Time and Space Efficient Multi-Model Convolution Vision Transformer for Tomato Disease Detection from Leaf Images with Varied Backgrounds

A consumption of 46.9 million tons of processed tomatoes was reported in 2022 which is merely 20%of the total consumption.An increase of 3.3%in consumption is predicted from 2024 to 2032.Tomatoes are also rich in iron,potassium,antioxidant lycopene,vitamins A,C and K which are important for preventing cancer,and maintaining blood pressure and glucose levels.Thus,tomatoes are globally important due to their widespread usage and nutritional value.To face the high demand for tomatoes,it is mandatory to investigate the causes of crop loss and minimize them.Diseases are one of the major causes that adversely affect crop yield and degrade the quality of the tomato fruit.This leads to financial losses and affects the livelihood of farmers.Therefore,automatic disease detection at any stage of the tomato plant is a critical issue.Deep learning models introduced in the literature show promising results,but the models are difficult to implement on handheld devices such as mobile phones due to high computational costs and a large number of parameters.Also,most of the models proposed so far work efficiently for images with plain backgrounds where a clear demarcation exists between the background and leaf region.Moreover,the existing techniques lack in recognizing multiple diseases on the same leaf.To address these concerns,we introduce a customized deep learning-based convolution vision transformer model.Themodel achieves an accuracy of 93.51%for classifying tomato leaf images with plain as well as complex backgrounds into 13 categories.It requires a space storage of merely 5.8 MB which is 98.93%,98.33%,and 92.64%less than stateof-the-art visual geometry group,vision transformers,and convolution vision transformermodels,respectively.Its training time of 44 min is 51.12%,74.12%,and 57.7%lower than the above-mentioned models.Thus,it can be deployed on(Internet of Things)IoT-enabled devices,drones,or mobile devices to assist farmers in the real-time monitoring of tomato crops.The periodicmonitoring promotes timely action to prevent the spread of diseases and reduce crop loss.

联合CTC和Transformer的轮式移动机器人语音控制研究

针对人机交互过程中手动控制轮式移动机器人步骤繁琐且双手受到限制的问题,提出并实现基于深度学习算法的轮式移动机器人语音控制系统。系统选取树莓派4B开发板作为主控制器,科大讯飞6阵列语音模块作为语音采集器,STM32单片机作为底层轮式移动机器人的控制器。语音识别算法部分设计基于Transformer的端到端语音识别模型,并加入连续时序分类算法来辅助模型的训练,模型的收敛速度和鲁棒性都得到相应的提升。模型在AISHELL-1语音数据集上测试的字错率为5.57%,相对于Transformer单独训练,字错率相对下降5.1%。经过平台搭建和实验,实现轮式移动机器人根据用户语音指令完成相对应动作的目的,有利于提高工作效率和解放用户的双手。

基于Mobile-VIT的旋转机械故障诊断方法

主要提出了一种基于Mobile-VIT神经网络技术的旋转机械故障的识别分析方法:首先,将旋转的机械构件在其各种连续运行工作状态条件下获得的高频振动信号通过短时的傅里叶变换方法转换为时频图像;然后,将图像输入到搭建好的Mobile-VIT网络模型中,通过其对时频图的识别以及特征提取实现旋转机械故障诊断。实验结果表明,提出的方法具有较高的故障识别准确率,能够有效识别旋转机械的运行状态。

The transition from traditional banking to mobile internet finance:an organizational innovation perspective-a comparative study of Citibank and ICBC

The development of Financial Technology(FinTech)in areas such as mobile Internet,cloud computing,big data,search engines,and blockchain technology have significantly changed the financial industry.FinTech is expected to overturn the traditional banking business model,forcing banks to upgrade and transform.This study adopts a comparative case study method to contrast and analyze the Industrial and Commercial Bank of China(ICBC)and Citibank.It analyzes the strategies,organizations,HR systems,and product innovations adopted by these two banks in response to the impact of FinTech.This paper proposes an“electric vehicle”mode for ICBC and an“airplane mode”for Citibank.Further,it describes the difficulties encountered by the Chinese banking industry and proposes some feasible ways to upgrade.“Technology power”will become the core competitive concept for the financial institutions of the future.

