检索结果-内蒙古大学图书馆

conference on signal processing, Sensor/Information Fusion, and Target Recognition xxVIII

作者： Walvoord, Derek J. Couwenhoven, Doug W. Harris Corp 400 Initiat Dr Rochester NY 14624 USA

ISBN: (纸本)9781510627024

Registration of image collections and video sequences is a critical component in algorithms designed to extract actionable intelligence from remotely sensed data. While methodologies for registration continue to evolve, the accuracy of alignment remains dependent on how well the approach tolerates changes in capture geometry, sensor characteristics, and scene content. Differences in imaging modality and field-of-view present additional challenges. Registration techniques have progressed from simple, global correlation-based algorithms, to higher-order model fitting using salient image features, to two-stage approaches leveraging high-fidelity sensor geometry, to new methods that exploit high-performance computing and convolutional neural networks (ConvNets). The latter offers important advantages by removing model assumptions and learning feature extraction directly through the minimization of a registration cost function. Deep learning approaches to image registration are still relatively unexplored for overhead imaging, and their ability to accommodate a large problem domain offers potential for several new developments. This work presents a new network architecture that improves accuracy and generalization capabilities over our modality-agnostic deep learning approach to registration that recently advanced the state of the art. A thoroughly tested ConvNet pyramid remains the core of our network approach, and has been optimized for registration and generalized to begin addressing derivative applications such as mosaic generation. Further modifications, such as objective function masking and reduced interpolation, have also been implemented to improve the overall registration process. As before, the trained network ingests image frames, applies a vector field, and returns a version of the input image that has been warped to the reference. Qualitative and quantitative performance of the new architecture is evaluated using several overhead still and full-motion video

关键词： deep learning convolutional neural networks registration stabilization

来源：评论

学校读者我要写书评

暂无评论

Scene Text-Line Extraction with Fully Convolutional Network and Refined Proposals

Scene Text-Line Extraction with Fully Convolutional Network ...

引用

Asia-Pacific signal and Information processing Association Annual Summit and conference (APSIPA)

作者： Guan-xin Zeng Yu-Hong Hou Po-Chyi Su Li-Wei Kang National Central University Taiwan National Taiwan Normal University Taiwan

ISBN: (数字)9789881476883

ISBN: (纸本)9781728181301

Texts appearing in images are often regions of interest. Locating such areas for further analysis can help to extract image-related information and facilitate many applications. Pixel-based segmentation and region-based object classification are two methodologies for identifying text areas in images and have their own pros and cons. In this research, a text detection scheme consisting of a pixel-based classification network and a supplemented region proposal network is proposed. The main network is a Fully Convolutional Network (FCN) employing Feature Pyramid Networks (FPN) and Atrous Spatial Pyramid Pooling (ASPP) to indicate possible text areas and text borders with high recall. Certain areas are further processed by the refinement network, i.e., a simplified Connectionist Text Proposal Network (CTPN) with high precision. Non-Maximum Suppression (NMS) is then applied to form appropriate text-lines. The experimental results show feasibility of the scheme.

关键词： Feature extraction Convolution Semantics Labeling Kernel Proposals image segmentation

来源：评论

学校读者我要写书评

暂无评论

Fast Optical Coherence Tomography image Enhancement using Deep Learning for Smart Laser Surgery: Preliminary Study in Bone Tissue 4

Fast Optical Coherence Tomography Image Enhancement using De...

引用

4th International conference on applications of Optics and Photonics (AOP)

作者： Bayhaqi, Yakub A. Rauter, Georg Navarini, Alexander Cattin, Philippe C. Zam, Azhar Univ Basel Dept Biomed Engn Biomed Laser & Opt Grp BLOG Gewerbestr 14 CH-4123 Allschwil Switzerland Univ Basel Dept Biomed Engn Bioinspired RObots MED Lab BIROMED Lab Gewerbestr 14 CH-4123 Allschwil Switzerland Univ Basel Dept Biomed Engn Dermatol Hebelstr 20 CH-4031 Basel Switzerland Univ Basel Dept Biomed Engn Ctr Med Image Anal & Nav CIAN Gewerbestr 14 CH-4123 Allschwil Switzerland

ISBN: (数字)9781510631649

ISBN: (纸本)9781510631649

One of the most common image denoising technique used in Optical Coherence Tomography (OCT) is the frame averaging method. Inherent to this method is that the more images are used, the better the resulting image. This approach comes, however, at the price of increased acquisition time and introduced sensitivity to motion artifacts. To overcome these limitations, we proposed an artificial neural network architecture able to imitate an averaging method using only a single image frame. The reconstructed image has an improvement in the signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) parameters compared to the original image. Additionally, we also observed an improvement in the sharpness of the denoised images. This result shows the possibility to use this method as a pre-processing step for real-time tissue classification in smart laser surgery especially in bone surgery.

关键词： Neural Network Optical Coherence Tomography Denoising Frame Averaging

来源：评论

学校读者我要写书评

暂无评论

LOSSLESS COMPRESSION OF CFA-SAMPLED imageS USING YDGCOCG TRANSFORMS WITH CDF wavelets 25

LOSSLESS COMPRESSION OF CFA-SAMPLED IMAGES USING YDGCOCG TRA...

