基于图像分层树的图像语义分割方法

针对传统语义分割模型缺乏空间结构信息,无法准确地描述对象轮廓的问题,提出了一种基于图像分层树的图像语义分割方法。分层树模型采用结构森林方法生成轮廓模型,为防止过度分割,运用超度量轮廓图算法得到多尺度轮廓图,然后利用支持向量机训练多尺度轮廓图生成图像分层树,通过随机森林精炼分层树,最终输出图像语义分割结果。在测试实验中,像素精确度达到82.1%,相比区域选择方法(Selecting Regions)提升了2.7%。并在较难区分的树和山脉的预测精确度上,相比层次标记方法(Stacked Labeling)分别提升了16%,25%,具有更高的稳定性。实验结果表明,在复杂的室外环境下,对图像语义分割的精确度、稳定性和速率均有明显改善。     

Strategic mining options optimization:Open pit mining,underground mining or both

Near-surface deposits that extend to considerable depths are often amenable to both open pit mining and/or underground mining. This paper investigates the strategy of mining options for an orebody using a Mixed Integer Linear Programming(MILP) optimization framework. The MILP formulation maximizes the Net Present Value(NPV) of the reserve when extracted with(i) open pit mining,(ii) underground mining, and(iii) concurrent open pit and underground mining. Comparatively, implementing open pit mining generates a higher NPV than underground mining. However considering the investment required for these mining options, underground mining generates a better return on investment than open pit mining. Also, in the concurrent open pit and underground mining scenario, the optimizer prefers extracting blocks using open pit mining. Although the underground mine could access ore sooner, the mining cost differential for open pit mining is more than compensated for by the discounting benefits associated with earlier underground mining.     

Segmentation of the image with multi-visual features for a traffic scene

Scene segmentations based on the pixel classifying calculation are complicated, and they use insufficient features, thus resulting in a low accuracy, so a new model is proposed to overcome these shortcomings,which is to learn these geometric classes based on multi-visual features of super-pixels. First, various features are extracted from the super-pixels of an input image. These features are used for classifying the super-pixels. Then the difference between the adjacent super-pixels is calculated to predict their consistency. The initial classification result and the consistency are synthesized to the Markov Random Field energy function, which is then minimized based on the graph-cuts algorithm to get the final labels of the super-pixels. Experimental results prove the effectiveness of the multi-visual features and the optimization method proposed, with superior performance achieved for traffic scenes.     

Prestructured MXene fillers with uniform channels to enhance CO2 selective permeation in mixed matrix membranes

The packing pattern of two-dimensional (2D) sheet-like fillers in membranes is relatively random, leading to the unfavorable permeability from tortuous diffusion pathway. A new strategy that using prestructured materials with uniform channels as fillers was proposed. In this work, Ti₃AlC₂ is etched to prepare multilayered MXene (m-MXene), the channels aggregate as a whole unit, ensure the impossibility of ineffective packing compared with traditional individual sheets, largely facilitating the selective permeation. Then, the m-MXene/Poly (amide-6-b-ethylene oxide) (Pebax) MMMs are synthesized. SEM images demonstrate the accordion shaped structure of filler, which is the multi-channels laminates. Furthermore, the results of gas permeation test exhibit enhanced performance of m-MXene/Pebax MMMs. MMM with 0.5 wt.% m-MXene behaved best, CO₂ permeability of 86.22 Barrer as well as CO₂/N₂ selectivity of 104.85, transcending the Robeson upper bound (2008). Having distinct enhancement for CO₂ separation, the m-MXene/Pebax MMMs in this work offer prospective practical applications.     

Fabrication process of smooth functionally graded materials through a real-time inline control of the component ratio

Functionally graded materials (FGMs) play an essential role in tissue engineering because of their satisfactory histocompatibility and excellent mechanical performance. While traditional manufacturing methods allow production of simple FGMs, precise control of composition and customized property at transition between the dissimilar materials is still a challenge. Here, an extrusion-based functionally graded additive manufacturing (FGAM) platform was developed to generate smooth graded parts by thrusting out monolithic cylindrical filaments with high viscosity. Furthermore, the rheological properties, hydrodynamic behavior, and mixed homogeneity of the non-Newtonian fluids were studied. Therefore, the appropriate solid contents, alternative energy-efficient mixers, and optimized printing parameters were proved to be beneficial for an outstanding deposition effect of the suspension. Ultimately, an object with smooth gradient was successfully manufactured. The validity of this strategy was verified via optical microscopy combined with an image processing method to gauge homogeneity and a scanning electron microscope to investigate graded composition and microstructure.     

