基于 Mask RCNN 的桥小脑角区脑膜瘤与听神经瘤分类定位研究

由于人体桥小脑角区的脑膜瘤与听神经瘤在影像学的表现以及发病位置极其相似,所以临床诊断极易发生误诊．针对此问题,本文应用掩膜区域卷积神经网络（Mask RCNN）对两类肿瘤进行分类定位研究．首先采集89名脑膜瘤与218名听神经瘤患者的T₁WI-SE序列的磁共振图像,对其进行预处理,再结合改进的特征金字塔网络（FPN）算法进行网络训练．本文对比了三种不同的Mask RCNN主干网络对两者分类定位的效果．结果表明,结合改进的FPN算法和ResNet101作为主干网络的Mask RCNN分类定位模型能够有效实现对两类肿瘤的分类定位,精确率为0.918 2、召回率为0.856 9、特异性为0.876 2、均值平均精度（mAP）为0.90．     

改进的SSD算法及其对遥感影像小目标检测性能的分析

针对以Faster R-CNN为代表的基于候选框方式的遥感影像目标检测方法检测速度慢,而现有SSD算法在小目标检测中性能低的问题,提出一种改进的SSD算法,综合利用现有基于候选框方式和一体化检测方式的优势,提升检测性能。该算法利用密集连接网络替换原有的VGGNet作为骨干网络,并且在密集连接模块之间构建特征金字塔,代替原有多尺度特征图。为验证所提算法的精度及性能,设计样本数据在线采集系统,并采集飞机及运动场目标样本集作为实验样本,通过对改进SSD算法的训练,验证了其网络结构的稳定性,在无迁移学习支持下依然能够达到良好效果,且训练过程不易发散。通过对比以101层的残差网络(ResNet101)作为基础网络的Faster R-CNN算法和R-FCN算法可知,改进SSD算法较Faster R-CNN算法和R-FCN算法的MAP在测试集上分别提升了9.13%和8.48%,小目标检测的MAP分别提升了14.46%和13.92%,检测单张影像耗时71.8 ms,较Faster R-CNN和R-FCN算法分别减少45.7 ms和7.5 ms。     

基于改进旋转区域生成网络的遥感图像目标检测

为了实现遥感图像中目标的快速准确检测,解决遥感图像目标带有旋转角度的问题,在卷积神经网络理论的基础上,将旋转区域网络生成融入到Faster R-CNN网络中,提出了一种基于Faster R-CNN改进的遥感图像目标检测方法。相对于主流目标检测方法,本文算法针对遥感图像中的大多数目标都具有方向性不定且相对聚集的特点,在区域候选网络中加入了旋转因子,以便能够生成任意方向的候选区域;同时,在网络的全连接层之前增加一个卷积层,以降低其特征图参数,增强分类器的性能,避免出现过拟合。将本文算法与几种主流目标检测方法进行对比分析后可知,本文算法因融合了多尺度特征及旋转区域网络的卷积神经网络所提取的特征,能得到更好的检测结果。     

基于改进Faster R-CNN的红外目标检测算法

为提升红外目标的检测精度,提出了一种引入频域注意力机制的Faster R-CNN红外目标检测算法。首先,针对红外图像边缘模糊和噪声问题,设计了一种并行的图像增强预处理结构;其次,在Faster R-CNN中引入频域注意力机制,设计了一种新型红外目标检测主干网络;最后,引入路径增强金字塔结构,融合多尺度特征进行预测,利用底层网络丰富的位置信息,提升检测精度。在红外飞机的数据集上进行实验,结果表明,改进后的Faster R-CNN目标检测框架比以ResNet50为主干的算法的AP提升了7.6%。此外,与目前主流算法对比,本文算法提高了红外目标的检测精度,验证了算法改进的有效性。     

