Improved Weighted Local Contrast Method for Infrared Small Target Detection

In order to address the problem of high false alarm rate and low probabilities of infrared small target detection in complex low-altitude background,an infrared small target detection method based on improved weighted local contrast is proposed in this paper.First,the ratio information between the target and local background is utilized as an enhancement factor.The local contrast is calculated by incorporating the heterogeneity between the target and local background.Then,a local product weighted method is designed based on the spatial dissimilarity between target and background to further enhance target while suppressing background.Finally,the location of target is obtained by adaptive threshold segmentation.As experimental results demonstrate,the method shows superior performance in several evaluation metrics compared with six existing algorithms on different datasets containing targets such as unmanned aerial vehicles(UAV).

Winter Wheat Crop Height Estimation Using Small Unmanned Aerial System (sUAS)

Deploying the small Unmanned Aerial System (sUAS) for data collection of high-resolution images is a big potential in determining crop physiological parameters. The advantage of using sUAS technology is the ability to acquire a high-resolution orthophoto and a 3D Model which is highly suitable for plant height monitoring. Plant height estimation has a big impact in the growth and development of wheat because it is essential for obtaining biomass, which is a factor for higher crop yield. Plant height is an indicator of high yield estimation and it correlates to biomass, nitrogen content, and other plant growth parameters. The study is aimed to determine an accurate height of wheat using the sUAS generated Digital Surface Model (DSM). A high-resolution imagery between 1.0 - 1.2 cm/pixel was obtained from a 35 m altitude with area coverage of 1.01 hectares. The DSM and orthophoto were generated from the sUAS, and the computed wheat heights were derived from the difference of Digital Elevation Model (DEM) and DSM data. Field measurement using steel tape was done for ground truth. The sUAS-based wheat height data were evaluated using the ground truth of 66 wheat-rows by applying correlation and linear regression analysis. Datasets were collected from three different flight campaigns (March 2018-May 2018). The sUAS-based wheat height data were significantly correlated, obtaining the result of R2 = 0.988, R2 = 0.996 and R2 = 0.944 for the month of March, April and May 2018 respectively. The significance of linear regression results was also validated by computing for the p-value. The p-value results were 0.00064, 0.0000824 and 0.0058 respectively. The main concern is the lodging of winter wheat, especially during the month of April which affects the recording of the plant’s height. Because some of the wheat plants are now lying on the ground, so measurements are done vertically. Nonetheless, the results showed that sUAS technology is highly suitable for many agricultural applications.

Online Multi-Object Tracking Under Moving Unmanned Aerial Vehicle Platform Based on Object Detection and Feature Extraction Network

In order to solve the problem of small object size and low detection accuracy under the unmanned aerial vehicle(UAV)platform,the object detection algorithm based on deep aggregation network and high-resolution fusion module is studied.Furthermore,a joint network of object detection and feature extraction is studied to construct a real-time multi-object tracking algorithm.For the problem of object association failure caused by UAV movement,image registration is applied to multi-object tracking and a camera motion discrimination model is proposed to improve the speed of the multi-object tracking algorithm.The simulation results show that the algorithm proposed in this study can improve the accuracy of multi-object tracking under the UAV platform,and effectively solve the problem of association failure caused by UAV movement.

Small objects detection in UAV aerial images based on improved Faster R-CNN

In order to solve the problem of small objects detection in unmanned aerial vehicle(UAV)aerial images with complex background,a general detection method for multi-scale small objects based on Faster region-based convolutional neural network(Faster R-CNN)is proposed.The bird’s nest on the high-voltage tower is taken as the research object.Firstly,we use the improved convolutional neural network ResNet101 to extract object features,and then use multi-scale sliding windows to obtain the object region proposals on the convolution feature maps with different resolutions.Finally,a deconvolution operation is added to further enhance the selected feature map with higher resolution,and then it taken as a feature mapping layer of the region proposals passing to the object detection sub-network.The detection results of the bird’s nest in UAV aerial images show that the proposed method can precisely detect small objects in aerial images.

