Transfer learning between crop types for semantic segmentation of crops versus weeds in precision agriculture

Agricultural robots rely on semantic segmentation for distinguishing between crops and weeds to perform selective treatments and increase yield and crop health while reducing the amount of chemicals used. Deep‐learning approaches have recently achieved both excellent classification performance and real‐time execution. However, these techniques also rely on a large amount of training data, requiring a substantial labeling effort, both of which are scarce in precision agriculture. Additional design efforts are required to achieve commercially viable performance levels under varying environmental conditions and crop growth stages. In this paper, we explore the role of knowledge transfer between deep‐learning‐based classifiers for different crop types, with the goal of reducing the retraining time and labeling efforts required for a new crop. We examine the classification performance on three datasets with different crop types and containing a variety of weeds and compare the performance and retraining efforts required when using data labeled at pixel level with partially labeled data obtained through a less time‐consuming procedure of annotating the segmentation output. We show that transfer learning between different crop types is possible and reduces training times for up to 80%. Furthermore, we show that even when the data used for retraining are imperfectly annotated, the classification performance is within 2% of that of networks trained with laboriously annotated pixel‐precision data.     

视频侦查中人像智能分析应用及算法优化

人像智能分析指的是对视频或录像中的人像进行结构化和可视化分析,对目标人物进行性别、年龄、发型等特征的智能识别,这项技术在视频侦查中有极高的应用价值。人像识别早期的算法是通过人工提取特征,通过学习低级视觉特征来针对不同属性进行分类学习,这种基于传统方法的模型表现常常不尽如人意。在计算机视觉领域,通过海量图像数据学习的神经网络比传统方法有更丰富的信息量和特征可以被提取。文章尝试通过深度学习技术训练神经网络模型对行人进行检测和识别,对于衣着不同的行人进行智能识别,具有更好的鲁棒性,提升了视频人像识别的准确率,拓展了人工智能技术在身份识别领域的应用。     

一种基于深度森林的恶意代码分类方法

针对当前恶意代码静态分析方法精度不足的问题，将恶意代码映射为无压缩的灰度图像，然后根据图像变换方法将图像变换为恒定大小的图像，使用方向梯度直方图提取图像的特征，最后提出一种基于深度森林的恶意代码分类方法.实验中选择不同家族的多个恶意代码样本进行分类，验证了该方法的有效性，并且实验结果优于近期提出的SPAM-GIST方法.     

基于改进奇异值分解滤波和谱峭度的滚动轴承故障诊断

针对含噪信号的有效奇异值个数难以确定的问题，提出了一种改进的奇异值分解降噪方法--奇异值累积法。该方法通过计算奇异值的实际下降值与奇异值平均下降速度累积量的差值，并取该差值最大值点的位置作为有效奇异值的分界点来确定有效奇异值的个数。在此基础上，提出了一种基于奇异值累积法与快速谱峭度的滚动轴承故障诊断方法。采用奇异值累积法对原信号进行降噪处理，然后利用快速谱峭度确定滤波器中心频率及带宽，通过分析频段包络谱中明显的频率成分来诊断故障。该方法可以有效去除信号中的噪声，使得到的峭度值所反映的故障冲击更接近实际情况。对含内圈、外圈故障的滚动轴承实验数据进行分析，实验结果表明，相比快速谱峭度的故障诊断方法，该方法具有更好的故障识别效果。     

红外菲涅尔透镜的注射超声辅助成型优化

为了提升聚合物红外菲涅尔透镜的光学性能,以其表面微沟槽的成型质量为目标,提出了一种高效的注射超声辅助成型方法,并对工艺参数进行了综合质量优化。首先分析了超声振动对聚合物的加热和加压效应,设计了一套一模两腔的对比试验模具;接着以红外菲涅尔透镜的调制传递函数MTF和齿形平均高度h为优化质量目标,设计了四步骤的多目标优化流程,通过试验设计、基于BP神经网络的质量目标与注射工艺参数关系建模、基于NSGA-Ⅱ的多目标优化和试验验证进行工艺参数的综合优化。实验结果表明:该多目标优化流程具有很高的精度,MTF和h的平均预测误差MPE分别为4.16%和3.32%;注射超声辅助成型的菲涅尔透镜微沟槽具有更高的复制质量,其齿沟槽平均高度h增加了15.6%,且h值的波动量随着h值的增大而增大,MTF值受齿高h均匀性的影响大于齿高h对其的影响。     

