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 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.

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。

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.

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。

地基云雷达与FY-4A卫星云顶高度联合反演方法

基于地基毫米波云雷达与FY-4A卫星对云顶同步观测的特点,分析云雷达垂直顶空观测的云顶高度与FY-4A卫星AGRI载荷通道数据之间关系,提出了地基云雷达与FY-4A卫星云顶高度联合反演方法,实现云雷达安装点周边区域卫星云顶高度的融合反演,并对反演结果进行验证分析。结果表明,AGRI载荷的第11~14通道值与云雷达观测云顶高度呈线性相关,且卫星通道值与云雷达观测云顶高度比值呈现冬季最小、春秋季次之、夏季最大的季节性变化特点;星地融合反演云顶高度与云雷达观测云顶高度相关系数0.84,融合后比融合前均方根误差减小了0.7 km,提高了卫星云顶高度的反演精度。

基于IMM-JPDA-ISTUKF的车载毫米波雷达多目标跟踪算法

为提高车载毫米波雷达多目标跟踪精度指标,提升道路车辆行驶安全性,文中在交互多模型无迹卡尔曼滤波(IMM-UKF)和联合概率数据关联(JPDA)融合的算法基础上,针对车辆运动状态突变处UKF鲁棒性差、滤波精度低的问题,提出了一种基于改进强跟踪UKF(ISTUKF)的IMM-JPDA-ISTUKF算法。通过模拟道路场景搭建的仿真环境对算法性能进行了验证,且为证明该算法在实际道路工况下跟踪精度的提升,还进行了雷达道路测试,通过雷达在道路上获取的车辆数据进一步验证了该算法的有效性。结果表明,该算法在目标车辆运动状态发生变化时的距离跟踪精度和速度跟踪精度方面均得到了提高。

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.

基于数据增强算法和CNN-LSTM的高精确度手势识别

近年来,基于雷达的手势识别技术在工业和生活中得到了广泛应用,但愈加复杂的应用场景对手势识别算法的准确率和鲁棒性提出了更高要求。对此,设计了一种基于毫米波雷达的高精确度手势识别算法。通过对已有分类算法的研究对比,构建了一种用于手势识别的卷积神经网络-长短期记忆网络(CNN-LSTM)深度学习算法模型;同时,运用布莱克曼窗抑制手势信号处理中的频谱泄露问题,并联合运用小波阈值和动态补零算法实现高效杂波抑制和数据增强。实测结果表明,设计的手势识别算法正确分类率达到97.29%,在不同的距离和角度情况下也可以保持较好的识别准确率,具有良好的鲁棒性。

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.

基于改进D-S证据理论的毫米波雷达与摄像头融合目标检测和识别

随着汽车保有量的增加和交通环境的日益复杂,汽车行驶安全性和交通事故发生率成为交通安全领域内重点关注的问题,通过智能辅助驾驶技术可以有效提升驾驶安全和降低交通事故发生率。为了提高智能驾驶辅助技术中环境感知的可靠性,提出了一种基于改进D-S证据理论的毫米波雷达与摄像头融合目标检测和识别方法。采用皮尔逊相关性系数改进了D-S证据理论,分析了雷达目标类别与雷达探测信息多参数耦合关系,建立了一种毫米波雷达稀疏点云的目标识别方法。在此基础上,对毫米波雷达信息和视觉检测信息进行加权,提升融合算法在不同工况下目标检测的准确性。实验结果表明,融合算法在不同工况下检测准确率达到93.29%,误检率和漏检率得到降低。相较于其他融合检测算法,本算法在视觉传感器发生漏检时,利用雷达检测算法仍可以对视觉漏检目标进行识别并给出相应的置信度。本工作为毫米波雷达与摄像头的数据融合提供了一种新方法。

基于RD-YOLO的毫米波雷达和视觉融合显著性目标检测

为了解决单传感器在复杂环境下目标检测精度低问题,提出了一种基于RD-YOLO的毫米波雷达和视觉融合的显著性目标检测方法。首先设计了能够将毫米波雷达点云转换为图像的方法,使毫米波雷达和视觉数据在模型输入时实现特征融合;然后通过动态互补注意力机制,对两个图像分支生成特征设置空间和通道动态注意力权重;最后采用YOLOv8检测融合后特征,引入改进损失函数Focal Loss以解决样本不均衡问题。在数据集nuScenes上开展的相关实验表明,与YOLOv5、YOLOv7、YOLOv8、Faster R-CNN和FCOS相比,所提方法目标检测综合性能良好,均值平均精度比原始YOLOv8提升了9.19%。

Compensation for the Influence of Radar-Target Relative Speed on Step Frequency MMW Target Identification

Aim To study the influence of radar-target relative speed on frequency MMW high-resolution ore-dimension distance profile and the compensation for it. Methods Based on the distance travelled by the electromagnetic wave, analyses were made for the compensation algorithm and the expression of the inverse FFT base distance was given.The relative importance of different compensation terms was studied in great detail. The concept of searching compensation was put forward. Results and Condclusion Dcm-△Dvimis the be distance of inverse FFT transformation, the effect caused by the distance △Dim on one-dimension profile is negligible, and the effect caused by the distance Dvim should not be neglected and must be compensated.

