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.

Multi-Branch Fault Line Location Method Based on Time Difference Matrix Fitting

The distribution network exhibits complex structural characteristics,which makes fault localization a challenging task.Especially when a branch of the multi-branch distribution network fails,the traditional multi-branch fault location algorithm makes it difficult to meet the demands of high-precision fault localization in the multi-branch distribution network system.In this paper,the multi-branch mainline is decomposed into single branch lines,transforming the complex multi-branch fault location problem into a double-ended fault location problem.Based on the different transmission characteristics of the fault-traveling wave in fault lines and non-fault lines,the endpoint reference time difference matrix S and the fault time difference matrix G were established.The time variation rule of the fault-traveling wave arriving at each endpoint before and after a fault was comprehensively utilized.To realize the fault segment location,the least square method was introduced.It was used to find the first-order fitting relation that satisfies the matching relationship between the corresponding row vector and the first-order function in the two matrices,to realize the fault segment location.Then,the time difference matrix is used to determine the traveling wave velocity,which,combined with the double-ended traveling wave location,enables accurate fault location.

Multi-branched carbazole derivatives for two-photon absorption and two-photon excited fluorescence

Carbazole-core multi-branched chromophores 9-ethyl- 3, 6-bis （ 2- { 4- [ 5- （4-tert-butyl-phenyl） - [ 1, 3, 4 ] oxadiazol-2-yl ] - phenyl }-vinyl） -carbazole（3） and 9-ethyl-3-（ 2- {4-[ 5-（4-tert-butyl- phenyl） -[ 1, 3, 4 ] oxadiazol-2-yl ] -phenyl }-vinyl ） -carbazole （ 2 ） are synthesized through Wittig reaction and characterized by nuclear magnetic resonance（NMR）and infrared（IR）. The two- photon absorption properties of chromophores are investigated. These chromophores exhibit large two-photon absorption crosssections and strong blue two-photon excited fluorescence. The cooperative enhancement of two-photon absorption（TPA） in the multi-branched structures is observed. This enhancement is partly attributed to the electronic coupling between the branches. The electronic push-pull structures in the arm and their cooperative effects help the extended charge transfer for TPA.

Wireless Interference Classification with Low Complexity Multi-Branch Networks

In non-cooperative communication systems,wireless interference classification(WIC)is one of the most essential technologies.Recently,deep learning(DL)based WIC methods have been proposed.However,conventional DL-based WIC methods have high computational complexity and unsatisfactory accuracy,especially when the interference-tonoise ratio(INR)is low.To this end,we propose three effective approaches.Firstly,we introduce multibranch convolutional neural networks(CNNs)for interference recognition.The multi-branch CNN is constructed by repeating a layer that aggregates several transformations with the same topology,and it notably improves the recognition ability for WIC.Our design avoids the carefully crafted selection of each transformation.Unfortunately,multi-branch CNNs are computationally expensive and memory-inefficient.To this end,we further propose Low complexity multibranch networks(LCMN),which are mathematically equivalent to multi-branch CNNs but maintain low computing costs and efficient inference.Thirdly,we present novel loss function,which encourages networks to have consistent prediction probabilities for samples with high visual similarities,resulting in increasing recognition accuracy of LCMN.Experimental results demonstrate the proposed methods consistently boost the classification performance of WIC without substantially increasing computational overhead compared to traditional DL-based methods.