Large scale cathodic exfoliation of graphite using deep eutectic solvent and water mixture

A short-time and low-cost synthesis route was used to produce large lateral size (from 2 to 15 μm) from monolayers to few layers of graphene by a two-step process of electrochemical exfoliation with a deep eutectic solvent in a mixture with water that can be reused, and ultrasonic bath. The graphene was characterized by SEM, TEM, AFM, Raman and electrochemical activity. During the electrochemical exfoliation, high expanded graphene particles were obtained and these were dispersed in a mixture of water with 5%wt ethylene glycol by an ultrasonic bath in order to complete the exfoliation process. An enhancement of the electrical conductivity of these dispersions was obtained with the increase of graphene concentration, 0.38 mg/mL, which best result was achieved with 30 wt% water and a DC voltage of 10 V. It was possible to add a conductive layer to a glass substrate with the graphene obtained and Tyndall effect was observed.     

整体壁板离散填料滚弯成形有限元模拟

具有一定曲率外形的铝合金整体壁板属于典型的多筋零部件,传统弯曲成形工艺成形困难,容易产生多种缺陷。提出了离散填料辅助滚弯成形整体2A12T4铝合金壁板滚弯工艺,并通过实验与整体填料滚变成形进行了对比,同时采用有限元模拟技术,探究了离散填料辅助整体壁板滚弯成形中的影响因素。研究结果表明:填料数量、上下垫板厚度等因素对成形性能影响较大,填料在3~6块时成形稳定,可以有效地避免筋条屈曲与失稳;随垫板厚度增加,壁板表面圆弧更加光滑、连续,但是回弹性增强;上下垫板分别在10~20 mm厚度时,综合性能较好。通过离散填料辅助滚弯技术可以成形出质量较高的整体壁板。     

基于深度学习的传感器优化布置方法

针对传感器优化布置(optimal sensor placement,简称OSP)问题,提出了一种新的使用深度神经网络的解决方案,并以简化的桥梁形状的桁架结构中的振动测试传感器优化为例进行了验证。首先,选择一种传统的传感器优化布置方法,对自动化生成的大量不同的桁架结构分别进行传感器优化布置计算,将所得优化布置结果在进行数据预处理后构建出深度学习方法所需要的训练集与验证集;其次,使用Python语言和深度学习框架TensorFlow设计实现与本研究问题适配的深度神经网络模型并训练;然后,随机生成了新的桁架结构参数;最后,将深度神经网络输出的传感器布置结果和传统方法的计算结果进行了比较,验证了本研究方法的有效性以及在速度上、可移植性与可扩展性方面的性能优势。     

多层圆片级堆叠THz硅微波导结构的制作

基于微电子机械系统(MEMS)工艺,提出一种多层圆片堆叠的THz硅微波导结构及其制作方法。为了验证该结构在制作THz无源器件中的优势,基于6层圆片堆叠的硅微波导结构,设计了一种中心频率365 GHz、带宽80 GHz的功率分配/合成结构,并对其进行了仿真。研究了制作该结构的工艺流程,攻克了工艺过程中的关键技术,包括硅深槽刻蚀技术和多层热压键合技术,并给出了工艺结果。最终实现了多层圆片堆叠功率分配/合成结构的工艺制作和测试。测试结果表明,尽管样品的插入损耗较仿真值增加3 dB左右,考虑到加工误差和夹具损耗等情况,样品主要技术指标与设计值较为一致。