Under the constraint of supplying materials,the dispatching of emergency materials with multiple supply points,multiple demand points,and multiple types needs to guarantee the high efficiency of distribution and improve the satisfaction degree of each demand point at the same time.Therefore,by establishing a scheduling model with transportation cost and demand satisfaction as the ob-jectives,we design an evolutionary learning algorithm(ELA)to improve the model's solution effect and accuracy,give an efficient scheduling plan,and introduce transshipment points to further reduce the transportation cost of each demand point so as to optimize the scheduling plan.The experimental analysis shows that the decision variable mapping encoding proposed in the first stage accelerates the solution speed by avoiding the generation of ineffective allocation schemes,and the designed ELA can reduce the transportation cost and improve the degree of demand satisfaction to a larger extent,and the given scheduling scheme reduces the transportation cost by 13.6%and improves the degree of demand satisfaction by 18.4%compared with the traditional GSA.In the second stage,after optimization using the savings method,the transportation cost of the scheduling plan is further reduced by 11.1%in the bi-objective scheduling plan.Combined with the actual scheduling demand,the two given schemes,the dual-objective scheduling scheme is suitable for the actual demand of reducing transportation cost,while the maximum demand satisfaction scheme is more helpful for improving the demand satisfaction,and the two schemes provide more valuable references for the actual scheduling demand.
To solve the problem of the poor performance of a flame detection algorithm in a complex forest background, such as poor detection performance, insensitivity to small targets, and excessive computational load, there is an urgent need for a lightweight, high-accuracy, real-time detection system. This paper introduces a lightweight object-detection algorithm called GS-YOLOv5s, which is based on the YOLOv5s baseline model and incorporates a multi-scale feature fusion knowledge distillation architecture. Firstly, the ghost shuffle convolution bottleneck is applied to obtain richer gradient information through branching. Secondly, the WIoU loss function is used to address the issues of GIoU related to model optimization, slow convergence, and inaccurate regression. Finally, a knowledge distillation algorithm based on feature fusion is employed to further improve its accuracy. Experimental results based on the dataset show that compared to the YOLOv5s baseline model, the proposed algorithm reduces the number of parameters and floating-point operations by approximately 26% and 36%, respectively. Moreover, it achieved a 3.1% improvement in mAP0.5 compared to YOLOv5s. The experiments demonstrate that GS-YOLOv5s, based on multi-scale feature fusion, not only enhances detection accuracy but also meets the requirements of lightweight and real-time detection in forest fire detection, commendably improving the practicality of flame-detection algorithms.
Aiming at the problems of the intrusion detection system based on deep learning,such as poor local feature extraction effect,single extraction dimension,prediction accuracy to be improved and high cost of model calculation,an intrusion detection model based on Dilated Convolutional Network(DCN)and Bidirectional Long Short Memory Network(BiLSTM)is proposed.Firstly,an improved three-layer DCN structure was used for local feature extraction,which solved the problem of poor local feature extraction performance and single extraction dimension.Secondly,the method combining expansion convolution with BiLSTM was used to solve the problem that the prediction accuracy needs to be improved.Finally,grouping convolution technology was introduced into the model to make the model lightweight,which solves the problem of high computational cost.The experiments were conducted on the public dataset CICIDS2017.After comparing traditional models with existing intrusion detection methods,the proposed model has excellent performance.The high prediction accuracy,recall rate,and F1 value of the proposed model demonstrate its effectiveness and feasibility.
In the underwater environment, the color distortion and low contrast of the image are caused by environmental problems such as light absorption and scattering, which leads to the degradation of image quality. In order to improve the visual effect of the image, this paper proposed a color correction underwater image enhancement algorithm based on convolutional neural network. Firstly, a new underwater imaging model was used to correct the color cast problem of underwater images. Then, the convolutional neural network is used to extract the channel features of the corrected image, and the channel features are re-weighted by the multi-scale attention module to enhance the consistency of different feature maps, and the feature fusion is performed with the color corrected image. Finally, the image enhancement effect was improved by the fusion of features through the reconstruction calculation module. Experimental results show that the proposed algorithm can better correct the color distortion of the image and improve the image contrast. The main advantage is that the running speed of the proposed algorithm is two times faster than other advanced underwater image enhancement methods.
