Lightweight Res-Connection Multi-Branch Network for Highly Accurate Crowd Counting and Localization

Crowd counting is a promising hotspot of computer vision involving crowd intelligence analysis,achieving tremendous success recently with the development of deep learning.However,there have been stillmany challenges including crowd multi-scale variations and high network complexity,etc.To tackle these issues,a lightweight Resconnection multi-branch network(LRMBNet)for highly accurate crowd counting and localization is proposed.Specifically,using improved ShuffleNet V2 as the backbone,a lightweight shallow extractor has been designed by employing the channel compression mechanism to reduce enormously the number of network parameters.A light multi-branch structure with different expansion rate convolutions is demonstrated to extract multi-scale features and enlarged receptive fields,where the information transmission and fusion of diverse scale features is enhanced via residual concatenation.In addition,a compound loss function is introduced for training themethod to improve global context information correlation.The proposed method is evaluated on the SHHA,SHHB,UCF-QNRF and UCF_CC_50 public datasets.The accuracy is better than those of many advanced approaches,while the number of parameters is smaller.The experimental results show that the proposed method achieves a good tradeoff between the complexity and accuracy of crowd counting,indicating a lightweight and high-precision method for crowd counting.

Deep Learning Based Efficient Crowd Counting System

Estimation of crowd count is becoming crucial nowadays,as it can help in security surveillance,crowd monitoring,and management for different events.It is challenging to determine the approximate crowd size from an image of the crowd’s density.Therefore in this research study,we proposed a multi-headed convolutional neural network architecture-based model for crowd counting,where we divided our proposed model into two main components:(i)the convolutional neural network,which extracts the feature across the whole image that is given to it as an input,and(ii)the multi-headed layers,which make it easier to evaluate density maps to estimate the number of people in the input image and determine their number in the crowd.We employed the available public benchmark crowd-counting datasets UCF CC 50 and ShanghaiTech parts A and B for model training and testing to validate the model’s performance.To analyze the results,we used two metrics Mean Absolute Error(MAE)and Mean Square Error(MSE),and compared the results of the proposed systems with the state-of-art models of crowd counting.The results show the superiority of the proposed system.

A multi-source information fusion layer counting method for penetration fuze based on TCN-LSTM

When employing penetration ammunition to strike multi-story buildings,the detection methods using acceleration sensors suffer from signal aliasing,while magnetic detection methods are susceptible to interference from ferromagnetic materials,thereby posing challenges in accurately determining the number of layers.To address this issue,this research proposes a layer counting method for penetration fuze that incorporates multi-source information fusion,utilizing both the temporal convolutional network(TCN)and the long short-term memory(LSTM)recurrent network.By leveraging the strengths of these two network structures,the method extracts temporal and high-dimensional features from the multi-source physical field during the penetration process,establishing a relationship between the multi-source physical field and the distance between the fuze and the target plate.A simulation model is developed to simulate the overload and magnetic field of a projectile penetrating multiple layers of target plates,capturing the multi-source physical field signals and their patterns during the penetration process.The analysis reveals that the proposed multi-source fusion layer counting method reduces errors by 60% and 50% compared to single overload layer counting and single magnetic anomaly signal layer counting,respectively.The model's predictive performance is evaluated under various operating conditions,including different ratios of added noise to random sample positions,penetration speeds,and spacing between target plates.The maximum errors in fuze penetration time predicted by the three modes are 0.08 ms,0.12 ms,and 0.16 ms,respectively,confirming the robustness of the proposed model.Moreover,the model's predictions indicate that the fitting degree for large interlayer spacings is superior to that for small interlayer spacings due to the influence of stress waves.

4种植物源性成分多重real-time PCR检测方法的建立及其在食用淀粉中的应用

建立一种可同时快速检测红薯、木薯、马铃薯、玉米源性成分的多重实时聚合酶链式反应(real-time polymerase chain reaction,real-time PCR)方法。分别以红薯g3pdh基因、木薯g3pdh基因、马铃薯UGPase基因、玉米zSSIIb基因为靶基因设计特异性引物和TaqMan探针,以18S rRNA基因为内参基因,建立多重real-time PCR方法,开展方法学验证,并对不同掺入比例模拟样品和实际淀粉样品进行检测。结果显示,该方法具有高通量、特异性强、灵敏度高等优点。与15种非目标源性均无交叉反应;对目标DNA的检测灵敏度可达到3×10^(-3) ng/μL,且具有良好的线性关系和扩增效率;对淀粉样品的检出限可达0.1%,对50份实际样品进行检测,结果与参比方法一致,说明建立的多重real-time PCR法可用于食用淀粉种类掺假鉴别检测。