MMRGait-1.0:多视角多穿着条件下的雷达时频谱图步态识别数据集

步态识别作为一种生物识别技术,在实际生活中通常被认为是一项检索任务。然而,受限于现有雷达步态识别数据集的规模,目前的研究主要针对分类任务且局限于单一行走视角和相同穿着条件,这限制了基于雷达的步态识别在实际场景中的应用。该文公开了一个多视角多穿着条件下的雷达步态识别数据集,该数据集使用毫米波雷达采集了121位受试者在多种穿着条件下沿不同视角行走的时频谱图数据,每位受试者共采集8个视角,每个视角采集10组,其中6组为正常穿着,2组为穿大衣,2组为挎包。同时,该文提出一种基于检索任务的雷达步态识别方法,并在公布数据集上进行了实验,实验结果可以作为基准性能指标,方便更多学者在此数据集上开展进一步研究。

基于RD-ANM的毫米波雷达动目标超分辨DOA估计方法

超分辨波达方位角估计是车载毫米波雷达实现目标精准定位及跟踪需要解决的关键问题。针对车载场景中常见的阵列孔径受限、少快拍、低信噪比以及信源相干的情况,该文提出了一种基于距离多普勒域原子范数最小化(RD-ANM)的车载毫米波雷达动目标超分辨DOA估计方法:首先,构建了基于动目标雷达回波的距离多普勒域阵列接收信号;其次,设计了动目标多普勒耦合相位补偿矢量,用以削弱目标运动对DOA估计的影响;最后,提出了基于原子范数框架的多目标超分辨DOA估计方法。相较于车载毫米波雷达现使用的DOA估计算法,该文算法能够在基于低信噪比条件和单快拍处理前提下获得较高的测角分辨率和估计精度,以及拥有不牺牲阵列孔径对相干信号进行处理的稳健性能。理论分析、数值仿真以及实测实验验证了该文算法的有效性。

基于Transformer-LSTM模型的毫米波雷达行人轨迹预测方法

在智能交通领域,对行人轨迹进行预测是一个非常有意义的研究课题。毫米波雷达凭借其强大的抗干扰能力和高隐私保密性,在行人轨迹预测方面具有良好的应用前景。为了解决传统预测算法准确度不高、长序列预测效果不佳的问题,文中提出一种基于Transformer-LSTM毫米波雷达的行人轨迹预测方法。该方法利用毫米波雷达采集行人轨迹数据,通过DBSCAN方法合成轨迹点,建立行人轨迹数据集,进行模型训练。该模型利用Transformer的动态权重分配和长短期记忆网络(Long Short-Term Memory,LSTM)的轨迹特征提取,深度挖掘不同层次的关键信息。实验结果表明,与经典运动学模型和传统神经网络模型相比,Transformer-LSTM在行人轨迹预测任务中预测精度更高、稳定性更好。

毫米波雷达微弱行人轨迹跟踪-预测一体化方法

针对城市道路场景微弱行人目标雷达回波信号极易被强背景杂波淹没,导致目标轨迹跟踪及预测失效难题,提出了一种基于调频连续波-多输入多输出毫米波雷达的微弱行人轨迹跟踪-预测一体化方法。首先,利用递归贝叶斯检测前跟踪算法,直接从未经阈值处理的雷达三维原始频谱数据中提取目标运动轨迹,解决了传统阈值决策信息丢失带来的跟踪性能下降问题,并在此基础上提出了一种基于Transformer的端到端轨迹预测模型,进一步挖掘隐藏在跟踪轨迹中的时空相关性,完成了微弱行人目标轨迹的精准预测。实验结果表明,方法在信噪比大于-20 dB时,预测轨迹的平均位移误差和最终位移误差分别小于0.706、1.215 m,均优于高斯过程、长短期记忆网络等传统方法。

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.

一次青藏高原低涡影响下的云-降水宏微观特征研究

为进一步认识高原低涡影响下云-降水的垂直结构和微物理特征,利用那曲毫米波雷达和雨滴谱仪资料,结合ERA-5再分析和FY-2G卫星资料,对2015年8月17-19日一次高原涡云系的天气背景、雷达回波和相关物理量进行分析。结果表明,18日凌晨高原涡发展达到最强,那曲上空对流云发展旺盛;随后,强度逐渐减弱,那曲上空过渡为层积云;至19日20时,高原涡减弱消散。此次高原涡降水包含对流云降水和层积云降水,前者回波呈火焰状的上冲结构,回波强度达30 dBZ,但持续时间短,距离地面1 km处存在明显回波亮带;后者在回波图上多为片状,雷达回波强度在20 dBZ以下,持续时间较长。两类降水的雨滴谱存在显著差异,对流云降水粒子直径更大、雨滴谱更宽。

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.