Design and implementation of unmanned aerial vehicle localization and light tracking system

Unmanned aerial vehicles(UAV)are applied widely and profoundly in various fields.Moreover,high-precision positioning and tracking in multiple scenarios are the core requirements for UAV usage.To ensure stable communication of UAVs in denial environments with substantial electromagnetic interference,a systematic solution is proposed based on a deep learning algorithm for target detection and visible light for UAV tracking.Considering the cost and computational power limitations on the hardware,the you only look once(YOLO)v4-Tiny model is used for static target detection of the UAV model.For UAV tracking,and a light tracker that can adjust the angle of emitted light and focus it on the target is used for dynamic tracking processing.Thus,achieving the primary conditions of UAV optical communication with good secrecy is also suitable for dynamic situations.The UAV tracker positions the UAV model by returning the coordinates and calculating the time delay,and then controls the spotlight to target the UAV.In order to facilitate the deployment of deep learning models on hardware devices,the lighter and more efficient model is selected after comparison.The trained model can achieve 99.25%accuracy on the test set.The dynamic target detection can reach 20 frames per second(FPS)on a computer with an MX520 graphics processing unit(GPU)and 6 GB of random access memory(RAM).Dynamic target detection on a Jetson Nano can reach 5.4 FPS.

MFE-YOLOX:无人机航拍下密集小目标检测算法

针对无人机航拍时物体尺度变化大,检测目标大多较小且物体较密集的问题,提出一种混合特征增强结构(mix feature enhancement, MFE)方法。通过在超分辨率方法中加入注意力机制以增强小目标信息提取,利用一种新的特征层融合计算方法,加强不同特征层间的融合效率,提高了中小型目标的检测精度;设计了尾端感受野扩大层以扩大尾端特征层感受野,使检测头可接收丰富的物体信息来定位并区分密集物体。实验在数据集VisDrone2021的测试集上进行测试,MFE-YOLOX网络的AP50结果为47.78%,在参数量、计算量与原网络相近的情况下精度提高了9.43个百分点。

基于上下文信息与特征细化的无人机小目标检测算法

无人机航拍图像中的目标检测是近年来研究的热点,针对无人机视角下目标小而密集、背景复杂导致检测精度低的问题,提出一种基于上下文信息与特征细化的无人机小目标检测算法。通过上下文特征增强模块,利用多尺度扩张卷积捕获与周围区域像素点的潜在关系,为网络补充上下文信息,并根据不同尺度的特征层自适应生成各层级特征图的输出权重,动态优化特征图表达能力;由于不同特征图细粒度不同,使用特征细化模块来抑制特征融合中冲突信息的干扰,防止小目标特征淹没在冲突信息中;设计了一种带权重的损失函数,加快模型收敛速度,进一步提高小目标检测精度。在VisDrone2021数据集进行大量实验表明,改进后的模型较基准模型mAP50提高8.4个百分点,mAP50:95提高5.9个百分点,FPS为42,有效提高了无人机航拍图像中小目标的检测精度。

高阶深度可分离无人机图像小目标检测算法

当前无人机图像中存在小目标数量众多、背景复杂的特点,目标检测中易造成漏检误检率较高的问题,针对这些问题,提出一种高阶深度可分离无人机图像小目标检测算法.首先,结合CSPNet结构与ConvMixer网络,深度可分离卷积核,获取梯度结合信息,并引入递归门控卷积C3模块,提升模型的高阶空间交互能力,增强网络对小目标的敏感度;其次,检测头采用两个头部进行解耦,分别输出特征图分类和位置信息,加快模型收敛速度;最后,使用边框损失函数EIoU,提高检测框精准度.在VisDrone2019数据集上的实验结果表明,该模型检测精度达到了35.1%,模型漏检率和误检率有明显下降,能够有效地应用于无人机图像小目标检测任务.在DOTA 1.0数据集和HRSID数据集上进行模型泛化能力测试,实验结果表明,该模型具有良好的鲁棒性.