At present, license plate recognition algorithm under restricted conditions is relatively mature and widely used in various license plate recognition system. Due to the influence of factors such as large differences in shooting angles and vehicle motion blur, Chinese license plate recognition is quite challenging. In response to the above problems, this research abandoned the single end-to-end deep learning license plate recognition method, and proposed a step-by-step license plate recognition algorithm that integrated detection and classification, and utilized a level-by-level object detection strategy combined with character classification to predict the characters of the license plate result. On the basis of the above, a multi-anchor character position regression algorithm was proposed to further accurately regress the local area position information of all license plate characters. At the same time, in order to meet the needs of character detection and character classification, as well as the imbalance of the existing license plate datasets, this study contributed a series of supporting license plate datasets. According to the published publications, this study contributed the first large-scale character-level annotated license plate dataset. Extensive experiments show that the method in this study can reach the current state-of-the-art on different datasets. If accepted, the dataset will be publicly available at https://gitee.com/wust30405/lpdataset.
For the underground integrated pipe gallery containing natural gas pipelines,the current daily management of leakage accident risk mainly includes real-time monitoring,regular detection,emergency plan preparation and risk assessment by using various sensing devices,and sometimes emergency drills under different scenarios based on the deduction of disaster chain and fault tree.These management methods urgently need to scientifically simulate and deduce the micro process of risk evolution under the set scenario based on the model of gas leakage diffusion and lack process details and quantitative consequence assessment in the prediction and early warning of accident development.We use the particle system simulation method in the game engine Unity3D to simulate the natural gas leakage and diffusion process inside the pipe gallery based on a certain diffusion physical model and realize the construction of the pipe gallery model and the setting of particle system parameters.The temporal and spatial dynamic visualization of the leakage natural gas concentration distribution in the pipe gallery under different leakage scenarios is realized by setting the scenario parameters.The simulation method can provide quantitative scientific basis for emergency plan formulation,emergency drill and emergency rescue of natural gas leakage accident in underground integrated pipe gallery.
Because pigmented skin lesion image classification based on manually designed convolutional neural networks (CNNs) requires abundant experience in neural network design and considerable parameter tuning, we proposed the macro operation mutation-based neural architecture search (OM-NAS) approach in order to automatically build a CNN for image classification of pigmented skin lesions. We first used an improved search space that was oriented toward cells and contained micro and macro operations. The macro operations include InceptionV1, Fire and other well-designed neural network modules. During the search process, an evolutionary algorithm based on macro operation mutation was employed to iteratively change the operation type and connection mode of parent cells so that the macro operation was inserted into the child cell similar to the injection of virus into host DNA. Ultimately, the searched best cells were stacked to build a CNN for the image classification of pigmented skin lesions, which was then assessed on the HAM10000 and ISIC2017 datasets. The test results showed that the CNN built with this approach was more accurate than or almost as accurate as state-of-the-art (SOTA) approaches such as AmoebaNet, InceptionV3 + Attention and ARL-CNN in terms of image classification. The average sensitivity of this method on the HAM10000 and ISIC2017 datasets was 72.4% and 58.5%, respectively.
Aiming at the problem that the traditional U-Net network has poor segmentation effect on tumors with small volume and fuzzy tissue edges. The residual structure is introduced into the up-sampling and down-sampling modules of the original U-Net network to extract more feature information and prevent network degradation; The attention mechanism is introduced before each jump connection to focus attention on the region of interest and suppress redundant features; Group Normalization (GN) is used to replace the commonly used Batch Normalization (BN) to reduce the impact of excessively small batch size on network accuracy; focal Tversky Loss function is used to improve the data imbalance in liver tumor segmentation. Finally, the improved model is trained and tested on LiTS2017 data sets. The experimental results show that compared with the traditional U-Net, the Dice index of liver and tumor segmentation of the improved model proposed in this paper has increased by 5.14% and 2.63% respectively, and the recall rate has increased by 1.8% and 9.05%.
