改进 YOLOv5 的 PDC 钻头复合片缺损识别

PDC 钻头复合片的缺损情况是影响钻进效率的重要因素,检测 PDC 钻头复合片是否缺损是修复 PDC 钻头的前提。 为 了减少对 PDC 钻头复合片的误检,提升检测准确率,提出了一种基于改进 YOLOv5 的目标检测算法。 该方法以 YOLOv5 网络为 基础,融合 RepVGG 重参数化模块增强网络的特征提取能力;在 C3 模块中引入坐标注意力机制,在通道注意力机制中嵌入位置 信息,提升对缺损复合片的目标检测能力;将边界框回归损失函数改进为 WIoU 损失函数,制定合适的梯度增益分配策略。 实 验结果表明,改进后的网络的精确率提升 2%,召回率提升 0. 9%,平均精度均值(mAP)提升了 1. 3%,达到了 98%,能够实现对 PDC 钻头复合片的缺损识别。     

一种基于特征重用金字塔的舰船检测算法

针对现有算法在SAR图像舰船目标检测场景中难以提取模糊目标特征的问题,提出一种基于特征重用金字塔的舰船目标检测算法。所提算法以YOLOV4-tiny为主体,首先将线性因子引入到K-means算法中整合初始锚框,加强网络对多尺度目标的适应性;其次在主干CSPDarknet53-tiny中添加注意力机制来抑制干扰信息,减弱复杂背景的影响;最后利用特征重用机制强化特征金字塔,提升网络对模糊目标特征的提取能力。实验结果表明,相较于YOLOV4-tiny网络,改进后的算法在SSDD数据集上的平均检测精度提升11.79%,证明了改进后算法在舰船检测中的有效性。     

改进型CenterNet的高铁无砟轨道板表面裂缝检测算法

针对传统的高铁无砟轨道板表面裂缝检测方法存在检测精度低、速度慢的问题，提出一种改进型CenterNet的高铁无砟轨道板表面裂缝检测算法。该算法在编解码网络之间加入空洞空间金字塔池化模块(ASPP)，以此扩大特征图的感受野，充分提取不同尺度的上下文信息；然后在特征提取网络中加入多光谱通道注意力模块（MCA），使网络可以更好学习每个通道的权重，捕获图像丰富的输入特征信息；最后使用α IoU损失函数来提高边界框预测的准确度。实验结果表明，本算法平均检测精度(mAP)达到8412%，相比传统算法平均检测精度提升了337%，对于轨道板表面裂缝具有较好的检测效果。     

基于注意力机制和跨尺度特征融合的摩托车头盔检测算法

在道路交通摩托车事故中，未佩戴头盔是导致骑乘人员受到致命伤害的主要原因。针对目前头盔检测中由于黑色头发、帽子和头盔的颜色和形状相似存在误检和漏检问题，提出了一种具有三重注意力机制和双向跨尺度特征融合的摩托车头盔检测算法。首先，在YOLOV5s的主干网络中引入三重注意力机制，提取了不同维度之间的语义依赖，消除了通道和权重的间接对应关系，通过关注相似样本的差异从而提升检测精度。其次，采用EIOU边框损失函数优化对遮挡和重叠目标的检测效果。最后，在特征金字塔中采用加权双向特征金字塔网络结构，实现高效的双向跨尺度连接和加权特征融合，增强了网络特征提取能力。实验结果表明，改进算法实现了98.7%的mAP@0.5、94.0%的mAP@0.5：0.95，与原算法相比，改进算法的mAP@0.5提升了3.9%以及mAP@0.5：0.95提升了7.6%，具有更高的精度和更强的泛化能力。     

结合注意力机制和密集连接网络的车辆检测方法

为提高算法对车辆检测的准确性,解决原有算法在复杂交通场景下对车辆检测效果不佳的问题,提出一种基于注意力机制和改进密集连接网络结构的车辆检测方法。首先在过渡层中使用SoftPool整合密集块之间的特征信息;其次通过轻量化通道注意力机制加强有效通道特征的表达,将其作为Darknet-53的深层特征提取层;引入CIOU损失作为模型的边界框位置预测损失项,使用深度可分离卷积缩减模型体积;与原算法相比mAP值提高2.6%,模型体积缩减为原来的42%,实验证明本算法在复杂交通场景下具有良好的检测性能。     