Multi-Branch Deepfake Detection Algorithm Based on Fine-Grained Features

With the rapid development of deepfake technology,the authenticity of various types of fake synthetic content is increasing rapidly,which brings potential security threats to people’s daily life and social stability.Currently,most algorithms define deepfake detection as a binary classification problem,i.e.,global features are first extracted using a backbone network and then fed into a binary classifier to discriminate true or false.However,the differences between real and fake samples are often subtle and local,and such global feature-based detection algorithms are not optimal in efficiency and accuracy.To this end,to enhance the extraction of forgery details in deep forgery samples,we propose a multi-branch deepfake detection algorithm based on fine-grained features from the perspective of fine-grained classification.First,to address the critical problem in locating discriminative feature regions in fine-grained classification tasks,we investigate a method for locating multiple different discriminative regions and design a lightweight feature localization module to obtain crucial feature representations by augmenting the most significant parts of the feature map.Second,using information complementation,we introduce a correlation-guided fusion module to enhance the discriminative feature information of different branches.Finally,we use the global attention module in the multi-branch model to improve the cross-dimensional interaction of spatial domain and channel domain information and increase the weights of crucial feature regions and feature channels.We conduct sufficient ablation experiments and comparative experiments.The experimental results show that the algorithm outperforms the detection accuracy and effectiveness on the FaceForensics++and Celeb-DF-v2 datasets compared with the representative detection algorithms in recent years,which can achieve better detection results.

Parameters Optimization of Multi-Branch Horizontal Well Basing on Streamline Simulation

As a highly efficient production method, the technique of multi-branch horizontal well is widely used in low permeability reservoirs, heavy oil reservoirs, shallow layer reservoirs and multi-layer reservoirs, because it can significantly improve the productivity of a single well, inhibit coning and enhance oil recovery. Study on sweep efficiency and parameters optimization of multi-branch horizontal well is at the leading edge of research. Therefore, the study is important for enhancing oil recovery and integral exploitation benefit of oil fields. In many applications, streamline simulation shows particular advantages over finite-difference simulation. With the advantages of streamline simulation such as its ability to display paths of fluid flow and acceleration factor in simulation, the flooding process is more visual. The communication between wells and flooding area has been represented appropriately. This method has been applied to the XS9 reservoir in Daqing Oilfield. The production history of this reservoir is about 10 years. The reservoir is maintained above bubble point so that the simulation meets the slight compressibility assumption. New horizontal wells are drilled following this rule.

Optimization for Hydroforming Loading Paths of Parallel Multi-branch Tubes Based on Grey System Theory

According to characteristic of hydroforming of parallel multi-branch tubes,multi-objective problems were transformed to single objective problem of relational grade comparison by grey system theory.Two different objectives were selected,according to the principle that process parameters were optimal which of grey relational grade were maximum,the optimal loading parameters under different objective condition were obtained,and loading paths were optimized.The results indicated that parallel multi-branch tubes hydroformed under loading paths optimized by grey system theory could meet with the requirement that objective was optimal.And the optimal loading paths under different objectives were different,and the appropriate objective should be selected according to forming characteristic.

Linear and Nonlinear Optical Properties of Novel Multi-branched Oligomers

We investigate the fluorene-vinylene unit dependent photo-physical properties of multi- branched truxene based oligomers （Tr-OFVn, n=1-4） employing steady-state absorption and emission spectroscopy, transient absorption spectroscopy, two-photon fluorescence, and z-scan technique. The results show that the increasing of fluorene-vinylene unit leads to a red-shift in the spectra of absorption and fluorescence, and shortens the excited state lifetime. Meanwhile, two-photon fluorescence efficiency and two-photon absorption cross section of truxene based oligolners gradually enhance in company with the extension of π- conjugated length. In addition, the values of two-photon absorption cross section modeled on the sum-over-state approach agree well with the experimental ones. The results indicate multi-branched truxene based oligomers bearing organic materials for two-photon applications.