Robust Iterated Sigma Point FastSLAM Algorithm for Mobile Robot Simultaneous Localization and Mapping

Simultaneous localization and mapping （SLAM） is a key technology for mobile robots operating under unknown environment. While FastSLAM algorithm is a popular solution to the SLAM problem, it suffers from two major drawbacks： one is particle set degeneracy due to lack of observation information in proposal distribution design of the particle filter; the other is errors accumulation caused by linearization of the nonlinear robot motion model and the nonlinear environment observation model. For the purpose of overcoming the above problems, a new iterated sigma point FastSLAM （ISP-FastSLAM） algorithm is proposed. The main contribution of the algorithm lies in the utilization of iterated sigma point Kalman filter （ISPKF）, which minimizes statistical linearization error through Gaussian-Newton iteration, to design an optimal proposal distribution of the particle filter and to estimate the environment landmarks. On the basis of Rao-Blackwellized particle filter, the proposed ISP-FastSLAM algorithm is comprised by two main parts： in the first part, an iterated sigma point particle filter （ISPPF） to localize the robot is proposed, in which the proposal distribution is accurately estimated by the ISPKF; in the second part, a set of ISPKFs is used to estimate the environment landmarks. The simulation test of the proposed ISP-FastSLAM algorithm compared with FastSLAM2.0 algorithm and Unscented FastSLAM algorithm is carried out, and the performances of the three algorithms are compared. The simulation and comparing results show that the proposed ISP-FastSLAM outperforms other two algorithms both in accuracy and in robustness. The proposed algorithm provides reference for the optimization research of FastSLAM algorithm.

Mobile Robot Hierarchical Simultaneous Localization and Mapping Using Monocular Vision

A hierarchical mobile robot simultaneous localization and mapping (SLAM) method that allows us to obtain accurate maps was presented. The local map level is composed of a set of local metric feature maps that are guaranteed to be statistically independent. The global level is a topological graph whose arcs are labeled with the relative location between local maps. An estimation of these relative locations is maintained with local map alignment algorithm, and more accurate estimation is calculated through a global minimization procedure using the loop closure constraint. The local map is built with Rao-Blackwellised particle filter (RBPF), where the particle filter is used to extending the path posterior by sampling new poses. The landmark position estimation and update is implemented through extended Kalman filter (EKF). Monocular vision mounted on the robot tracks the 3D natural point landmarks, which are structured with matching scale invariant feature transform (SIFT) feature pairs. The matching for multi-dimension SIFT features is implemented with a KD-tree in the time cost of O(lbN). Experiment results on Pioneer mobile robot in a real indoor environment show the superior performance of our proposed method.

A Simple Deep Learning Network for Classification of 3D Mobile LiDAR Point Clouds

Automatic and accurate classification is a fundamental problem to the analysis and modeling of LiDAR(Light Detection and Ranging)data.Recently,convolutional neural network(ConvNet or CNN)has achieved remarkable performance in image recognition and computer vision.While significant efforts have also been made to develop various deep networks for satellite image scene classification,it still needs to further investigate suitable deep learning network frameworks for 3D dense mobile laser scanning(MLS)data.In this paper,we present a simple deep CNN for multiple object classification based on multi-scale context representation.For the pointwise classification,we first extracted the neighboring points within spatial context and transformed them into a three-channel image for each point.Then,the classification task can be treated as the image recognition using CNN.The proposed CNN architecture adopted common convolution,maximum pooling and rectified linear unit(ReLU)layers,which combined multiple deeper network layers.After being trained and tested on approximately seven million labeled MLS points,the deep CNN model can classify accurately into nine classes.Comparing with the widely used ResNet algorithm,this model performs better precision and recall rates,and less processing time,which indicated the significant potential of deep-learning-based methods in MLS data classification.