面向嵌入式平台的轻量级光学遥感图像舰船检测

针对现有基于深度学习的轻量级目标检测算法对复杂遥感场景图像中舰船目标检测精度低、检测速度慢的问题,提出了一种面向嵌入式平台的轻量级光学遥感图像舰船实时检测算法（STYOLO）。首先,针对主干网络内存访问成本较高的问题,利用高效网络架构ShuffleNet v2作为主干网络对图像进行特征提取,降低内存访问成本,提高网络并行度;其次,利用Slim-neck特征融合结构作为特征增强网络,以融合较低层级特征图中的细节信息,增强对小目标的特征响应,在多尺度信息融合区域施加坐标注意力机制,强化目标关注以提高较难样本检测以及抗背景干扰能力;最后,提出一种跨域迁移和域内迁移相结合的学习策略,减少源域与目标域的差异性,提升迁移学习效果。实验结果表明：基于光学遥感图像舰船检测公开数据集HRSC2016,与同类型快速检测算法YOLOv5s相比,所提算法的检测精度提高了2.7个百分点,参数量减少了61.77%,在嵌入式平台Jetson Nano上检测速度达到102.8 frame/s,能够有效实现对光学遥感图像中舰船目标的实时、准确检测。     

多尺度特征融合的目标检测算法研究

针对Faster R-CNN在多尺度目标检测时易出现小目标漏检和误检的问题,提出一种改进的多尺度目标检测算法。将利于小目标检测的低层网络和利于大尺度目标检测的高层网络进行多尺度特征融合;在训练阶段,采用在线难例样本挖掘算法维护难例样本分类池,加速神经网络模型迭代收敛,解决训练样本不均衡、训练效率低下的问题;计算并统计待检测目标的尺度大小,合理控制用于生成候选区域的锚框尺寸,提高模型泛化能力。采用PASCAL VOC2012公开数据集和类人足球机器人自建数据集进行算法验证,实验结果表明,相比Faster R-CNN算法,本算法的平均检测精度在上述数据集下分别提高了8.61和5.47个百分点。     

基于密集卷积和域自适应的高光谱图像分类

针对常规的高光谱图像分类算法不能很好地解决不同图像中的频谱偏移的问题,提出了一种基于密集卷积和域自适应的高光谱图像分类算法,首先在源域中使用密集卷积进行深度特征学习,然后应用域自适应技术转移到目标域。目前的域自适应高光谱图像分类框架中常用卷积神经网络进行特征学习,但是当深度增加时会出现因梯度消失而导致分类精度下降的情况,因此本文通过引入密集卷积进行深度特征学习,提高域自适应高光谱图像分类的精度。在Indiana高光谱数据集和Pavia高光谱数据集上验证所提算法的有效性,整体分类精度分别为61.06%和89.63%,与其他域自适应高光谱图像分类方法对比,所提方法具有更好的分类精度。     

基于改进Faster R-CNN的空中目标检测

相比传统图像目标检测算法,基于大数据和深度学习的检测算法无须人工设计特征,且检测性能更稳健。在防空应用背景下,自建了空中目标静态和视频图像数据集用于训练和测试,改进了基于深度学习的目标检测框架Faster R-CNN,将其专用于空中目标检测。结合空中目标检测任务的特点和需求,提出膨胀积累、区域放大、局部标注、自适应阈值、时空上下文等改进策略,弥补了Faster R-CNN对弱小目标和被遮挡目标不敏感的缺陷,提高了检测速度和精度。实验表明,改进后的Faster R-CNN在应对弱小目标、多目标、杂乱背景、光照变化、模糊、大面积遮挡等检测难度较大的情况时,均能获得很好的效果。数据集上测试结果的平局准确率均值较改进之前提高了16.7%,检测速度提高了3倍。     

深度神经网络在自动瞄靶技术中的应用研究

基于Mask R-CNN模型，对惯性约束聚变实验研究中靶图进行关键结构的语义分割，通过计算中心来定位目标靶点位置，通过模拟靶图数据来验证算法的可行性。在算法验证过程中通过程序批量生成不同结构参数、不同角度的模拟靶图作为数据集，然后通过增噪、成像位置变换等操作拓展样本数量。在大样本数量基础上对算法模型进行训练测试，实现语义分割，通过计算注入孔的中心得到目标靶点位置。测试结果显示准确率和召回率均在97%以上，在靶点识别精度上优于10个像素点。     

