基于轻量化YOLOv4的交通信息实时检测方法

针对日常道路场景下的车辆目标检测问题,提出一种轻量化的YOLOv4交通信息实时检测方法。首先,制作了一个多场景、多时段的车辆目标数据集,并利用K-means++算法对数据集进行预处理;其次,提出轻量化YOLOv4检测模型,利用MobileNet-v3替换YOLOv4的主干网络,降低模型的参数量,并引入深度可分离卷积代替原网络中的标准卷积;最后,结合标签平滑和退火余弦算法,使用LeakyReLU激活函数代替MobileNet-v3浅层网络中原有的激活函数,从而优化模型的收敛效果。实验结果表明,轻量化YOLOv4的权值文件为56.4 MB,检测速率为85.6 FPS,检测精度为93.35%,表明所提方法可以为实际道路中的交通实时信息检测及其应用提供参考。

Underwater Sea Cucumber Target Detection Based on Edge-Enhanced Scaling YOLOv4

Sea cucumber detection is widely recognized as the key to automatic culture.The underwater light environment is complex and easily obscured by mud,sand,reefs,and other underwater organisms.To date,research on sea cucumber detection has mostly concentrated on the distinction between prospective objects and the background.However,the key to proper distinction is the effective extraction of sea cucumber feature information.In this study,the edge-enhanced scaling You Only Look Once-v4(YOLOv4)(ESYv4)was proposed for sea cucumber detection.By emphasizing the target features in a way that reduced the impact of different hues and brightness values underwater on the misjudgment of sea cucumbers,a bidirectional cascade network(BDCN)was used to extract the overall edge greyscale image in the image and add up the original RGB image as the detected input.Meanwhile,the YOLOv4 model for backbone detection is scaled,and the number of parameters is reduced to 48%of the original number of parameters.Validation results of 783images indicated that the detection precision of positive sea cucumber samples reached 0.941.This improvement reflects that the algorithm is more effective to improve the edge feature information of the target.It thus contributes to the automatic multi-objective detection of underwater sea cucumbers.

基于Yolov4算法的交通标志检测

为了提高交通标志识别的速度和精度,提出了一种采用Yolov4(You only look once version4)深度学习框架的交通标志识别方法,并将该方法与SSD(single shot multi box detector)和Yolov3(You only look once version 3)算法进行对比,所提算法模型参数量显著增加。进一步对Yolov4的主干特征提取网络和多尺度输出进行调整,提出轻量化的Yolov4算法。仿真实验表明,此算法能够快速有效检测交通标志,具有实时性和适用性。

基于改进YOLOv4的轻量化目标检测算法

针对当前YOLOv4目标检测网络结构复杂、参数多、训练所需的配置高以及实时检测每秒传输帧数(FPS)低的问题,提出一种基于YOLOv4的轻量化目标检测算法ML-YOLO。首先,用MobileNetv3结构替换YOLOv4的主干特征提取网络,从而通过MobileNetv3中的深度可分离卷积大幅减少主干网络的参数量;然后,用简化的加权双向特征金字塔网络(Bi-FPN)结构替换YOLOv4的特征融合网络,从而用Bi-FPN中的注意力机制提高目标检测精度;最后,通过YOLOv4的解码算法来生成最终的预测框,并实现目标检测。在VOC2007数据集上的实验结果表明,ML-YOLO算法的平均准确率均值(mAP)达到80.22%,与YOLOv4算法相比降低了3.42个百分点,与YOLOv5m算法相比提升了2.82个百分点;而ML-YOLO算法的模型大小仅为44.75 MB,与YOLOv4算法相比减小了199.54 MB,与YOLOv5m算法相比,只高了2.85 MB。实验结果表明,所提的ML-YOLO模型,一方面较YOLOv4模型大幅减小了模型大小,另一方面保持了较高的检测精度,表明该算法可以满足移动端或者嵌入式设备进行目标检测的轻量化和准确性需求。

基于Siamese-YOLOv4的印刷品缺陷目标检测

在印刷工业生产中,针对直接使用YOLOv4网络进行印刷缺陷目标检测精度低、所需训练样本数量大的问题,提出了一种基于Siamese-YOLOv4的印刷品缺陷目标检测方法。首先,使用了一种图像分割和随机参数变化的策略对数据集进行增强;然后,在主干网络中增加了孪生相似性检测网络,并在相似性检测网络中引入Mish激活函数来计算出图像块的相似度,在此之后将相似度低于阈值的区域作为缺陷候选区域;最后,训练候选区域图像,从而实现缺陷目标的精确定位与分类。实验结果表明:Siamese-YOLOv4模型的检测精度优于主流的目标检测模型,在印刷缺陷数据集上,Siamese-YOLOv4网络对卫星墨滴缺陷的检测准确率为98.6%,对脏点缺陷的检测准确率为97.8%,对漏印缺陷的检测准确率为93.9%;检测的平均精度均值(mAP)达到了96.8%,相较于YOLOv4算法、Faster R-CNN算法、SSD算法、EfficientDet算法分别提高了6.5个百分点、6.4个百分点、14.9个百分点、10.6个百分点。所提Siamese-YOLOv4模型一方面在印刷品缺陷检测中有较低的误检率和漏检率,另一方面通过相似性检测网络计算图像块的相似度从而提高了检测的精度,表明所提缺陷检测方法可应用于印刷质检以提高印刷企业的缺陷检测水平。

