Enhancing Pneumonia Detection in Pediatric Chest X-Rays Using CGAN-Augmented Datasets and Lightweight Deep Transfer Learning Models

Pneumonia ranks as a leading cause of mortality, particularly in children aged five and under. Detecting this disease typically requires radiologists to examine chest X-rays and report their findings to physicians, a task susceptible to human error. The application of Deep Transfer Learning (DTL) for the identification of pneumonia through chest X-rays is hindered by a shortage of available images, which has led to less than optimal DTL performance and issues with overfitting. Overfitting is characterized by a model’s learning that is too closely fitted to the training data, reducing its effectiveness on unseen data. The problem of overfitting is especially prevalent in medical image processing due to the high costs and extensive time required for image annotation, as well as the challenge of collecting substantial datasets that also respect patient privacy concerning infectious diseases such as pneumonia. To mitigate these challenges, this paper introduces the use of conditional generative adversarial networks (CGAN) to enrich the pneumonia dataset with 2690 synthesized X-ray images of the minority class, aiming to even out the dataset distribution for improved diagnostic performance. Subsequently, we applied four modified lightweight deep transfer learning models such as Xception, MobileNetV2, MobileNet, and EfficientNetB0. These models have been fine-tuned and evaluated, demonstrating remarkable detection accuracies of 99.26%, 98.23%, 97.06%, and 94.55%, respectively, across fifty epochs. The experimental results validate that the models we have proposed achieve high detection accuracy rates, with the best model reaching up to 99.26% effectiveness, outperforming other models in the diagnosis of pneumonia from X-ray images.

Transfer Learning Approach to Classify the X-Ray Image that Corresponds to Corona Disease Using ResNet50 Pre-Trained by ChexNet

The COVID-19 pandemic has had a widespread negative impact globally. It shares symptoms with other respiratory illnesses such as pneumonia and influenza, making rapid and accurate diagnosis essential to treat individuals and halt further transmission. X-ray imaging of the lungs is one of the most reliable diagnostic tools. Utilizing deep learning, we can train models to recognize the signs of infection, thus aiding in the identification of COVID-19 cases. For our project, we developed a deep learning model utilizing the ResNet50 architecture, pre-trained with ImageNet and CheXNet datasets. We tackled the challenge of an imbalanced dataset, the CoronaHack Chest X-Ray dataset provided by Kaggle, through both binary and multi-class classification approaches. Additionally, we evaluated the performance impact of using Focal loss versus Cross-entropy loss in our model.

Fault Estimation and Accommodation for Networked Control Systems with Transfer Delay

在这份报纸，差错评价和差错的一个方法为有转移延期和进程噪音的联网的控制系统(NCS ) 的容忍的控制被介绍。首先，联网的控制系统作为有转移的分离时间的系统推迟的 multiple-input-multiple-output (MIMO ) 被建模，处理噪音，并且为无常建模。在这个模型下面并且在一些条件下面，一个差错评价方法被建议估计系统差错。根据差错评价和滑动模式控制理论的信息，一个差错容忍的控制器被设计恢复系统性能。最后，模拟结果被用来验证方法的效率。

基于迁移学习与残差网络的快递包裹X光图像识别

针对快递包裹违禁物品识别存在种类繁多、依赖人力和X光图像获取难度大等问题,为提高快递包裹违禁物品识别的效率和准确度,本研究提出一种迁移学习与残差网络相结合的快递包裹X光图像识别方法(TL-ResNet18)。首先构建了相似度高的源领域数据集和目标领域数据集;其次,选用ResNet18作为预训练模型,调整初始化参数结构,并将ResNet18学习到的内容作为初始化参数迁移到目标领域,实现快递包裹X光图像分类;最后,将相同数据集作为三种模型的输入并对结果进行对比。实验结果表明,TL-ResNet18模型的局部微调和全局微调的识别准确率分别为93.5%、95.0%,相比于ResNet18模型提高了7%、8.5%,且精确度、召回率和F1值都优于ResNet18模型,该方法性能更优,且不受小型数据集对深层网络训练的限制,有利于快递包裹X光图像识别的智能化发展。

