An end-to-end framework for the detection of mathematical expressions in scientific document images

The detection of mathematical expressions is a prerequisite step for the digitisation of scientific documents. Many different multistage approaches have been proposed for the detection of expressions in document images, that is, page segmentation and expression detection. However, the detection accuracy of such methods still needs improvement owing to errors in the page segmentation of complex documents. This paper presents an end-to-end framework for mathematical expression detection in scientific document images without requiring optical character recognition (OCR) or document analysis techniques applied in conventional methods. The novelty of this paper is twofold. First, because document images are usually in binary form, the direct use of these images, which lack texture information as input for detection networks, may lead to an incorrect detection. Therefore, we propose the application of a distance transform to obtain a discriminating and meaningful representation of mathematical expressions in document images. Second, the transformed images are fed into the faster region with a convolutional neural network (Faster R-CNN) optimized to improve the accuracy of the detection. The proposed framework was tested on two benchmark data sets (Marmot and GTDB). Compared with the original Faster R-CNN, the proposed network improves the accuracies of detection of isolated and inline expressions by 5.09% and 3.40%, respectfully, on the Marmot data set, whereas those on the GTDB data set are improved by 4.04% and 4.55%. A performance comparison with conventional methods shows the effectiveness of the proposed method.     

基于改进Faster R-CNN的嘴部检测方法

在通过嘴部进行人机交互的场景下,外界光线变化、小目标检测的复杂性、检测方法的不通用性等因素给不同场景下嘴部的识别带来了很大困难.该文以不同场景下的人脸图像为数据源,提出了一种基于改进Faster R-CNN的人脸嘴部识别算法.该方法在Faster R-CNN框架中结合多尺度特征图进行检测,首先将同一卷积块不同卷积层输出的特征图结合,然后对不同的卷积块按元素进行求和操作,在输出的特征图上进行上采样得到高分辨率的表达能力更强的特征,从而提高了嘴部这种小目标的检测性能.在网络训练试验中运用多尺度训练和增加锚点数量增强网络检测不同尺寸目标的鲁棒性.实验表明,相比于原始的Faster R-CNN,对嘴部的检测准确率提高了8%,对环境的适应性更强.     

基于LabVIEW和Mask R-CNN的柱塞式制动主缸内槽表面缺陷检测

为了解决传统图像处理方法对于铸铝材料表面缺陷检测通用性不高、准确度低等问题,研究了一种基于Mask R-CNN神经网络的缺陷检测系统。首先,采用自主研发的缺陷检测装置采集柱塞式制动主缸内槽表面图像,对其进行预处理,制作成Microsoft COCO格式数据集;其次,搭建适用于该数据集的Mask R-CNN神经网络结构,并绘制训练过程损失函数与平均精度均值曲线;最后,将检测结果与基于SVM和Faster R-CNN模型的检测结果进行比较,统计了3种神经网络模型的单图检测平均时间和识别率。试验结果表明,在相同样本条件下,该方法的识别率比另外2种方法高,达到了93.6%,能够更精确地检测柱塞式制动主缸内槽的表面缺陷。     

基于改进的Mask R-CNN的行人细粒度检测算法

针对复杂场景下行人检测效果差的问题,采用基于深度学习的目标检测中领先的研究成果,提出了一种基于改进Mask R-CNN框架的行人检测算法。首先,采用K-means算法对行人数据集的目标框进行聚类得到合适的长宽比,通过增加一组长宽比（2：5）使12种anchors适应图像中行人的尺寸;然后,结合细粒度图像识别技术,实现行人的高定位精度;其次,采用全卷积网络（FCN）分割前景对象,并进行像素预测获得行人的局部掩码（上半身、下半身）,实现对行人的细粒度检测;最后,通过学习行人的局部特征获得行人的整体掩码。为了验证改进算法的有效性,将其与当前具有代表性的目标检测方法（如更快速的区域卷积神经网络（Faster R-CNN）、YOLOv2、R-FCN）在同数据集上进行对比。实验结果表明,改进的算法提高了行人检测的速度和精度,并且降低了误检率。     

基于Mask R-CNN与SG滤波的手势识别关键点特征提取方法

手势识别是人机交互的重要手段.为了精确识别手势并摒除光照等环境干扰,同时减除由于手部高维运动造成的关键点剧烈抖动的问题,提出一种基于基于蒙版区域的卷积神经网络(Mask Region-based convolutional neural network,Mask R-CNN)与多项式平滑算法(Savitzky-Golay,SG)的手势关键点提取方法.该方法首先对输入的红绿蓝(RGB)三通道图像进行特征提取与区域分割,获得手部的实例分割与掩码.然后利用ROIAling及功能性网络进行目标匹配,标记出22个关键点(21个骨骼点+1个背景点).将标记后结果送入SG滤波器进行数据平滑,并进行骨骼点的重新标定.从而得到稳定的手势提取特征.对模型进行对比实验,结果表明,该方法能够最大程度摒除环境干扰,并精准提取关键点.与传统基于轮廓分割的手势关键点提取相比,模型的鲁棒性大大提高,识别精度达到93.48％.     