基于低空视觉的直升机旋翼轴心轨迹测量法

为满足直升机飞行状态旋翼轴心轨迹测量需求,提出了一种基于低空双目视觉的轴心轨迹测量方法。首先,针对高速旋转体的小量测量,基于立体视觉测量原理设计构建了以无人机为低空移动平台的旋翼轴心轨迹视觉测量系统。其次,针对动-动测量环境条件,提出了地面标定与空中动态实时标定联合的测量系统标定方法。然后,基于空间几何关系,论述了旋翼轴心静态估计与动态轨迹测量方法;最后,进行了测量系统地面精度验证试验与直升机地面开车状态的旋翼轴心轨迹测量试验。试验结果表明,该方法准确可靠,点间距离精度优于1‰,可实现轴心轨迹的非接触精确测量,还可推广应用于其他动态小量测量任务。

无线紫外光中继协作无人机集结编队方法

为了解决无人机集结编队时机间通信易受强电磁干扰问题,采用紫外信标模型完成集群无人机的信息交互,从而实现集结定位,将一致性理论与无线紫外光虚拟势场避障法相结合,在空中集结时使无人机编队的飞行姿态、航向角以及速度保持一致,以跟随无人机作为中继节点,提升相邻无人机发现概率,提高编队队形准确性;同时提出无线紫外光通信协作无人机中继集结编队方法,实现了无人机编队快速准确形成期望编队队形。结果表明,所提方法使得无人机四边形编队、五边形编队、六边形编队的通信可靠性分别提升了7.69%、9.41%和10.0%。该研究为无人机编队集结方法和无人机编队中的机间通信提供了一定的理论参考。

从飞行汽车看低空经济新业态

近年来,我国低空经济呈现蓬勃发展态势,相关政策密集出台,企业也作出了积极响应,行业发展确定性进一步增强,相关产业取得长足发展,特别是无人机领域处于世界先进水平,但我国低空经济整体水平与欧美发达国家相比仍有一定差距。通用航空、无人机、飞行汽车是低空经济的主要产业领域,特别是近年出现的飞行汽车被视为低空经济发展的新引擎。在飞行汽车研发方面,国内外均尚处于起步阶段,我国在若干局部技术上领先,但并未形成明显优势。未来,需推动飞行汽车品质和关键技术的不断突破,同时,也要加紧筹划和完善配套的法律法规、基础设施、行业标准、运行环境及监管政策以推动城市空中交通(UAM)体系的建设和低空经济的整体发展。

改进YOLOv5s的无人机视角下小目标检测算法

针对无人机飞行时与目标距离较远,被拍摄的目标大小有明显的差异且存在被物体遮挡等问题,提出一种基于YOLOv5s的无人机视角下小目标检测改进算法BD-YOLO。在特征融合网络中采用双层路由注意力(bi-level routing attention,BRA),其以动态稀疏的方式过滤特征图中最不相关的特征,保留部分重要区域特征,从而提高模型特征提取的能力;由于特征图经过多次下采样后会丢失大量位置信息和特征信息,因此采用一种结合注意力机制的动态目标检测头DyHead(dynamic head),该检测头通过尺度感知、空间感知和任务感知的三者统一,以实现更强的特征表达能力;使用Focal-EIoU损失函数,来解决YOLOv5s中CIoU Loss计算回归结果不准确的问题,从而提高模型对小型目标的检测精度。实验结果表明,在VisDrone2019-DET数据集上,BD-YOLO模型较YOLOv5s模型在平均精度(mAP@0.5)指标上提高了0.062,对比其他主流模型对于小目标的检测都有更好的效果。