Optical Properties and Response Mechanism Analysis of Multi-branched Fluorescent Probes Based on Intramolecular Charge Transfer

In this work, the optical properties of fluorescent probes used for detection of biothiol were studied by employing time-dependent density functional theory. By calculating the single photon absorption and emission properties of probe Mol.1, Mol.2 and Mol.3 before and after reaction with cysteine and homocysteine, we have investigated the effect of carboncarbon triple bond and benzene ring on the properties of fluorescent probes. It is found that the oscillator strength of probe molecules increases gradually with the improvement of the structure of the electron donor triphenylamine and the addition of carbon-carbon triple bonds, and better properties of fluorescence probes have also been demonstrated. At the same time, the effect of different number of side branches on the molecular properties of the probe was also studied. The results showed that compared with single-branched molecule Z1 and tribranched probe Mol.3, two side probe molecules Z2 had higher oscillator strength and better detection effect. In addition, the new single-branched probe Mol.4 with the addition of carbon-carbon triple bonds and benzene rings has better probe properties and simpler structure than the tribranched probe Mol.3.

Proppant transport law in multi-branched fractures induced by volume fracturing

To further clarify the proppant transport and placement law in multi-branched fractures induced by volume fracturing, proppant transport simulation experiments were performed with different fracture shapes, sand ratios, branched fracture opening time and injection sequence of proppants in varied particle sizes. The results show that the settled proppant height increases and the placement length decreases in main fractures as the fracturing fluid diverts gradually to the branched fractures at different positions. The flow rate in branched fractures is the main factor affecting their filling. The diverion to branched fractures leads to low flow rate and poor filling of far-wellbore branched fractures. The inclined fracture wall exerts a frictional force on the proppant to slow its settlement, thus enhancing the vertical proppant distribution in the fracture. The increase of sand ratio can improve the filling of near-wellbore main fracture and far-wellbore branched fracture and also increase the settled proppant height in main fracture. Due to the limitation of fracture height, when the sand ratio increases to a certain level, the increment of fracture filling decreases. When branched fracture is always open(or extends continuously), the supporting effect on the branched fractures is the best, but the proppant placement length within the main fractures is shorter. The fractures support effect is better when it is first closed and then opened(or extends in late stage) than when it is first opened and then closed(or extends in early stage). Injecting proppants with different particle sizes in a specific sequence can improve the placement lengths of main fracture and branched fracture. Injection of proppants in an ascending order of particle size improves the near-wellbore fracture filling, to a better extent than that in a descending order of particle size.

基于多分支增强和融合注意力机制的水下图像增强算法

由于水对光的折射和吸收,水下图像通常会出现严重的退化,如色偏、模糊、能见度低等.为了提高水下图像的可视性,提出了一种基于多分支增强和融合注意力机制的水下图像增强网络MBFA-GAN.首先,通过分析水下图像的色彩退化和模糊因素,设计了青品色温修复模块和模糊恢复模块对水下图像进行色彩矫正和模糊恢复.然后,基于对多个分支特征的互补性考虑,采用循环合并策略将多个分支增强的特征利用自适应融合模块进行融合,逐步增强图像细节.最后,设计了融合注意力模块,用于深度挖掘图像在通道维度和像素维度的相关性矩阵,以提高增强图像的真实性.实验结果表明,与现有算法相比,提出的水下图像增强算法去模糊效果较好且颜色更真实,可以有效改善水下图像色偏和模糊的问题.

基于特征增强的双重注意力去雾网络

针对现有去雾方法处理的图像细节模糊和色彩偏差等问题,提出了一种基于特征增强的双重注意力去雾网络。该网络采用编码器-解码器结构,设计了一个双重注意力特征增强模块,其中,利用Ghost模块替代非线性卷积,实现模型轻量化处理,通过RFB充分融合不同尺度的特征,实现均匀去雾,引入双重注意力实现信息跨通道与空间交互,保证模型性能和抑制噪声特征。使用RESIDE数据集对网络进行训练和测试。实验结果表明,所提算法在主观视觉和客观评价指标上均有优异表现,能有效地提升网络的特征提取能力,实现对不同场景雾图的色彩恢复,增强图像的对比度和清晰度。

