人脸表情识别中滤波算法的研究

去噪滤波技术是数字图像处理中的一个重要内容,是计算机视觉技术、模式识别、图像数据分析的基础,因而受到广泛的研究;并在人脸识别、表情识别、遥感成像等领域获得广泛的应用。通过对多种去噪滤波算法进行比较,选定了一种改进的二维中值滤波快速算法将其应用到表情识别中,得到较好效果。     

植物叶片智能分析系统的设计

为避免农作物病害智能诊断过程中人为主观因素的影响,客观准确的表达叶片信息,利用数字图像处理技术和农业植保专家知识相结合,设计了适合于大田作物的植物叶片图像处理与分析系统。该系统主要包括植物叶片几何失真校正模块,几何特征计算模块、颜色识别模块以及病害区域识别模块;以校正后非线性失真现象的叶片图像为基础,实现了叶片几何特征值和颜色值的计算,并提取其病斑区域图像。试验结果表明,该方法满足病害智能诊断要求,具有良好的适应性和实用性。     

基于最小阈值法的棉花幼苗识别研究

 面对纯机械式分苗机构经常出现的漏分和多分的问题,本文提出使用基于最小阈值法的图像识别技术对棉花幼苗进行识别和分离。使用了HSV颜色空间内与光照条件无关的H值将RGB真彩色图像转化为灰度图像,以提高图像处理的实时性。基于最小值阈值法,提出一种能够有效地将处于复杂背景下的棉花幼苗茎部图像提取的方法,使用棉花幼苗茎部图像的面积和长宽比等几何信息能有效识别棉花幼苗,并得到棉花幼苗茎部的重心坐标。本文的研究证明了用图像处理中的最小阈值法识别棉花幼苗是可行的。     

Uterine biology in pigs and sheep

There is a dialogue between the developing conceptus (embryo-fetus and associated placental membranes) and maternal uterus which must be established during the peri-implantation period for pregnancy recognition signaling, implantation, regulation of gene expression by uterine epithelial and stromal cells, placentation and exchange of nutrients and gases. The uterus provide a microenvironment in which molecules secreted by uterine epithelia or transported into the uterine lumen represent histotroph required for growth and development of the conceptus and receptivity of the uterus to implantation. Pregnancy recognition signaling mechanisms sustain the functional lifespan of the corpora lutea (CL) which produce progesterone, the hormone of pregnancy essential for uterine functions that support implantation and placentation required for a successful outcome of pregnancy. It is within the peri-implantation period that most embryonic deaths occur due to deficiencies attributed to uterine functions or failure of the conceptus to develop appropriately, signal pregnancy recognition and/or undergo implantation and placentation. With proper placentation, the fetal fluids and fetal membranes each have unique functions to ensure hematotrophic and histotrophic nutrition in support of growth and development of the fetus. The endocrine status of the pregnant female and her nutritional status are critical for successful establishment and maintenance of pregnancy. This review addresses the complexity of key mechanisms that are characteristic of successful reproduction in sheep and pigs and gaps in knowledge that must be the subject of research in order to enhance fertility and reproductive health of livestock species.     

Pattern recognition receptors in equine endotoxaemia and sepsis

Pattern recognition receptors (PRRs) on host cells detect pathogens to activate innate immunity which, in turn, initiates inflammatory and adaptive immune responses. Successful activation of PRRs is, therefore, critical to controlling infections and driving pathogen-specific adaptive immunity, but overactivity of PRRs causes systemic inflammation, which is detrimental to the host. Here we review the PRR literature as it relates to horses and speculate on the role PRRs may play in sepsis and endotoxaemia.     

基于鱼体特征点检测的淡水鱼种类识别

针对传统机器视觉技术对淡水鱼种类进行检测时特征提取过程复杂的问题,该研究提出了基于特征点检测的淡水鱼种类识别方法。以鳊、鳙、草鱼、鲢、鲤5种大宗淡水鱼为对象,构建了淡水鱼特征点检测数据集;以AlexNet模型为基础,通过减小卷积核尺寸、去除局部响应归一化、引入批量归一化、更换损失函数,构建了改进AlexNet模型用于特征点检测;并以特征点为依据提取特征值、构造特征向量,使用Fisher判别分析方法实现了淡水鱼的种类识别。试验结果表明：改进AlexNet模型在测试集上的归一化平均误差的均值为0.0099,阈值δ为0.02和0.03时的失败率F_0.02、F_0.03分别为2.50%和0.83%,具有较好的精准度和误差分布情况;基于该模型和Fisher判别分析的淡水鱼种类识别方法对5种淡水鱼的识别准确率为98.0%,单幅图像的平均识别时间为0.368 s,保证了时效性。由此可知,提出的改进AlexNet模型能实现淡水鱼的特征点检测并具有较高的精度,可为淡水鱼种类识别、尺寸检测、鱼体分割等提供条件,该方法可为淡水鱼自动化分类装置的研发奠定基础。     