城市森林结构多样性预测冠下地面温度的潜力研究

城市森林冠层具有调控城市森林微气候的能力,但现有研究尚未阐明冠层结构对冠下地面温度的影响及其预测潜力。文章基于无人机机载激光雷达(UAV-LiDAR)提取哈尔滨林业示范基地的城市森林冠层结构多样性特征指标,探究单一结构多样性特征对冠下地面温度的影响,以及结构多样性多因子组合对温度的预测潜力。结果表明:1)城市森林结构多样性的8个特征因子与冠下地面温度呈显著相关关系(P<0.05),其中深间隙(DG)、深间隙分数(DGF)、覆盖分数(CF)、间隙分数分布(GFP)表征了结构多样性的覆盖/开放度特征;冠层高度标准差(H_(std))、冠层高度最大值(H_(max))、95%分位点高度(ZQ_(95))表征了高度特征;垂直复杂指数(VCI)表征了异质性特征。2)城市森林冠层结构多样性的覆盖/开放度特征对冠下地面温度的响应更强(R^(2)为0.15~0.5),强于高度指标(R^(2)为0.14~0.19)以及异质性指标(R^(2)=0.14)。3)结合高度指标、覆盖/开放度指标以及异质性指标的多因子预测模型2(R^(2)=0.61,RMSE=0.51,MSE=0.26,AIC=62.74),对于冠下地面温度的预测性能更优。研究明晰了城市森林结构多样性的多因子变量及其特征组合预测冠下地面温度的潜力,为城市森林冠层结构调控内部小气候环境研究提供了科学参考。

基于改进YOLOX的无人机航拍图像小目标检测算法

在无人机上运用目标检测技术具有广泛的应用前景,但和自然拍摄的图像不同,无人机航拍的图像更加复杂,且大多数为小目标。而现有的检测算法缺乏对小目标的特征提取能力,从而导致严重的误检和漏检问题。针对上述问题,提出一种基于YOLOX框架的高效的无人机小目标检测算法。首先,在特征融合网络中增加一层检测微小目标的特征融合结构,通过利用浅层特征图中丰富的位置信息和轮廓信息来加强网络对小目标的识别能力;同时,为了防止额外参数的增加,将减少头网络中的一层卷积层并缩减其通道数。其次,提出一种通道-空间注意力机制模块(Channel Spatial Attention Module,CSAM),利用最优的权重分配使网络聚焦于特征图中小目标密集的区域。最后,提出一种带位置引导的标签分配策略(LB-SimOTA),根据每个预测框和真实框的交并比(IOU)的大小,分别配以不同的权重,以改善网络中整体预测框的质量。在小目标居多的数据集VisDrone2019上的实验结果表明,文中提出的算法和YOLOX-S相比,针对车和人的检测精度提升了8.63%,检测速度FPS也可达到86,因此更适合在无人机对地面小目标检测的场景下部署。

基于改进YOLOv4的低慢小无人机实时探测算法

针对低慢小无人机探测任务中精度不高、在嵌入式平台上部署实时性能差的问题,提出了一种基于改进YOLOv4的小型无人机目标检测算法。通过增加浅层特征图、改进锚框、增强小目标,提高网络对小目标的检测性能,通过稀疏训练和模型修剪,大大缩短了模型运行时间。在1080Ti上平均精度(mAP)达到85.8%,帧率(FPS)达75 frame/s,实现了网络轻量化。该模型部署在Xavier边缘计算平台上,可实现60 frame/s的无人机目标检测速度。实验结果表明:与YOLOv4和YOLOv4-tiny相比,该算法实现了运行速度和检测精度的平衡,能够有效解决嵌入式平台上的无人机目标检测问题。

面向无人机视角下小目标检测的YOLOv8s改进模型

从无人机视角进行目标检测,面临图像目标小、分布密集、类别不均衡等难点,且由于无人机的硬件条件限制了模型的规模,导致模型的准确率偏低。提出一种融合多种注意力机制的YOLOv8s改进模型,在骨干网络中引入感受野注意力卷积和CBAM(concentration-based attention module)注意力机制改进卷积模块,解决注意力权重参数在感受野特征中共享问题的同时,在通道和空间维度加上注意力权重,增强特征提取能力;通过引入大型可分离卷积注意力思想,改造空间金字塔池化层,增加不同层级特征间的信息交融;优化颈部结构,增加具有丰富小目标语义信息的特征层;使用inner-IoU损失函数的思想改进MPDIoU(minimum point distance based IoU)函数,以innerMPDIoU代替原损失函数,提升对困难样本的学习能力。实验结果表明,改进后的YOLOv8s模型在VisDrone数据集上mAP、P、R分别提升了16.1%、9.3%、14.9%,性能超过YOLOv8m,可以有效应用于无人机平台上的目标检测任务。