Mobile robot simultaneous localization and map building based on improved particle filter

We present an investigation into the use of pan tilt zoom camera and sonar sensors for simuhaneous localization and mapping with artificial colored landmarks. An improved particle filter is applied to estimate a posterior of the pose of the robot, in which each particle has associated it with an entire map. The distributions of landmarks are also represented by particle sets, where separate particles are used to represent the robot and the landmarks. Hough transform is used to extract line segments from sonar observations and build map simultaneously. The key advantage of our method is that the full posterior over robot poses and landmarks can be nonlinearly approximated at every point in time by particles. Especially the landmarks are affixed on the moving robots, which can reduce the impact of the depletion problem and the impoverishment problem produced by basic particle filter. Experimental results show that this approach has advantages over the basic particle filter and the extended Kalman filter.

Phase field calculation of interface mobility in a ternary alloy

A novel approach based on the quantitative phase field model was proposed to calculate the interface mobility and applied to the α/β interface of a ternary Ti-6Al-4V alloy.Phase field simulations indicate that the higher interface mobility leads to the faster transformation rate,but only a unique value of interface mobility matches the diffusion equation under the diffusion-controlled condition.By comparing the transformation kinetics from phase field simulations with that from classical diffusion equation,the interface mobility at different temperatures can be obtained.The results show that the calculated interface mobility increases with increasing temperature and accords with Arrhenius equation very well.

Peak-to-Average Power Ratio Performance Analysis for Orthogonal Chirp Division Multiplexing Multicarrier Systems Based on Discrete Fractional Cosine Transform

In doubly selective fading channels, the orthogonal frequency division multiplexing (OFDM) multicarrier system may fail. Chirp like basis (fractional Fourier transform-fractional cosine transform) may be used instead of complex exponential basis in this case to improve the system performance. However, in multicarrier transmission, the high peak to average power ratio (PAPR) of the transmitted signal is one of the difficult problems that face both the chirp and the exponential basis. In this paper, an evaluation for the PAPR performance of a multicarrier system based on the fractional cosine transform (FrCT) is introduced and then compared with DFrFT and FFT. Moreover, applying the SLAM technique over these systems is provided to understand the behaviour of these systems when applying SLAM. Simulations verify that this system obtains a better PAPR performance. Moreover, further PAPR reduction can be gained using the well-known PAPR reduction methods. Moreover, applying SLAM technique improves the performance of (dB) by 4 dB to 5 dB and all systems become as competitive to each other when SLAM is applied. Finally, BER performance comparison among OFDM, Discrete Cosine Transform MCM (DCT- MCM), Discrete Hartley Transform MCM (DHT-MCM), DFrFT-OCDM and DFrCT- OCDM MCM systems was done by means of simulation over 100,000 multicarrier blocks for each one and showed that our proposed scenario gave the best performance.

Multi-sensor Data Fusion by Improved Hough Transformation

In this paper we present an evidence-gathering approach to slove the multi-sensor data fusion problem. It uses an improved Hough transformation method rather than the usual statistical or geometric approach to extract the directions and positions of the walls in a room and update the location (orientation and position)of a mobile robot. The simulation results show that the proposed method is of practical importance since it is very simple and easy to implement.

0.2~2.0GHz100W超宽带GaN功率放大器

设计并实现了一款基于0.25μm GaN高电子迁移率晶体管(HEMT)工艺的100 W超宽带功率放大器。基于SiC无源工艺设计了集成无源器件(IPD)输入预匹配电路芯片;设计了基于陶瓷基片的T型集成输出预匹配电路;基于建立的传输线变压器(TLT)的精确模型,设计了宽带阻抗变换器,在超宽频带内将50Ω的端口阻抗变换至约12.5Ω,再通过多节微带电路与预匹配后的GaN HEMT芯片实现阻抗匹配。最终,以较小的电路尺寸实现了功率放大器的超宽带性能指标。测试结果表明,功率放大器在0.2~2.0 GHz频带内,在漏极电压36 V、输入功率9 W、连续波的工作条件下,输出功率大于103 W,漏极效率大于50%,输入电压驻波比(VSWR)≤2.5。

