一种新型的无人机牧草分割网络—LMS-DeeplabV3+

目前草原环境复杂、牧草分散且与背景颜色差异小,无法实现高效精准的分割,因此本文提出了一种新型的轻量化多尺度DeeplabV3+网络(lightweight and multi-scale DeeplabV3+network, LMS-DeeplabV3+)。该网络以DeeplabV3+为基础网络,首先选用轻量级的MobilenetV2作为骨干网络用于初步特征提取,并为了适应牧草分割任务做了网络配置上的调整;其次在加强特征提取模块和解码模块中均使用深度可分离卷积代替普通卷积以轻量化网络;此外利用密集空洞空间金字塔池化(dense atrous spatial pyramid pooling, DASPP)模块捕获更大的感受野,加强各特征之间的交互;又引入卷积注意力机制(convolutional block attention module, CBAM)重分配权重加强特征提取。实验证明,提出的新网络与原始网络相比平均交并比(mean intersection over union,mIOU)提升了8.06个百分点、平均像素精度(mean pixel accuracy,mPA)提升了6.7...     

用于空间的三反射镜光学系统设计

现代空间光学的发展需要很大的线视场、像质接近衍射极限或很大相对口径的无中心遮拦反射系统。能满足这些要求的非三反射系统莫属,虽然现在已有功能十分齐全的光学系统自动优化软件,但求出一个合理的初始结构尺寸还是十分重要的。文章主要探讨了用于空间的三反射镜光学系统设计方法。     

Colored timed Petri nets for modeling and analysis of cluser tools

In this paper, a colored timed resource‐oriented Petri net (CTROPN) is developed to model the cluster tools in semiconductor fabrication. It is structurally simple and concise, since each module in a cluster tool corresponds to a unique place in CTROPN. It is also powerful in modeling the qualitative and temporal behavior of the cluster tools with a single‐blade or dual‐blade robot. The initial transient behavior, steady‐state behavior and the final transient behavior can all be investigated through a single model. Copyright © 2010 John Wiley and Sons Asia Pte Ltd and Chinese Automatic Control Society     

杂交空间结构的预应力全过程分析方法

杂交空间结构系由不同类型的单一结构形式组合而成的空间结构体系,预应力全过程分析是其关键和难点。结合几类典型的杂交空间结构工程实践,详细讨论了不同结构形式中采用的预应力模拟和全过程分析方法,提出了初始形变迭代分析法、初始形变顺序分析法、升温降温法、模拟千斤顶法、影响矩阵迭代法等预应力分析方法,完善了杂交空间结构的预应力分析理论,并通过对实际工程的分析验证了这些方法的准确性。     

三电平逆变器SHEPWM多组解特性比较及实验

研究了应用于三电平NPC逆变器的特定消谐(Selective Harmonic Eliminated PWM, SHEPWM)技术.传统SHE均假设波形1/4周期对称,这有助于减小方程组的维数,但同时也减少了可行解的空间,而消除低次谐波并不严格要求波形1/4周期对称.针对1/4周期对称SHE,以N=7为例给出四组解,然后进一步研究1/2周期对称下的解,从理论上证明可以存在无数组解,进一步将其分为两种不同的类型.分析和比较了1/4周期对称解和1/2周期对称解的区别和联系,以及不同类型的解在谐波特性方面的差异,并通过实验进行了验证.     

动态场景下基于视觉的SLAM技术研究

针对视觉同时定位与地图构建(SLAM)技术在动态环境中存在定位精度低、地图虚影等问题,提出了一种基于深度学习的动态SLAM算法。该算法利用网络参数少且目标识别率高的YOLOv8n改善系统的视觉前端,为视觉前端增加语义信息,提取动态区域特征点。然后采用LK光流法识别动态区域的动态特征点,剔除动态特征点并保留动态区域内的静态特征点,提高特征点利用率。此外,该算法通过增加地图构建线程,剔除YOLOv8n提取的动态物体点云,接收前端提取的语义信息,实现静态语义地图构建,消除由动态物体产生的虚影。实验结果显示,在动态环境下该算法与ORB-SLAM3相比,定位精度提升92.71%,与其他动态视觉SLAM算法相比,也有小幅度改善。     

THEORETICAL ANALYSIS OF COORDINATES MEASUREMENT BY FLEXIBLE 3D MEASURING SYSTEM

The system mathematical model of flexible 3D measuring system is built by theoretical analysis,and the theoretical formula for measuring space point coordinate is also derived. Frog-jumping based coordinate transform method is put forward in order to solve measuring problem for large size parts. The frog-jumping method is discussed,and the coordinate transform mathematical model is method of the space point coordinate compared to original value,and an advanced method is provided. Form the space point coordinate transform formula can derive the calculation measuring method for measuring large size parts.     