基于改进Double-Head RCNN的无人机航拍图像小目标检测算法

为解决无人机航拍图像中小目标特征信息少且容易被噪声干扰导致现有算法漏检率和误检率高的问题,提出一种改进Double-Head Region-卷积神经网络(RCNN)的无人机航拍图像小目标检测算法。在骨干网络ResNet-50上引入Transformer和可变形卷积(DCN)模块,更有效提取小目标特征信息和语义信息;提出一种基于内容感知特征重组(CARAFE)的特征金字塔网络(FPN)结构模块,解决特征融合过程中小目标被背景噪声干扰而丢失特征信息的问题;在区域建议网络中针对小目标尺度分布特点重新设置Anchor生成尺度,进一步提升小目标检测性能。在VisDrone-DET2021数据集上的实验结果表明:所提算法能提取更具有表征能力的小目标特征信息和语义信息,对比Double-Head RCNN算法,所提算法的参数量增加了9.73×10^(6),FPS损失了0.6,但是AP、AP50和AP75分别提升了2.6%、6.2%和2.1%,APs提升了3.1%。

基于无人机航拍的苎麻倒伏信息解译研究

茎杆倒伏是苎麻三麻培育中最常见的灾害,传统的监测方法具有耗时耗力、不及时等局限性。提出了一种基于无人机航拍获取苎麻倒伏信息的方法,首先利用Pix4D Mapper软件生成苎麻的冠层正射影像和数字表面模型(digital surface model,DSM),基于正射影像提取苎麻光谱、纹理及形状特征,基于DSM提取苎麻株高指标,最后结合3种机器学习算法构建正常/倒伏苎麻分类模型。结果表明,基于DSM提取的株高信息可以有效代替大田实测株高,模型R2为0.899。倒伏和正常苎麻在光谱、纹理、形状及株高特征上具有差异。在3种机器学习算法中,支持向量机和决策树模型的性能最好,准确率达到99%,能够高效地识别苎麻倒伏地块。以上研究结果为准确、快速评估作物倒伏情况提供了技术支撑。

无人机激光雷达技术和倾斜摄影技术在地形测绘中的应用

针对地形落差大、植被茂密、通视条件差的区域,本文提出了一种综合利用无人机激光雷达(LiDAR)技术与倾斜摄影技术完成测图的方法。利用激光雷达穿透性强,高密度、高精度获取数据的优势,可准确获取地面高程信息;利用倾斜摄影建模技术可准确地获取地物的平面位置信息,实现比较直观的可视化效果,可快速进行矢量数据采集工作。通过两种技术手段的结合,可实现高精度地形图的生产。本文从实际出发,以黑龙江某项目为例,论证了此方法能够完全满足1∶500数字地形图的精度要求。

高分辨率特征增强的无人机航拍小目标检测

针对无人机航拍图像背景复杂、小尺寸目标分布密集等造成的检测精度低等问题,提出一种高分辨率特征增强的无人机航拍小目标检测算法。首先,提出了高分辨率特征增强网络,通过减少主干网络的下采样倍数来扩大输出特征图的尺度,同时引入双线性插值法来减少采样后特征信息的丢失,从而保留更多语义特征与细节特征。其次,在主干网络嵌入一种结合局部跨阶段结构的快速空间金字塔池化(Spatial pyramid pooling fast cross stage partial construction,SPPFCSPC)模块,增强局部与全局特征的信息融合,从而获得更大的感受野。最后,通过马赛克混合数据增强方法来增强图像背景的复杂度,提高模型的泛化能力。在公开数据集VisDrone 2019上的实验结果表明,与“你只需看一次”(You only look once,YOLO)系列等其他主流算法相比,本文算法的平均精度均值有显著的提高,在不同场景下均验证了本文算法的优越性,表明本文算法对无人机航拍图像的密集小目标检测任务有较强的实用性。