基于感受野增强与并行坐标注意力的航拍小目标检测

针对航拍图像小目标占比多、背景复杂、存在检测精度低的问题。提出一种基于感受野增强与并行坐标注意力的航拍小目标检测算法，设计一种感受野增强模块，使用不同大小的空洞卷积扩大感受野范围并融合有效通道注意力机制来提高网络的特征提取能力；改进特征融合结构来提高算法对小目标的检测能力；设计一种并行坐标注意力模块用以提高航拍密集小目标检测以及抗背景干扰能力。采用不同输入分辨率在VisDrone数据集上进行实验，实验结果表明，所提算法的mAP0.5相较于YOLOv5算法提高了5.4%,mAP0.5:0.95提高了4.2%,在输入分辨率1 536×1 536下的mAP0.5可达54.9%,可以实现较好的小目标检测效果。     

基于改进YOLOX-S的交通标志识别

交通标志是车辆在规范行驶过程中的一个重要指导,交通标志识别是无人驾驶汽车环境感知中必不可少的重要内容。本文基于YOLOX-S算法进行改进,通过在主干网络末端添加CBAM注意力机制模块,强化特征提取网络所得到的特征;使用Focal Loss函数,更好地消除正负样本不均衡问题,挖掘难例样本;使用GIOU损失函数,解决了原损失函数存在的优化不一致和尺度敏感的问题,进一步提高模型的识别准确率。本文基于TT100K数据集对提出算法进行了实验,对比了几种主流算法与本文算法的识别精度,实验结果表明在具有较高FPS的前提下,本文算法对大部分目标类别的检测精度都有所提升。相较于YOLOX-S模型,所提模型的coco精度评价指标mAP＿50提升1.9%,mAP＿50:95提升2.1%,FPS为35.6。证明了所做改进的有效性。     

基于特征融合的自适应多尺度无锚框目标检测算法

为了提高 CenterNet 无锚框目标检测网络的目标检测能力,提出一种基于注意力特征融合和多尺度特征提取网络的改 进 CenterNet 目标检测网络。 首先,为了提升网络对多尺度目标的表达能力,设计了自适应多尺度特征提取网络,利用空洞卷积 对特征图进行重采样获取多尺度特征信息,并在空间维度上进行融合;其次,为了更好地融合语义和尺度不一致的特征,提出了 一种基于通道局部注意力的特征融合模块,自适应地学习浅层特征和深层特征之间的融合权重,保留不同感受域的关键特征信 息。 最后,通过在 VOC 2007 测试集上对本文算法进行验证,实验结果表明,最终算法的检测精度达到 80. 94%,相较于基线算法 CenterNet 提升了 3. 82%,有效提升了无锚框目标检测算法的最终性能     

基于改进YOLOv7模型的变电站异物入侵识别

针对无人值守的变电站频繁出现鸟巢、塑料袋等异物入侵的问题,现有的区域卷积神经网络(R-CNN)和YOLO算法不能满足实时、准确检测识别的要求。因此提出一种基于改进YOLOv7的图像识别方法。首先,利用灰度化处理和高斯噪声增强图片数据,然后在特征提取网络中增加分支以提高小目标的检测能力。在此基础上,采用卷积块注意模块(CBAM)分支注意力机制来加强通道数和空间自适应学习能力,并在输出路径上引入SimAM注意力机制,使其在不增加参数的情况下提高模型对异物特征的提取能力,最后通过某变电站实际监控图进行算法验证。结果表明改进后的YOLOv7异物识别平均精度均值(mAP)为94.40%,比原YOLOv7mAP提升了3.69%,比Faster R-CNN和SSD分别提高了17.46%和16.66%,可以做到对变电站异物的实时检测识别,所提方法具有较好的可行性和工程使用价值。     

基于改进Centernet的变电设备红外检测方法

变电站红外图像中小目标众多并且环境复杂,导致现有检测算法精度较低,因此本文提出一种基于改进Centernet的变电设备红外检测方法。首先以Centernet作为基础模型,将FPN结构引入上采样网络以充分利用小目标特征信息,从而解决小目标难以被精确检测的问题;然后,为提升网络在复杂环境中检测的鲁棒性,通过在主干网络resnet50中嵌入注意力机制来提升网络对重要目标的关注;最后,采用CIOU损失替换中心点偏移损失和宽高损失的训练策略以加速网络收敛、提升训练效果。实验结果表明,本文方法在小目标检测和复杂环境检测中都能有较好的检测效果,检测精度相比改进前提升3.1%,达到92.7%,相比Faster R-CNN等现有方法精度更高,在变电设备红外检测中具有一定参考价值。     

