Wavelet enabled convolutional autoencoder based deep neural network for hyperspectral image denoising

Denoising of hyperspectral images (HSIs) is an important preprocessing step to enhance the performance of its analysis and interpretation. In reality, a remotely sensed HSI experiences disturbance from different sources and therefore gets affected by multiple noise types. However, most of the existing denoising methods concentrates in removal of a single noise type ignoring their mixed effect. Therefore, a method developed for a particular noise type doesn’t perform satisfactorily for other noise types. To address this limitation, a denoising method is proposed here, that effectively removes multiple frequently encountered noise patterns from HSI including their combinations. The proposed dual branch deep neural network based architecture works on wavelet transformed bands. The first branch of the network uses deep convolutional skip connected layers with residual learning for extracting local and global noise features. The second branch includes layered autoencoder together with subpixel upsampling that performs repeated convolution in each layer to extract prominent noise features from the image. Two hyperspectral datasets are used in the experiment to evaluate the performance of the proposed method for denoising of Gaussian, stripe and mixed noises. Experimental results demonstrate the superior performance of the proposed network compared to other state-of-the-art denoising methods with PSNR 36.74, SSIM 0.97 and overall accuracy 94.03?%.

     

Feature learning using convolutional denoising autoencoder for activity recognition

Neural Computing and Applications - Wearable technology offers a prospective solution to the increasing demand for activity monitoring in pervasive healthcare. Feature extraction and selection are...     

Development of a wearable guide device based on convolutional neural network for blind or visually impaired persons

Multimedia Tools and Applications - This study proposes a design for a wearable guide device for blind or visually impaired persons on the basis of video streaming and deep learning. This work...     

基于降噪自编码神经网络的化合物毒性预测方面的研究

常规毒理学实验方法周期长,耗资高,对现代药物研发和环境化合物安全性评估具有局限性,通过对化合物毒理性研究,提取1047维分子指纹特征,提出去噪自编码神经网络无监督学习机制及对腐败特征的自联想学习特性提取隐含毒性化合物特征,实现化合物毒性预测和毒性化合物的活性预测。该方法在化合物毒性预测和活性预测中的预测精度分别为79.825%,80.85%, 敏感性分别为79.62%,80.25%,特异性分别为80.03%,81.45%。实验结果表明,去噪自编码网络较浅层机器学习更适用于高通量化合物毒性预测, 较传统自编码网络更具优越性。     

Developing a way-finding system on mobile robot assisting visually impaired people in an indoor environment

Multimedia Tools and Applications - A way-finding system in an indoor environment consists of several components: localization, representation, path planning, and interaction. For each component,...     

基于卷积神经网络的驾驶辅助系统设计

为了减轻驾驶员在行驶过程中的操作负担,进而降低误差判断事件的出现几率,设计一种基于卷积神经网络的驾驶辅助系统。在执行良好的汽车导航架构中,限定Learning Navigation模块与Learning Controller模块的连接位置,再根据辅助驾驶传感器对于行驶画面的采集情况,对车辆巡航能力进行定向控制,抑制监测仪表中辅助波的过渡振动,完成驾驶辅助系统的需求与设计分析。在此基础上,确定辅助激活函数、约束仪表中的行车图像,建立标准化的卷积神经网络。按照驾驶辅助数据的学习结果,对其进行传输处理,进而连接驾驶辅助系统的Job请求,实现系统的顺利运行。利用卷积神经网络平台设计实车实验结果表明,应用驾驶辅助系统后,车辆监测仪表中辅助波振动幅度的最小值处于36-61Hz之间,平均波长偏移量明显减小,驾驶员的行驶操作负担得到有效缓解。     

Multi-scale dilated convolution of convolutional neural network for image denoising

Multimedia Tools and Applications - Convolutional Neural Network has achieved great success in image denoising. The conventional methods usually sense those beyond scope contextual info at the...     

基于卷积神经网络的行人目标检测系统设计

为获得最直观的行人目标检测结果,避免运动姿态不确定性对实时检测造成的影响,设计基于卷积神经网络的行人目标检测系统。以CNN计算框架作为硬件结构主体,分级连接目标传感器与神经型卷积分类器,按照并行检测原理及卷积神经架构搭建检测体系结构。建立训练文件体系,通过迎合目标训练环境的方式,配置必要的检测文件参数,完成待检测行人目标的样本训练处理。在检测节点架构中,规定与访问接口关联的配置条件,借助增设的模块复用加速结构,直接获取行人目标检测结果,实现行人目标的样本重构,完成基于卷积神经网络的行人目标检测系统设计。实验结果表明,与PCA、SVM算法相比,应用卷积神经网络型检测系统后,单位时间内的行人目标检测量达到9.6×109T,目标数据堆积速率降低至1.14×109T/s,能够直观获取行人目标检测结果,有效抑制了运动姿态不确定性对系统实时检测的影响。     

