基于部分亲和场的行走奶牛骨架提取模型

在奶牛关键点预测的基础上,通过点、线重构奶牛骨架结构,可为奶牛跛行检测、发情行为分析、运动量估测等提供重要参考。本研究基于部分亲和场,以养殖场监控摄像头拍摄的视频为原始数据,使用1 600幅图像训练了奶牛骨架提取模型,实现了奶牛站立、行走状态下关键点信息和部分亲和场信息的预测,并通过最优匹配连接对奶牛骨架结构进行准确提取。为了验证该模型的性能,采用包含干扰因素的100幅单目标奶牛和100幅双目标奶牛图像进行了测试。结果表明,该模型对单目标行走奶牛骨架提取的置信度为78.90%,双目标行走奶牛骨架提取的置信度较单目标下降了10.96个百分点。计算了不同关键点相似性(Object keypoint similarity,OKS)下的模型准确率,当OKS为0.75时,骨架提取准确率为93.40%,召回率为94.20%,说明该模型具有较高的准确率。该方法可以提取视频中奶牛骨架,在无遮挡时具有高置信度和低漏检率,当遮挡严重时置信度有所下降。该模型的单目标和双目标图像帧处理速度分别为3.30、3.20 f/s,基本相同。本研究可为多目标奶牛骨架提取提供参考。     

家禽弃尸重制骨骼囊腔铸塑干教具的实验探索

本文笔者将解剖实验用后的家禽弃尸重新研制成实用的家禽骨骼囊腔铸塑干教具，不仅解决了实验动物尸体二次污染环境的问题，而且因为教具形态逼真、精细牢固而使解剖教学更形象生动，教学成果显著，同时还节省了很大一笔解剖教学实验经费。     

基于Web的森林防火专家系统知识库的构建

介绍了基于森林防火专家系统的知识表示、知识获取和知识库组织的基本原理和方法，阐述了一种基于网络的知识库构建的一种新方案，使推理机的设计实现容易、知识库的维护更加灵活，实现了知识的获取功能以及网络多用户访问时的并发控制功能，克服了传统专家系统的弊端。     

吉林省日光温室钢骨架的受力性能研究

针对目前吉林省内稍大跨度的日光温室产品抗风雪能力弱的现状,阐述了7.5~9m跨度钢骨架的几何设计参数、形状参数及选材方法。通过对比分析管材和钢筋的耗钢量、抗弯刚度及抗腐蚀性等特征,确定管材替代钢筋投入日光温室使用能有效提高其经济效益。同时,利用同济大学3D 3 S软件,以8m跨度钢骨架为例,分析了给定结构形式的受力性能及选材截面尺寸,为新建者提供参考。     

基于二叉树结构的面状目标主骨架生成算法

按照面状目标的数学定义将其分为简单目标和复连通目标2大类,应用基于Delaunay三角网自动提取骨架技术各类面状目标的主骨架进行提取。结果表明,该算法对目标主骨架的提取效果较好,且内插结果基本符合内插要求。     

Soil nutrient supply and biomass production in a mixed forest on a skeleton‐rich soil and an adjacent beech forest

In the natural forest communities of Central Europe, beech (Fagus sylvatica L.) predominates in the tree layer over a wide range of soil conditions. An exception with respect to the dominance of beech are skeleton‐rich soils such as screes where up to 10 broad‐leaved trees co‐exist. In such a Tilia‐Fagus‐Fraxinus‐Acer‐Ulmus forest and an adjacent mono‐specific beech forest we compared (1) soil nutrient pools and net nitrogen mineralization rates, (2) leaf nutrient levels, and (3) leaf litter production and stem increment rates in order to evaluate the relationship between soil conditions and tree species composition. In the mixed forest only a small quantity of fine earth was present (35 g l^—1) which was distributed in patches between basalt stones; whereas a significantly higher (P < 0.05) soil quantity (182 g l^—1) was found in the beech forest. In the soil patches of the mixed forest C and N concentrations and also concentrations of exchangeable nutrients (K, Ca, Mg) were significantly higher than in the beech forest. Net N mineralization rates on soil dry weight basis in the mixed forest exceeded those in the beech forest by a factor of 2.6. Due to differences in fine earth and stone contents, the volume related soil K pool and the N mineralization rate were lower in the mixed forest (52 kg N ha^—1 yr^—1, 0—10 cm depth) than in the beech forest (105 kg N ha^—1 yr^—1). The leaf N and K concentrations of the beech trees did not differ significantly between the stands, which suggests that plant nutrition was not impaired. In the mixed forest leaf litter fall (11 %) and the increment rate of stem basal area (52 %) were lower than in the beech forest. Thus, compared with the adjacent beech forest, the mixed forest stand was characterized by a low volume of patchy distributed nutrient‐rich soil, a lower volume related K pool and N mineralization rate, and low rates of stem increment. Together with other factors such as water availability these patterns may contribute to an explanation of the diverse tree species composition on Central European screes.     

