Text in natural scenes is often affected by complex text background, irregular text arrangement and other factors, resulting in a significant increase in the difficulty of text detection. In order to further improve the accuracy of text detection, an improved text detection method AC-DBNet for complex natural scenes is proposed based on the segment-based detection technology DBNet. The main points of improvement are to strengthen the feature extraction capability of DBNet network model, add the following extraction module CEM module before the feature map fusion, increase the receptive file through hole convolution, and add the attention guidance module AM to enhance the detection of small target text area, and improve the accuracy of the detection network through these improvements. The accuracy of the improved algorithm in the test set reached 92.1%, which was 0.7% higher than that before the improvement. The improved algorithm improves the accuracy of detection under the task of text detection in real scenes, showing the superiority of the detection algorithm.
在公共网络上广泛传播的视频不可避免地包含有隐私信息.为此很多针对视频的加密方案相继提出.文章针对HEVC视频感兴趣区域(ROI)提出一种自同步混沌流密码加密算法,以人脸所在区域作为感兴趣区域为例,实现对HEVC视频隐私区域的保护.首先,根据动力系统反控制原理,构建出无简并三维离散超混沌系统.在此基础上,提出一种自同步混沌流密码系统.然后,对HEVC视频ROI实现加密,具体如下.对于HEVC视频序列中的I帧,检测人脸所在区域并作为ROI;对于P帧,则利用视频流中压缩域信息追踪ROI.并将ROI信息转化为Tiles索引.利用自同步混沌流密码系统生成的伪随机序列对Tiles索引以及Tile内的语法元素MVD符号位、MVD值和QTC符号位进行加密.解码端的解密算法也同时详细给出.同时,借助补充增强信息(SEI)实现编解码端对ROI信息的同步.实验结果验证了所提方案的有效性.最后,对所提方法的性能和安全性进行了分析,并对所提方法的嵌入对编码器所带来的影响进行了讨论.
针对现有传统荧光材料防伪技术存在的可靠性较低、容易被仿制、防伪成本高昂等问题,文章提出了基于Faster R-CNN的荧光材料防伪图案识别方案,该方法使用了深度学习技术和区域提议网络(RPN)来实现对防伪图案的自动检测和识别.该方法首先使用区域提议网络生成候选图片区域,然后使用卷积神经网络对这些候选区域进行分类和定位,从而提高识别网络的分类准确率,缩减训练时间,图像识别准确率达 96%,获得了能对荧光材料防伪图案精准识别的网络模型,该方法在防伪图案识别方面具有很高的准确性和鲁棒性,可以有效地应用于荧光材料防伪图案识别领域,该研究为防伪技术的发展提供了新思路和方法.
由于现有的安全防护方法能耗量大,为此研究智能电网无线通信设备安全防护方法.根据不同的属性值生成决策树,计算信息的增益,得到最优的标签计算参数后获得其数组划分子集.对子集中的网络信息属性特征进行提取.选择最大增益值的数组,添加节点属性特征.进行攻击行为路径的预测,找到信息来源后识别安全威胁.对侧链区块数据进行预处理,将数据传递给到太坊主链上,并不断对其进行跟踪,数据异常时主链进行退回操作从而完成安全防护.实验结果表明,8 个小组的数据包在转发过程中,产生的能耗均小于 365MJ,使得通信数据在传输和转发中能够得到相对安全的空间,实现较好安全防护.
The construction of new electric power system is the key development direction of our national energy strategy,and its security is the top priority.Due to unstable energy supply,power load growth,meteorological disasters,man-made damage and other factors,the threats and challenges faced by the power system are increasingly complex.These challenges make tradi-tional approaches to power system security increasingly ineffective.Artificial intelligence technology is providing new ideas and solutions for power system security.To this end,this paper investigates the power system security application algorithms based on artificial intelligence technology,analyzes,compares and summarizes the power system security algorithms and their applications for three types of needs,namely,power system stability assessment,power security event recognition and process-ing,and power equipment detection and control.Then,the common technical challenges of current power system security algo-rithms are discussed.Finally,the development trend of power system security algorithm is predicted.
