Real-Time Monitoring Method for Cow Rumination Behavior Based on Edge Computing and Improved MobileNet v3

[Objective]Real-time monitoring of cow ruminant behavior is of paramount importance for promptly obtaining relevant information about cow health and predicting cow diseases.Currently,various strategies have been proposed for monitoring cow ruminant behavior,including video surveillance,sound recognition,and sensor monitoring methods.How‐ever,the application of edge device gives rise to the issue of inadequate real-time performance.To reduce the volume of data transmission and cloud computing workload while achieving real-time monitoring of dairy cow rumination behavior,a real-time monitoring method was proposed for cow ruminant behavior based on edge computing.[Methods]Autono‐mously designed edge devices were utilized to collect and process six-axis acceleration signals from cows in real-time.Based on these six-axis data,two distinct strategies,federated edge intelligence and split edge intelligence,were investigat‐ed for the real-time recognition of cow ruminant behavior.Focused on the real-time recognition method for cow ruminant behavior leveraging federated edge intelligence,the CA-MobileNet v3 network was proposed by enhancing the MobileNet v3 network with a collaborative attention mechanism.Additionally,a federated edge intelligence model was designed uti‐lizing the CA-MobileNet v3 network and the FedAvg federated aggregation algorithm.In the study on split edge intelli‐gence,a split edge intelligence model named MobileNet-LSTM was designed by integrating the MobileNet v3 network with a fusion collaborative attention mechanism and the Bi-LSTM network.[Results and Discussions]Through compara‐tive experiments with MobileNet v3 and MobileNet-LSTM,the federated edge intelligence model based on CA-Mo‐bileNet v3 achieved an average Precision rate,Recall rate,F1-Score,Specificity,and Accuracy of 97.1%,97.9%,97.5%,98.3%,and 98.2%,respectively,yielding the best recognition performance.[Conclusions]It is provided a real-time and effective method for monitoring cow ruminant behavior,and the proposed federated edge intelligence model can be ap‐plied in practical settings.

改进Tiny-YOLOv3的工业钢材瑕疵检测算法

深度学习网络模型参数量大,不适用于嵌入式或移动设备上。针对工业钢材生产过程中的实时检测问题,提出了一种改进的R-Tiny-YOLOv3工业钢材瑕疵检测算法。首先,在Tiny-YOLOv3结构中加入残差网络结构,提高检测的精度。增加了空间金字塔SPP网络模块,提高网络特征提取能力。结合不同网络层的特征信息,将检测提高到三个尺度。然后,选取CIOU作为损失函数,使目标检测框的回归更加稳定。最后对数据集进行数据增强,并在Cambricon 1H8嵌入式平台进行测试。实验结果表明改进的R-Tiny-YOLOv3算法能够实时地检测出瑕疵目标,平均准确率提高了10.8%,运算速度可达39.8帧/s,为工业钢材瑕疵检测的嵌入式应用提供了参考。

Improved Z^(1/3) Law of Nuclear Charge Radius

An improved Z^1/3 law of nuclear charge radius is presented. The comparison between the calculated and experimental nuclear charge radii now available shows that this new formula is better than the other conventional formulae.

Improved Polarization Retention of BiFeO3 Thin Films Using GdScO3(110)Substrates

Epitaxial ferroelectric one direction over the thin fihns on single-crystal substrates generally show a preferred domain orientation in other in demonstration of a poor polarization retention. This behavior will affect their application in nonvolatile ferroelectric random access memories where bipolar polarization states are used to store the logic 0 and 1 data. Here the retention characteristics of BiFe03 thin films with Srftu03 bottom electrodes on both GdSc03 （110） and SrTiO3 （100） substrates are studied and compared, and the results of piezoresponse force microscopy provide a long time retention property of the films on two substrates. It is found that bismuth ferrite thin films grown on GdScO3 substrates show no preferred domain variants in comparison with the preferred downward polarization orientation toward bottom electrodes on SrTi03 substrates. Tile retention test from a positive-up domain to a negative-down domain using a signal generator and an oscilloscope coincidentally shows bistable polarization states on the GdSeOa substrate over a measuring time of 500s, unlike the preferred domain orientation on SrTi03, where more than 65~o of upward domains disappear after 1 s. In addition, different sizes of domains have been written and read by using the scanning tip of piezoresponse force microscopy, where the polarization can stabilize over one month. This study paves one route to improve the polarization retention property through the optimization of the lattice-mismatched stresses between films and substrates.

