Non-intrusive soil carbon content quantification methods using machine learning algorithms:A comparison of microwave and millimeter wave radar sensors

Agricultural and forestry biomass can be converted to biochar through pyrolysis gasification,making it a significant carbon source for soil.Applying biochar to soil is a carbon-negative process that helps combat climate change,sustain soil biodiversity,and regulate water cycling.However,quantifying soil carbon content conventionally is time-consuming,labor-intensive,imprecise,and expensive,making it difficult to accurately measure in-field soil carbon’s effect on storage water and nutrients.To address this challenge,this paper for the first time,reports on extensive lab tests demonstrating non-intrusive methods for sensing soil carbon and related smart biochar applications,such as differentiating between biochar types from various biomass feedstock species,monitoring soil moisture,and biochar water retention capacity using portable microwave and millimeter wave sensors,and machine learning.These methods can be scaled up by deploying the sensor in-field on a mobility platform,either ground or aerial.The paper provides details on the materials,methods,machine learning workflow,and results of our investigations.The significance of this work lays the foundation for assessing carbon-negative technology applications,such as soil carbon content accounting.We validated our quantification method using supervised machine learning algorithms by collecting real soil mixed with known biochar contents in the field.The results show that the millimeter wave sensor achieves high sensing accuracy(up to 100%)with proper classifiers selected and outperforms the microwave sensor by approximately 10%–15%accuracy in sensing soil carbon content.

A Survey of Gesture Recognition Using Frequency Modulated Continuous Wave Radar

With technology advances and human requirements increasing, human-computer interaction plays an important role in our daily lives. Among these interactions, gesture-based recognition offers a natural and intuitive user experience that does not require physical contact and is becoming increasingly prevalent across various fields. Gesture recognition systems based on Frequency Modulated Continuous Wave (FMCW) millimeter-wave radar are receiving widespread attention due to their ability to operate without wearable sensors, their robustness to environmental factors, and the excellent penetrative ability of radar signals. This paper first reviews the current main gesture recognition applications. Subsequently, we introduce the system of gesture recognition based on FMCW radar and provide a general framework for gesture recognition, including gesture data acquisition, data preprocessing, and classification methods. We then discuss typical applications of gesture recognition systems and summarize the performance of these systems in terms of experimental environment, signal acquisition, signal processing, and classification methods. Specifically, we focus our study on four typical gesture recognition systems, including air-writing recognition, gesture command recognition, sign language recognition, and text input recognition. Finally, this paper addresses the challenges and unresolved problems in FMCW radar-based gesture recognition and provides insights into potential future research directions.

A Method for Improving Ranging Accuracy of Short-range Millimeter Wave Radar

The range-velocity ambiguity caused by moving target influences on the ranging accuracy of a short-range millimeter wave radar greatly.A new method was presented in this paper to reduce the range-velocity ambiguity and improve the ranging accuracy by estimating parameters of the echo signal with fractional Fourier transform and self-correlation.And,a new quick searching algorithm was given also to increase the calculation speed.Compared to the Chinese remainder theorem method,the proposed method is excellent for its simplicity and reducing the computation complexity.The simulation results show its validity.

Backscattering by sea of centimeter and millimeter waves at small grazing angle

Using experimental data reflected by the sea on specific radar cross-section （SRCS） at millimeter and centimeter waves, the approximations of the wind speed, angle of the sea surface radiation and polarization of the incident field can be calculated. The simulation model of the scattered signal has been proposed on the basis of the semi-Markov nested processes. For the first time it has been proved that for the description of reflections at spikes and pauses, it is possible to use finite atomic functions. The proposed model allows us to estimate the baekscatter intensity of millimeter and centimeter radio waves by the sea at grazing angle of surface radiation, as well as to simulate scattered signal.

