高斯颜色模型在瓷片图像分类中的应用

由于RGB颜色空间不能很好贴近人的视觉感知,同时也缺少对空间结构的描述,因此采用兼顾颜色信息和空间信息的高斯颜色模型以获取更全面的特征,提出了一种基于高斯颜色模型和多尺度滤波器组的彩色纹理图像分类法,用于瓷器碎片图像的分类。首先将原始图像的RGB颜色空间转换到高斯颜色模型;再用正规化多尺度LM滤波器组对高斯颜色模型的3个通道构造滤波图像,并借助主成分分析寻找主特征图,接着选取各通道的最大高斯拉普拉斯和最大高斯响应图像,与特征图联合构成特征图像组用以进行参数提取;最后以支持向量机作为分类器进行学习和分类。实验结果表明,与基于灰度的、基于RGB模型的和基于RGB_bior 4.4小波的方法相比,本文方法具有更好的分类结果,其中在Outex纹理图像库上获得的分类准确率为96.7%,在瓷片图像集上获得的分类准确率为94.2%。此方法可推广应用到其他彩色纹理分类任务。     

基于视觉的多特征手势识别

手势是一种自然直观的交互方式,基于视觉的手势识别是实现新一代人机交互的关键技术。本文在已有的手势识别技术基础上,从手势分割及手势表示两方面着手,提出了一种单目视觉下的手势识别方法。利用颜色特征检测肤色区域,成功分割出人手;利用人手的轮廓及凸缺陷检测指尖,再利用指尖的数目和方位来表示一个手势,进而结合轮廓长度和面积等几何特征完成手势识别。传统的指尖检测方法需要遍历并扫描手掌外轮廓,计算量大,本文通过凸缺陷检测指尖,减少了计算量,提高了指尖检测的速度。实验结果表明,本文的方法具有很好的鲁棒性及实时性,能适应环境的变化。     

基于检索决策树的螨种分类鉴定专家系统

应用传统的分类检索方法进行粉螨亚目螨种的分类鉴定是一个复杂、繁琐的过程。随着人工智能技术的发展,采用专家系统技术编制螨种分类鉴定软件将改变传统的分类鉴定方式。通过学习粉螨亚目的分类学和专家系统的相关知识,将螨种的分类鉴定与专家系统有机地结合起来,采用面向对象方法,设计并实现了基于决策树的螨种分类鉴定专家系统。通过构建基于面向对象的决策树知识表示模型,提炼出螨种知识库的记录结构,根据节点对象的逻辑关系,设计了基于决策树的推理机制,并按照一定的推理控制策略实现了螨种的分类鉴定。结果证明,满足了用户的需要,具有重要的实践价值。     

The Complex and Critical Role of Glycine 12 (G12) in Beta-Connexins of Human Skin

Glycine is an amino acid with unique properties because its side chain is composed of a single hydrogen atom. It confers conformational flexibility to proteins and conserved glycines are often indicative of protein domains involving tight turns or bends. All six beta-type connexins expressed in human epidermis (Cx26, Cx30, Cx30.3, Cx31, Cx31.1 and Cx32) contain a glycine at position 12 (G12). G12 is located about halfway through the cytoplasmic amino terminus and substitutions alter connexin function in a variety of ways, in some cases altering protein interactions and leading to cell death. There is also evidence that alteration of G12 changes the structure of the amino terminus in connexin- and amino acid- specific ways. This review integrates structural, functional and physiological information about the role of G12 in connexins, focusing on beta-connexins expressed in human epidermis. The importance of G12 substitutions in these beta-connexins is revealed in two hereditary skin disorders, keratitis ichthyosis and erythrokeratodermia variabilis, both of which result from missense mutations affecting G12.     

