卷积神经网络下的高分二号卫星影像道路提取

本文针对深度神经网络对高分二号遥感影像道路提取时细节信息丢失较多、道路周围环境考虑不充分等情况,在已有的研究成果上,提出一种基于全卷积神经网络遥感影像道路提取的改进方案.方案创新研究了全卷积神经网络的算法原理,将预调色后的高分二号影像按一定尺寸分幅输出,将输出图像及标签对应输入于以全卷积神经网络为基础的改进网络,通过结合残差单元以及增加网络层数得到识别精度较高的道路提取图像.实验表明,该方法在同一样本中对高分二号卫星影像道路提取的效果有所提升,道路的完整性和准确性有所提高.     

基于深层卷积随机配置网络的电熔镁炉工况识别方法研究

为解决电熔镁炉工况识别模型泛化能力和可解释性弱的缺陷, 提出一种基于深层卷积随机配置网络(Deep convolutional stochastic configuration networks, DCSCN)的可解释性电熔镁炉异常工况识别方法. 首先, 基于监督学习机制生成具有物理含义的高斯差分卷积核, 采用增量式方法构建深层卷积神经网络(Deep convolutional neural network, DCNN), 确保识别误差逐级收敛, 避免反向传播算法迭代寻优卷积核参数的过程. 定义通道特征图独立系数获取电熔镁炉特征类激活映射图的可视化结果, 定义可解释性可信度评测指标, 自适应调节深层卷积随机配置网络层级, 对不可信样本进行再认知以获取最优工况识别结果. 实验结果表明, 所提方法较其他方法具有更优的识别精度和可解释性.     

Economic production of probiotics from kitchen waste

In the present work, a novel treatment method of kitchen waste for economic production of probiotics was investigated. This required the selection of suitable probiotic microorganisms. Based on the pure cultures of probiotic strains for the fermentation kitchen waste, 5 strains of microorganisms including 1 strain of Lactobacillus, 2 strains of Bacillus, and 2 strains of yeast were selected, respectively. These probiotic microorganisms were mixed at the same ratio and cultured using the kitchen waste as culture medium at pH of 7.2 and temperature of 37°C. After 24 h, the total count of the viable cells reached 2.24 ×10¹⁰ CFU/g, which was higher than that obtained in any single probiotic strain pure culture. It was found that the presence of yeasts and Bacillus species enhanced the growth of Lactobacillus strain. Bench scale experiments were also done in a self-designed rotating drum type bioreactor. The experimental results indicate that there is a good possibility of utilizing kitchen waste for the economic production of probiotics.     

Studies on thermodynamic nature of steroselectivity for ruthenium(II) polypyridyl complex binding to DNA

A pair of ruthenium(II) complex isomers, Δ- and Λ-[Ru(bpy)₂(p-mpip)]²⁺ {bpy = 2,2′-bipyridine, p-mpip = 2-(4-methylphenyl)imidazo[4,5-f]1,10-phenanthroline} have been synthesized and characterized. Thermodyanmics of binding of the two isomers to calf thymus DNA (CT DNA) has been investigated by isothermal titration calorimetry (ITC). It is rare that the Λ isomer binds to DNA more strongly than the Δ isomer, which has been verified by viscosity measurements, and can be explained by their different thermodynamic driving forces of DNA binding for the first time. This suggests that the dominant factor governing the stereoselectivity of DNA binding of metal complex may be the different thermodynamic driving forces between interactions of its different isomer with DNA. This new finding may be very helpful to understand the nature of steroselective DNA binding of small chiral molecules, and be useful to the development of DNA molecular probes and new DNA targeting therapeutic drugs.     

多糖的磷酸化修饰及其生物活性研究进展

多糖是由多个单糖分子经脱水缩合,藉由糖苷键连接而成的天然大分子化合物,在生物中有储存能量、信息传递和组成结构的作用。多糖的理化性质受其空间结构、分子量、侧链取代基类型、数量和位置的影响。选择合适的改性方法对多糖进行分子修饰,提高其生物活性,这是研究多糖构效关系及开发多糖产品的重要途径。磷酸化多糖是多糖经特定的化学方法进行磷酸基团结构修饰制备的衍生物。本文重点阐述了酸法、磷酰氯法和磷酸盐法3种磷酸化多糖的修饰工艺和工艺条件,分析目前存在的问题,归纳了磷酸化多糖在抗氧化、抗肿瘤、抗病毒和免疫调节等方面的生物活性作用,旨在为多糖的构效研究中提供工艺参考和理论依据。     

基于HALCON的字符识别及缺陷检测

人类日常生活或生产中,利用字符来识别物体已成为主要的途径,因此字符印刷的正确性影响人类的认知及产品的质量.论文的实验针对这一问题,设计一个包括霍尔传感器元件及大恒摄像机的旋转的实验平台,霍尔开关传感器在固定的位置触发摄像机拍照,经过图像采集卡将图像传到计算机,然后利用HALCON软件和基于组件的模板匹配技术在旋转实验台上模拟对在线产品进行字符检测,剔徐废品,提高成品的合格率.结果表明该试验方法可以准确地进行字符识别及缺陷检测.     

基于数控机床的超声自动检测系统

研究了基于数控机床的超声C扫描成像技术。设计试验分析数控机床的运动特性,建立其运动控制的数学模型以实现超声信号采集与C扫描成像的同步。利用RS-232实现数控机床与工控机的串口通信,建立超声自动检测试验平台。通过大量超声C扫描试验表明,该系统具有通用性强、操作方便、开发成本低、检测精度高等优点。