基于深度学习的铸件CT图像分割算法

针对现有方法分割弱边缘铸件CT图像难度大、精度低、鲁棒性差的问题,提出一种融合残差模块与混合注意力机制的U型网络分割算法(AttRes-U-Nets)。该算法以U-Net网络为基础,首先构建深度残差网络ResNets作为算法的编码网络,解决传统U-Net网络特征提取能力不足的问题;然后,引入改进后的混合注意力机制,突出分割目标区域与通道的特征响应,提高网络灵敏度;最后,将Focal loss与Dice loss结合为一种新损失函数FD loss缓解样本不平衡带来的负面影响。使用120阀体数据集对算法性能进行验证,实验结果表明,本文算法对铸件分割的像素准确率(PA)和交互比(IoU)分别达到98.72%和97.40%,优于传统U-Net算法与其他主流语义分割算法,为弱边缘分割提供了新思路。     

基于CT图像的肺部肿瘤三维分割研究

肺部CT图像的三维分割计算机肺部肿瘤三维辅助诊断系统的关键技术之一,肿瘤与周围组织的复杂性造成分割困难。利用Connected Threshold及Fast Marching算法,通过两种途径实现了肺部肿瘤的三维分割,并经过实验证明这两种三维分割方法的有效性,为医务人员提供了更为直观逼真的肿瘤立体图像。     

基于改进Res-UNet的视网膜图像血管分割

精准的视网膜血管分割可以辅助诊疗如糖尿病、高血压等疾病。眼睛血管结构和病理特征的复杂性导致血管分割的精度和速度都存在很多局限。为了克服这一问题,提出了一种改进的U-net分割方法,该方法将U-net网络解码器和编码器中的卷积模块改为残差模块,使用非局部注意模块连接编码器和解码器。网络模型在不增加参数量的情况下,通过添加残差模块和注意力机制提高了像素之间的信息相关性以及模型提取特征的能力。最后,采用DRIVE数据集对所提模型与原U-net网络进行对比评价,新模型在测试集上的特征检测准确率、特异性、灵敏度和Dice系数分别达到了0.9679、0.9896、0.8245和0.8281。实验结果证明,所提网络模型可对视网膜进行精确地血管分割。     

结合先验稀疏字典和空洞填充的CT图像肝脏分割

针对复杂多变的肝脏图像,提出了一种基于先验稀疏字典和空洞填充的三维肝脏图像分割方法。对腹部CT图像进行Gabor特征提取,并分别在Gabor图像和灰度图像的肝脏金标准边界上选择大小相同的图像块作为两组训练集,利用训练集得到两种查询字典及稀疏编码。将金标准图像与待分割图像配准,并将配准后的肝脏边界作为待分割图像的肝脏初始边界;在初始边界点上的十邻域内选择大小相同的两组图像块作为测试样本,利用测试样本与查询字典计算稀疏编码及重构误差,并选择重构误差最小的图像块的中心作为待分割肝脏的边界点;最后,设计一种空洞填充方法对肝脏边界进行补全和平滑处理,得到最终分割结果。利用医学图像计算和计算机辅助介入国际会议中提供的肝脏数据进行了实验验证。结果表明,该方法对肝脏分割图像具有较好的适用性和鲁棒性,并获得了较高的分割精度。其中,平均体积重叠率误差为(5.21±0.45)%,平均相对体积误差为(0.72±0.12)%,平均对称表面距离误差为(0.93±0.14)mm。     

基于改进注意力W-Net的工业烟尘图像分割

针对小目标烟尘尺寸小、边缘稀薄和U-Net在提取小目标烟尘特征效果不佳等原因导致的烟尘漏检、误检和分割精度低等问题,提出一种基于改进注意力W-Net(IAW-Net)的烟尘图像分割网络。采用注意力机制将U-Net扩展为W-Net,在W-Net的基础上引入改进的注意力机制,增强了小目标烟尘的特征;针对小目标烟尘特点对焦点损失进行改进,增加了小目标烟尘的分割比重。实验结果表明,IAW-Net能够在不影响大目标烟尘分割的基础上更加关注小目标烟尘的分割效果,从而提升了烟尘图像的整体分割能力,相比现有语义分割网络具有更好的分割效果。     