基于无线传感器网络的桥梁结构健康监测系统

基于桥梁结构健康监测和无线传感器网络基本概念,设计出用于桥梁结构健康监测的无线传感器节点模型和软硬件结构体系。硬件基于Intel公司研制的Imote2平台,软件采用TinyOS操作系统。最后对基于无线传感器网络的桥梁结构健康监测系统的优势进行了分析与总结。无线传感器网络以其覆盖范围广、测量精度高和分布式、自组织等特点解决了有线系统的一些固有缺点,还可解决结构健康监测系统鲁棒性、自身寿命以及数据泛滥等问题。     

Design of Intelligent Agricultural Greenhouse Environment Monitoring System

Aiming at the problem of greenhouse environmental information collection in agricultural production, this paper designs a greenhouse monitoring system based on STM32 microcontroller. The system uses STM32F103 series MCU as the main control unit, which is used to receive the detection results of the sensor. The host computer software displays and processes the monitoring data. The temperature and humidity sensors monitor air temperature and humidity. The capacitive soil moisture detection sensor monitors soil moisture. The carbon dioxide sensor monitors the concentration of carbon dioxide. The photosensitive sensor monitors light intensity. This design has low hardware cost and high data acquisition accuracy, which can be applied to the actual greenhouse agricultural production and has wide practicability.     

Accurate contour control of mechatronic servo systems without parameter adjustment by use of modified input data method

This paper describes a technique of accurate contour control by a modified input data method for industrial mechatronic servo systems with unadjusted servo parameters. Adjustment of the servo parameters in order to achieve high‐performance contour control in mechatronic servo systems is a tedious task. The modified input data method can realize satisfactory contour control performance accuracy within rather large tolerances for servo parameter errors. The effectiveness of the proposed method has been verified by simulation and experimental studies of contour control of an actual mechatronic servo system. © 2007 Wiley Periodicals, Inc. Electr Eng Jpn, 160(4): 60–69, 2007; Published online in Wiley InterScience ( www.interscience.wiley.com ). DOI 10.1002/eej.20322     

A time-varying transformer outage model for on-line operational risk assessment

The failure probabilities of system components may vary with changes in the operating conditions. Performing a probabilistic risk assessment in real-time is challenging, since component failure probabilities are difficult to predict. Accordingly, this paper introduces a delayed semi-Markov process that incorporates real-time data from advanced sensors, as a means of efficiently calculating time-varying or condition-based failure probabilities. To demonstrate the feasibility of the procedure, a time-varying transformer outage model with numerical examples is presented. In the proposed technique, an analytic random model is developed to accommodate the impact of real-time dissolved gas analysis data, as well as other conditions pertaining to the failure probabilities of system components.     

基于分裂Bregman方法的加权频差电阻抗成像算法

Sleep apnea (SA) is a common sleep disorder. Identifying patients at risk by means of comprehensive monitoring that requires overnight stay at professional sleep clinics are costly and inconvenient and can lead to unreliable results in view of the unfamiliar sleep environment. Existing wearable devices for sleep monitoring, which can be used in a familiar home environment, do not provide the same comprehensive monitoring as through clinical monitoring. The larger objective of the present work is to develop a sleep monitoring system for home use, which can provide comprehensive monitoring. In the development in this paper, machine learning (ML) models are explored for the classification of SA and sleep stages using multisensory data, without neglecting any of the required signals. The data acquired through the sensors are normalized, their features are extracted using Composite Multiscale Sample Entropy (CMSE) and are standardized using a robust scaling algorithm. Processed features are classified using a Neural Network (NN) and the obtained results for the SA classification are compared with those obtained by using a Support Vector Machine (SVM) approach. The impact of neglecting signals when classifying sleep stages is analyzed as well. The results are presented in the paper and observations are made. The NN model trained with the Bayesian regularization algorithm has provided an overall average accuracy of 94.5% and performed slightly better than when trained using the scaled conjugate gradient backpropagation algorithm (93.2%). The SVMs have yielded lower accuracy levels compared to the NNs (<92%). It is observed that the use of all 14 signals for SS classification yields an overall test accuracy of 72.3%, which is higher than that when one or few signals are used. It is concluded that ML models are effective in classifying sleep data from multiple sensors. Accuracy levels are higher when fused multisensory data are used as inputs. Furthermore, NN models are found to be better suitable in practical application and can be incorporated into an inexpensive and convenient wearable device that can carry out comprehensive monitoring.     