基于GIS的公交换乘网络构建及可达性分析

为了提高公交换乘效率、优化公交系统,基于GIS软件构建公交换乘网络,运用该网络对换乘可达性进行了测度和分析.结合Space-P模型和网络分析法,以拉萨市城关区为研究区域,基于公交线路路径、站点、交叉口等基本信息构建同站换乘子网络.结合公交站点服务范围、步行通道路径、交叉口等信息构建异站换乘子网络.二者协同实现了基于ArcGIS的公交换乘网络构建,并依据该网络对公交线路的乘客在车时间和换乘系数进行测度和分析.结果表明:构建的换乘网络能够对乘客在车时间进行良好的测度,乘客在车时间最大值为68.68 min,最小值为2.00 min,乘客换乘在车时间平均值为29.90 min.该换乘网络能够对换乘系数进行良好的测度,得到有效换乘线路90 300条,换乘系数最大为4条(线路为62条),最小为0条(线路为1 354条).采用可达性度量模型,可实现对公交站点时间可达性和换乘可达性的良好测度和分析.

Neonatal Transfer Situation Following Implementation of a Perinatal Network: An Analysis in Douala, Cameroon

Background: Postnatal transfer (PT) is interhospital transport of care-needing newborns. In 2016, a perinatal network was implemented to facilitate PT in the town of Douala, Cameroon. The network was supposed to improve PT-related care standards. This study aimed at determining characteristics of PT five years following the implementation of this network. Methods: A cross-sectional study was carried out from February to May 2021 at neonatology wards of six hospitals in Douala. Medical records of newborns transferred to the hospitals were scrutinized to document their characteristics. Parents were contacted to obtain information on PT route and itinerary. Data were analyzed using Epi Info software and summarized as percentages, mean and odds ratio. Results: In total, 234 of the 1159 newborns admitted were transferred, giving a PT prevalence of 20.2% (95% CI 17.9% - 22.6%). Male-to-female ratio of the transferred newborns was 1.3. Neonatal infection (26.5%), prematurity (23.5%) and respiratory distress (15.4%) were the main reasons for transfer. Only 3% of the PT was medicalized while only 2% of the newborns were transferred through perinatal network. On admission, hypothermia and respiratory distress were found in 31% and 35% of the newborns, respectively. The mortality rate among babies was 20% and these had a two-fold risk of dying (95% CI 1.58 - 3.44, p Conclusion: PT and the perinatal network are lowly organized and implemented in Douala. Sensitization of medical staff on in utero transfer, creating center for coordination of the network, and implementation of neonatal transport system could improve the quality of PT.

基于TCN和迁移学习的混凝土坝变形预测方法

混凝土坝变形测点数据丢失或者新增测点测量时间太短都会导致这部分测点的数据量不足,使得变形预测精度受到影响。为了提高这些小数据量测点的变形预测精度,提出了将时域卷积网络(TCN)与迁移学习相结合的变形预测方法。以数据量充足的测点为源域,以缺少数据的测点为目标域,将在源域上训练好的TCN模型的结构和参数迁移到目标域模型中,固定其中的冻结层参数,利用目标域中的数据对目标域模型可调层参数进行调整。同时,采用动态时间规整选择与目标域数据序列相似度最高的监测数据作为最佳源域数据,提升迁移学习效果。工程实例分析表明:迁移学习后的目标域模型的均方根误差和平均绝对误差与利用足量数据训练的TCN模型的预测误差相比,差异仅分别为1.73%和8.09%,小数据量情况下TCN预测模型的精度得到了提高。

Routing in Delay Tolerant Networks (DTN)<br>—Improved Routing with MaxProp and the Model of “Transfer by Delegation” (Custody Transfer)

In this paper, we address the problem of routing in delay tolerant networks (DTN). In such networks there is no guarantee of finding a complete communication path connecting the source and destination at any time, especially when the destination is not in the same region as the source, which makes traditional routing protocols inefficient in that transmission of the messages between nodes. We propose to combine the routing protocol MaxProp and the model of “transfer by delegation” (custody transfer) to improve the routing in DTN networks and to exploit nodes as common carriers of messages between the network partitioned. To implement this approach and assess those improvements and changes we developed a DTN simulator. Simulation examples are illustrated in the article.