多分支加权的Transformer霍克斯过程

时序点过程作为一种异步事件序列建模的重要方法,目前已经在地震、医疗等领域得到了广泛的应用。Transformer等深度学习模型的引入使得模型的预测性能得到了突破性进步,为了解决基于Transformer的霍克斯过程模型在对事件序列建模时出现的学习偏差问题,提出了多分支加权的Transformer霍克斯过程模型(multi-branch weighted transformer Hawkes process,MWTHP)。基于多分支的思想,通过为不同角度下学习到的依赖关系赋予差异化的重要性,提高模型对事件序列的建模能力;为了应对基于Transformer的霍克斯过程模型的局部感知能力较差问题,构建了一种基于因果卷积的局部感知增强网络,改善了模型对事件序列局部上下文信息的关注能力。通过在多个合成数据集和真实世界数据集上进行实验,采用对数似然值、时间均方根误差、事件类型准确率等指标进行综合评价。实验结果验证了所提模型的性能优于其他基准模型;通过消融实验,证明了局部感知增强网络的有效性。

Facile Synthesis of Multi-Branched Gold Nanostructures through a TBAB-Assisted Route in Aqueous Solution and Their SERS Property

Facile synthesis of multi-branched gold nanostructures by using the tetrabutyl ammonium bromide （TBAB） as a capping agent is described. The reaction is carried out in a one-step process at mild temperature. Gold nanostructures with more than six sharp branches ranging from 70 to 130 nm in length are synthesized in high yield. It is proposed that the relative weak adsorption capacity of TBAB leads to the incompletely covered gold surface and the growth of nanoparticles occurs on the uncovered gold surface, and therefore short branches appear consequently. Then positively charged TBAB layers on the gold surfaces prevent the branches from aggregating with each other which stimulates the branch growth. The prepared branched gold nanoparticles show efficient surface-enhanced Raman scattering （SERS） properties. Low temperature （4 ℃） is unfavorable to the formation of multi-branched gold nanostructures, and only thin small irregular plate-like nanoparticles are produced. The addition of SDS in TBAB aqueous solution results in forming SDS micelles at much lower concentration of SDS （0.4 mmol/L） as compared to that in pure water, and short branched gold nanoparticles are obtained in the SDS-TBAB system.

Thermodynamic Transitions of Antiferromagnetic Ising Model on the Fractional Multi-branched Husimi Recursive Lattice

The multi-branched Husimi recursive lattice is extended to a virtual structure with fractional numbers of branches joined on one site. Although the lattice is undrawable in real space, the concept is consistent with regular Husimi lattice. The Ising spins of antiferromagnetic interaction on such a set of lattices are calculated to check the critical temperatures(Tc) and ideal glass transition temperatures(Tk) variation with fractional branch numbers. Besides the similar results of two solutions representing the stable state(crystal) and metastable state(supercooled liquid)and indicating the phase transition temperatures, the phase transitions show a well-defined shift with branch number variation. Therefore the fractional branch number as a parameter can be used as an adjusting tool in constructing a recursive lattice model to describe real systems.

结合特征融合和通道注意力的多分支换装行人重识别

换装行人重识别(CC Re-ID)是行人重识别中的一个新兴研究课题,旨在找出被换衣的行人。当前方法主要集中在使用多模态数据辅助解耦表征学习,如通过脸、步态、身体轮廓等辅助数据解耦行人自身属性以减少服装影响,但这些方法泛化能力较差,需要大量额外工作。此外,仅使用原始数据的方法对于相关信息的提取不够充分,性能较弱。针对CC Re-ID存在的上述问题,提出一种结合特征融合和通道注意力的多分支换装行人重识别方法(MBFC)。通过在主干网络中融入通道注意力机制,在特征通道层面学习关键信息,设计局部与全局特征融合方法以提高网络对行人细粒度特征的提取能力。此外,MBFC模型采用多分支结构,使用服装对抗损失、交叉熵标签平滑损失等多种损失函数引导模型学习与服装无关的信息,减少服装对模型的影响,从而提取到更有效的行人信息。在PRCC和VC-Clothes数据集上进行广泛实验,结果表明,所提模型在RANK-1和平均精度均值(mAP)指标上优于对比的CC Re-ID方法。