变压器局部放电超声定位技术改进研究

介绍了一种基于超声波检测的变压器局部放电定位优化方法。在理论上分析了传统的电声定位不足之处,并通过对超声波传感器安装位置的分析和优化布置,提高了定位精度。     

硬币检测传感器

文中以材质、直径、厚度三种参数作为硬币检测的表征值,综合多种传感器功能,提出智能硬币检测传感器.该传感器以微处理器为基础,融合多个参数,通过专家系统对硬币的真伪和币值作出判断.文中给出的相关的测试数据表明,系统对参数的阀值设定是合适的.     

Apple Grading System Based on Weight and Colour

This paper involves the systematic, integrated and unified design of an automatic apple grading system. Apples are placed on a constant-speed conveyor system and graded based on 2 parameters: weight and colour. For weight detection, a piezo-resistive load cell (ELAF-T1-M-10L) is used. For colour detection, an RGB colour sensor (46CLR-D5LAC3-D5) is used. If the apples are equal to or above a pre-set weight threshold of 0.08kg and if the colour of the apple is red, green or yellow, they are graded in to three bins (bin 1: red apples, bin 2: green apples, bin 3: yellow apples). If the weight is below the threshold and/or if the colour of the apple is not red, green or yellow, they are sent to a reject bin. Both weight and colour requirements have to be met to be successfully graded, otherwise the apple is re-jected. Electric linear actuators (PA-15), actuated by combined signals from load cell, colour sensor and photoelectric sensors (OBR2000-R2) positioned very close to the actuators, are used to eject the apples to the correct bins. A digital magnetic speed sensor (DIGISPEC 0090/0091) is used to monitor the speed of the motor and maintain the required constant conveyor speed. Food grade conveyor belt (NS20 UFMT) is used to design the conveyor and direct drive DC motor (DMS07G) is used to move the conveyor belt. In this paper, the functionality, operation, important parameters and justifications for choosing individual components have been provided as well as detailed explanations of the overall system operation, along with potential drawbacks of proposed system. Relevant calculations necessary for system design are also provided.     

基于嵌入式与Web技术的激光测距远程监控系统

为了满足工业生产中,不同场合对距离过程控制的需求,本文提出了一种以激光传感器做为检测端,嵌入式系统作为数据处理和控制的远程监控系统方案。系统采用新一代的测距设备-DLS—B激光测距传感器,以基于ARM—Linux的嵌入式的平台为核心,使用RS485工业总线技术与Web服务器技术,对多台激光测距传感器进行监控,实现了激光测距系统的数据采集与处理。     

A control algorithm for the magnetic drive of an air-pulse valve of an internal combustion engine

This paper proposes an adaptive control algorithm for the fast-acting electromagnetic air pulse valve (APV) of an internal combustion engine (ICE) without specialized position and speed sensors. The APV generates wave activity in the intake system via fast adjustment of the APV to open and close the intake passage and generates resonance in the cylinder-intake manifold in the whole range of ICE rotation frequencies. The electromagnetic drive is installed directly before the ICE intake valve and adjusted via a control system to ensure the required dynamical performance of the drive under external influences. The existing systems and methods of control to ensure required dynamical performance under external influences presuppose the use of specialized position or speed sensors, which makes the system more complicated; increases its weight, dimensions, and cost; and makes it less reliable in general. A complex mathematical model of the ARV mechatronic drive has been designed to analyze the operating performance of the magnetic drive and control methods. The suggested control method makes it possible to indirectly determine in real time the position in space and speed of the armature approaching the terminal by the type of armature voltage and changes in current, taking account of all destabilizing factors. This paper also presents the results of studying the influence of the magnet armature on failures of current as well as the influence of the current pickup on the armature seating speed. The method of calculating the armature startup time by the voltage alteration in the control magnet winding is described. As proven by the experiments, the suggested control algorithm ensures a final valve seating speed of <0.4 mps for a switchover time of about 3–4 ms; thus the efficiency of the suggested algorithm has been proven.