一种基于GAN和自适应迁移学习的样本生成方法

研究了基于生成式对抗网络(GAN)和跨域自适应迁移学习的样本生成和自动标注方法。该方法利用自适应迁移学习网络,基于已有的少量可见光图像样本集,挖掘目标在红外和可见光图像中特征内在相关性,构建自适应的转换迁移学习网络模型,生成标注好的目标图像。提出的方法解决了红外图像样本数量少且标注费时的问题,为后续多频段协同目标检测和识别获得了足够的样本数据。实验结果表明:自动标注算法对实际采集的装甲目标图像和生成的装甲目标图像各1 000张进行自动标注测试,对实际装甲目标图像的标注准确率达到95%以上,对生成的装甲目标标注准确率达到83%以上;利用真实图像和生成图像的混合数据集训练的分类器的性能和使用纯真实图像时基本一致。     

基于Mask R-CNN结合边缘分割的颗粒物图像检测

对颗粒物的尺寸检测是生产中重要的环节,使用相机采集图像并处理是常用的非接触检测方法。围绕颗粒物的识别与尺寸检测需求,选用沙粒为检测对象,提出了一种改进颗粒物边界掩膜的Mask R-CNN模型。该模型结合经典的边缘检测技术,并利用深度学习模型预测掩膜,根据边缘分割的结果来得到更高精度的掩膜。使用DenseNet作为检测网络的主干网络,使得整体网络参数量更少,并利用通道注意力机制加强网络的特征提取能力。实验结果表明,改进的网络可以提高检测的精度,且结合图像处理的方式能够改善掩膜尺寸检测的准确度,为颗粒物的工业检测提供了一种有意义的方法。     

A Crop Image Segmentation and Extraction Algorithm Based on Mask RCNN

The wide variety of crops in the image of agricultural products and the confusion with the surrounding environment information makes it difficult for traditional methods to extract crops accurately and efficiently. In this paper, an automatic extraction algorithm is proposed for crop images based on Mask RCNN. First, the Fruits 360 Dataset label is set with Labelme. Then, the Fruits 360 Dataset is preprocessed. Next, the data are divided into a training set and a test set. Additionally, an improved Mask RCNN network model structure is established using the PyTorch 1.8.1 deep learning framework, and path aggregation and features are added to the network design enhanced functions, optimized region extraction network, and feature pyramid network. The spatial information of the feature map is saved by the bilinear interpolation method in ROIAlign. Finally, the edge accuracy of the segmentation mask is further improved by adding a micro-fully connected layer to the mask branch of the ROI output, employing the Sobel operator to predict the target edge, and adding the edge loss to the loss function. Compared with FCN and Mask RCNN and other image extraction algorithms, the experimental results demonstrate that the improved Mask RCNN algorithm proposed in this paper is better in the precision, Recall, Average precision, Mean Average Precision, and F1 scores of crop image extraction results.     

Instance Segmentation of Multiple Myeloma Cells Using Deep-Wise Data Augmentation and Mask R-CNN

Multiple myeloma is a condition of cancer in the bone marrow that can lead to dysfunction of the body and fatal expression in the patient. Manual microscopic analysis of abnormal plasma cells, also known as multiple myeloma cells, is one of the most commonly used diagnostic methods for multiple myeloma. However, as it is a manual process, it consumes too much effort and time. Besides, it has a higher chance of human errors. This paper presents a computer-aided detection and segmentation of myeloma cells from microscopic images of the bone marrow aspiration. Two major contributions are presented in this paper. First, different Mask R-CNN models using different images, including original microscopic images, contrast-enhanced images and stained cell images, are developed to perform instance segmentation of multiple myeloma cells. As a second contribution, a deep-wise augmentation, a deep learning-based data augmentation method, is applied to increase the performance of Mask R-CNN models. Based on the experimental findings, the Mask R-CNN model using contrast-enhanced images combined with the proposed deep-wise data augmentation provides a superior performance compared to other models. It achieves a mean precision of 0.9973, mean recall of 0.8631, and mean intersection over union (IOU) of 0.9062.     

