剂量CT中的噪声和伪影;边缘信息编码流部分主要关注低剂量CT图像的边缘信息,利用边缘信息保护图像结构。为充分利用编码器特征,本文引入噪声基投影模块,构建基于编码器和解码器特征的基,并利用该基将编码器提取的特征投影到对应的子空间,获取更好的特征表示。本文在公开数据集上进行实验以验证提出网络的有效性,实验结果表明,相较于其他低剂量CT去噪网络,SPEENet可以取得更好的去噪效果。      

基于模型融合的有限角度CT迭代重建方法

针对有限角度扫描的CT重建,提出一种基于模型融合的CT迭代重建方法。模型来源于患者的早期Dicom图像。对扫描角度有限的投影数据,用统计迭代算法进行初步重建,得到预重建图像;将预重建图像与模型进行融合,得到融合图像;然后再次投影,补全原始投影缺失的部分,根据补全的投影数据重建出中间结果,之后重复投影、融合、重建过程直到满足终止条件。仿真实验表明,该算法能完整重建整个目标,在有效保留原目标特征的同时提高了小角度投影数据重建的质量。     

基于相对TV最小的CT图像重建算法

总变差（TV）最小算法是一种有效的CT图像重建算法,可以对稀疏或含噪投影数据进行高精度重建。然而,在某些情况下,TV算法会产生阶梯状伪影。在图像去噪领域,相对TV算法展现了优于TV算法的性能。鉴于此,将相对TV模型引入图像重建,提出相对TV最小优化模型,并在自适应梯度下降-投影到凸集（ASD-POCS）框架下设计对应的求解算法,以进一步提升重建精度。以Shepp-Logan、FORBILD及真实CT图像仿真模体进行重建实验,验证了该算法的正确性并评估了算法的稀疏重建能力和抗噪能力。实验结果表明,相对TV算法可以实现逆犯罪,可以对稀疏或含噪投影数据进行高精度重建,与TV算法相比,该算法可以取得更高的重建精度。      

基于CNN和Transformer耦合网络的低剂量CT图像重建方法

在投影角度个数不变的情况下,降低每个角度下的射线剂量,是一种有效的低剂量CT实现方式,然而,这会使得重建图像的噪声较大。当前,以卷积神经网络（CNN）为代表的深度学习图像去噪方法已经成为低剂量CT图像去噪的经典方法。受Transformer在计算机视觉任务中展现的良好性能的启发,本文提出一种CNN和Transformer耦合的网络（CTC）,以进一步提高CT图像去噪的性能。CTC网络综合运用CNN的局部信息关联能力和Transformer的全局信息捕捉能力,构建8个由CNN部件和一种改进的Transformer部件构成的核心网络块,并基于残差连接机制和信息复用机制将之互联。与现有4种去噪网络比较,CTC网络去噪能力更强,可以实现高精度低剂量CT图像重建。      

基于新的非局部先验模型的Bayesian低剂量CT重建算法

为了改善低剂量CT重建图像质量,在传统非局部先验的基础上,提出了一种基于投影对称性的改进非局部先验模型。基于该先验模型构造了一种贝叶斯（Bayesian）重建算法,并将其应用到低剂量CT投影数据降噪中,通过滤波反投影算法重建出图像。仿真实验结果表明,本文所提出的算法较基于传统先验模型的重建算法,能在去除噪声与保持边缘之间取得较好的平衡。     

多层球状物角度稀疏投影CT重建仿真研究

本文主要针对高密度差多层球状物体的角度稀疏投影CT重建问题,进行基于FBP算法的仿真研究。参考特殊工件材料结构特征设计建立检测对象模型,采用C＋＋语言开发完成平行束圆周扫描FBP重建仿真软件平台。分别采用直接补零和紧邻填充两种方法补全缺失的稀疏投影数据,并进行仿真重建和比较分析。对重建图像进行了有针对性地基于边缘和区域的增强处理方法研究,使缺陷特征得以凸显。本文的研究内容和方法可为具有类似特征的CT无损检测提供技术参考。     

基于非局部全核变分方法的稀疏角多能CT重建

稀疏角采样与减小X射线源电流可有效降低多能谱CT低辐射剂量,然而会导致投影数据不足且包含较大噪声,重建图像会严重降质。针对这一问题,本文对传统全核变分（TNV）正则化方法进行推广,利用非局部梯度向量构成的雅克比矩阵的低秩特性,提出非局部全核变分（NLTNV）正则化方法。该方法用单个正则项同时建模能谱CT图像的结构相似性、梯度域稀疏性与非局部自相似性3种先验信息,能恢复稀疏角度投影含较大噪声（剂量较低）时图像的结构特征,并且有效缓解了用多正则项建模多能谱CT图像不同先验信息所导致的正则化参数过多问题。此外,基于NLTNV的重建模型为凸优化模型,保证了算法的稳定性与收敛性。实验结果表明,与TNV正则化方法相比,本方法显著提升重建图像的整体质量。      

一种基于数据外插改进的ART迭代算法

本文针对有限角度的投影数据的CT图像重建问题,提出了一种基于数据外插改进的ART算法.该算法的基本思想足运用已知角度的投影数据来补全未知角度的投影数据,再用ART算法进行图像重建.最后用模拟的投影数据进行了重建图像的数值实验.实验结果表明该算法不但提高了重建图像质量,同时也提高了图像达代的收敛速度.     