基于热红外图像的光伏板热斑检测方法研究

光伏板长期处于室外环境,极易因污渍遮挡而产生热斑效应,进而影响光伏电站的安全和高效运行。针对该问题,对基于传统图像处理和基于机器学习的目标检测算法的热斑检测开展研究。基于传统图像处理,利用区域分割算法和边缘检测算法进行试验,并研究热斑检测的效果。基于机器学习,提出了一种改进型你只看一次第四版本(YOLOv4)的热斑检测方法。其中,数据集通过实地拍摄光伏板热斑搭配模拟热斑的方法来获取。试验结果表明,改进的YOLOv4模型对数据集中的热斑检测指标交并比(IoU)达到92.31%、平均精度(AP)达到93.42%,均优于YOLOv4模型的效果。该研究具有一定的工程应用价值。

Delivery Invoice Information Classification System for Joint Courier Logistics Infrastructure

With the growth of the online market,demand for logistics and courier cargo is increasing rapidly.Accordingly,in the case of urban areas,road congestion and environmental problems due to cargo vehicles are mainly occurring.The joint courier logistics system,a plan to solve this problem,aims to establish an efficient logistics transportation system by utilizing one joint logistics delivery terminal by several logistics and delivery companies.However,several courier companies use different types of courier invoices.Such a system has a problem of information data transmission interruption.Therefore,the data processing process was systematically analyzed,a practically feasible methodology was devised,and delivery invoice information processing standards were established for this.In addition,the importance of this paper can be emphasized in terms of data processing in the logistics sector,which is expected to grow rapidly in the future.The results of this study can be used as basic data for the implementation of the logistics joint delivery terminal system in the future.And it can be used as a basis for securing the operational reliability of the joint courier logistics system.

基于ADSS光缆弧垂的在线监测及环境隐患预警算法

由于恶劣的自然环境和光缆本身弧度的变化会影响全介质自承式(ADSS)光缆传输效率,同时一些如施工或超高车辆及烟火等环境隐患也会严重威胁其正常运行,提出了基于ADSS光缆弧垂在线监测及环境隐患预警算法。首先,采用倾角测量法对弧度进行实时监测;然后,对YOLOv4检测算法进行改进,并设计了可视化智能融合终端;最后,将所提算法与其它算法进行对比实验。实验结果表明:所提算法的平均精度为96.43%,实现了对通信光缆运行状态的实时监测、数据处理与故障分析和智能隐患异常报警,并集成融合为ADSS光缆在线监测可视化智能终端。

基于密集连接卷积神经网络的道路车辆检测与识别算法

针对现有道路车辆检测识别算法中存在的检测精度不高、实时性差以及小目标车辆漏检等问题,提出一种基于密集连接卷积神经网络的道路车辆检测与识别算法。首先,基于YOLOv4网络框架,通过采用密集连接的深度残差网络结构,加强特征提取阶段的特征复用,实现对浅层复杂度较低的特征的利用;然后,在多尺度特征融合网络引入跳跃连接结构,强化网络的特征信息融合和表征能力,以降低车辆漏检率;最后,采用维度聚类算法重新计算先验框尺寸,并按照合理的策略分配给不同检测尺度。实验结果表明,该算法在KITTI数据集上获得了98.21%的检测精度和48.05 frame/s的检测速度,对于BDD100K数据集中复杂恶劣环境中的车辆也有较好的检测效果,在满足实时检测要求的同时有效提升检测精度。

Real-time hand tracking based on YOLOv4 model and Kalman filter

Aiming at the shortcomings of current gesture tracking methods in accuracy and speed, based on deep learning You Only Look Once version 4(YOLOv4) model, a new YOLOv4 model combined with Kalman filter real-time hand tracking method was proposed. The new algorithm can address some problems existing in hand tracking technology such as detection speed, accuracy and stability. The convolutional neural network(CNN) model YOLOv4 is used to detect the target of current frame tracking and Kalman filter is applied to predict the next position and bounding box size of the target according to its current position. The detected target is tracked by comparing the estimated result with the detected target in the next frame and, finally, the real-time hand movement track is displayed. The experimental results validate the proposed algorithm with the overall success rate of 99.43% at speed of 41.822 frame/s, achieving superior results than other algorithms.