模糊控制技术在大空间二次回风空调系统中的应用

介绍了模糊控制技术在一交响乐厅二次回风空调系统中的应用。针对该工程实际情况提出了完整的控制策略,设计了新风阀、二次回风阀和冷水阀的模糊控制器,以保证室内温湿度和空气质量达到设计要求。以建筑物集成管理系统为操作平台,在中央管理计算机上实现了各设定值的优化,由现场DDC完成闭环控制。     

Single-Step Fast Tissue Clearing of Thick Mouse Brain Tissue for Multi-Dimensional High-Resolution Imaging

Recent advances in optical clearing techniques have dramatically improved deep tissue imaging by reducing the obscuring effects of light scattering and absorption. However, these optical clearing methods require specialized equipment or a lengthy undertaking with complex protocols that can lead to sample volume changes and distortion. In addition, the imaging of cleared tissues has limitations, such as fluorescence bleaching, harmful and foul-smelling solutions, and the difficulty of handling samples in high-viscosity refractive index (RI) matching solutions. To address the various limitations of thick tissue imaging, we developed an Aqueous high refractive Index matching and tissue Clearing solution for Imaging (termed AICI) with a one-step tissue clearing protocol that was easily made at a reasonable price in our own laboratory without any equipment. AICI can rapidly clear a 1 mm thick brain slice within 90 min with simultaneous RI matching, low viscosity, and a high refractive index (RI = 1.466), allowing the imaging of the sample without additional processing. We compared AICI with commercially available RI matching solutions, including optical clear agents (OCAs), for tissue clearing. The viscosity of AICI is closer to that of water compared with other RI matching solutions, and there was a less than 2.3% expansion in the tissue linear morphology during 24 h exposure to AICI. Moreover, AICI remained fluid over 30 days of air exposure, and the EGFP fluorescence signal was only reduced to ~65% after 10 days. AICI showed a limited clearing of brain tissue >3 mm thick. However, fine neuronal structures, such as dendritic spines and axonal boutons, could still be imaged in thick brain slices treated with AICI. Therefore, AICI is useful not only for the three-dimensional (3D) high-resolution identification of neuronal structures, but also for the examination of multiple structural imaging by neuronal distribution, projection, and gene expression in deep brain tissue. AICI is applicable beyond the imaging of fluorescent antibodies and dyes, and can clear a variety of tissue types, making it broadly useful to researchers for optical imaging applications.     

浮选图像中泡沫边缘分割

矿物浮选过程中泡沫的表面特征是浮选性能的重要指标,它可以实时、直观地反映浮选效果的变化,准确分割泡沫的边缘信息是浮选过程中一项重要的任务。近年来,研究人员提出了各种浮选泡沫图像分割算法,但浮选泡沫图像中存在泡沫数量多、泡沫间粘连严重以及边缘不清晰等问题,现存的方法由于其特征提取能力有限,无法精确的分割泡沫边缘。据此,本文利用深度学习提出了一种基于多尺度融合的浮选泡沫图像边缘分割算法,该算法通过引入一种深度高分辨率的编码结构以及一种基于注意力的分层融合方法来增强模型的特征提取能力,从而提高对于浮选泡沫边缘的分割效果。具体而言,深度高分辨率的编码结构可以在不同分辨率层级上同时维护特征信息,使我们的网络模型可以有效地捕捉不同尺度的信息,在提高图像语义理解能力的同时能够保持更多的细节信息,提高处理高分辨率以及密集任务图像的能力。除此之外,本文设计了一种基于注意力的分层融合方法来充分融合深层和浅层的特征图,使融合得到的特征图趋向于更重要的特征信息,从而提高识别浮选泡沫的边界和精确定位浮选泡沫的能力。该算法在泡沫边界分割数据集上凭借58.25的泡沫IoU以及73.62的泡沫Fscore取得了最佳的性能,证明了我们提出的算法可以更加准确地分割浮选泡沫边缘。