On Intersection Problem for Perfect Binary Codes

The main result is that to any even integer q in the interval 0 ≤ q ≤ 2^{n+1-2log(n+1)}, there are two perfect codes C₁ and C₂ of length n = 2^m − 1, m ≥ 4, such that |C₁ ∩ C₂| = q.     

Bounds on Codes Derived by Counting Components in Varshamov Graphs

We are interested in improving the Varshamov bound for finite values of length n and minimum distance d. We employ a counting lemma to this end which we find particularly useful in relation to Varshamov graphs. Since a Varshamov graph consists of components corresponding to low weight vectors in the cosets of a code it is a useful tool when trying to improve the estimates involved in the Varshamov bound. We consider how the graph can be iteratively constructed and using our observations are able to achieve a reduction in the over-counting which occurs. This tightens the lower bound for any choice of parameters n, k, d or q and is not dependent on information such as the weight distribution of a code. This work is taken from the author’s thesis [10]     

Two-point AG codes from the Beelen-Montanucci maximal curve

In this paper we investigate two-point algebraic-geometry codes (AG codes) coming from the Beelen-Montanucci (BM) maximal curve. We study properties of certain two-point Weierstrass semigroups of the curve and use them for determining a lower bound on the minimum distance of such codes. AG codes with better parameters with respect to comparable two-point codes from the Garcia-Güneri-Stichtenoth (GGS) curve are discovered.     

Two new bounds on the size of binary codes with a minimum distance of three

We prove two new upper bounds on the size of binary codes with a minimum distance of three, namelyA(10, 3)76 andA(11, 3)152.     

基于TW3-C RUS法的骨龄评估方法研究

经典TW3-C RUS (Tanner and Whitehouse 3-Chinese RUS)法将手骨的关键骨骺区域严格划分为9个等级,未充分考虑骨骺发育的连续性,导致骨龄评估存在一定误差。针对该问题,本文提出一种基于TW3-C RUS法的改进骨龄评估方法。采用阈值法的思想,动态选择网络模型输出的前N个等级概率值,并将前N个概率值作为权值计算手骨的加权得分,降低由于手骨单一等级判定引起的误差。针对网络模型冗余问题,采用跨阶段局部网络(cross stage partial network, CSP-Net)轻量化深度残差网络(residual network 50,Resnext50)。实验表明,改进后的方法对男性骨龄评估的平均绝对误差(mean absolute error,MAE)为0.421 4岁,女性MAE为0.412 8岁,相比于经典TW3-C RUS法,骨龄评估准确率有明显提升。轻量化后的网络模型参数量为46.28 MB,相比Resnext50网络模型有明显降低。     

基于transformer自适应特征向量融合的图像分类

针对目前基于transformer的图像分类模型直接应用在小数据集上性能较差的问题,本文提出了transformer自适应特征向量融合网络,该网络在特征提取器中将不同阶段的特征进行融合,减少特征信息丢失的同时获得更多不同感受野下的信息,同时利用最大池化来去除特征中的冗余信息,从而使提取的特征更具有判别性。此外,为了充分利用图像的各级特征信息来进行分类预测,本文将网络各阶段产生的特征向量进行融合,使融合后的特征向量更具有表征能力,从而减少网络对大数据集的依赖,使网络在小数据集中也能获得很好的性能。实验表明,本文提出的 算法在数据集Mini-ImageNet-100、CIFAR-100和ImageNet-1k上的TOP-1准确率分别达到了74.22%、85.86%和81.4%。在没有增加计算量的情况下,在baseline上分别提高了6.0%、3.0%和0.1%,且参数量减少了18.3%。本文代码开源在“https://github.com/xhutongxue/afvf”。     

基于图卷积的视网膜血管轮廓及高不确定度区域细化框架

针对传统卷积神经网络(convolutional neural network, CNN)受感受野大小的限制,无法直接有效地获取空间结构及全局语义等关键信息,导致宽血管边界及毛细血管区域特征提取困难,造成视网膜血管分割表现不佳的问题,提出一种基于图卷积的视网膜血管分割细化框架。该框架通过轮廓提取及不确定分析方法,选取CNN粗分割结果中潜在的误分割区域,并结合其提取的特征信息构造出合适的图数据,送入残差图卷积网络(residual graph convolutional network, Res-GCN)二次分类,得到视网膜血管细化分割结果。该框架可以作为一个即插即用模块接入任意视网膜血管分割网络的末端,具有高移植性和易用性的特点。实验分别选用U型网络(U-neural network, U-Net)及其代表性改进网络DenseU-Net和AttU-Net作为基准网络,在DRIVE、STARE和CHASEDB1数据集上进行测试,本文框架的Sp分别为98.28%、99.10%和99.04%,Pr分别为87.97%、88.87%和90.25%,证明其具有提升基准网络分割效果的细化能力。     

基于CTCNet对球结膜进行糖尿病视网膜病变分类

糖尿病视网膜病变(diabetic retinopathy, DR)是一种糖尿病性微血管病变,会在球结膜微血管上有所体现,球结膜血管图像的获取比眼底图像更加便捷,但微血管的特征变化微小且难以量化。为了能够对患者进行早期辅助诊断,本文依据球结膜微血管形态与DR的关联,首先对球结膜图像进行预处理,使用限制对比度自适应直方图均衡(contrast limited adaptive histogram equalization, CLAHE)算法进行图像增强,随机处理使数据增强,然后结合卷积神经网络(convolutional neural network, CNN)和Transformer各自的网络优势构建CTCNet,对处理后的球结膜血管图像进行DR分类,分类准确率达到了97.44%,敏感度97.69%,特异性97.11%,精确度97.69%,通过实验对比CNN和Transformer, CTCNet网络性能优于其他模型,能够有效识别DR。     

基于卷积神经网络的随机因子重采样图像检测

图像重采样检测是图像取证领域的重要任务,其目的是检测图像是否经过重采样操作。现有的基于深度学习的重采样检测方法大多只针对特定的重采样因子进行研究,而较少考虑重采样因子完全随机的情况。本文根据重采样操作中所涉及的插值技术原理设计了一组高效互补的图像预处理结构以避免图像内容的干扰,并通过可变形卷积层和高效通道注意力机制(efficient channel attention, ECA)分别提取和筛选重采样特征,从而有效提高了卷积神经网络整合提取不同重采样因子的重采样特征的能力。实验结果表明,无论对于未压缩的重采样图像还是JPEG压缩后处理的重采样图像,本文方法都可以有效检测,且预测准确率相比现有方法均有较大提升。     

Feature generation based on relation learning and image partition for occluded person re-identification

In order to solve the challenging tasks of person re-identification(Re-ID) in occluded scenarios, we propose a novel approach which divides local units by forming high-level semantic information of pedestrians and generates features of occluded parts. The approach uses CNN and pose estimation to extract the feature map and key points, and a graph convolutional network to learn the relation of key points. Specifically, we design a Generating Local Part (GLP) module to divide the feature map into different units. Based on different occluded conditions, the partition mode of GLP has high flexibility and variability. The features of the non-occluded parts are clustered into an intermediate node, and then the spatially correlated features of the occluded parts are generated according to the de-clustering operation. We conduct experiments on both the occluded and the holistic datasets to demonstrate its effectiveness.