Counting of alpha particle tracks on imaging plate based on a convolutional neural network

Imaging plates are widely used to detect alpha particles to track information,and the number of alpha particle tracks is affected by the overlapping and fading effects of the track information.In this study,an experiment and a simulation were used to calibrate the efficiency parameter of an imaging plate,which was used to calculate the grayscale.Images were created by using grayscale,which trained the convolutional neural network to count the alpha tracks.The results demonstrated that the trained convolutional neural network can evaluate the alpha track counts based on the source and background images with a wider linear range,which was unaffected by the overlapping effect.The alpha track counts were unaffected by the fading effect within 60 min,where the calibrated formula for the fading effect was analyzed for 132.7 min.The detection efficiency of the trained convolutional neural network for inhomogeneous ^(241)Am sources(2π emission)was 0.6050±0.0399,whereas the efficiency curve of the photo-stimulated luminescence method was lower than that of the trained convolutional neural network.

ICON/MIGHTI与TIMED/SABER探测温度数据的对比

ICON卫星为临近空间环境特性研究、建模和预报提供了新数据.通过对ICON/MIGHTI与TIMED/SABER在90~105 km高度探测温度数据的比较,计算两者的年平均温度偏差和均方根误差,同时分析月平均温度偏差在不同月份中随高度和纬度的分布情况,为MIGHTI和SABER温度探测数据在临近空间大气建模和预报应用提供参考依据.结果表明,MIGHTI和SABER的温度垂直廓线变化趋势基本吻合,数值上有所差异.在12°S-42°N范围内,MIGHTI探测温度与SABER相比,在90~93 km时偏低,偏差最大值约2.5 K,在93~105 km偏高,偏差的绝对值最大约10 K.在不同季节,白天的温度偏差通常高于夜晚.SABER和MIGHTI的月平均温度偏差随季节和纬度的变化显著,夏季时的月平均温度偏差最大,且温度的均方根误差最大.

基于Count Sketch的预处理贪婪Kaczmarz方法

在贪婪Kaczmarz方法中,通过对系数矩阵进行正交三角分解引入右预处理子能够提高贪婪Kaczmarz方法的收敛速率。但在系数矩阵的行数远大于列数的情况下,正交三角分解的成本过高。为降低预处理的成本,通过引入Count Sketch变换,提出了基于Count Sketch的预处理贪婪Kaczmarz方法,并对新方法进行了收敛性分析。理论分析说明了新方法在系数矩阵条件数较大时比已有方法具有更好的收敛速率。数值实验验证了新方法的有效性。

TimeGAN-Informer长时机场能见度预测

能见度的预测对机场的业务决策、保障飞机的安全起降具有重要的意义。针对现有能见度预测模型预测时间较短的问题,提出一种基于TimeGAN Informer(Time Generative Adversarial Network-Informer)的机场能见度预测方法。利用2018—2022年气象和污染物数据,通过相关系数法和递归特征消除法提取出能见度的主要影响因素,使用TimeGAN时间序列生成对抗网络对数据进行扩充,并将Informer长时间序列预测模型应用于能见度预测。结果显示:当预测步长为1 d、2 d、3 d时,TimeGAN Informer的绝对误差(Mean Absolute Error,MAE)分别为2.42、3.13、3.57,比Informer分别降低了0.29、0.27、0.28,比长短时记忆网络(Long Short-Term Memory,LSTM)分别降低了0.28、0.49、0.63;均方根误差(Root Mean Square Error,RMSE)分别为3.03、3.7、4.09,比Informer分别降低了0.38、0.22、0.24,比长短时记忆网络(LSTM)分别降低了0.3、0.5、1.04;百分误差小于30%的分别占测试样本集的78.07%、70.68%、63.84%。尽管随着步长的增加预测效果变差,但在预测步长为3 d时,多数样本的预测误差仍小于30%,实现了对机场区域较为准确的长时能见度预测。

Modeling of Fluid Turbulence Modification Using Two-time-scale Dissipation Models and Accounting for the Particle Wake Effect

Presently developed two-phase turbulence models under-predict the gas turbulent fluctuation, because their turbulence modification models cannot fully reflect the effect of particles. In this paper, a two-time-scale dis- sipation model of turbulence modification, developed for the two-phase velocity correlation and for the dissipation rate of gas turbulent kinetic energy, is proposed and used to simulate sudden-expansion and swirling gas-particle flows. The proposed two-time scale model gives better results than the single-time scale model. Besides, a gas tur- bulence augmentation model accounting for the finite-size particle wake effect in the gas Reynolds stress equation is proposed. The proposed turbulence modification models are used to simulate two-phase pipe flows. It can prop- erly predict both turbulence reduction and turbulence enhancement for a certain size of particles observed in ex- periments.