BEV-radar:毫米波雷达-相机双向融合的三维目标检测

在自动驾驶场景下的3D目标检测任务中,探索毫米波雷达数据作为RGB图像输入的补充正成为多模态融合的新兴趋势。然而,现有的毫米波雷达-相机融合方法高度依赖于相机的一阶段检测结果,导致整体性能不够理想。本文提供了一种不依赖于相机检测结果的鸟瞰图下双向融合方法(BEV-radar)。对于来自不同域的两个模态的特征,BEV-radar设计了一个双向的基于注意力的融合策略。具体地,以基于BEV的3D目标检测方法为基础,我们的方法使用双向转换器嵌入来自两种模态的信息,并根据后续的卷积块强制执行局部空间关系。嵌入特征后,BEV特征在3D对象预测头中解码。我们在nu Scenes数据集上评估了我们的方法,实现了48.2 m AP和57.6 NDS。结果显示,与仅使用相机的基础模型相比,不仅在精度上有所提升,特别地,速度预测误差项有了相当大的改进。代码开源于https://github.com/Etah0409/BEV-Radar。

Vital Sign Signal Extraction Based on mmWave Radar

This research develops an algorithm for health monitoring and uses the Infineon BGT60TR13C shield 60GHz radar to build a set of software and hardware demonstration platforms for algorithm verification. The algorithm can monitor the position, breathing, heartbeat and other information of the elderly in real time, which is used to judge the health status of the elderly and improve the convenience for early warning and rescue of abnormal situations in time. However, limited by the transmission power of this radar and the number of transmitting and receiving antennas, the current detection range of the algorithm is relatively small, and the ability to distinguish between multiple persons is limited. This problem can be optimized by replacing other types of radars in the future.

Convolution neural network and 77 GHz millimeter wave radar based intelligent liquid classification system

An intelligent liquid classification system based on 77 GHz millimeter wave radar and convolution neural network are proposed in this paper.The data are collected by the AWR1843 radar platform and processed by the neural network on the host PC in real-time.The doppler heatmap generated by radar signal processing is tried for the first time as the input of the system.The information carried by the heatmap in 2 dimensions is analyzed and the reason why the doppler heatmap could be used for classification is explained.The feasible experiment proved that the proposed method can successfully classify 8 kinds of common liquid with high accuracy.The result of the experiment is explained and the limitations of the experiment are discussed.It can be drawn that the combination of FMCW millimeter wave radar and convolution neural network is a method with great potential for liquid classification.The advantages of real time,non-invasive and unilateral measurement can also be used for the detection of dangerous liquids.

融合毫米波与激光雷达的障碍物检测与跟踪方法

在园区环境中,无人车装载单一毫米波雷达或激光雷达传感器进行障碍物检测跟踪时存在探测范围有限、准确率低及稳定性差等问题。为此,提出一种基于毫米波雷达与激光雷达融合的多障碍物检测跟踪方法。利用改进欧氏聚类算法对道路内激光点云目标进行提取,基于信息筛选策略获得毫米波雷达数据中的有效目标;基于目标检测交并比与可靠性分析,对2种目标进行自适应融合,并利用跟踪门与联合概率数据关联(JPDA)算法完成前后帧数据匹配;应用多运动模型交互与无迹卡尔曼滤波实现障碍物跟踪。实车实验表明:相比单一毫米波雷达与激光雷达障碍物检测跟踪,所提方法有更好的准确性与稳定性。

基于DenseNet和卷积注意力模块的高精度手势识别

非接触的手势识别是一种新型人机交互方式,在增强现实(AR)/虚拟现实(VR)、智能家居、智能医疗等方面有着广阔的应用前景,近年来成为一个研究热点。由于需要利用毫米波雷达进行更精确的微动手势识别,该文提出一种新型的基于MIMO毫米波雷达的微动手势识别方法。采用4片AWR1243雷达板级联而成的毫米波级联(MMWCAS)雷达采集手势回波,对手势回波进行时频分析,基于距离-多普勒(RD)图和3D点云检测出人手目标。通过数据预处理,提取手势目标的距离-时间谱图(RTM)、多普勒-时间谱图(DTM)、方位角-时间谱图(ATM)和俯仰角-时间谱图(ETM),更加全面地表征手势的运动特征,并形成混合特征谱图(FTM),对12种微动手势进行识别。设计了基于DenseNet和卷积注意力模块的手势识别网络,将混合特征谱图作为网络的输入,创新性地融合了卷积注意力模块(CBAM),实验表明,识别准确率达到99.03%,且该网络将注意力放在手势动作的前半段,实现了高精度的手势识别。