基于神经免疫生长可免域网络的红外光谱图像分割算法

管道运输对远距离输送石油天然气有着较大优势,而与之伴随的管道安全问题使得管道安全检测至关重要。为确保任何时间下管道状况的有效检测,红外成像技术由于其根据对象的热辐射信息反映目标特征的特殊性,能够忽视可见光的影响检测管道状态,因而在管道检测领域有重要意义。但由于户外环境的多样性,交错的管道和复杂环境使得采集的红外管道图像具有目标特征分布不均匀,目标遮挡和背景类目标干扰等问题。这些问题增加了提取管道目标的难度,不利于管道的分割和检测。生物免疫系统在抗原检测、提取和消除上表现出识别、学习、记忆、耐受和协调配合等目前复杂系统优化策略所缺乏的优异特性,借鉴生物神经系统调控免疫系统的机理,设计一种基于神经免疫网络的复杂背景下红外管道目标的检测与提取算法。根据生物神经网络在免疫系统中的调控机制,利用基础管道形状特征模型构建用于红外管道目标定位的神经网络,并将最优神经免疫可免域和区域种子生长结合,解决管道遮挡影响提取目标完整性的问题。选择三种典型的红外管道图像,将传统目标检测算法与基于神经免疫网络的算法进行了效果对比分析。结果表明,传统算法的平均真阳性率为40.56%,Jaccard相似性指数为27.18%,绝对误差率为11.75%,而基于神经免疫网络算法的真阳性率为98.05%,Jaccard相似性指数为94.44%,绝对误差率为1.18%。对比可知,神经免疫网络算法的真阳性率比传统方法高57.49%,绝对误差率则低10.57%,验证了复杂背景下,本文算法相比传统方法能够更加准确地提取完整的红外管道目标,这对管道安全检测效率的提高有着重要意义。     

基于Mask RCNN的桥梁裂缝检测方法设计及研究北大核心CSCD

裂缝是桥梁道路上常见的一种病害,针对其检测准确率有待提高的问题,提出了基于Mask RCNN(region-based convolutional neural networks)的桥梁裂缝检测算法,设计了语义增强模块(semantic enhancement module,SEM),将该模块与特征金字塔网络(feature pyramid network,FPN)相结合,通过特征融合Add计算得到新的多尺度特征图feature maps。针对裂缝形态复杂多样存在识别困难的问题,将裂缝做了两类划分进行检测,并制定了两种策略进行对比实验。实验结果表明:该文中改进的方法可以得到更好的检测结果,检测准确率Accuracy可达99.8%,平均检测精度(mean average precision,mAP)提高了12.6%。     

An object tracking method based on guided filter for night fusion image

Online object tracking is a challenging problem as it entails learning an effective model to account for appearance change caused by intrinsic and extrinsic factors. In this paper, we propose a novel online object tracking with guided image filter for accurate and robust night fusion image tracking. Firstly, frame difference is applied to produce the coarse target, which helps to generate observation models. Under the restriction of these models and local source image, guided filter generates sufficient and accurate foreground target. Then accurate boundaries of the target can be extracted from detection results. Finally timely updating for observation models help to avoid tracking shift. Both qualitative and quantitative evaluations on challenging image sequences demonstrate that the proposed tracking algorithm performs favorably against several state-of-art methods.     