添加含碳熔剂细化镁合金晶粒的方法

提出了镁合金用粉末状晶粒细化剂加入的新方式-子母球混合粉压块加入法，以常用镁合金AZ31为基，将采用普通方式加入细化剂和采用子母球混合粉压块加入细化剂的镁合金微观组织进行了定量比较，并对熔炼时碳粉在熔体中的分散方式进行了简单的模拟，最后采用纯镁对该加入方式的有效性进行了进一步的验证，试验证明，该加入方式对以改善碳粉对镁及其合金的细化效果，并且操作简单易行。     

基于混合参数蚁群算法的移动机器人路径规划北大核心

针对传统蚁群算法因初期信息素分布不均导致算法初期路径选择概率随机性大、搜索速度慢等缺陷,设计一种使用混合参数的蚁群改进算法。在算法开始阶段引入遗传算法,对遗传算法的适应度函数进行改进;设置一个评价点使遗传算法在合适的时机进入蚁群算法,并对算法的信息素挥发因子p采用一种自适应调整方式;对遗传算法的交叉率和变异率以及蚁群算法的信息因子和期望因子采用变异的混合参数,发挥4个参数因子在算法中的优点;在蚁群算法后面设置一个路径进化率的评价点判定是否再次进行遗传变异操作,目的是使蚁群算法跳出局部最优;算法最后引入B样条曲线光滑机制。实验结果表明:改进算法在简单和复杂的环境里找到的路径长度和路径拐点数明显减少,有比其他3种算法更快更准的寻找全局最优能力。     

Surface finish generated in hard turning of quenched alloy steel parts using conventional and wiper ceramic inserts

Significant progress has already been achieved in green manufacturing including dry and hard, often high-speed, machining technologies. For instance, the demand for higher productivity has resulted in the wider application of ceramic and PCBN tools with special multi-radii (wiper) geometry. This paper reports some important characteristics of the surface roughness produced in the turning of a hardened low-chromium alloy steel using mixed alumina–titanium carbon (TiC) ceramic cutting tools equipped with both conventional and wiper inserts. The characteristic geometrical features of surfaces obtained in both these turning operations have been assessed by means of representative two-dimensional (2D) surface roughness parameters, and some 3D visualizations, which allowed more complete characterization of the surface topography and prediction of its service properties. Results show that keeping equivalent feed rates, i.e. 0.1 mm/rev for conventional and 0.2 mm/rev for wiper tools, the surfaces obtained have similar 3D height roughness parameters, and comparable values of skew and kurtosis. At defined cutting parameters, surfaces produced by wiper tools contain blunt peaks with distinctly smaller slopes resulting in better bearing properties. Only marginal changes of Ra parameter were recorded during 15 min machining trials.     

基于CNN-RF的嵌入式数控系统故障诊断研究北大核心

采用Stacking集成策略,融合卷积神经网络(CNN)和随机森林(RF)法提出一种故障诊断方法CNN-RF。该方法不仅能准确提取数据集中的数据特征,而且针对数据集中故障数据数量不足的问题能提供平衡数据集误差的有效方法,以提高诊断的准确性。分别采用单独模型和集成后的模型对采集到的嵌入式数控系统实时运行数据进行分析处理。结果表明:利用CNN-RF模型进行嵌入式数控系统故障诊断的准确度较高,验证了该模型的正确性。     

用钒(V)-PAR-NH_2OH-DDMAA混合显色体系测定钒

研究了钒(V)-PAR-羟胺-DDMAA(Dodecyldimethylamino acetic acid)混合显色配位体系的特性.该体系在0.5～0.9mol/L硫酸溶液介质中,其最大吸收峰λ_(max)为560nm波长处,在560nm波长处空白值几乎等于零,摩尔吸光系数∈_560为3.7× 10~4L·mol~(-1)·cm~(-1).配合物的组成比,V(v):PAR:NH_2OH:DDMAA为1:1:1:1:1.0～30μgV~(5+)+/25ml范围内.服从比尔定律,用该方法直接测定钨精矿和钨矿中微量钒,获得了满意的结果.