Mixture reality-based assistive system for visually impaired people

Individuals with visual impairments often face challenges in their daily lives, particularly in terms of independent mobility. To address this issue, we present a mixed reality-based assistive system for visually impaired individuals, which comprises a Microsoft Hololens2 device and a website and utilizes a simultaneous localization and mapping (SLAM) algorithm to capture various large indoor scenes in real-time. This system incorporates remote multi-person assistance technology and navigation technology to aid visually impaired individuals. To evaluate the effectiveness of our system, we conducted an experiment in which several participants completed a large indoor scene maintenance task. Our experimental results demonstrate that the system is robust and can be utilized in a wide range of indoor environments. Additionally, the system enhances environmental perception and enables visually impaired individuals to navigate independently, thus facilitating successful task completion.     

A wearable system for mobility improvement of visually impaired people

Degradation of the visual system can lead to a dramatic reduction of mobility by limiting a person to his sense of touch and hearing. This paper presents the development of an obstacle detection system for visually impaired people. While moving in his environment the user is alerted to close obstacles in range. The system we propose detects an obstacle surrounding the user by using a multi-sonar system and sending appropriate vibrotactile feedback. The system aims at increasing the mobility of visually impaired people by offering new sensing abilities.     

基于尺度特征卷积神经网络的高分对地观测系统设计

针对高分对地观测系统使用过程中会受到不同活动项目的 约束影响,出现系统成像、回传及活动完成率低的问题,导致观测效果不佳,为此提出了基于尺度特征卷积神经网络的高分对地观测系统设计;该系统通过管控中心服务器推送系统运行状态信息,实现三维显示任务的功能;利用CMOS图像传感器实现成像面对应点的传送和FPGA控制器控制其数据存储时间;采用BCM5464千兆交换机,实现数据高速传输;构建并训练尺度特征卷积神经网络,利用RPN网络识别目标区域特征,通过划分目标的前景和背景确定了该区域内的训练兴趣区域坐标,从而使RPN网络权值学习达到了预期目标,提升了目标检测识别的准确性,设计对地观测信息管理流程,完成系统设计;由实验结果可知,该系统最高成像、回传概率、活动完成率分别为83％、99.9％和100％,具有良好的观测效果.     

iNavigation: an image based indoor navigation system

This paper presents a novel image-based indoor navigation web application designed for mobile phone. It is inspired by Google Street View that features 360° imagery for navigation. Ordinary data collection of image based navigation systems implements panorama cameras, so it is difficult to be extended to indoor environment. On the other hand, they cannot provide timely updates because it requires immense image data. This paper introduces a ‘proof of concept’ which only uses ordinary organized photo collections instead of panoramic photo to guide people through the building. It implements SIFT (scale-invariant feature transform) feature detection and ANN (approximately nearest neighbor) search to provide positioning service. People can upload query images to obtain current position. It also enables information sharing by using IPM (inverse perspective mapping) technique to figure out distance from a single query image, and update the query image into the image collection correctly based on the distance calculation.     

一种高效的稀疏卷积神经网络加速器的设计与实现

针对卷积神经网络计算硬件化实现困难的问题,之前大部分卷积神经网络加速器的设计都集中于解决计算性能和带宽瓶颈,忽视了卷积神经网络稀疏性对加速器设计的重要意义,近来少量的能够利用稀疏性的卷积神经网络加速器设计也往往难以同时兼顾计算灵活度、并行效率和资源开销。本文首先比较了不同并行展开方式对利用稀疏性的影响,分析了利用稀疏性的不同方法,然后提出了一种能够利用激活稀疏性加速卷积神经网络计算的同时,相比于同领域其他设计,并行效率更高、额外资源开销更小的并行展开方法,最后完成了这种卷积神经网络加速器的设计并在FPGA上实现。研究结果表明:运行VGG-16网络,在ImageNet数据集下,该并行展开方法实现的稀疏卷积神经网络加速器和使用相同器件的稠密网络设计相比,卷积性能提升了108.8%,整体性能提升了164.6%,具有明显的性能优势。     