Effects of the structure of the branches on the two-photon absorption properties for the multi-branched molecules with nitrogen (N) as coupling center

In order to investigate the effects of the structure of branches on the TPA properties for multi-branched molecules, the TPA cross section is calculated by using ZINDO/SOS method. The investigated mole- cules have different branches (chomorfores based on stilbene, dithienothiophene and flourene) with nitrogen(N) as coupling center. The results show that the cooperative enhancement in multi-branched molecules depends on the structures of the branches and the structures of branches play an important role in the enhancement of the TPA cross section. The designed molecules with stilbene and dithie- nothiophene as branched possess relatively larger two-photon absorption cross sections.

Airflow distribution law of multi-branch pipe of pneumatic rice direct seeder based on dimensional analysis

In order to investigate the flow characteristics and distribution law of airflow in multi-branch pipe of pneumatic rice precision direct seeder and obtain the mathematical model between airflow parameters and pipe geometry structure.In this study,the airflow flow law of the multi-branch pipeline of the pneumatic system was studied,the mechanism of airflow flow in the multi-branch pipe was analyzed,and it was clarified that the main factors affecting the airflow flow in the pipe,namely,air density,air dynamic viscosity,the total flow rate of the inlet branch pipe,the length of the closed end of the header,the inner diameter of the outlet branch pipe,and the outlet branch pipe spacing.Numerical simulations were carried out using Fluent simulation software to elucidate the cause of multi-branch pipes of uneven distribution of airflow in multi-branch pipes,the empirical equation among these factors and the flow velocity of the outlet branch pipe are established by dimensional analysis method.The bench test results show that the established empirical equations are applicable in the following ranges:0.018 m3/s≤Q≤0.054 m3/s,0.045 m≤d≤0.05 m,0.075 m≤L≤0.125 m,0.7 m≤Y 1≤0.875 m(0.5 m≤Y 2≤0.75 m,0.36 m≤Y 3≤0.45 m),the prediction accuracy can be controlled within 10%of the empirical formula,which can provide a reference for the prediction and optimization design of outlet velocity of the multi-branch pipe.

基于融合MBAM与YOLOv5的PCB缺陷检测方法

随着电子信息产业迅速发展,PCB行业作为电子信息产业的基础,其产品质量对后续生产的电子产品有着决定性影响。针对PCB缺陷目标较小,缺陷类型多,特征不明显,在实际生产过程中易产生误检、漏检等问题,提出了一种多分支注意力MBAM模块方法,在3个不同维度对特征图进行关注,以增强特征提取的能力,对缺陷区域给予更多的注意力表示。通过改进YOLOv5结构,将MBAM与YOLOv5网络结合,有效的提升了对PCB中小目标的检测性能。最后通过在网络不同位置添加MBAM模块进行对比实验,选取了最佳的添加位置。通过在PCB缺陷数据集上的实验结果表明,改进后的PCB缺陷检测算法具有良好的检测性能,优于其他对比算法,最终的AP达到了96.7%,对比标准YOLOv5的94.7%提高了2个百分点,其他项指标均有涨点,在保持检测速度基本不变的情况下,精准地识别PCB缺陷类型。

基于“大数据+人工智能”的多院区医院行政管理工具探索

公立医院大多自有大数据平台和办公自动化系统,但两个系统独立提供服务,且缺乏使用指南和智能查询等功能,造成职工办事效率和平台使用率双低的局面;在多院区医院管理中,跨院区业务办理存在诸多不便,亟须升级行政公共服务能力,改革创新管理模式。在智慧医疗发展的大背景下,多数医院医疗服务的智慧化程度较高而行政办公的信息化程度不足。本研究拟构建基于“大数据+人工智能”的医院行政助手,以期利用信息化手段提升管理效率、便捷职工办事,推动医院精细化管理和决策。同时为多院区医院的行政办公服务提供经验参考。