Edge detection of gravity anomaly with an improved 3D structure tensor

Edge detection plays an important role in geological interpretation of potential field data,which can indicate the subsurface faults,contact,and other tectonic features.A variety of methods have been proposed to detect and enhance the edges.3 D structure tensor can well delineate the edges of geological bodies,however,it is sensitive to noise and additional false edges need to be removed artificially.In order to overcome these disadvantages,this paper redefines the 3 D structure tensor with a Gaussian envelop and proposes a new normalized edge detector,which can remove the additional false edges and reduce the influence of noise effectively,and balance the edges of different amplitude anomalies completely.This method has been tested on the synthetic and measured gravity data,showing that the new improved method achievesbetter results and reveals more details.

Clinical application of improved 2D computer-assisted fluoroscopic navigation through simulating a 3D vertebrae image to guide pedicle screw internal fixation

Objective To study the effect of using improved 2D computer-assisted fluoroscopic navigation through simulating 3D vertebrae image to guide pedicle screw internal fixation.Methods Posterior pedicle screw internal fixation,distraction

基于Tiny-YOLOV3的无人机地面目标跟踪算法设计

为了提高四旋翼无人机对地面目标跟踪的稳定性和跟踪精度,提出了一种结合Tiny-YOLOV3和卡尔曼滤波的跟踪算法;首先分析了Tiny-YOLOV3的原理和网络结构,并基于Tiny-YOLOV3的目标检测结果,结合无人机状态和目标的几何关系建立了目标跟踪系统的数学模型;接着对目标相对运动关系进行分析,建立目标的运动学模型,考虑到目标检测结果受干扰影响较大,应用卡尔曼滤波器实现对目标轨迹的滤波和预测,进而提升目标跟踪的精度;最后根据经过卡尔曼滤波后的目标轨迹信息设计无人机控制律,在轨迹控制的同时引入对无人机偏航角的控制,从而实现无人机对目标的稳定跟踪;仿真结果表明无人机对目标的位置跟踪精度在0.5 m以内,速度跟踪误差在0.2 m/s以内,偏航角跟踪误差在3°以内,跟踪效果良好,从而论证了所提算法的有效性。

基于Tiny-YOLOv3的小目标检测仿真

针对轻量级神经网络模型检测精度不高,容易对小目标物体产生漏检的问题,该文提出了基于Tiny-YOLOv3的目标检测改进算法。将Tiny-YOLOv3模型中的池化层用卷积核为3×3、步长为2×2的卷积层代替,对输入图像的尺寸进行调整,对特征提取网络最后4层的特征图尺寸与通道数进行修改,并在原有模型的基础上添加了一层特征融合层。在VOC2007数据集上进行仿真实验,改进后的模型mAP上升了3.79%,瓶子这类物体的AP值提高了14%,说明小目标物体的检测效果得到了提升,降低了中小目标检测过程中的漏检率。

基于Tiny-YOLOv3改进算法的工件识别

针对Tiny-YOLOv3算法在工件识别实时检测中存在漏检率高的问题,提出了在Tiny-YOLOv3基础上加以改进实现了对工件更加快速、准确地识别。主要改进的方式是在Tiny-YOLOv3的特征提取网络中增加3个网络模块,即SPP结构、SE模块和Ghost模块,并用卷积层代替池化层,改进后的网络结构平均精度均值、准确率和网络模型大小都有着显著的改善。试验结果表明,改进后的算法能够更好的提升工件识别的效率,并同时满足在嵌入式设备中进行实时检测的要求。

提高马钢焦炉K_(3)系数的生产实践

通过分析影响马钢焦炉K_(3)系数的因素,采取优化焦炉生产组织、提高设备保产能力、加强炼焦过程监控、全面对标找差、强化炉体维护等具体措施,使得焦炉K_(3)系数得到了稳定提升。

基于Tiny-YOLOv3的网络结构化压缩与加速

针对特定应用场景下,Tiny-YOLOv3(You Only Look Once v3)网络在嵌入式平台部署时存在资源开销大、运行速度慢的问题,文中提出了一种结合剪枝与量化的结构化压缩方案,并搭建了针对压缩后网络的卷积层加速系统。结构化压缩方案使用稀疏化训练与通道剪枝来减少网络中的计算量,使用激活值定点数量化和权重二的整数次幂量化来减少网络卷积层中的参数存储量。在卷积层加速系统中,可编程逻辑部分按照并行加流水线方法设计了一个卷积层加速器核,处理系统部分负责卷积层加速系统调度。实验结果表明,Tiny-YOLOv3经过结构化压缩后的网络平均准确度为0.46,参数压缩率达到了5%。卷积层加速系统在Xilinx的ZYNQ芯片进行部署时,硬件可以稳定运行在250 MHz时钟频率下,卷积运算单元的算力为36 GOPS。此外,加速平台整体功耗为2.6 W,且硬件设计节约了硬件资源。