基于联合标定的图像分割

针对普适环境下人体区域的提取问题,提出一种基于彩色摄像头和深度摄像头联合标定的图像分割方法。首先根据深度摄像头采集的深度信息易处理和受外界环境影响小的优点,利用大律法从深度图像中快速提取出人体区域;然后建立世界坐标系中彩色图像坐标系和深度图像坐标系的对应关系,实现彩色信息和深度信息的相对位置映射;最后根据分割的深度图像及其在彩色图像中的映射,在彩色图像中完成人体区域分割。通过在复杂背景和不同光照条件下的图像分割实验,验证了本方法具有良好的精度和鲁棒性。     

基于智能视觉的机械零件图像分割技术

为有效获取零件特征以提高现代生产智能化、精密化水平,基于智能视觉的机械零件分割研究起着关键作用。针对航天机器人等装配车间流水线零部件智能感知问题,研究基于智能视觉的零部件图像分割算法,实现机械零部件分割识别。基于Deeplabv3图像分割算法提出一种增加自定义Encoder-Decoder特征提取模块的网络结构Deeplabv3-d,采用掩膜标记特征区域,基于该改进网络结构,采用mobile Net和Resnet101两种骨干网络进行零件图像分割对比实验,证明了该图像分割算法在零件图像分割应用领域的实用性。     

基于高层特征融合的图像语义分割

在深度学习下的图像语义分割中,为了探究高层特征对于户外场景语义分割性能的影响,在对高层特征进行分析的基础上,提出了融合高层特征的图像语义分割方法。在目前主流深度学习框架Caffe下搭建的分割模型,并采用斯坦福8类户外场景数据集对模型进行了训练和测试。测试结果验证了该方法的有效性和准确性。     

基于CV模型优化的肝脏MRI图像分割法

目的:为解决肝脏病灶边缘变化不明显的磁共振成像(MRI)图像分割问题,提出了一种基于无边缘主动轮廓(CV)模型的优化图像分割方法对病灶区图像进行分割.方法:首先通过传统CV模型与用新边缘函数优化CV模型进行分割,再分别采用Jaccard、Dice系数对图像分割结果定量评价,并对两种模型的分割时间、迭代次数进行分析.结果...     

融合下补偿结构的眼底血管图像分割网络研究

眼底血管图像在临床中通常被用于眼部疾病的诊断及监测,其中血管的形态结构能够反映疾病的重要特征,因此,眼底血管图像的分割处理对眼部疾病的诊断和预防具有十分重要的医学意义。针对目前人工智能主流算法中卷积和池化操作会导致很多特征丢失,提取特征时会忽视图像中的空间信息,图像中的细小血管很难分割出来等问题,基于U-net模型进行了相关研究,结合空间注意力模块对空间特征进行细化,同时提出了一种下补偿结构LCSAnet。该结构能够减少网络提取特征信息过程中的特征损失,从而提高分割精度。研究实验在DRIVE数据集上完成,LC-SAnet的分割准确率达到96.97%,F1值达到74.36%。结果证明,LC-SAnet表现出更好的分割性能,对细小血管的结构识别更加准确。     

Automatic setae segmentation from Chaetoceros microscopic images

A novel image processing model Grayscale Surface Direction Angle Model (GSDAM) is presented and the algorithm based on GSDAM is developed to segment setae from Chaetoceros microscopic images. The proposed model combines the setae characteristics of the microscopic images with the spatial analysis of image grayscale surface to detect and segment the direction thin and long setae from the low contrast background as well as noise which may make the commonly used segmentation methods invalid. The experimental results show that our algorithm based on GSDAM outperforms the boundary‐based and region‐based segmentation methods Canny edge detector, iterative threshold selection, Otsu's thresholding, minimum error thresholding, K‐means clustering, and marker‐controlled watershed on the setae segmentation more accurately and completely. Microsc. Res. Tech. 77:684–690, 2014. © 2014 Wiley Periodicals, Inc.     

Brain tumor segmentation in multi‐spectral MRI using convolutional neural networks (CNN)

A tumor could be found in any area of the brain and could be of any size, shape, and contrast. There may exist multiple tumors of different types in a human brain at the same time. Accurate tumor area segmentation is considered primary step for treatment of brain tumors. Deep Learning is a set of promising techniques that could provide better results as compared to nondeep learning techniques for segmenting timorous part inside a brain. This article presents a deep convolutional neural network (CNN) to segment brain tumors in MRIs. The proposed network uses BRATS segmentation challenge dataset which is composed of images obtained through four different modalities. Accordingly, we present an extended version of existing network to solve segmentation problem. The network architecture consists of multiple neural network layers connected in sequential order with the feeding of Convolutional feature maps at the peer level. Experimental results on BRATS 2015 benchmark data thus show the usability of the proposed approach and its superiority over the other approaches in this area of research.     