TCP/IP is a typical network layer protocol,because of the TCP/IP protocol is basic in the related operating system ker-nel,so for layered protocol working strategy it is difficult.Via IP packet in the Linux kernel to restructure and the segmentation process,can be defined in the related application layer to the layered protocol,UDP is used below the condition that the user at the same time the socket restructuring and piecewise implementation to application layer.This simulation program is IP restruc-turing and clear segmentation process key processing activities,and layered protocol the entire process intuitive,this method used in computer network protocol related teaching effect is better.In this paper,the network application protocol messages re-structuring and segmented scenario analysis,Discuss the instance teaching method of network protocol layering
in recent years,with the development of technology,people put forward higher and higher requirements for the per-formance of FM transmitting system,formulate and implement the corresponding upgrading and transformation plan.Based on this background,this paper first introduces what is antenna polarization,and the characteristics of single-polarization and mixed-polarization antenna,and then analyzes the practical case of using the mixed-polarization antenna in FM transmitting system,finally,the advantages of hybrid polarization antenna are analyzed in detail with a practical case.Hope to inspire people,for the future construction,upgrading of wireless transmission systems and other work to provide useful reference.
With the reform of the national economic system and the establishment of a socialist market economic system,joint ventures in agricultural land inventory are an important means to strengthen land registration and land asset management.Based on embedded development technology,geographic information system technology,and database management technology,the system realizes the data archiving files generated during the field verification process,helping field personnel to quickly com-plete the farmland verification and parcel data check and supplement work;In the processing of industrial data,the system rea-lizes the functions of fast query,browsing,editing,saving and output of parcel information,realizes the close connection be-tween internal and external industries and the accuracy and timeliness of information transmission,and promotes the entire ag-ricultural land inventory.The completion process of the capital verification work has formed a new pattern of farmland inventory and capital verification with internal and external interoperability and dynamic management.
In order to reduce the signal interruption of UAV communication and improve the quality of search and path planning for optimizing multi UAV cooperation,a combined algorithm based on UAV self positioning and improved ant colony genetic algorithm is proposed.By transforming the multiple traveling salesman problem(MTSP)of multiple UAVs into a combinatorial optimization(TSP)problem of multiple independent paths,The search optimization operator of local pheromone is introduced.In order to avoid premature convergence of path addressing,the selection operator and crossover operator are designed and app lied to perform crossover operations on population individuals of different UAV path planning.On the basis of analyzing the routing selection of drone communication networks,by adjusting the path planning method of drones,optimizing communication network performance,reducing communication delay,and achieving reliable information transmission of multiple drone targets in the region.
为实现对网络通信数据特征的精准检测,以云计算为背景,提出云计算下网络通信大数据混合属性特征检测.根据云计算下不同信息的分布情况,构建通信属性信息流模型;将节点信息按照时序录入信息流模型.考虑到云计算下网络通信大数据混合信息具有较强的逆相似性,进行字符串传输速率的控制;确定大数据混合属性边缘提取算子;将集成的通信数据团进行分解,通过对混合属性特征的计算与重要性排序,实现对网络通信大数据混合属性特征的检测.实验结果证明:所提方法可以优化检测方法约简效果,提高特征检测结果的准确性.
为了达到智能家居负荷控制中降低用电量和电价的目的,提出一种多目标智能家居人机交互负荷控制算法,利用边际成本建立考虑负载率;对开关电器和温控电器进行分类,通过智能设备采集人体活动、室内外温度和光照强度,实现多参数舒适度模型设计,以电价最小化为目标,构建舒适度和电价多目标模型和多参数舒适度,利用基于适应度值的距离比改进粒子群优化算法求解模型,得到最优的智能家居人机交互负载控制方案,构建智能家居远程控制系统的功能模块,并建立频时参数跟踪学习模型,优化多目标智能家居人机交互负载控制算法.实验结果表明,该算法能够合理降低电价,实现智能家居中低耗电量和低用电负荷的控制.
Faced with huge amount of data calculation problems brought by the developments of the artificial intelligence,the big data technology and so forth,the approximate computing,especially the design of approximate adders,has become an im-portant research direction.A new-type dynamic segmentation approximate adder is proposed combined with the parallelism of carry state and the flexibility of dynamic segmentation,because of the probability distribution of carry propagate chain lengths in adders and the peculiarity that lengths of carry propagation chains are shorter than the bit width of the adder itself.The ex-perimental analysis and the application verification show that the proposed design scheme is superior and feasible.
常规的无线网络密钥管理方法,由于管理措施的不到位,导致管理效率低.提出基于K-means算法的无线网络密钥管理方法.通过对无线网络的特征量分析出密钥的运行环境,通过LKH++算法建立密钥树,方便后期基于K-means算法无线网络密钥的聚类设计,通过验证密钥的存储功能、安全性和更新性来实现管理方法.实验通过对比不同方法下密钥的重构概率,获得在模拟实验中基于K-means算法的管理方法于样本 1 和样本 5 中可达到 100%的管理执行效率,纵向增强管理性能,能满足具有良好管理要求.