SCHEDULE ARRANGEMENT AND OPTIMIZATION OF THE FILE TRANSFER NETWORK

This project was designated as Meritorious of Mathematical Contest inModeling (MCM'94). We have been required tu solve a problem of findins thebest schedule of a file transfer network in order to niake the niaktispan the smallestone. Three situations with

Research on Surface Information Extraction Based on Deep Learning and Transfer Learning

The land cover types in South China are varied, and the terrain is undulating, and the area of different land types is small, and the remote sensing monitoring work was difficult. In order to solve these problems, an automatic classification method based on transfer learning and convolutional neural network model was established in this paper, with a total classification accuracy of 98.1611%. This paper proposes a land use classification remote sensing method based on deep learning, which improved the automation level and monitoring accuracy of complex land surface remote sensing monitoring in South China, and it provided technical support for the land consolidation work in China.

基于机器深度学习的小麦播种机控制系统研究

针对我国小麦播种机自动控制系统的可靠性及灵敏度不高的问题,基于机器深度学习对小麦播种机的控制系统进行了设计和改进。小麦播种机的主要组成包括控制系统、排种系统、监控系统、电力系统、机架和驾驶室、覆土镇压和排肥装置。为了使播种机的控制系统能有效进行图像检测识别,提升播种机的控制精度,采用机器深度学习中的卷积神经网络算法对控制系统进行设计,并采用迁移学习的方式对模型进行训练和检测。为了验证播种机控制系统的性能,对其进行播种精度控制和播种性能测试试验,结果表明:播种机的精度和性能均符合播种机的设计要求。

Spatial patterns nitrogen transfer models of ectomycorrhizal networks in a Mongolian scotch pine plantation

Ectomycorrhizal(EM)networks provide a variety of services to plants and ecosystems include nutrient uptake and transfer,seedling survival,internal cycling of nutrients,plant competition,and so on.To deeply their structure and function in ecosystems,we investigated the spatial patterns and nitrogen(N)transfer of EM networks usingN labelling technique in a Mongolian scotch pine(Pinus sylvestris var.mongolica Litv.)plantation in Northeastern China.In August 2011,four plots(20 × 20 m)were set up in the plantation.125 ml 5 at.%0.15 mol/LNHNOsolution was injected into soil at the center of each plot.Before and 2,6,30 and 215 days after theN application,needles(current year)of each pine were sampled along four 12 m sampling lines.Needle total N andN concentrations were analyzed.We observed needle N andN concentrations increased significantly over time afterN application,up to 31 and0.42%,respectively.There was no correlation between needle N concentration andN/N ratio(R2=0.40,n=5,P=0.156),while excess needle N concentration and excess needleN/N ratio were positively correlated across different time intervals(R~2=0.89,n=4,P\0.05),but deceased with time interval lengthening.NeedleN/N ratio increased with time,but it was not correlated with distance.NeedleN/N ratio was negative with distance before and 6th day and 30th day,positive with distance at 2nd day,but the trend was considerably weaker,their slop were close to zero.These results demonstrated that EM networks were ubiquitous and uniformly distributed in the Mongolian scotch pine plantation and a random network.We found N transfer efficiency was very high,absorbed N by EM network was transferred as wide as possible,we observed N uptake of plant had strong bias forN andN,namely N fractionation.Understanding the structure and function of EM networks in ecosystems may lead to a deeper understanding of ecological stability and evolution,and thus provide new theoretical approaches to improve conservation practices for the management of the Earth’s ecosystems.