A Fast One-Step Scanning Thinning Algorithm

A new quick thinning algorithm was presented in this paper,which assigneddifferent weights to different pixels near the skeleton pixels and rules out the im possible pixels in thethinning procedure. It needs neither iteration on pixel panel nor detection on every pixel such thatthe time consumed is greatly lessened. A filling algorithm is also offered for better results.Experiment on variety of binary patterns showed that it could get both a high speed and a goodskeleton shape compared with other algorithms,This new algorithm reaches perfect result.     

基于深度图像和生猪骨架端点分析的生猪步频特征提取

为高效提取生猪的行走快慢特征,以微软公司Kinect作为图像采集设备,采集生猪运动深度图像序列。在对各帧深度图像进行骨架提取、剪枝的基础上,采用基于路径相似性骨架图匹配法确定生猪前后肢骨架端点,进一步以骨架端点所属骨架枝子集像素值特征判定端点远近侧属性。以生猪前后肢远、近侧端点的帧间相对坐标变化建立了生猪运动模型,提出了通过帧间坐标变化点集拟合正弦曲线计算生猪行走完整步的方法。最后,通过计算序列完整步与序列采集时间长度比值提取生猪步频特征。通过对采集的28个生猪运动深度图像序列及其镜像序列共56个图像序列进行的试验,表明该文提出方法的正确率达到82.1%。该项研究对于开展生猪异常步态分析,进一步建立生猪多源特征融合的计算机视觉异常监测系统,提高生猪异常行为预警可靠性具有重要意义。     

基于深度学习与激光点云的橡胶林枝干重建及参数反演

树木的几何建模在林木性状评价、森林动态经营管理与可视化研究中具有重要意义。现今,从激光雷达（Light Detection And Ranging,LiDAR）数据中重建树体三维模型并精准获取林木空间枝干结构参数是数字林业发展的必然趋势。该研究提出了一种深度学习与计算机图形学相融合的树木骨架重建与参数反演方法。该方法以PR107、CATAS 7-20-59、CATAS 8-79三个品种的橡胶树为实验对象,首先,采用背包移动激光雷达获取三个橡胶树品种的样地数据,并通过体素剖分和数据增广策略来构建橡胶树训练样本集。其次,构造由四层特征编码层和特征解码层所组成的点云分类深度学习网络,并包含优化的PointConv模块与不同尺度的特征插值模块,以实现在多尺度条件下,全面考虑点云的全局和局部优化特征,引导网络实现枝叶点云的精确分类。最后,面向分类后的枝干点云,运用计算机图形学的空间连通性算法与圆柱拟合策略,重建树木骨架模型,并自动解决叶子点云与对应的一级枝干归属问题,进而在叶团簇尺度下开展对单株树的精细描述与参数反演。通过对三块橡胶树测试样地的验证和与实测值的比对表明,该研究提出的深度学习网络枝叶分类总体准确率在90.32%以上。骨架重建与叶团簇分析结果显示,PR107品种橡胶树具有较为发散的树冠、最大的分枝夹角和叶团簇体积;CATAS 7-20-59品种橡胶树冠呈花瓶型,分枝夹角和叶团簇体积较小;而CATAS 8-79品种橡胶树尽管胸径最粗,但不耐寒害处于落叶期导致冠积最小。同时,反演得到的橡胶树一级枝干直径与实测值比对为：决定系数R2不低于0.94,均方根误差（Root Mean Square Error,RMSE）小于3.01 cm;主枝干与一级枝干的分枝角为：决定系数R2不低于0.91,均方根误差RMSE不高于4.94°。同时发现橡胶树一级枝干的直径与对应的叶团簇体积呈正相关分布。该研究将人工智能的理论模型应用于林木的激光点云数据处理中,为林木激光点云的智能化分析与处理提供了新颖的解决思路。