基于TimeGAN数据增强的复杂过程故障分类方法

针对传统基于重构的故障分类方法在故障样本稀疏或失衡情况下效果不佳、故障子空间区分能力弱的问题,提出基于TimeGAN数据增强的复杂过程故障分类方法.针对小子样故障,使用TimeGAN对历史故障数据进行数据增强,生成与历史数据分布相似的虚拟故障样本;采用马氏距离评估虚拟样本的质量,剔除不可信样本,构造平衡的故障样本集.将故障样本映射到高维核空间,并在核空间中提取故障子空间.设计故障分类策略并定义4种故障分类性能评估指标以定量衡量算法的分类性能.Tennessee Eastman应用结果表明,所提数据增强方法可以有效扩充故障样本,进而提高故障重构率.与WGAN-GP和SMOTE方法进行对比,发现基于TimeGAN数据增强的故障分类方法具有更好的分类性能.

RPNet: Rice plant counting after tillering stage based on plant attention and multiple supervision network

Rice is a major food crop and is planted worldwide. Climatic deterioration, population growth, farmland shrinkage, and other factors have necessitated the application of cutting-edge technology to achieve accurate and efficient rice production. In this study, we mainly focus on the precise counting of rice plants in paddy field and design a novel deep learning network, RPNet, consisting of four modules: feature encoder, attention block, initial density map generator, and attention map generator. Additionally, we propose a novel loss function called RPloss. This loss function considers the magnitude relationship between different sub-loss functions and ensures the validity of the designed network. To verify the proposed method, we conducted experiments on our recently presented URC dataset, which is an unmanned aerial vehicle dataset that is quite challenged at counting rice plants. For experimental comparison, we chose some popular or recently proposed counting methods, namely MCNN, CSRNet, SANet, TasselNetV2, and FIDTM. In the experiment, the mean absolute error(MAE), root mean squared error(RMSE), relative MAE(rMAE) and relative RMSE(rRMSE) of the proposed RPNet were 8.3, 11.2, 1.2% and 1.6%, respectively,for the URC dataset. RPNet surpasses state-of-the-art methods in plant counting. To verify the universality of the proposed method, we conducted experiments on the well-know MTC and WED datasets. The final results on these datasets showed that our network achieved the best results compared with excellent previous approaches. The experiments showed that the proposed RPNet can be utilized to count rice plants in paddy fields and replace traditional methods.

Lightweight Fish Bait Particle Counting Method Based on Pruning and Shift Quantization

In the process of aquaculture,monitoring the number of fish bait particles is of great significance to improve the growth and welfare of fish.Although the counting method based on onvolutional neural network(CNN)achieve good accuracy and applicability,it has a high amount of parameters and computation,which limit the deployment on resource-constrained hardware devices.In order to solve the above problems,this paper proposes a lightweight bait particle counting method based on shift quantization and model pruning strategies.Firstly,we take corresponding lightweight strategies for different layers to flexibly balance the counting accuracy and performance of the model.In order to deeply lighten the counting model,the redundant and less informative weights of the model are removed through the combination of model quantization and pruning.The experimental results show that the compression rate is nearly 9 times.Finally,the quantization candidate value is refined by introducing a power-of-two addition term,which improves the matches of the weight distribution.By analyzing the experimental results,the counting loss at 3 bit is reduced by 35.31%.In summary,the lightweight bait particle counting model proposed in this paper achieves lossless counting accuracy and reduces the storage and computational overhead required for running convolutional neural networks.

LabVIEW-based auto-timing counts virtual instrument system with ORTEC 974 Counter/Timer

In order to achieve the auto-timing counts measurement of nuclear radiation using ORTEC 974 Counter/Timer, an auto-timing counts virtual instrument system based on the LabVIEW virtual instrument development platform and GPIB instrument control and transmission bus protocol is designed in this paper. By introducing software timing technique, the minimum time base of factory setting improves from 0.1 s to 0.03 s. The timing counts performance and longtime stability are also discussed in detail. The automatic data recording and saving facilitates data analysis and processing. Its real-time display and statistic function is very convenient for monitoring the nuclear radiation.