我国人口流动变化的新趋势及其对乡村转型的影响

乡村人口流动是我国现代化进程中乡村转型和社会变迁最具解释力的现象,因为它不仅是劳动力资源的流动,而且是城乡各种体制变革的重要推动力量。改革开放以来,我国乡村人口大规模向城市流动,并呈持续增长态势,但进入2010年以后,随着我国农业税费改革、农业补贴政策、农村土地承包流转等一系列促农政策的出台和国家关于乡村发展战略的优化,乡村人口流动的相关指标值也在发生着阶段性的变化。为了梳理这种变化趋势及特征,以国家人口普查数据、统计年鉴的人口数据和课题组实证调查的数据为依据进行统计描述和分析。结果表明,随着乡村人口流动呈现出阶段性变化新趋势,其带给乡村的影响已经从原来的乡村空心化、老龄化等显性问题转移到乡村转型中的结构失衡问题上来,由此也带来了对城乡一体化背景下乡村角色和功能重新定位问题的思考,在此基础上提出了相应的对策政策,以促进乡村人口流动在乡村转型发展中发挥积极作用。

卡夫卡文学世界“儿子们”群像塑造

该文聚焦于卡夫卡文学世界“儿子们”的群像塑造。通过童年经验分析卡夫卡心理长期处于恐惧状态的深层原因,并将其独特的“父子冲突”书写与奥匈帝国现代化进程中的不均衡转型相联系,提出卡夫卡笔下“自我矮化”的儿子群像正是他所置身其中的时代语境、族群境遇与家庭环境的文学呈现。作为表现主义德语文学运动边缘的作家,卡夫卡塑造的儿子群像具有特殊性,不是“儿子罢黜父亲”,而是“父亲罢黜儿子”。“儿子们”在水平社会流动与垂直社会流动层面取得的成就在父权之下毫无力量。这样的文学呈现也超越了个体体验上升为特定时代的群体体验。

平邑甜茶发根农杆菌介导转化体系构建与优化

【目的】建立并优化平邑甜茶(Malus hupehensis Rehd.)幼苗简单、高效、快速的发根农杆菌转化体系。【方法】以不同苗龄的平邑甜茶幼苗为材料,使用携带GFP及GUS过表达质粒的发根农杆菌K599,通过注射法、活化菌液浸染及菌株涂抹方法侵染平邑甜茶根颈部诱导毛状根,利用GFP荧光检测、GFP蛋白印迹检测、DNA检测和GUS染色方法进行转基因株系验证,同时对成功转化株系根、茎、叶组织的表达量进行分析,检测发根农杆菌在转化平邑甜茶株系中的迁移性。【结果】不同苗龄平邑甜茶幼苗诱导毛状根能力具有较大差异,其中三叶龄幼苗诱导毛状根率最高,为96%,五叶龄幼苗最高的毛状根诱导率为65%,以使用菌株涂抹方式转化植株的毛状根分布为最佳、毛状根诱导率为最高。发根农杆菌在成功转化株系中存在随机向地上部迁移的现象,并能够整合目的基因至叶片叶柄及部分主叶脉基因组中,但在培养30 d时无法通过蛋白印迹方式在蛋白水平上检测到蛋白信号。【结论】建立并优化了简单、高效的发根农杆菌介导的平邑甜茶转化体系,鉴定了发根农杆菌在转化株系中随机向地上部迁移规律,为发根农杆菌技术的进一步利用提供理论依据。

数字基础设施建设对共同富裕的影响研究

我国正值数字化转型和推进共同富裕的交叠期,数字基础设施建设成为缓解“数字鸿沟”、促进共同富裕的新契机。本文在阐释数字基础设施建设和共同富裕二者理论关系的基础上,以2011~2020年290个地级及以上城市为样本进行了实证检验。研究发现,数字基础设施建设对共同富裕存在积极影响。进一步分析显示,移动通信基站、互联网宽带基础设施、5G基站、超大型和大型数据中心、智能计算中心对共同富裕的正向作用程度依次递减。机制分析表明,数字基础设施建设通过提升资本配置效率、促进劳动力流动助推共同富裕,其影响效应在非中心城市和城市群内部城市更明显。研究为新一轮数字基础设施建设优化布局,促进地区间要素流动,助力全体人民共同富裕具有借鉴意义。