基于迁移学习和逻辑回归模型的花卉分类研究

【目的】不同种花卉之间的相似性以及同种花卉内部的多变性加大了花卉图像分类难度,其难点是要人工设计出能充分体现花卉颜色、形状和花瓣形态等特征的特征提取方法。传统的花卉图像分类方法的精度不高且模型的泛化能力较差,这些问题亟待解决。【方法】课题组提出一种基于数据增强的VGG16迁移学习卷积神经网络提取花卉图像特征,再训练多类逻辑回归模型的花卉图像分类识别方法;并且通过在flowers17和flowers102花卉数据集上进行测试,来验证课题组所提出的花卉分类识别方法的有效性。【结果】课题组所提出的花卉分类识别方法在flowers17和flowers102数据集中分别达到了97.89%和92.10%的分类精度,高于现有其他花卉图像分类方法。【结论】通过预训练的深度人工神经网络提取的高区分度的花卉图像特征,优于人工设定的花卉图像特征,能训练出更高效精准的花卉识别分类器。基于本研究内容,下一步可对VGG16网络进行降维改进,让模型参数减少,从而实现快速实时应用。

Mid-Range Wireless Power Transfer and Its Application to Body Sensor Networks

It has been reported that, through the evanescent near fields, the strongly coupled magnetic resonance is able to achieve an efficient mid-range Wireless Power Transfer (WPT) beyond the characteristic size of the resonator. Recent studies on of the relay effect of the WPT allow more distant and flexible energy transmission. These new developments hold a promise to construct a fully wireless Body Sensor Network (wBSN) using the new mid-range WPT theory. In this paper, a general optimization strategy for a WPT network is presented by analysis and simulation using the coupled mode theory. Based on the results of theoretical and computational study, two types of thin-film resonators are designed and prototyped for the construction of wBSNs. These resonators and associated electronic components can be integrated into a WPT platform to permit wireless power delivery to multiple wearable sensors and medical implants on the surface and within the human body. Our experiments have demonstrated the feasibility of the WPT approach.

基于机器学习和迁移学习的地震预警震级估计

震级估计是地震预警的重要环节之一。快速且可靠的震级估计可以为地震预警系统提供有效的地震预警信息。传统的地震预警震级估计方法主要利用从P波信号中提取单一的特征建立震级估计经验预测方程,且震级估计误差较大。为了提高地震预警震级估计的可靠性以及探索机器学习方法在中国川滇地区地震预警震级估计的可行性,基于日本K-NET台网记录的强震动数据建立用于震级估计的机器学习预训练模型(CRnet-M),CRnet-M模型结合了卷积神经网络和循环神经网络;结合迁移学习和中国川滇地区的强震动数据,对预训练的CRnet-M模型进行微调和训练,进而建立了用于中国川滇地区的震级估计模型(TLCRnet-M)。研究结果表明:对于日本的测试数据集,在P波触发后3 s,和传统的地震预警震级估计方法相比,预训练的CRnet-M模型有更小的震级估计误差,且绝对误差在0~0.5震级单位范围内的百分比达到了86.89%;对于川滇地区的测试数据集,在P波触发后3 s,和传统的地震预警震级估计方法以及未使用迁移学习的CRnet-M模型相比,使用迁移学习的TLCRnet-M模型提高了震级估计的可靠性,且绝对误差在0~0.5震级单位范围内的百分比为76.25%。方法在一定程度上提高了地震预警震级估计的可靠性,且对于地震预警系统有一定意义。

基于迁移学习和CNN-LSTM的水轮机空化状态识别方法

针对水轮机空化声发射信号中包含较多噪声、依赖人工降噪与特征提取以及深度学习模型准确率极度依赖海量训练数据的问题,提出一种基于迁移学习和卷积神经网络-长短时记忆网络(CNN-LSTM)的水轮机空化状态识别方法。首先,将数据输入CNN中提取隐含特征;然后,在LSTM中提取特征隐含的时序信息并输出空化类型,通过训练网络参数建立基于CNN-LSTM的空化状态识别模型;最后,引入迁移学习对类似工况进行空化状态识别。结果表明:该模型能准确识别出3种不同的水轮机空化类型,其平均识别准确率达到较高水平;与传统深度学习模型相比,该模型在极少样本学习任务中的识别准确率具有明显优势。