Protein structure determines function, and the dihedral angles formed by N-C and C-C bonds in the protein backbone play an important role in protein three-dimensional structure and spatial conformation. Starting from the protein primary sequence, prediction of backbone dihedral angle can accelerate the effective sampling of low-energy structural conformational space and greatly advance the 3D structure prediction, which can be used as an effective and rapid aid for biological experiments. With more protein biological samples and improved computational performance, deep learning methods have been widely applied to protein dihedral angle prediction in recent years. To deeply understand this work, what are detailed introduced are feature representations of protein residues, computational methods for dihedral angle processing, evaluation metrics and common datasets. And the recent research progress based on deep learning is rigorously reviewed in terms of network structure design, input feature representation, model generalization performance, etc. The effectiveness and shortcoming of each algorithm are also compared and analyzed. Upon above analysis, the future research field and application prospect are presented.
命名实体是构建产业企业画像和产业知识图谱的重要依据,为解决现有方法在有色冶金领域命名实体识别任务当中无法充分提取文本语义特征、没有充分利用标签当中的先验知识和嵌套命名实体识别效果不佳的问题,提出了一种基于机器阅读理解框架(MRC)和知识增强语义表示模型(ERNIE)的MEAB(MRC-ERNIE-Attention-BiLSTM)模型结构.该模型在MRC框架的基础上,引入了基于Attention的信息融合策略,将两种不同结构的数据在ERNIE预训练模型进行特征提取之后转换为向量,并在信息融合层进行向量融合,使模型能够学习到标签当中的先验知识.随后BiLSTM模型对具有语义信息的向量从两个方向进行特征提取,并在一种多层嵌套命名实体识别器中进行输出,提高了嵌套命名实体的识别准确率.在构建的有色冶金领域命名实体识别数据集上的实验表明,MEAB模型的精确率、召回率和F1 值分别达到了78.77%、79.76%和79.26%,证明了该模型的有效性.
近些年来,目标检测算法在移动机器人环境感知领域表现出了突出的性能.但是目标检测算法存在模型庞大和计算复杂的问题,制约了目标检测算法在移动嵌入式设备上的部署和发展.YOLO是一种单阶段的目标检测算法,具有较高的准确度和较快的运行速度.该文提出了一种基于YOLOv5s改进后适用于嵌入式设备的移动机器人目标检测算法DID-YOLO.首先,使用深度可分离卷积和倒置残差模块对YOLOv5s的backbone网络进行重构,降低模型复杂度和计算量,达到轻量化的目的;其次,利用特征层和输出层结合的知识蒸馏训练提高重构后目标检测网络的精度.在目标检测通用数据集PASCAL VOC上实验表明:DID-YOLO模型尺寸为3.63 MB,相较原网络模型尺寸减小了48.65%;经过特征层和输出层蒸馏后,DID-YOLO 的mAP@0.5 提升至73.83%;DID-YOLO在Jetson AGX Xavier上实现了每秒31.2 帧的实时图像处理速度.提出的DID-YOLO性能显著,满足了移动机器人嵌入式平台的实时高精度检测需求.
先天性心脏病(先心病)是严重威胁儿童健康的常见疾病,尽早进行先心病筛查对于该疾病的及时诊断和手术治疗十分重要,但这在医疗资源相对匮乏的偏远地区却难以实现.针对上述问题,为实现儿童先心病的智能化早期筛查,提出了一种基于复Morlet小波谱图与Transformer架构深度神经网络分类器的异常心音识别方法,其中小波谱图可以兼顾非平稳心音信号特征描述的时间分辨率与频率分辨率,而心音分类模型则是在经典ResNet50 骨干网络的基础上嵌入了Transformer架构自注意力模块,可有效提升对时序信号谱图的特征提取能力.此外,还实现了基于Hilbert变换的心音包络检查,以及基于XMLRPC协议与Redis队列的并发心音识别服务部署,便于和前端机器人整合使用.实验测试显示识别准确率在现场心音数据集上达到87.5%,在PhysioNet 2016 心音数据集上达到94.5%,表明该方法不仅在公开数据集上优于已有方法,即使是针对现场复杂环境下的心音识别任务也能取得较为理想的效果.