Melanin Distribution in Human Skin: Influence of Cytoskeletal,Polarity, and Centrosome-Related Machinery of Stratum basale Keratinocytes

Melanin granules cluster within supra-nuclear caps in basal keratinocytes (KCs) of the human epidermis, where they protect KC genomic DNA against ultraviolet radiation (UVR) damage. While much is known about melanogenesis in melanocytes (MCs) and a moderate amount about melanin transfer from MC to KC, we know little about the fate of melanin once inside KCs. We recently reported that melanin fate in progenitor KCs is regulated by rare asymmetric organelle movement during mitosis. Here, we explore the role of actin, microtubules, and centrosome-associated machinery in distributing melanin within KCs. Short-term cultures of human skin explants were treated with cytochalasin-B and nocodazole to target actin filaments and microtubules, respectively. Treatment effects on melanin distribution were assessed by the Warthin–Starry stain, on centrosome-associated proteins by immunofluorescence microscopy, and on co-localisation with melanin granules by brightfield microscopy. Cytochalasin-B treatment disassembled supra-nuclear melanin caps, while nocodazole treatment moved melanin from the apical to basal KC domain. Centrosome and centriolar satellite-associated proteins showed a high degree of co-localisation with melanin. Thus, once melanin granules are transferred to KCs, their preferred apical distribution appears to be facilitated by coordinated movement of centrosomes and centriolar satellites. This mechanism may control melanin’s strategic position within UVR-exposed KCs.     

Multi-Scale Dilated Convolutional Neural Network for Hyperspectral Image Classification

基于多尺度空洞卷积神经网络的高光谱图像分类

郑姗姗¹,刘文¹,单锐¹,赵静一²,江国乾³,张智⁴

（1. 燕山大学理学院,河北 秦皇岛 066004;2. 燕山大学机械工程学院,河北 秦皇岛 066004;3. 燕山大学电气工程学院,河北 秦皇岛 066004;4. 北京航天研究所,北京 100094）

创新点说明：

1）将图像分割方法——空洞卷积用于卷积神经网络进行高光谱图像分类,提取更加广泛、抽象的图像特征。

2）构建基于多尺度空洞卷积神经网络的高光谱图像分类方法。搭建多尺度聚合结构,在每个通道中使用快捷连接和空洞卷积结构,有效提取图像特征,避免信息丢失。

研究目的：

针对图像信息丢失问题,得到高精度的高光谱图像分类方法。

研究方法：

在Indian Pines和Pavia University数据集上,与4个已有的高光谱图像分类方法进行对比实验,比较OA, AA和Kappa值。

研究结果：

多尺度空洞卷积神经网络在Indian Pines和Pavia University数据集上OA值分别达到了99.58%,99.92%。AA值分别达到了99.57%,99.90%。Kappa分别达到了99.52%,99.89%。

结论：

1）在卷积神经网络中引入空洞卷积,可以有效避免图像信息丢失。

2）多尺度空洞卷积神经网络能提取更佳的判别性特征,实现高分类性能。

关键词：多尺度聚合;空洞卷积;高光谱图像分类;快捷连接

     

西南山区交通工程弃渣的工程特性评价及其分类

山区交通工程弃渣力学参数的准确测定,一直是弃渣场边坡稳定性评价面临的基础科学问题。在山区交通工程弃渣场运行特点和弃渣固体废弃物特征认知的基础上,探讨弃渣取样和试验代表性问题的物理根源,依托实际工程累计数据尝试提出简单实用的弃渣工程特性评价和分类方法。结果表明:①常规交通工程弃渣的密度、颗粒分析和强度试验,由于弃渣的粒径范围差异大、颗粒空间分布不均、弃渣来源复杂等固体废弃物特征造成取样和试验代表性难题; 用多阶段坡角测量和颗粒分析试验代替传统的弃渣边坡试验,解决了试验和取样代表性难题。②利用弃渣粗细比可将弃渣分为细粒弃渣、混合型弃渣和粗粒弃渣。③根据弃渣粗细比、天然休止角和整形坡率,将西南山区交通工程弃渣分为细粒弃渣、混合型弃渣和粗粒弃渣等3类。细粒弃渣,弃渣粗细比小于0.3,天然休止角小于31.5°,整形坡率为1.00:2.00; 混合型弃渣,弃渣粗细比为0.3～1.0,天然休止角为31.5°～39.5°,整形坡率为1.00:2.00～1.00:1.50; 粗粒弃渣,弃渣粗细比大于1.0,天然休止角大于39.5°,整形坡率为1.00:1.50。④用多阶段坡角测量、颗粒分析试验和无黏性土边坡稳定性系数计算公式代替传统的弃渣边坡试验和稳定性系数计算方法,方法简单,结果保守,可以作为弃渣边坡稳定性评价和安全控制技术的有益补充。     