基于斑马鱼图像特征的鱼群检测算法北大核心CSCD

计算机视觉方法越来越多地应用于斑马鱼的群体行为研究;但是,由于斑马鱼游动过程形体变化大,遮挡多,准确与鲁棒地检测出斑马鱼仍然是一件非常具有挑战性的问题。为了解决该问题,提出一种基于斑马鱼图像特征的鱼群检测算法。首先通过分析目标特性,提出使用鱼头和鱼尾替代全鱼的检测方法,解决了传统整鱼检测在鱼群交叉遮挡时失效的难题;然后基于斑马鱼图像特征自动构建训练集,避免了深度学习手动标注的费时费力问题。通过对实际斑马鱼视频进行处理验证,与现有的算法相比,本文提出的方法在标注率、召回率(recall,R)与遮挡检测率(occlusion detection rate,ODR)等性能指标上有更好的实验效果。其中,在标注性能方面,本文提出的自动标注方法在总标注率上达到87.40%;在训练集效果方面,本文自动标注算法结合人工校正在标注时间上相比于人工标注方法减少93.11%,均值平均精度(mean average precision,mAP)达到79.80%;在目标检测方面,在目标遮挡率为42.72%的情况下,本文检测算法能够获得82.0%的召回率及58.02%的遮挡检测率。     

Ricci Curvature-Based Semi-Supervised Learning on an Attributed Network

In recent years, on the basis of drawing lessons from traditional neural network models, people have been paying more and more attention to the design of neural network architectures for processing graph structure data, which are called graph neural networks (GNN). GCN, namely, graph convolution networks, are neural network models in GNN. GCN extends the convolution operation from traditional data (such as images) to graph data, and it is essentially a feature extractor, which aggregates the features of neighborhood nodes into those of target nodes. In the process of aggregating features, GCN uses the Laplacian matrix to assign different importance to the nodes in the neighborhood of the target nodes. Since graph-structured data are inherently non-Euclidean, we seek to use a non-Euclidean mathematical tool, namely, Riemannian geometry, to analyze graphs (networks). In this paper, we present a novel model for semi-supervised learning called the Ricci curvature-based graph convolutional neural network, i.e., RCGCN. The aggregation pattern of RCGCN is inspired by that of GCN. We regard the network as a discrete manifold, and then use Ricci curvature to assign different importance to the nodes within the neighborhood of the target nodes. Ricci curvature is related to the optimal transport distance, which can well reflect the geometric structure of the underlying space of the network. The node importance given by Ricci curvature can better reflect the relationships between the target node and the nodes in the neighborhood. The proposed model scales linearly with the number of edges in the network. Experiments demonstrated that RCGCN achieves a significant performance gain over baseline methods on benchmark datasets.     

Fuzzy analysis of community detection in complex networks

A snowball algorithm is proposed to find community structures in complex networks by introducing the definition of community core and some quantitative conditions. A community core is first constructed, and then its neighbors, satisfying the quantitative conditions, will be tied to this core until no node can be added. Subsequently, one by one, all communities in the network are obtained by repeating this process. The use of the local information in the proposed algorithm directly leads to the reduction of complexity. The algorithm runs in O(n+m) time for a general network and O(n) for a sparse network, where n is the number of vertices and m is the number of edges in a network. The algorithm fast produces the desired results when applied to search for communities in a benchmark and five classical real-world networks, which are widely used to test algorithms of community detection in the complex network. Furthermore, unlike existing methods, neither global modularity nor local modularity is utilized in the proposal. By converting the considered problem into a graph, the proposed algorithm can also be applied to solve other cluster problems in data mining.     

基于流形学习的声目标特征提取方法研究

为解决目前声目标识别面临的鲁棒性不足问题,提出将流形学习应用到声目标的特征提取中,在经典流形学习算法的基础上,研究讨论了目标声信号频域中存在的低维流形,通过两种实际的地面和低空飞行声目标数据集进行对比识别实验,分析了基于流形学习的声目标特征提取方法的性能,结果表明基于流形学习的特征提取方法可以发现声信号的本质特征,提高了声目标识别系统的准确性和鲁棒性. 关键词： 声目标识别 特征提取 流形学习