考虑信息泄露影响的光纤传感网络传输数据安全导入方法

为了提高对光纤传感网络传输数据安全传输能力,提出基于信息泄漏加密传输的光纤传感网络传输数据安全导入方法。采用离散混沌时间序列同步调制方法实现对光纤传感网络传输数据导入过程中调制编码处理,根据混沌随机编码特性实现对光纤传感网络传输数据过程中的同步输出稳定性调节和自主随机编码,构建光纤传感网络信息传输泄漏抑制的加密密钥,通过Logistics映射实现对光纤传感网络传输数据导入过程中的信息泄漏加密重传,根据光纤传感网络传输数据的混合敏感密钥表征和算术编码,实现信息泄漏控制和数据安全导入。实验结果表明,采用该方法进行光纤传感网络传输数据导入的加密性能较好,其光纤传感网络传输数据已加密比特序列的识别率为98%,误码率仅为2.7∗10-9%,抗泄漏能力达到了0.970,并且导入后的数据完整度达到了0.996,说明该方法具备了较强的加密效果、抗泄漏能力,实现数据的安全导入和加密传输。

Pattern recognition and data mining software based on artificial neural networks applied to proton transfer in aqueous environments

In computational physics proton transfer phenomena could be viewed as pattern classification problems based on a set of input features allowing classification of the proton motion into two categories： transfer ＇occurred＇ and transfer ＇not occurred＇. The goal of this paper is to evaluate the use of artificial neural networks in the classification of proton transfer events, based on the feed-forward back propagation neural network, used as a classifier to distinguish between the two transfer cases. In this paper, we use a new developed data mining and pattern recognition tool for automating, controlling, and drawing charts of the output data of an Empirical Valence Bond existing code. The study analyzes the need for pattern recognition in aqueous proton transfer processes and how the learning approach in error back propagation （multilayer perceptron algorithms） could be satisfactorily employed in the present case. We present a tool for pattern recognition and validate the code including a real physical case study. The results of applying the artificial neural networks methodology to crowd patterns based upon selected physical properties （e.g., temperature, density） show the abilities of the network to learn proton transfer patterns corresponding to properties of the aqueous environments, which is in turn proved to be fully compatible with previous proton transfer studies.

基于计算机视觉的采摘机械臂控制系统设计

为了提高机械臂采摘的效率及增强安全自主性,提出了一种基于改进卷积神经网络和迁移学习的计算机图像识别模型。首先,建立了由可采摘与不可采摘图像组成的样本数据集,将每幅图像的像素设置为256×256;然后,构建基于改进卷积神经网络和迁移学习的计算机图像识别模型,并将自动编码机网络结构与卷积神经网络运算方法相结合,利用自动编码机网络结构具有编码和解码的环节,通过卷积神经网络运算方式构建出一种改进的卷积神经网络;通过卷积层挖掘图片信息中具有采摘信息的特征,同时消除随机环境对图片的干扰,解码部分能够对特征图像进行上采样并判断是否应该进行采摘与采摘姿势;最后,将构建网络模型与迁移学习相结合进行实验,分析迁移学习方法、数据集样本大小、网络参数对实验结果的影响。结果表明:采摘机械臂识别模型整体识别率更高,能够构建出效率更高、鲁棒性更强的采摘控制系统。

Forecasting Loss of Ecosystem Service Value Using a BP Network: A Case Study of the Impact of the South-to-north Water Transfer Project on the Ecological Environmental in Xiangfan, Hubei Province, China

Objective To recognize and assess the impact of the South-to-north Water Transfer Project (SNWTP) on the ecological environment of Xiangfan, Hubei Province, situated in the water-out area, and develop sound scientific countermeasures. Methods A three-layer BP network was built to simulate topology and process of the eco-economy system of Xiangfan. Historical data of ecological environmental factors and socio-economic factors as inputs, and corresponding historical data of ecosystem service value (ESV) and GDP as target outputs, were presented to train and test the network. When predicted input data after 2001 were presented to trained network as generalization sets, ESVs and GDPs of 2002, 2003, 2004... till 2050 were simulated as output in succession. Results Up to 2050, the area would have suffered an accumulative total ESV loss of RMB 104.9 billion, which accounted for 37.36% of the present ESV. The coinstantaneous GDP would change asynchronously with ESV, it would go through an up-to-down process and finally lose RMB89.3 billion, which accounted for 18.71% of 2001. Conclusions The simulation indicates that ESV loss means damage to the capability of socio-economic sustainable development, and suggests that artificial neural networks (ANNs) provide a feasible and effective method and have an important potential in ESV modeling.