基于改进 Cascade R-CNN 的输电线路多目标检测

针对无人机巡检图像中小目标难以检测、障碍物遮挡目标、正负样本不平衡等问题,提出基于改进Cascade R-CNN的输电线路多目标检测方法.改进了Cascade R-CNN的特征提取网络,基于ResNet101基础网络结构,设计6层新型特征金字塔网络(FPN)与之融合,提高了对小目标、重叠目标的检测能力;引入了高斯形式的软非极大值抑制(Soft-NMS)方法,降低了受遮挡影响的目标的漏检率;利用Focal损失改进损失函数,缓解了正负样本不平衡对检测精度的影响.训练过程中,基于添加噪声、亮度变换、尺度放缩等数据增强方法扩充数据集,提升了训练模型的泛化性能.实验结果表明,改进的模型在复杂背景下能够对3种瓷质绝缘子、瓷质绝缘子缺陷、相间棒、防震锤以及鸟窝同时检测,平均精度均值(mAP)达到94.1％,为输电线路的智能巡检提供了一种新思路.     

A Secure Method for Data Storage and Transmission in Sustainable Cloud Computing

Cloud computing is a technology that provides secure storage space for the customer’s massive data and gives them the facility to retrieve and transmit their data efficiently through a secure network in which encryption and decryption algorithms are being deployed. In cloud computation, data processing, storage, and transmission can be done through laptops and mobile devices. Data Storing in cloud facilities is expanding each day and data is the most significant asset of clients. The important concern with the transmission of information to the cloud is security because there is no perceivability of the client’s data. They have to be dependent on cloud service providers for assurance of the platform’s security. Data security and privacy issues reduce the progression of cloud computing and add complexity. Nowadays; most of the data that is stored on cloud servers is in the form of images and photographs, which is a very confidential form of data that requires secured transmission. In this research work, a public key cryptosystem is being implemented to store, retrieve and transmit information in cloud computation through a modified Rivest-Shamir-Adleman (RSA) algorithm for the encryption and decryption of data. The implementation of a modified RSA algorithm results guaranteed the security of data in the cloud environment. To enhance the user data security level, a neural network is used for user authentication and recognition. Moreover; the proposed technique develops the performance of detection as a loss function of the bounding box. The Faster Region-Based Convolutional Neural Network (Faster R-CNN) gets trained on images to identify authorized users with an accuracy of 99.9% on training.     

基于改进Mask R-CNN的输电线路防外破目标检测方法研究

随着现代电力系统的不断发展,电网规模越来越大,外破原因引发的故障已经成为架空输电线路故障的主要原因之一。在架空输电线路视频监控中,使用传统的边界框式目标检测方法进行外破预警时,误报或漏报的情况时有发生。掩模实例分割神经网络(Mask-RCNN)训练时使用的像素级掩模标注数据集成本较高,限制了该算法的大规模应用。针对这些问题,将改进的Mask-RCNN网络应用到输电线路外破目标检测领域,在数据集标注过程中,使用边界框标注代替部分掩模标注。训练时,将检测分支的特征迁移到掩模分支。实验结果表明,改进后的算法能够在掩模标注样本占比80%的条件下,对常见外破类别的平均识别准确率高于91%,为输电线路外破隐患的准确识别与分割提供了一种可行的思路。     

基于R-CNN的学生课堂管理系统

随着计算机智能化的发展,计算机技术已经深入到人们日常生活.利用计算机技术可以辅助教学,但目前大多数计算机技术都是应用于检测学生课堂疲劳情况.由于学生课堂中的行为有较多种,现有的方法不能准确评估学生的课堂表现,以及衡量老师的教学水平.为了解决这个问题,提出了一种新的学生课堂管理系统,在这个系统中可以检测学生的多种课堂行为...     

基于对抗神经网络和神经网络模型的筒子纱抓取方法

为提高筒子纱抓取和上纱过程的自动化和柔性化程度,应用单目视觉系统引导机器人完成上纱过程。采用工业相机获取单个筒子纱不同形态的数据信息,应用GAN(生成式对抗神经网络)对筒子纱数据集扩充,提高筒子纱数据集多样性。将所得的数据集加载到Faster R-CNN(更快速区域卷积神经网络)模型里进行训练,应用训练好的神经网络识别和定位筒子纱,引导机器人完成上纱任务。应用搭建的单目视觉系统实验平台对结果进行测试,结果表明,经过标定后的视觉系统可以完成多个筒子纱的抓取任务,以满足筒子纱上纱过程的自动化和柔性化要求。