Feature enhancement framework for brain tumor segmentation and classification

Automatic medical image analysis is one of the key tasks being used by the medical community for disease diagnosis and treatment planning. Statistical methods are the major algorithms used and consist of few steps including preprocessing, feature extraction, segmentation, and classification. Performance of such statistical methods is an important factor for their successful adaptation. The results of these algorithms depend on the quality of images fed to the processing pipeline: better the images, higher the results. Preprocessing is the pipeline phase that attempts to improve the quality of images before applying the chosen statistical method. In this work, popular preprocessing techniques are investigated from different perspectives where these preprocessing techniques are grouped into three main categories: noise removal, contrast enhancement, and edge detection. All possible combinations of these techniques are formed and applied on different image sets which are then passed to a predefined pipeline of feature extraction, segmentation, and classification. Classification results are calculated using three different measures: accuracy, sensitivity, and specificity while segmentation results are calculated using dice similarity score. Statistics of five high scoring combinations are reported for each data set. Experimental results show that application of proper preprocessing techniques could improve the classification and segmentation results to a greater extent. However, the combinations of these techniques depend on the characteristics and type of data set used.     

改进Mask R-CNN的交通场景多目标快速检测与分割

针对智能驾驶中出现的交通场景多目标检测与分割效率低、鲁棒性差等问题,提出一种改进的Mask R-CNN交通场景多目标快速检测与分割方法。首先采用轻量级MobileNet作为骨干网络,有效减少网络参数并压缩模型体积,提升后续嵌入式端的算法移植能力,其次通过优化FPN与骨干网络卷积结构,保证高底层之间特征信息的完整传递,通过调整超参数得到交通场景多目标检测与分割改进网络模型。设计不同交通场景下的对比实验,改进网络能够准确实现多目标的检测与分割,平均检测精度可达85.2%。在ApolloScape和NuScence数据集上进行迁移实验,改进网络展示出良好的泛化能力。本文所提出的改进骨干网络与网络结构优化,能够适应多种复杂交通场景,完成交通场景多目标的快速检测与分割,为智能驾驶提供了理论依据与技术方案。     

基于视觉注意的随机游走图像分割

传统随机游走图像分割需要多次交互设置种子点以获得理想的分割结果。在视觉注意的基础上,提出了一种新的自动确定种子点的随机游走图像分割算法。首先对图像进行超像素分割,并生成概率边界图(PBM);然后基于Itti模型,通过视觉注意焦点的转移搜寻待分割的关键区域;为确定关键分割区域种子点,以当前注意焦点作为极点对概率边界图进行极坐标变换,在获得的极坐标概率边界图上建立关于焦点区域边界的能量函数,采用图论max-flow min-cut算法最小化能量函数检测焦点区域的最优边界,焦点区域边界内的超像素即为种子点;最后以超像素为节点构造图,在图上随机游走完成图像分割。在Berkeley Segmentation Data Set上的实验表明本文方法能有效分割复杂图像。     

Computerized detection and segmentation of mitochondria on electron microscope images

Mitochondrial function plays an important role in the regulation of cellular life and death, including disease states. Disturbance in mitochondrial function and distribution can be accompanied by significant morphological alterations. Electron microscopy tomography (EMT) is a powerful technique to study the 3D structure of mitochondria, but the automatic detection and segmentation of mitochondria in EMT volumes has been challenging due to the presence of subcellular structures and imaging artifacts. Therefore, the interpretation, measurement and analysis of mitochondrial distribution and features have been time consuming, and development of specialized software tools is very important for high-throughput analyses needed to expedite the myriad studies on cellular events. Typically, mitochondrial EMT volumes are segmented manually using special software tools. Automatic contour extraction on large images with multiple mitochondria and many other subcellular structures is still an unaddressed problem. The purpose of this work is to develop computer algorithms to detect and segment both fully and partially seen mitochondria on electron microscopy images. The detection method relies on mitochondria's approximately elliptical shape and double membrane boundary. Initial detection results are first refined using active contours. Then, our seed point selection method automatically selects reliable seed points along the contour, and segmentation is finalized by automatically incorporating a live-wire graph search algorithm between these seed points. In our evaluations on four images containing multiple mitochondria, 52 ellipses are detected among which 42 are true and 10 are false detections. After false ellipses are eliminated manually, 14 out of 15 fully seen mitochondria and 4 out of 7 partially seen mitochondria are successfully detected. When compared with the segmentation of a trained reader, 91% Dice similarity coefficient was achieved with an average 4.9 nm boundary error.