基于迁移学习与YOLOv8n的田间油茶果分类识别

为降低视觉引导式油茶果采摘机器人采摘被遮挡油茶果时造成的果树和抓取装置损伤,该研究提出了一种基于迁移学习和YOLOv8n算法的油茶果分类识别方法,将油茶果分成无遮挡和遮挡两类。首先,采用COCO128目标检测数据集作为源域,苹果数据集为辅助域的迁移学习方法训练模型。其次,将学习方法、训练数据量、学习率和训练轮数这4种因素组合,共进行了52组YOLOv8n检测性能的消融试验。最后,将YOLOv8n模型与YOLOv3-tiny、YOLOv5n和YOLOv7-tiny等模型进行比较。试验结果表明,随机权重初始化方式受训练数据量和学习率影响较大,学习率为0.01时模型检测效果最好;而迁移学习方法仅用随机权重初始化1/2的数据量即可达到与其相当的平均精度均值;迁移学习方式下,YOLOv8n模型的平均精度均值最高达到92.7%,比随机权重初始化方式提升1.4个百分点。与YOLOv3-tiny、YOLOv5n和YOLOv7-tiny等模型相比,YOLOv8n模型的平均精度均值分别提高24.0、1.7和0.4个百分点,研究结果可为YOLOv8n模型训练参数优化和油茶果分类识别提供参考。     

Partial discharge pattern recognition based on S transform and two-directional 2DPCA

A new feature extraction method is proposed to recognize different types of partial discharge (PD) signals. Firstly,four typical categories of PD artificial defect models are made and S transform (ST) is employed to obtain a time-frequency representation of the recorded UHF signals. Then,two-directional two-dimensional principal component analysis ((2D) 2PCA) is applied to compress the ST amplitude (STA) matrix to extract features. Finally,support vector machine (SVM) combined with particle swarm optimization (PSO) algorithm is introduced to accomplish the recognition of experimental samples. Classification results demonstrate that the average recognition rate of (10,5) combination is the highest while the one of (5,5) combination is the lowest among four kinds of feature dimension combinations. Moreover,PSO can obviously improve the classification performance of SVM. Specifically,all the average recognition rates of PSO-SVM are higher than 94.43%and the maximum value comes to 97.67%. Therefore,the feature sets extracted by ST and (2D) 2PCA can not only achieve dramatic dimension reduction,but also retain the major information of original data. It is proved that the proposed algorithm can obtain ideal results in PD pattern recognition.     

基于矛盾解决理论与物质——场分析的实木地板节子识别新方法研究

针对传统区域生长方法识别实木地板节子存在准确率低且速度慢的问题,运用TRIZ中矛盾解决理论分析与物质——场分析,提出一种结合分水岭、区域生长以及边缘检测的新的实木地板节子识别算法。算法首先将原图像转换为灰度图像;其次,运用形态学分水岭的方法对灰度图像进行分割;再次,选取满足条件的种子区域进行区域生长,得到节子区域;最后,运用Sobel算子对图像进行梯度运算,并找到节子的边缘。仿真实验表明,该算法较传统方法能够找到更合适的种子区域和区域生长的阈值,实现了对节子的快速、完整提取,节子分割平均用时60ms,平均辨识准确率在90％以上。     

基于多重分形和LVQ神经网络的小麦病害智能识别

针对不同小麦病害有不同的形状特征,利用多重分形分析提取小麦病害图像的8个多重分形谱值作为小麦病害的形状特征参数,并利用这8个特征参数来索引图像数据库作为学习向量量化(LVQ)神经网络的输入,进行样本训练、分类识别.试验结果表明,该算法对小麦病害的识别率可达90.0％以上.