A new software for automated counting of glistenings in intraocular lenses in vivo

AIM:To assess the performance of a bespoke software for automated counting of intraocular lens(IOL)glistenings in slit-lamp images.METHODS:IOL glistenings from slit-lamp-derived digital images were counted manually and automatically by the bespoke software.The images of one randomly selected eye from each of 34 participants were used as a training set to determine the threshold setting that gave the best agreement between manual and automatic grading.A second set of 63 images,selected using randomised stratified sampling from 290 images,were used for software validation.The images were obtained using a previously described protocol.Software-derived automated glistenings counts were compared to manual counts produced by three ophthalmologists.RESULTS:A threshold value of 140 was determined that minimised the total deviation in the number of glistenings for the 34 images in the training set.Using this threshold value,only slight agreement was found between automated software counts and manual expert counts for the validating set of 63 images(κ=0.104,95%CI,0.040-0.168).Ten images(15.9%)had glistenings counts that agreed between the software and manual counting.There were 49 images(77.8%)where the software overestimated the number of glistenings.CONCLUSION:The low levels of agreement show between an initial release of software used to automatically count glistenings in in vivo slit-lamp images and manual counting indicates that this is a non-trivial application.Iterative improvement involving a dialogue between software developers and experienced ophthalmologists is required to optimise agreement.The results suggest that validation of software is necessary for studies involving semi-automatic evaluation of glistenings.

A Computer Vision-Based System for Metal Sheet Pick Counting

Inventory counting is crucial to manufacturing industries in terms of inventory management,production,and procurement planning.Many companies currently require workers to manually count and track the status of materials,which are repetitive and non-value-added activities but incur significant costs to the companies as well as mental fatigue to the employees.This research aims to develop a computer vision system that can automate the material counting activity without applying any marker on the material.The type of material of interest is metal sheet,whose shape is simple,a large rectangular shape,yet difficult to detect.The use of computer vision technology can reduce the costs incurred fromthe loss of high-value materials,eliminate repetitive work requirements for skilled labor,and reduce human error.A computer vision system is proposed and tested on a metal sheet picking process formultiple metal sheet stacks in the storage area by using one video camera.Our results show that the proposed computer vision system can count the metal sheet picks under a real situation with a precision of 97.83%and a recall of 100%.

A Deep Learning-Based Crowd Counting Method and System Implementation on Neural Processing Unit Platform

In this paper, a deep learning-based method is proposed for crowdcountingproblems. Specifically, by utilizing the convolution kernel densitymap, the ground truth is generated dynamically to enhance the featureextractingability of the generator model. Meanwhile, the “cross stage partial”module is integrated into congested scene recognition network (CSRNet) toobtain a lightweight network model. In addition, to compensate for the accuracydrop owing to the lightweight model, we take advantage of “structuredknowledge transfer” to train the model in an end-to-end manner. It aimsto accelerate the fitting speed and enhance the learning ability of the studentmodel. The crowd-counting system solution for edge computing is alsoproposed and implemented on an embedded device equipped with a neuralprocessing unit. Simulations demonstrate the performance improvement ofthe proposed solution in terms of model size, processing speed and accuracy.The performance on the Venice dataset shows that the mean absolute error(MAE) and the root mean squared error (RMSE) of our model drop by32.63% and 39.18% compared with CSRNet. Meanwhile, the performance onthe ShanghaiTech PartB dataset reveals that the MAE and the RMSE of ourmodel are close to those of CSRNet. Therefore, we provide a novel embeddedplatform system scheme for public safety pre-warning applications.

Improvement of Counting Sorting Algorithm

By analyzing the internal features of counting sorting algorithm. Two improvements of counting sorting algorithms are proposed, which have a wide range of applications and better efficiency than the original counting sort while maintaining the original stability. Compared with the original counting sort, it has a wider scope of application and better time and space efficiency. In addition, the accuracy of the above conclusions can be proved by a large amount of experimental data.

Robust Counting in Overcrowded Scenes Using Batch-Free Normalized Deep ConvNet

The analysis of overcrowded areas is essential for flow monitoring,assembly control,and security.Crowd counting’s primary goal is to calculate the population in a given region,which requires real-time analysis of congested scenes for prompt reactionary actions.The crowd is always unexpected,and the benchmarked available datasets have a lot of variation,which limits the trained models’performance on unseen test data.In this paper,we proposed an end-to-end deep neural network that takes an input image and generates a density map of a crowd scene.The proposed model consists of encoder and decoder networks comprising batch-free normalization layers known as evolving normalization(EvoNorm).This allows our network to be generalized for unseen data because EvoNorm is not using statistics from the training samples.The decoder network uses dilated 2D convolutional layers to provide large receptive fields and fewer parameters,which enables real-time processing and solves the density drift problem due to its large receptive field.Five benchmark datasets are used in this study to assess the proposed model,resulting in the conclusion that it outperforms conventional models.