Human Activity Recognition Based on Frequency-Modulated Continuous Wave and DenseNet

With the development of wireless technology, Frequency-Modulated Continuous Wave (FMCW) radar has increased sensing capability and can be used to recognize human activity. These applications have gained wide-spread attention and become a hot research area. FMCW signals reflected by target activity can be collected, and human activity can be recognized based on the measurements. This paper focused on human activity recognition based on FMCW and DenseNet. We collected point clouds from FMCW and analyzed them to recognize human activity because different activities could lead to unique point cloud features. We built and trained the neural network to implement human activities using a FMCW signal. Firstly, this paper presented recent reviews about human activity recognition using wireless signals. Then, it introduced the basic concepts of FMCW radar and described the fundamental principles of the system using FMCW radar. We also provided the system framework, experiment scenario, and DenseNet neural network structure. Finally, we presented the experimental results and analyzed the accuracy of different neural network models. The system achieved recognition accuracy of 100 percent for five activities using the DenseNet. We concluded the paper by discussing the current issues and future research directions.

基于毫米波雷达的生命体征信号检测实验教学设计

为了提高学生嵌入式系统的软硬件设计能力,设计基于毫米波雷达的非接触式生命体征信号测量实验。一方面,基于调频连续波(FMCW)雷达机制,采用MATLAB进行生命体征信号建模及呼吸率与心率时频信号提取的仿真实验;另一方面,采用60 GHz毫米波雷达模块,基于STM32单片机进行系统电路设计、PCB设计与实物制作,完成参试者呼吸率及心率的数据采集与OLED显示。上位机采用Qt Creator进行软件界面与功能设计,通过蓝牙模块与单片机进行无线数据通信,以30 s为周期执行呼吸率和心率数据的读取与显示。最后,搭建测试环境,对系统功能进行验证及其设计指标的测试与评估。该实验项目与生活实际紧密相关,注重理论与实践相结合,在系统整体设计与跨平台设计中培养学生的综合实践能力与协作能力。

高精度毫米波柱面孔径全息成像算法研究

近年来,国际恐怖袭击事件时有发生,使得安全形势变得极为严峻,因此迫切需要引入新型的安全检查手段,防范和应对重大安全事故的发生。在此背景下,毫米波全息三维成像凭借其安全性、穿透性和高分辨率成像等优势,被广泛地应用在安检领域。由于人体安检需要在短时间内对目标进行精确的检测,具有实时性的特点,即需要在实时情况下快速做出判断和决策,因此对成像精度有着更高的要求。本文提出了一种高精度毫米波柱面孔径全息成像算法,给出了柱面孔径三维成像所需要满足的采样准则;首先,将成像区域划分为序列不同半径的成像柱面,依据成像精度和相位误差控制范围设置成像参考柱面,简化重复计算匹配滤波函数的过程;然后,将回波信号变换至三维波数域中,并与匹配滤波函数相乘;接着,沿距离向对匹配滤波的结果进行积分,聚焦获得相应半径柱面上目标的散射信息;最后,对不同半径柱面上的目标进行聚焦,获取目标的全部散射信息,进而完成对整个观测区域的三维图像重建。本文给出了参考面最佳分配策略和算法成像复杂度分析,并通过点目标仿真详细地分析了相位误差对成像质量的影响,并与现有的方法进行了比较,以及在同样的系统参数下对人体模特进行成像实验,仿真和实验的结果表明本文方法的有效性、精确性和正确性。