基于Chambolle-Pock框架的核TV多通道图像重建算法

总变差最小化算法是一种基于压缩感知理论的图像重建算法,能够从稀疏投影或含噪投影数据中高精度地重建图像,已经被广泛应用于计算机断层成像、磁共振成像、电子顺磁共振成像。能谱CT、T1或T2加权的MRI及EPRI均属于多通道成像。逐通道TV算法可以实现较高精度的图像重建,然而忽略了各通道图像之间的相似性。核TV算法是一种考虑了通道间图像相似性的TV类算法,可以实现高精度图像重建。面向多通道图像重建,以CT重建为研究范例,本文提出一种Chambolle-Pock算法框架下的核TV多通道图像重建算法。通过仿真模体和真实CT图像模体的重建实验,验证算法的正确性,分析算法的收敛性,探索算法参数对收敛速率的影响,评估算法的稀疏重建能力及含噪投影重建能力。结果表明,相对于逐通道TV算法,所提算法可以取得更高的重建精度。核TV算法是一种高精度的多通道图像重建算法,可以应用于各种成像模态的多通道重建场合。      

一种基于频域外推的不完全角度CT重建算法

不完全角度重建问题一直是CT图像重建领域研究的重点和难点。目前,通常的不完全角度重建方法是基于空域的迭代方法,但由于正反投影的高计算复杂度,空域迭代方法存在计算耗时,对硬件资源需求大等问题。本文提出了一种基于外推的邻近网格迭代算法（INNG-TV）。首先,平行束采样的数据通过傅里叶变换和样条插值到频域空间,然后在迭代的过程中,傅立叶空间投影已知部分的数据始终不变,缺失部分数据通过对重建图像进行INNG外推得到,同时在图像空间对重建图像做非负、最小化总变分等先验及优化约束。     

SCIENCE CHINA Earth Sciences SUBJECT INDEX TO VOLUME 57(2014)

正~~     

Distribution of throughfall and stemflow in multi‐strata agroforestry,perennial monoculture,fallow and primary forest in central Amazonia,Brazil

The partitioning of rain water into throughfall, stemflow and interception loss when passing through plant canopies depends on properties of the respective plant species, such as leaf area and branch angles. In heterogeneous vegetation, such as tropical forest or polycultural systems, the presence of different plant species may consequently result in a mosaic of situations with respect to quantity and quality of water inputs into the soil. As these processes influence not only the water availability for the plants, but also water infiltration and nutrient leaching, the understanding of plant effects on the repartitioning of rain water may help in the optimization of land use systems and management practices. We measured throughfall and stemflow in a perennial polyculture (multi‐strata agroforestry), monocultures of peach palm (Bactris gasipaes) for fruit and for palmito, a monoculture of cupuaçu (Theobroma grandiflorum), spontaneous fallow and primary forest during one year in central Amazonia, Brazil. The effect on rain water partitioning was measured separately for four useful tree species in the polyculture and for two tree species in the primary forest. Throughfall at two stem distances, and stemflow, differed significantly between tree species, resulting in pronounced spatial patterns of water input into the soil in the polyculture system. For two tree species, peach palm for fruit (Bactris gasipaes) and Brazil nut trees (Bertholletia excelsa), the water input into the soil near the stem was significantly higher than the open‐area rainfall. This could lead to increased nutrient leaching when fertilizer is applied close to the stem of these trees. In the primary forest, such spatial patterns could also be detected, with significantly higher water input near a palm (Oenocarpus bacaba) than near a dicotyledonous tree species (Eschweilera sp.). Interception losses were 6·4% in the polyculture, 13·9 and 12·3% in the peach palm monocultures for fruit and for palmito, respectively, 0·5% in the cupuaçu monoculture and 3·1% in the fallow. With more than 20% of the open‐area rainfall, the highest stemflow contributions to the water input into the soil were measured in the palm monocultures and in the fallow. Copyright © 1999 John Wiley & Sons, Ltd.     

A Monte Carlo study of rainfall forecasting with a stochastic model

A procedure for short-term rainfall forecasting in real-time is developed and a study of the role of sampling on forecast ability is conducted. Ground level rainfall fields are forecasted using a stochastic space-time rainfall model in state-space form. Updating of the rainfall field in real-time is accomplished using a distributed parameter Kalman filter to optimally combine measurement information and forecast model estimates. The influence of sampling density on forecast accuracy is evaluated using a series of a simulated rainfall events generated with the same stochastic rainfall model. Sampling was conducted at five different network spatial densities. The results quantify the influence of sampling network density on real-time rainfall field forecasting. Statistical analyses of the rainfall field residuals illustrate improvement in one hour lead time forecasts at higher measurement densities.     

Anthropogenic Eutrophication and Red Tide Outbreak in Lacustrine Systems of the Kashmir Himalaya

Red tide, a recurrent phenomenon has become conspicuous in several Kashmir lake ecosystems since 1991. The responsible organism (Euglena pedunculata), a rare flagellate rediscovered in the Kashmir Himalaya (Khan 1993) caused first and unprecedented red tide outbreak, constituting a maximum of 96% of resident numerical phytoplankton density in Dal Lake. At present, conflicting hypotheses exist on the generation of causal assemblage(s) imparting redness to waters: Jeeji Bai (1991) linked its origin to acid precipitation – a fallout of burning oil‐fields during the Gulf War – whilst Khan (1993) holds local factor(s) responsible. Field/experimental studies support the latter contention that the influx of untreated sewage, in unison with warm temperatures, high levels of PhAR, iron and interruption to hydrological flow‐pattern together with absence/or reduction in grazing activity created conducive environmental milieu for red tide outbreak. Dal Lake “red tide” drifted the bloom‐inoculum to other waters, including Lake Wular, where additional ecological niches were carved out, threatening the aesthetic value and biological diversity of Kashmir lakes. Ecological monitoring indicates frequent seasonal red tide occurrence in Dal Lake (including summer‐autumn event of 1998) which testifies its unabated eutrophication status. Further studies are needed on ecological adaptability and biogeographic distribution of this rare and unique red tide‐causing flagellate.