层次概念能够有效解释语言模型的隐含知识,并且提升语言模型的结构化优化性能.针对层次概念表示和学习模型的多样化发展,该文分析了层次概念表示的线性空间、概率空间和文本蕴含性质,梳理了概念学习模型的层次结构和优化原理,用于促进概念学习模型的应用效果.通过阐述概念学习过程和语义空间的层次性质,归纳层次概念学习的四类计算模型:基于条件熵的文本层次概念抽取;建立语言资源的概念层次和神经网络的隐含层次之间的映射;通过迭代的随机过程拓展主题模型;在正则化因子中添加语义关系约束.通过概念学习综述得出如下结论:层次性的语言模型广泛结合了显明和隐含的概念表示方法;统计模型和语言资源的语义映射是拓展层次结构的主要路径;层次结构具有双曲空间和嵌套球形结构;层次结构分析可以提升统计模型的解释水平.
为了提升云网安全服务的可靠性,降低监测耗时,提出基于SRv6 技术的云网安全服务可视化监测方法.首先,基于SRv6 技术构建云网架构,通过设置软件-硬件路由装置对云网的安全服务做分区处理,通过分区处理提高云网安全服务质量;其次,云网安全服务智能分区后,采用小波变换算法对云网中的流量数据实行降噪处理,避免噪声对可视化监测过程产生影响;然后,通过奇数序列来预测偶数序列,通过偶数序列来预测奇数序列,引入预测算子作为预测方向,并通过线性搜索方法在校正方向与预测方向的多目标成像中寻找最佳预测点;最后,采用三维可视化的监测机制,结合信息熵挖掘算法,实现云网安全服务的可视化监测.实验结果显示,该方法的监测效果好、监测时间短.所提的可视化监测方法可以解决流量差异性、不同服务类型、冗余性等问题的干扰.
目前,基于信道状态信息(Channel State Information,CSI)的室内摔倒检测(Fall Detection,FD)系统已被证明拥有巨大潜力,但是,不同室内布局带来的多径效应的差异往往使其无法实现跨场景使用.因此,该文提出了DA-Fall(Domain-adaptive Fall),通过结合两种自适应策略的域自适应方法来改进未标记噪声信号的泛化,从而提高对目标域的检测精度.在提出的摔倒检测系统中,引入了域鉴别器和域混淆自适应层来进行对抗性训练.首先,该算法通过引入依赖于相对值的相对鉴别器来优化对抗训练,从而更好地反映域间差异.其次,将基于多核架构的最大均值差异(Multiple Kernel Maximum Mean Difference,MK-MMD)作为域对抗损失的正则化项,进一步减小域间的边缘分布距离.实验分析表明,DA-Fall取得了比WiFall,RT-Fall,SignGAN更好的效果,在原场景与新场景中分别达到了96.83%和91.03%的检测精度.
传统芒果病虫害防治,需要人工进行识别,现引入深度学习技术,可快速准确地对芒果病虫害进行识别.以攀西地区芒果的12 种病虫害为研究对象,采用的数据集一部分来自公开数据集MangoLeafBD,另一部分由爬虫技术获得的网络图片组成,共获取图片6 769 张,其中4 879 张为训练集,1 220 张为验证集,670 张为测试集.为迎合实际应用的需要,选择了MobileNetV3、MobileViT等4 种不同规模的轻量级深度学习网络模型,结合迁移学习训练策略进行对比实验,比较了各个模型的参数量、精确率、召回率等参数.实验结果显示,MobileViT模型用于芒果病虫害分类识别效果最佳,该模型的精确率为96.31%,召回率为96.12%,F1 为96.20%,均优于其他模型.由此表明,模型具有较好的鲁棒性和识别性能,可为芒果病虫害分类识别提供技术参考.