融合图像去雾与Tiny-YOLOv3的护帮板状态检测研究

为解决液压支架工长时间作业过程中,因身体疲劳不能及时发现护帮板未护帮的问题,采用实时性高的Tiny-YOLOv3算法检测护帮板状态,但检测任务会受到综采工作面尘雾的影响。因此,提出一种融合图像去雾与Tiny-YOLOv3的目标检测算法,并在此基础上优化图像去雾算法的CUDA实现,首先将暗通道图像用RGB单通道图像代替,然后按列分组求大气光值,合并初始透射率的kernel函数并优化精细化透射率计算方式,提升图像去雾速度,保证算法的实时性。实验结果表明,在煤矿护帮板状态检测场景中,融合算法比Tiny-YOLOv3算法的准确率提高了22.8%,且满足实时检测的要求。

Identification of Convective and Stratiform Clouds Based on the Improved DBSCAN Clustering Algorithm

A convective and stratiform cloud classification method for weather radar is proposed based on the density-based spatial clustering of applications with noise(DBSCAN)algorithm.To identify convective and stratiform clouds in different developmental phases,two-dimensional(2D)and three-dimensional(3D)models are proposed by applying reflectivity factors at 0.5°and at 0.5°,1.5°,and 2.4°elevation angles,respectively.According to the thresholds of the algorithm,which include echo intensity,the echo top height of 35 dBZ(ET),density threshold,andεneighborhood,cloud clusters can be marked into four types:deep-convective cloud(DCC),shallow-convective cloud(SCC),hybrid convective-stratiform cloud(HCS),and stratiform cloud(SFC)types.Each cloud cluster type is further identified as a core area and boundary area,which can provide more abundant cloud structure information.The algorithm is verified using the volume scan data observed with new-generation S-band weather radars in Nanjing,Xuzhou,and Qingdao.The results show that cloud clusters can be intuitively identified as core and boundary points,which change in area continuously during the process of convective evolution,by the improved DBSCAN algorithm.Therefore,the occurrence and disappearance of convective weather can be estimated in advance by observing the changes of the classification.Because density thresholds are different and multiple elevations are utilized in the 3D model,the identified echo types and areas are dissimilar between the 2D and 3D models.The 3D model identifies larger convective and stratiform clouds than the 2D model.However,the developing convective clouds of small areas at lower heights cannot be identified with the 3D model because they are covered by thick stratiform clouds.In addition,the 3D model can avoid the influence of the melting layer and better suggest convective clouds in the developmental stage.

Indices of El Nio and El Nio Modoki:An Improved El Nio Modoki Index

In recent years, El Nino Modoki （a type of pseudo-El Nino） has been distinguished as a unique large-scale ocean warming phenomenon happening in the central tropical Pacific that is quite different from the traditional El Nino. In this study, EOF analysis was used to successfully separate El Nino and El Nino Modoki. The abilities of the NINO3 index, NINO3.4 index, NINO1＋2 index and NINO4 index in characterizing El Nino were explored in detail. The resulting suggestion was that, comparatively, NINO3 is the optimal index for monitoring El Nino among the four NINO indices, as the other NINO indices were found to be less good at distinguishing between El Nino and El Nino Modoki signals, or were easily disturbed by El Nino Modoki signals. Further, an improved El Nino Modoki index （IEMI） was introduced in the current paper to better represent the El Nino Modoki that is captured by the second leading EOF mode of monthly tropical Pacific sea surface temperature anomalies （SSTAs）. The IEMI is an improvement of the El Nino Modoki index （EMI） through adjustments made to the inappropriate weight coefficients of the three boxes of EMI. The IEMI therefore overcomes the EMI’s inability to monitor the two historical El Nino Modoki events, as well as avoids the possible risk （present in the EMI） of excluding the interference of the El Nino signal. The realistic and potential advantages of the IEMI are clear.

基于Tiny-YOLOv3的田间绿色柑橘目标检测方法

目的:为准确、快速地识别田间绿色柑橘,提出一种基于Tiny-YOLOv3网络的目标检测方法。方法:采用卷积层替换Tiny的池化层以减少目标信息丢失,借鉴DenseNet网络,在Tiny网络中嵌入2个多层密集块,提出Tiny-Dense-YOLOv3网络。结果:在2个数据集上试验以验证改进模型的有效性,在果园柑橘数据集中,Tiny-Dense-YOLOv3的准确率、召回率和F 1值分别为88.98%、95.29%和92.03%,相比于Tiny-YOLOv3模型分别提高3.55%、4.81%和4.15%;在MSCOCO集的柑橘数据集中,Tiny-Dense-YOLOv3的F 1值为52.83%,相比于Tiny-YOLOv3模型,F 1值提高了6.33%。Tiny、Darknet53和Tiny-Dense等网络输出特征图的可视化结果表明,不同网络均能提取果实目标特征,其中Tiny网络未能有效抑制树叶、枝干等背景特征的干扰。结论:Tiny-Dense-YOLOv3轻量化卷积网络可实现对田间绿色柑橘的高精度实时检测。