Analysis of the constraint relation between ground and self-adaptive mobile mechanism of a transformable wheel-track robot

To maneuver in unstructured terrains where the ground might be soft, hard, flat or rough, a transformable wheel-track robot (NEZA-I) with a self-adaptive mobile mechanism is proposed and developed. The robot consists of a control system unit, two symmetric transformable wheel-track (TWT) units, and a rear-wheel unit. The TWT unit is the main mobile mechanism for the NEZA-I robot, with the rear-wheel unit acting as an assistant mechanism. Driven only by one servomotor, each TWT unit can efficiently select between track mode and wheel mode for optimal locomotion, autonomously switching locomotion mode and track configuration with changes in the terrain. In this paper, the mechanism structure, the self-adaptive drive system, the locomotion mode and posture of the NEZA-I robot are presented, the kinematic relation of the inside parts of the TWT unit is analysed, and the mathematic model of the constraint relation between the mobile mechanism and the ground, abbreviated to "MGCR model" is set up for the NEZA-I robot to go through some typical unstructured environments. The mechanism parameters, which influence the self-adaptability of the NEZA-I robot, are found and optimized. Basic experiments show that the mobile mechanism has the self-adaptability to navigate in unstructured terrains and has superior obstacle-negotiating performance, and that the MGCR model and the analysis method of mechanism parameters are reasonable. From a mechanism point of view, it can provide an idea for research on the adaptive control of the robot.

一种轮腿式越障机器人的研究

针对移动机器人在复杂多变地形环境下实现高机动性、强越障等需求,提出了一种被动变形式的轮腿式越障机器人设计方案。该机器人的变形轮转换过程是由外力操作得到的,因此,不需要任何驱动器,减少了机构的复杂性。在完成机器人整体三维建模的基础上,对变形轮的结构、原理及受力情况进行了分析;以变形过程中的触发转矩和展开前后半径之比为指标进行结构优化;分析机器人变形阶段受力情况,并对机器人平台的相关参数进行调整以实现稳定越障;使用Adams软件对机器人变形、越障过程进行运动学仿真,并制作物理样机对整机结构设计的合理性进行了实验验证。

基于多尺度注意力机制的实时激光雷达点云语义的分割

为既能提高分割精度,又能克服车载计算资源局限,提出一种面向移动机器人平台的车载实时点云语义分割方法,并进行了综合实验。该方法采用基于投影的激光雷达语义分割方法,将三维点云投影到球面图像,并结合二维卷积进行分割。引入多头注意力机制(MHSA),实现轻量级语义分割模型,以一种全新的方式,将一种深度学习模型架构Transformer映射到卷积。将Transformer的MHSA迁移至卷积,以形成多尺度自注意力机制(MSSA)。结果表明:与当前主流方法CENet、FIDNet、PolarNet相比,本方法在NVIDIA JETSON AGX Xavier计算平台上保持了较高的分割精度(平均交并比为63.9%)及较高的检测速率(41帧/s),从而证明了其对移动机器人平台的适用性。

基于移动平台的便携式FTIR灾害现场气体检测技术

傅里叶红外光谱技术(FTIR)的引入极大地提升了气体的检测速度和准确性。然而在石油化工泄露和爆炸、森林火灾等突发危险现场,存在着人力无法现场检测的情况。为了应对这种场景,开发了基于移动平台的便携式FTIR灾害现场气体检测方法。通过将云服务器和树莓派的优势相结合实现远程操作移动平台,将搭载便携式FTIR光谱仪的移动平台远程操作驶入现场,并及时通过4G/5G模块传回现场位置数据、视频数据和光谱气体浓度数据。最后,在野外基地利用该平台进行了实测,对CO_(2)、CO、SO_(2)、C_(6)H_(6)等多种污染气体进行了检测,结果验证了该系统在野外工况下的可靠性。本工作为FTIR光谱仪或其他光谱类检测系统在移动平台上的集成应用提供了思路和解决方案。