由于传统方法在窄带物联网环境下电力物资供应信息自动共享中应用效果不佳,不仅共享延迟时间比较长,而且丢包率比较高,无法达到预期的共享效果,为此提出窄带物联网环境下电力物资供应信息自动共享方法.根据信息欧氏距离与标准差采集电力物资供应信息,利用关联规则对电力物资供应信息融合调度,实现对供应信息预处理,根据相似度匹配供应信息,通过窄带物联网环境对信息传输到信息接收端,以此实现窄带物联网环境下电力物资供应信息自动共享.经实验证明,设计方法共享延迟时间在1s以内,丢包率也在 1%以内,在电力物资供应信息自动共享方面具有良好的应用前景.
At present,the rapid growth of iot services has brought about complex card security issues,and the recognition accu-racy of traditional machine learning algorithms is low in the case of unbalanced samples.From the data point of view,implement the Spark-Smote oversamplingalgorithm in the distributed environment,and select the characteristic data based on Relief-Fmethod.Combined with the improved value normalization method,and finally send to the classifier to learn and output the risk level type of the iot card,so as to realize the risk monitoring of the iot card in the China Unicom live network.The experimental results show that the proposed method achieves better classification accuracy and execution efficiency.
This article proposes a method of using the DETR model for intelligent road recognition in intelligent vehicle autonomous driving,and improves the DETR model.based on the DETR model,the shortcomings in dealing with complex problems in road intelligent recognition were elaborated.In response to the existing problems,the Swin Transformer module was added to DETR to improve the performance of road object detection.At the same time,a multi head self attention mechanism was adopted to achieve high-precision recognition of road multiple targets,achieving the goal of model optimization,this article compares the improved DETR network with other common recognition algorithms through experiments and result analysis.The results show that the improved DETR network outperforms other models in terms of accuracy,recall,and average accuracy.
This paper,by deploying and testing three university-level student innovation and entrepreneurship training program projects at Guangzhou University of Technology,provides a detailed account of the process of deploying a Java project to a light-weight application server using the Baota Control Panel.The process includes server procurement and configuration,domain registration and resolution,opening relevant ports and configuring project-specific environments,importing project databases and source code,binding domains,deploying projects,and conducting online testing.The research findings can serve as a ref-erence and guide for university students and Java enthusiasts who wish to deploy their projects on servers after system develop-ment.
although the combination of deep learning and face recognition can improve the response speed and the accuracy of face recognition,the computational complexity and parameters of the algorithm are very large,and the requirements for hard-ware performance are very strict,as a result,it has not been popularized.In recent years,in order to control the cost of computing,researchers have developed a variety of lightweight networks,the article as a background,first explained the principle of face recognition,and then introduced the recognition technology based on CNN,it includes MTCNN,Mobile Net and so on.At last,it discusses the concrete application of CNN in face recognition,which involves constructing data set,selecting detection meth-od,processing data and so on.
由于传统数字媒体移动端界面的设计过于单薄,导致目前用户的发展需求得不到满足.因此,现提出基于人机交互技术的数字媒体移动端界面设计方法.首先,基于人机交互技术优化界面的图像,获取图像的边缘轮廓特征,需要通过边缘检测来确定二值图像的边界曲线,进而控制了移动端界面的图像设计.其次,增强界面细节信息传输,成功实现对移动端界面图像的视觉传递优化,并对人机交互界面进行辅助控制.为了证明以图像处理技术为基础的基于人机交互技术的数字媒体移动端界面设计方法的可行性,进行测试后,各个任务的测试结果与实验所设定的预期结果相符合,证明了基于人机交互技术的数字媒体移动端界面设计方法的功能拥有较强的稳定性,满足此次软件测试的要求.
Using data mining technology to analyze abnormal data of potential thieves can effectively improve the early warning of theft and optimize the allocation of public security forces.This article will study how to use information gain decision trees to quickly classify intelligence information suspected of theft.The classification process involves continuously generating new branches starting from the suspected theft behavior attribute of the root node.Generating each branch requires calculating the information gain of different attributes and selecting split attributes.The method in the article can quickly classify a large amount of intelligence information basic data suspected of theft,and can be combined with other data mining methods such as association analysis,clustering analysis,and anomaly detection in practical work.