基于卷积降噪自编码器的苹果树种鉴别模型研究

结合卷积降噪自编码器与随机森林算法,提出一种新型的卷积降噪自编码器-随机森林（CDAE-RF）模型,并基于可见-近红外光谱数据集来识别苹果树种。首先,通过网格式搜索、平行实验的方法优化了L1范数等参数,提高了模型的鲁棒性;然后,对比实验分析了CDAE-RF、主成分分析-随机森林模型（PCA-RF）、K最近邻分类算法等方法在不同噪声水平下光谱识别的准确性和鲁棒性。实验结果表明,相对于传统算法,新提出的CDAE-RF模型识别准确率达97.92%,在加噪情况下具有更高的鲁棒性。CDAE-RF模型降低了随机森林算法对噪声的敏感性,提高了噪声光谱图像识别的准确性,为地物波谱识别提供了一种新的方法。     

An edge detection framework conjoining with IMU data for assisting indoor navigation of visually impaired persons

Smartphone applications based on object detection techniques have recently been proposed to assist visually impaired persons with navigating indoor environments. In the smartphone, digital cameras are installed to detect objects which are important for navigation. Prior to detect the interested objects from images, edges on the objects have to be identified. Object edges are difficult to be detected accurately as the image is contaminated by strong image blur which is caused by camera movement. Although deblurring algorithms can be used to filter blur noise, they are computationally expensive and not suitable for real-time implementation. Also edge detection algorithms are mostly developed for stationary images without serious blur. In this paper, a modified sigmoid function (MSF) framework based on inertial measurement unit (IMU) is proposed to mitigate these problems. The IMU estimates blur levels to adapt the MSF which is computationally simple. When the camera is moving, the topological structure of the MSF is estimated continuously in order to improve effectiveness of edge detections. The performance of the MSF framework is evaluated by detecting object edges on video sequences associated with IMU data. The MSF framework is benchmarked against existing edge detection techniques and results show that it can obtain comparably lower errors. It is further shown that the computation time is significantly decreased compared to using techniques that deploy deblurring algorithms, thus making our proposed technique a strong candidate for reliable real-time navigation.     

基于集成降噪自编码的在线网络入侵检测模型

针对神经网络在线入侵检测模型训练时易出现过拟合和泛化能力弱的问题,提出基于改进的集成降噪自编码在线入侵检测模型以区分正常和异常的流量模式。降噪自编码减少了训练数据与测试数据的差别,缓解过拟合问题,提高模型的性能。同时阈值的选择方法直接影响网络入侵检测模型检测精度,该阈值采用随机方法确定,无须于离线入侵检测,不需通过完整的数据集即可选择最佳的阈值。采用CICIDS2017中的异常的数据流对模型进行测试,准确率分别为90.19%。结果表明,作为一种在线检测模型,提出的异常检测模型优于其他异常检测方法。     

Multi-class object detection system using hybrid convolutional neural network architecture

Multimedia Tools and Applications - Object detection in computer vision has been a significant research area for the past decade. Identifying objects with multiple classes from an image has...     

Comprehensive assessment for motor and visually impaired people using a hierarchical model

Universal Access in the Information Society - The evolution of various modern technologies has inspired researchers to assess the effectiveness of these technologies for people with diversified...     

结合降噪卷积神经网络和条件生成对抗网络的图像双重盲降噪算法

针对图像降噪中降噪效果差、计算效率低的问题,提出了一种结合降噪卷积神经网络（DnCNN）和条件生成对抗网络（CGAN）的图像双重盲降噪算法。首先,使用改进的DnCNN模型作为CGAN的生成器来对加噪图片的噪声分布进行捕获;其次,将剔除噪声分布后的加噪图片和标签一同送入判别器进行降噪图像的判别;然后,利用判别结果对整个模型的隐层参数进行优化;最后,生成器和判别器在博弈中达到平衡,且生成器的残差捕获能力达到最优。实验结果表明,在Set12数据集上,当噪声水平分别为15、25、50时：所提算法与DnCNN算法相比,基于像素点间误差评价指标,其峰值信噪比（PSNR）值分别提升了1.388 dB、1.725 dB、1.639 dB;所提算法与三维块匹配（BM3D）、加权核范数最小化（WNNM）、DnCNN、收缩场级联（CSF）和一致性神经网络（CSNET）等现有算法相比,结构相似性（SSIM）评价指标值平均提升了0.000 2~0.104 1。实验结果验证了所提算法的优越性。     

Modelling of navigation based LNA parameters using neural network technique

An approach to model the parameters of LNA which is ideal for GLONASS navigation system. To design LNA, multilayer perceptron architecture is used. The parameters of LNA are calculated using Levenberg Marquardt Backpropagation Algorithm for the frequency range 300 MHz to 18 GHz. ANN model is trained using Agilent MGA 71543 Low Noise Amplifier datasheet and this model shows high regression. The smith and polar charts are plotted for frequency range 300 MHz to 18 GHz and parameters are calculated for center frequency of L1 band of GLONASS, which is 1.602 GHz.