Brain Time Stack图像融合技术在CT中的应用

目的:分析Brain Time Stack图像融合技术在CT中的应用。方法:选取2021年3月—2022年9月衡水市第四人民医院收治的50例CT检查患者作为研究对象。所有患者进行CT检查并进行Brain Time Stack后处理。比较四组不同部位CT值、标准差(SD)、信噪比(SNR)。比较四组图像主观质量评分。分析不同部位CT值、SD、SNR与图像主观质量评分的相关性。结果:B组的延髓、额叶灰质、额叶白质、小脑内侧、小脑外侧、颞肌肌肉CT值明显低于A组;C组的延髓、脑室、额叶白质、小脑内侧、小脑外侧、颞肌肌肉CT值高于A组;D组延髓、额叶灰质、颞肌肌肉CT值明显低于A组,脑室、额叶白质、小脑外侧CT值明显高于A组;C组延髓、额叶灰质、额叶白质、小脑内侧、小脑外侧、颞肌肌肉CT值明显高于B组;D组延髓、脑室、额叶白质、小脑内侧、小脑外侧、颞肌肌肉CT值明显高于B组;D组延髓、额叶灰质、额叶白质、小脑内侧、小脑外侧、颞肌肌肉CT值明显低于C组;D组脑室CT值明显高于C组,差异有统计学意义(P<0.05)。B组、C组、D组延髓、脑室、额叶灰质、额叶白质、小脑内侧、小脑外侧、颞肌肌肉SD值明显低于A组;C组延髓、脑室、额叶白质、小脑内侧、小脑外侧、颞肌肌肉SD值均明显高于B组;C组额叶灰质SD明显低于B组;D组延髓、脑室、额叶灰质、额叶白质、小脑内侧、小脑外侧、肌肉SD均明显低于B组、C组,差异有统计学意义(P<0.05)。B组、C组、D组延髓、脑室、额叶灰质、额叶白质、小脑内侧、小脑外侧、颞肌肌肉SNR均明显高于A组;C组、D组延髓、额叶灰质、额叶白质、小脑内侧、小脑外侧、颞肌肌肉SNR值明显高于B组;C组、D组脑室SNR明显低于B组;D组延髓、脑室、额叶灰质、额叶白质、小脑内侧、小脑外侧、颞肌肌肉SNR明显高于C组,差异有统计学意义(P<0.05)。D组图像主观质量评分最高,差异有统计学意义(P<0.05)。延髓、脑室、额叶灰质、额叶白质、小脑内侧、小脑外侧及颞肌肌肉SD与主观质量评分呈明显负相关,SNR与主观质量评分间呈明显正相关,差异有统计学意义(P<0.05)。结论:利用Brain Time Stack图像融合技术对头部CT扫描检查图像处理,动脉期结合前一期及后一期的图像数据在处理后具有更好的质量和更少的噪音。

Association of daily sitting time and leisure-time physical activity with body fat among U.S.adults

Background:Prolonged sitting and reduced physical activity lead to low energy expenditures.However,little is known about the joint impact of daily sitting time and physical activity on body fat distribution.We investigated the independent and joint associations of daily sitting time and physical activity with body fat among adults.Methods:This was a cross-sectional analysis of U.S.nationally representative data from the National Health and Nutrition Examination Survey2011-2018 among adults aged 20 years or older.Daily sitting time and leisure-time physical activity(LTPA)were self-reported using the Global Physical Activity Questionnaire.Body fat(total and trunk fat percentage)was determined via dual X-ray absorptiometry.Results:Among 10,808 adults,about 54.6%spent 6 h/day or more sitting;more than one-half reported no LTPA(inactive)or less than 150 min/week LTPA(insufficiently active)with only 43.3%reported 150 min/week or more LTPA(active)in the past week.After fully adjusting for sociodemographic data,lifestyle behaviors,and chronic conditions,prolonged sitting time and low levels of LTPA were associated with higher total and trunk fat percentages in both sexes.When stratifying by LTPA,the association between daily sitting time and body fat appeared to be stronger in those who were inactive/insuufficiently active.In the joint analyses,inactive/insuufficiently active adults who reported sitting more than 8 h/day had the highest total(female:3.99%(95%confidence interval(95%CI):3.09%-4.88%);male:3.79%(95%CI:2.75%-4.82%))and trunk body fat percentages(female:4.21%(95%CI:3.09%-5.32%);male:4.07%(95%CI:2.95%-5.19%))when compared with those who were active and sitting less than 4 h/day.Conclusion:Prolonged daily sitting time was associated with increased body fat among U.S.adults.The higher body fat associated with 6 h/day sitting may not be offset by achieving recommended levels of physical activity.