基于多模型神经网络的湿度廓线反演研究

为提升微波辐射计对大气廓线探测的精度,利用ARM大气观测站提供的地基微波辐射计、毫米波测云雷达以及探空数据,构建了两种添加不同云信息的反向传播神经网络(back propagation neural network,BPNN)模型(添加入云和出云高度的C-BPNN模型与添加雷达反射率因子的Z-BPNN模型)与一种未添加云信息的BPNN模型(记为BPNN0),并对反演结果进行了对比,结果表明:C-BPNN模型和Z-BPNN模型在任何天气下(有云或无云),得到的反演误差都小于BPNN0模型;C-BPNN相较于另外两种模型反演结果具有更高的稳定性。对3种模型各自反演结果最好的个例分析发现,C-BPNN与Z-BPNN模型主要的误差存在于高空无云但是相对湿度却出现跃变的情况,说明神经网络模型对初始权值与阈值较为敏感,因此通过遗传算法(genetic algorithms,GA)对BPNN模型进行优化。经GA优化后的反演结果表明:BPNN0模型与C-BPNN模型具有明显优化效果,而Z-BPNN模型优化效果则不明显。

AF-CenterNet:基于交叉注意力机制的毫米波雷达和相机融合的目标检测

对于自动驾驶领域而言,确保在各种天气和光照条件下精确检测其他车辆目标是至关重要的。针对单个传感器获取信息的局限性,提出一种基于cross-attention注意力机制的融合方法(AF),用于在特征层面上融合毫米波雷达和相机信息。首先,将毫米波雷达和相机进行空间对齐,并将对齐后的点云信息投影成点云图像。然后,将点云图像在高度和宽度方向上进行扩展,以提高相机图像和点云图像之间的匹配度。最后,将点云图像和相机图像送入包含AF结构的CenterNet目标检测网络中进行训练,并生成一个空间注意力权重,以增强相机中的关键特征。实验结果表明,AF结构可以提高原网络检测各种大小目标的性能,特别是对小目标的检测提升更为明显,且对系统的实时性影响不大,是提高车辆在多种场景下检测精度的理想选择。

光斑密度峰值的毫米波雷达目标检测

毫米波雷达目标检测任务是车辆环境感知的重要组成部分,对恶劣天气条件下的智能驾驶具有重要意义.虽然现有雷达目标检测方法已取得不错的研究成果,但雷达数据仍存在手工设计的特征信息量不足、特征提取不充分以及时序特征未充分利用的问题.为解决这些问题,本文提出光斑密度峰值与神经网络相融合的两阶段目标检测方法.第一个阶段光斑密度峰值聚类算法,对射频图像中的目标进行粗略估计,并生成聚类簇对应的目标候选.将聚类生成的候选目标特征融合到原始射频图像.第二阶段基于通道融合的3D自编码器目标检测网络进一步提取目标多普勒速度和时序特征并分类.实验表明,所提出的两阶段方法与基准实验RODNet(CDC)相比,平均精度指标提升4.3%,平均召回率提升2.3%.

基于毫米波深度传感的抗伴生干扰的眨眼检测

眨眼检测在多种实际应用场景中起着关键作用,如眼病检测、人机交互、疲劳驾驶预防等。针对来自人体的伴生干扰会严重影响眨眼信号的特征提取问题,本文提出一种非线性独立分量分析框架的自监督深度对比学习方法来分离眨眼和伴生干扰。本文设计一个基于时间相关性的分离网络ES-Net1,该网络将具有时间相关和时间不相关的两个正负样本序列作为网络的输入,通过ES-Net1内部的特征提取器恢复眨眼和伴生干扰信号的时间结构,从而实现非线性混合信号的分离。本文基于TI公司的AWR1642毫米波雷达平台实现mmBlinkSEN原型系统,通过14 000组数据验证mmBlinkSEN的有效性。实验结果表明,在存在人体伴生干扰情况下,mmBlinkSEN对眨眼频率的检测精度高达88%。