基于深度级联模型工业安全帽检测算法

在工业生产中,安全帽对人体头部提供了较好的安全保障。在现场环境中,检验施工人员是否佩戴安全帽主要依靠人工检查,因而效率非常低。为了解决施工现场安全帽检测识别难题,提出一种基于深度级联网络模型的安全帽检测方法。首先通过You Only Look Once version 4(YOLOv4)检测网络对施工人员进行检测;然后运用注意力机制残差分类网络对人员ROI区域进行分类判断,识别其是否佩戴安全帽。该方法在Ubuntu18.04系统和Pytorch深度学习框架的实验环境中进行,在自主制作工业场景安全帽数据集中进行训练和测试实验。实验结果表明,基于深度级联网络的安全帽识别模型与YOLOv4算法相比,准确率提高了2个百分点,有效提升施工人员安全帽检测效果。

Methods and Means for Small Dynamic Objects Recognition and Tracking

A literature analysis has shown that object search,recognition,and tracking systems are becoming increasingly popular.However,such systems do not achieve high practical results in analyzing small moving living objects ranging from 8 to 14 mm.This article examines methods and tools for recognizing and tracking the class of small moving objects,such as ants.To fulfill those aims,a customized You Only Look Once Ants Recognition(YOLO_AR)Convolutional Neural Network(CNN)has been trained to recognize Messor Structor ants in the laboratory using the LabelImg object marker tool.The proposed model is an extension of the You Only Look Once v4(Yolov4)512×512 model with an additional Self Regularized Non–Monotonic(Mish)activation function.Additionally,the scalable solution for continuous object recognizing and tracking was implemented.This solution is based on the OpenDatacam system,with extended Object Tracking modules that allow for tracking and counting objects that have crossed the custom boundary line.During the study,the methods of the alignment algorithm for finding the trajectory of moving objects were modified.I discovered that the Hungarian algorithm showed better results in tracking small objects than the K–D dimensional tree(k-d tree)matching algorithm used in OpenDataCam.Remarkably,such an algorithm showed better results with the implemented YOLO_AR model due to the lack of False Positives(FP).Therefore,I provided a new tracker module with a Hungarian matching algorithm verified on the Multiple Object Tracking(MOT)benchmark.Furthermore,additional customization parameters for object recognition and tracking results parsing and filtering were added,like boundary angle threshold(BAT)and past frames trajectory prediction(PFTP).Experimental tests confirmed the results of the study on a mobile device.During the experiment,parameters such as the quality of recognition and tracking of moving objects,the PFTP and BAT,and the configuration parameters of the neural network and boundary line model were analyzed.The results showed an increased tracking accuracy with the proposed methods by 50%.The study results confirmed the relevance of the topic and the effectiveness of the implemented methods and tools.

轻量化的PCB表面缺陷检测算法

针对印刷电路板(PCB)表面缺陷检测存在的速度低和准确率不高等问题,提出了一种基于改进YOLOv4-tiny模型的PCB表面缺陷检测算法。首先,采用了优化后的聚类方法对缺陷数据集进行聚类,以解决初始先验框不适合PCB表面缺陷检测的问题;其次,为了解决主干网络在下采样时可能丢失小尺度目标信息的问题,引入了切片操作;接着,在特征融合网络中,引入了软池化卷积结构,以提高模型感受野,增强对小目标特征的表达能力;最后,通过引入改进后的交叉熵损失函数优化了损失函数。在北京大学开源的印刷电路板缺陷数据集上验证了所提算法的效果,结果表明,相较于其他经典算法,所提算法在检测速度、精度和模型参数量等指标上都有较大的提升。

Using deep learning in an embedded system for real-time target detection based on images from an unmanned aerial vehicle: vehicle detection as a case study

For a majority of remote sensing applications of unmanned aerial vehicles(UAVs),the data need to be downloaded to ground devices for processing,but this procedure cannot satisfy the demands of real-time target detection.Our objective in this study is to develop a real-time system based on an embedded technology for image acquisition,target detection,the transmission and display of the results,and user interaction while providing support for the interactions between multiple UAVs and users.This work is divided into three parts:(1)We design the technical procedure and the framework for the implementation of a real-time target detection system according to application requirements.(2)We develop an efficient and reliable data transmission module to realize real-time cross-platform communication between airborne embedded devices and ground-side servers.(3)We optimize the YOLOv4 algorithm by using the K-Means algorithm and TensorRT inference to improve the accuracy and speed of the NVIDIA Jetson TX2.In experiments involving static detection,it had an overall confidence of 89.6%and a rate of missed detection of 3.8%;in experiments involving dynamic detection,it had an overall confidence and a rate of missed detection of 88.2%and 4.6%,respectively.