D-2-DenseNet噪音鲁棒的城市音频分类模型

为了提高噪音环境下城市音频分类系统的鲁棒性,提出了一种双特征2阶密集卷积神经网络（D-2-DenseNet）噪音鲁棒的城市音频分类模型.首先介绍了噪音添加和噪音鲁棒处理,阐述了一种双特征互补偿的算法;然后结合2阶密集卷积神经网络与自适应机制提出了一种噪音鲁棒音频分类模型：双特征2阶密集卷积神经网络.模型采用双特征互补偿自适应算法,可在特征提取与模型训练中更有针对性地提取有效音频信息,降低噪音干扰,以提高噪音鲁棒性.最后,基于Dcase2016数据集开展噪音环境下城市音频分类测试.实验结果表明,模型分类准确率分别可达77.12%、75.52%,与基线模型相比,平均分类准确率分别提高了8.51%和10.38%,验证了模型良好的噪音鲁棒性.     

基于改进MobileNet v2的垃圾图像分类算法

针对现有的垃圾图像分类模型实时性能差和分类精度低的问题,提出基于改进MobileNet v2的垃圾图像分类方法,构建以MobileNet v2为核心的轻量级特征提取网络. 通过调整宽度因子降低模型的参数量;在模型中嵌入通道和空间注意力模块,增强网络对特征的细化能力;设计多尺度特征融合结构,增强网络对尺度的适应性;利用迁移学习的方式优化模型参数,进一步提高模型精度. 实验结果表明,算法在自建数据集上的平均准确率为94.6%,分别高于MobileNet v2、VGG16、GoogleNet、ResNet50、ResNet101模型2.0%、3.4%、3.2%、2.3%、1.2%;所提算法在2种公共图像分类数据集CIFAR-100和tiny-ImageNet中均取得不错表现;模型参数量仅为0.83 M,体积约为基础模型的2/5,在边缘设备JETSON TX2上的单次推理耗时68 ms,实现了推理速度和预测准确率的提升.     

hTERT-Driven Immortalization of RDEB Fibroblast and Keratinocyte Cell Lines Followed by Cre-Mediated Transgene Elimination

The recessive form of dystrophic epidermolysis bullosa (RDEB) is a crippling disease caused by impairments in the junctions of the dermis and the basement membrane of the epidermis. Using ectopic expression of hTERT/hTERT + BMI-1 in primary cells, we developed expansible cultures of RDEB fibroblasts and keratinocytes. We showed that they display the properties of their founders, including morphology, contraction ability and expression of the respective specific markers including reduced secretion of type VII collagen (C7). The immortalized keratinocytes retained normal stratification in 3D skin equivalents. The comparison of secreted protein patterns from immortalized RDEB and healthy keratinocytes revealed the differences in the contents of the extracellular matrix that were earlier observed specifically for RDEB. We demonstrated the possibility to reverse the genotype of immortalized cells to the state closer to the progenitors by the Cre-dependent hTERT switch off. Increased β-galactosidase activity and reduced proliferation of fibroblasts were shown after splitting out of transgenes. We anticipate our cell lines to be tractable models for studying RDEB from the level of single-cell changes to the evaluation of 3D skin equivalents. Our approach permits the creation of standardized and expandable models of RDEB that can be compared with the models based on primary cell cultures.