基于毫米波雷达的非视距目标探测和定位方法研究

隐蔽非直视目标探测和定位技术在城市巷战、安防、防暴等领域具有广泛的应用。本文针对L形非视距区域内单目标的探测与定位问题进行研究。首先,对雷达布设位置及可探测区域进行了分析;其次,针对共同可探测区域,设计了多路径联合检测器;最后,基于设计的多径雷达应用系统,利用所提检测定位方法进行实验验证。与现有研究相比,本文的创新点在于:在L形非视距场景下,对宽波束毫米波雷达布设位置及可探测区域进行分析并给出合理建议;其次利用所提多路径联合检测器将不同反射面的多条路径联合使用,提高了检测性能。本文通过实验验证了系统的可靠性及雷达位置对可探测区域的影响,通过仿真与实验探究了所提检测器的性能改善。

机器学习的高精度毫米波雷达测距信号误差补偿方法

毫米波雷达是一种常用的非接触式测距技术,受环境因素以及测量过程中存在的各种误差影响,测距结果可能存在一定的误差。研究误差补偿方法可以有效提高毫米波雷达的测距精度,从而更准确地获取目标物体的距离信息。为此,提出了机器学习的高精度毫米波雷达测距信号误差补偿方法。通过高斯滤波器去除雷达测距信号中的噪声,完成信号的去噪处理。利用模拟插入脉冲计数法和四象限光斑定位法,测量目标物体的距离和角度信息,通过自适应惯性权重与收敛因子优化粒子群算法,并利用优化后的粒子群算法改进BP神经网络,将测量的距离和角度信息输入到改进的BP神经网络中展开训练,即可得到补偿后的雷达测距信号。实验结果表明,该方法的信号处理效果好,补偿后的毫米波雷达测距信号方位角和俯仰角误差接近于0,且信号平滑度较高。

基于扩展卡尔曼滤波的交互式多模型跟踪算法研究

在辅助驾驶系统中,行人轨迹跟踪一直是一项有挑战性的任务,因为行人的回波信号中往往存在着许多干扰噪声。此外,行人在运动过程中可能会做出突然转身或其他改变方向的行为,这将直接导致行人运动轨迹呈现出非线性特征。针对上述问题,文中提出一种基于扩展卡尔曼滤波的交互式多模型跟踪(IMM-EKF)方法,适用于毫米波雷达对行人进行轨迹跟踪。首先,在扩展卡尔曼滤波算法(EKF)的基础上重构状态预测协方差矩阵,来补偿EKF非线性化过程中引入的误差;然后将改进的EKF作为交互式多模型算法(IMM)中的滤波器,根据行人运动特性选择匀速模型和协调转弯模型作为跟踪模型,利用所提出的IMM-EKF算法进行轨迹跟踪。实验结果表明,所提出的滤波算法较典型的EKF和改进的EKF算法,在跟踪滤波精度方面均有所提升,同时具备更优的跟踪鲁棒性。

基于调频连续波雷达的人体生命体征检测算法

针对现有雷达非接触生命体征检测精度低、实时性差等问题,提出一种基于调频连续波(FMCW)雷达的人体生命体征检测算法。首先,通过毫米波雷达获取生命体征信号;其次,利用改进的经验小波变换(EWT)算法,实现生命体征信号的自适应分解和重构,通过引入麻雀搜索算法(SSA)和模糊熵(FE)寻找频谱分割线的最优值;最后,通过改进频率插值的估计算法计算心率和呼吸频率。通过与医用重症监护仪进行对比实验验证所提算法的优越性和鲁棒性。实验结果表明,所提算法相较于小波变换(WT)算法、CEEMD(Complementary Ensemble Empirical Mode Decomposition)算法和VMD(Variational Mode Decomposition)算法,均方误差(MSE)分别减小了77.65、27.25和21.05,平均绝对百分比(MAPE)分别减小了7.33、4.33和3.42个百分点,实时性分别提高了0.72 s, 16.74 s和1.87 s。同时,利用所提算法也实现了对心率变异性(HRV)的检测。