基于改进DeepLabv3+算法的起重机锈迹检测

室外工作的起重机金属结构易产生锈蚀现象,严重的锈蚀会导致结构承载能力显著降低,从而引发灾难性事故。文中针对当前起重机人工锈迹巡检中存在的漏检、误检和费时等问题,提出一种基于改进DeepLabv3+算法的自动化锈迹检测方法。该方法依托于机器视觉,将原始DeepLabv3+的骨干网络替换为幽灵网络(GhostNet)以提升网络的轻量化程度;使用特征金字塔网络(FPN)进行特征提取,用于抑制噪声和背景对锈迹提取的不良干扰;引入空间感知独立自注意机制(SSA)来提高网络区域感知性能;最后使用特征融合(Add)代替原始网络的特征堆叠来降低算法参数量。将所提方法应用于室外起重机锈迹检测,结果表明所提算法的检测性能优于原始算法和其他经典语义分割算法,具有重要的工程应用价值。

基于改进Tiny-YOLOv3的人数统计方法

卷积神经网络已经成为了计算机视觉处理最为广泛的技术方法,基于卷积神经网络的目标检测技术也是一个热门的研究话题。本文通过引入通道注意力机制,对目标检测算法Tiny-YOLOv3进行改进,训练人体头部的目标检测模型,从而统计当前监控环境下的人数。实验结果表明该方法能取得较好的头部检测效果,人数统计准确率高。

基于改进Tiny-YOLOv3的烟雾检测算法

实时准确的烟雾检测对森林火灾的预警至关重要,特别是面向计算力有限的嵌入式设备。为在这类设备上稳定检测烟雾,本文选择Tiny-YOLOv3作为基准模型。受Mobilenet的启发,设计一种改进Tiny-YOLOv3的新型网络结构,使用深度可分离卷积替换普通卷积,增加网络层数和输出通道数提高模型的精度。实验结果表明,所提出模型的体积小于Tiny-YOLOv3,在公开烟雾数据集的精度高于Tiny-YOLOv3,且性能优于经典的主流检测模型。这证明了本文算法的有效性。

Optimization of buckling load for laminated composite plates using adaptive Kriging-improved PSO:A novel hybrid intelligent method

An effective hybrid optimization method is proposed by integrating an adaptive Kriging(A-Kriging)into an improved partial swarm optimization algorithm(IPSO)to give a so-called A-Kriging-IPSO for maximizing the buckling load of laminated composite plates(LCPs)under uniaxial and biaxial compressions.In this method,a novel iterative adaptive Kriging model,which is structured using two training sample sets as active and adaptive points,is utilized to directly predict the buckling load of the LCPs and to improve the efficiency of the optimization process.The active points are selected from the initial data set while the adaptive points are generated using the radial random-based convex samples.The cell-based smoothed discrete shear gap method(CS-DSG3)is employed to analyze the buckling behavior of the LCPs to provide the response of adaptive and input data sets.The buckling load of the LCPs is maximized by utilizing the IPSO algorithm.To demonstrate the efficiency and accuracy of the proposed methodology,the LCPs with different layers(2,3,4,and 10 layers),boundary conditions,aspect ratios and load patterns(biaxial and uniaxial loads)are investigated.The results obtained by proposed method are in good agreement with the literature results,but with less computational burden.By applying adaptive radial Kriging model,the accurate optimal resultsebased predictions of the buckling load are obtained for the studied LCPs.

Improvement of Binocular Reconstruction Algorithm for Measuring 3D Pavement Texture Using a Single Laser Line Scanning Constraint

The dense and accurate measurement of 3D texture is helpful in evaluating the pavement function.To form dense mandatory constraints and improve matching accuracy,the traditional binocular reconstruction technology was improved threefold.First,a single moving laser line was introduced to carry out global scanning constraints on the target,which would well overcome the difficulty of installing and recognizing excessive laser lines.Second,four kinds of improved algorithms,namely,disparity replacement,superposition synthesis,subregion segmentation,and subregion segmentation centroid enhancement,were established based on different constraint mechanism.Last,the improved binocular reconstruction test device was developed to realize the dual functions of 3D texture measurement and precision self-evaluation.Results show that compared with traditional algorithms,the introduction of a single laser line scanning constraint is helpful in improving the measurement’s accuracy.Among various improved algorithms,the improvement effect of the subregion segmentation centroid enhancement method is the best.It has a good effect on both overall measurement and single pointmeasurement,which can be considered to be used in pavement function evaluation.