随着互联网时代的不断发展,互联网上的信息量不断增多,"信息过载"等相关问题愈发严重,从而导致用户很难快速地获取到有用的信息,因此推荐系统应运而生.推荐系统可以预测用户的需求并推荐给用户其最可能喜欢的内容,来缓解人们从海量信息中做出选择的烦恼.推荐算法是推荐系统的核心,它完全可以决定一个推荐系统的性能.推荐准确度及可解释性是推荐算法目前面临的两大难题.可评判推荐算法是对话推荐算法的一种,在预测出项目的同时,也及时给出推荐项目的理由,并且为用户提供一个重新推荐的机会,用户通过对解释项进行评判来使推荐系统重新预测出商品,可有效解决上述两个问题.该文首先基于变分推断与神经协同过滤相结合的思想,对算法和模型进行了形式化的定义和理论推导,并且从概率的角度出发使用贝叶斯神经网络实现了该模型.通过与其他可评判推荐算法进行实验对比,证实了该模型的许多推荐指标已经达到了目前最先进的水平.
差异数据定位在数据增量同步等领域得到了很好的应用.当下学者们将数据分块算法应用在差异数据定位中,提出了很多优秀的数据分块算法,并对各自算法的效率进行了论述和实验,但并没有从理论上论述数据分块算法在差异数据定位中的正确性.此外,定位到的差异数据的大小与数据分块算法的关系也没有进行理论分析,多是以实验结果来辅助说明.为此,文中对数据差异定位的过程进行抽象,对数据分块算法应用于该过程的正确性进行了论证,并对数据分块算法在数据差异定位中的作用进行分析.通过理论推导的方式,证明了数据分块算法在定位差异数据时的正确性,同时得出差异数据的大小与数据分块算法的关系.文中结论对设计应用于差异数据定位的数据分块算法有一定的参考意义.
现有目标检测算法消耗大量算力资源、参数量大、占用内存空间多,不利于在小型设备上推广使用.因此,基于结构重参数化技术并结合YOLO系列算法的研究成果,提出了一种轻量化目标检测模型Rep-YOLO.使用结构重参数化技术实现模型在训练时的多分支结构和推理时的线性结构之间的转换,从而减少模型推理时对算力资源的消耗.另外,为了降低模型的参数量,利用深度可分离卷积、网络裁剪等方法,重新设计了多尺度特征融合网络和检测头.实验结果表明:在PASCAL VOC上,Rep-YOLO-s1 精度可达82.7%,Rep-YOLO-s1 与YOLOv6s相比,在参数量减少54.8%的情况下,精度提高了2.4 百分点,推理速度提升了 6%.在NVIDIA RTX 3060 GPU上,Rep-YOLO-s0 比YOLOv6s的推理速度快10%,Rep-YOLO-nano比YOLOv7-tiny快4%,精度提高了0.5 百分点.Rep-YOLO与规模类似的模型相比,体积更小,精度更高,更加利于资源有限的部署应用.
急性缺血性脑卒中病灶很容易在磁共振成像(MRI)上表现为高信号区域.相较于MRI,计算机断层扫描(CT)成像速度快、价格低,不易受金属植入物干扰,但CT对缺血性脑卒中病灶不敏感,通常在CT上难以确定病灶的位置,且CT包含的信息量比MRI少.考虑到速度与可用性的提升以及成本的降低,为了以CT生成的MRI代替真实的MRI对急性缺血性脑卒中进行诊断,提出一种CT到MRI的跨模态图像生成算法.首先,利用影像组学在CT上确定病灶区域并提取影像组学特征,筛选出信息增益最大的特征并可视化,然后将该特征图与CT一同作为生成对抗网络的输入.生成对抗网络在pix2pix生成器中引入残差块,鉴别器采用PatchGAN.最后在损失函数中引入病灶特征相似性损失函数,更加关注病灶区域的相似性.经两名放射科医生的主观判断与评估指标的客观分析,结果表明,该算法生成的MRI与真实MRI相似性极高,且病灶位置正确,形状相似,可为医生的诊疗提供帮助.