引用

25th IEEE International conference on image processing (ICIP)

作者： Suzuki, Taizo Univ Tsukuba Fac Engn Informat & Syst Tsukuba Ibaraki Japan

ISBN: (纸本)9781479970612

This paper discusses reversible color transforms, in which a raw camera image captured using a color filter array (CFA) is from red, green, and blue (RGB) color space to YDgCoCg color space for loss-less compression. We found that conventional reversible color transforms (YDgCoCg transforms) are composed of three Haar wavelets, which are simple type of wavelet. Since the finding means that the YDgCoCg transforms on the basis of other wavelets that can more accurately predict pixels of interest in the predict steps have the potential to generate more sparse signals, we replaced Haar wavelets in the YDgCoCg transforms with Cohen-Daubechies-Feauveau (CDF) 5/3 and 9/7 wavelets, which were customized on the basis of the original pixel positions in two-dimensional (2D) space. The experimental results show that the extended YDgCoCg (YDgCoCg-x) transforms, which achieve higher sparsity, outperformed the conventional transforms when they are applied to the color transform parts in the JPEG 2000 and JPEG extended range (xR) lossless modes.

关键词： Color filter array lossless compression reversible color transforms wavelets YDgCoCg color space

来源：评论

学校读者我要写书评

暂无评论

An Always-On 3.8 μJ/86% CIFAR-10 Mixed-signal Binary CNN Processor With All Memory on Chip in 28-nm CMOS

引用

IEEE JOURNAL OF SOLID-STATE CIRCUITS 2019年第1期54卷 158-172页

作者： Bankman, Daniel Yang, Lita Moons, Bert Verhelst, Marian Murmann, Boris Stanford Univ Dept Elect Engn Stanford CA 94305 USA Katholieke Univ Leuven ESAT MICAS Dept Elect Engn B-3001 Heverlee Belgium

The trend of pushing inference from cloud to edge due to concerns of latency, bandwidth, and privacy has created demand for energy-efficient neural network hardware. This paper presents a mixed-signal binary convolutional neural network (CNN) processor for always-on inference applications that achieves 3.8 mu J/classification at 86% accuracy on the CIFAR-10 image classification data set. The goal of this paper is to establish the minimum-energy point for the representative CIFAR-10 inference task, using the available design tradeoffs. The BinaryNet algorithm for training neural networks with weights and activations constrained to + 1 and -1 drastically simplifies multiplications to xNOR and allows integrating all memory on-chip. A weight-stationary, data-parallel architecture with input reuse amortizes memory access across many computations, leaving wide vector summation as the remaining energy bottleneck. This design features an energy-efficient switched-capacitor (SC) neuron that addresses this challenge, employing a 1024-bit thermometer-coded capacitive digital-to-analog converter (CDAC) section for summing pointwise products of CNN filter weights and activations and a 9-bit binary-weighted section for adding the filter bias. The design occupies 6 mm(2) in 28-nm CMOS, contains 328 kB of on-chip SRAM, operates at 237 frames/s (FPS), and consumes 0.9 mW from 0.6 V/0.8 V supplies. The corresponding energy per classification (3.8 mu J) amounts to a 40x improvement over the previous low-energy benchmark on CIFAR-10, achieved in part by sacrificing some programmability. The SC neuron array is 12.9x more energy efficient than a synthesized digital implementation, which amounts to a 4x advantage in system-level energy per classification.

关键词： Binarized neural networks deep learning mixed-signal processing near-memory computing switched-capacitor (SC)

来源：评论

学校读者我要写书评

暂无评论

Segmentation of Teeth Region via Machine Learning in Panoramic x-Ray Dental images

Segmentation of Teeth Region via Machine Learning in Panoram...

引用

IEEE signal processing and Communications applications (SIU)

作者： Ali Güven Imam Samil Yetik Ahmet Çulhaoğlu Kaan Orhan Mehmet Kılıçarslan Kılıçarslan TOBB Ekonomi ve Teknoloji Üniversitesi Ankara Diş Hekimliği Fakültesi Kırıkkale Üniversitesi Kırıkkale Diş Hekimliği Fakültesi Ankara Üniversitesi Ankara

ISBN: (数字)9781728172064

ISBN: (纸本)9781728172071

Segmentation of teeth region from the dental panoramic x-Ray images is an important task in determining various diseases. The main goal of this article is to be able to automatically segment the region of teeth in panoramic x-ray images. First, the center point of the teeth area in the images was determined automatically. Then, a feature set was developed including intensity values of pixels, x-coordinate relative to this center point, y-coordinate relative to this point, and the pixel values obtained by subtraction of maximum and minimum values in 3×3 window. CatBoost algorithm was used for machine learning. When creating the machine learning model, k-fold cross validation of training data set and grid search optimization of hyper parameters, were applied to avoid over fitting of data set. The results were analyzed using the learning curve, F1, accuracy, recall, and precision scores.

关键词： x-ray imaging Teeth image segmentation Machine learning Dentistry Training data Task analysis

来源：评论

学校读者我要写书评

暂无评论

Kernel Computations from Large-Scale Random Features Obtained by Optical processing Units

Kernel Computations from Large-Scale Random Features Obtaine...

引用

IEEE International conference on Acoustics, Speech and signal processing

作者： Ruben Ohana Jonas Wacker Jonathan Dong Sebastien Marmin Florent Krzakala Maurizio Filippone Laurent Daudet Laboratoire de Physique de l’Ecole Normale Superiéure ENS Université PSL CNRS Sorbonne Université Université de Paris Paris France Department of Data Science EURECOM Biot France LightOn Paris France

ISBN: (数字)9781509066315

ISBN: (纸本)9781509066322

Approximating kernel functions with random features (RFs) has been a successful application of random projections for nonparametric estimation. However, performing random projections presents computational challenges for large-scale problems. Recently, a new optical hardware called Optical processing Unit (OPU) has been developed for fast and energy-efficient computation of large-scale RFs in the analog domain. More specifically, the OPU performs the multiplication of input vectors by a large random matrix with complexvalued i.i.d. Gaussian entries, followed by the application of an element-wise squared absolute value operation - this last nonlinearity being intrinsic to the sensing process. In this paper, we show that this operation results in a dot-product kernel that has connections to the polynomial kernel, and we extend this computation to arbitrary powers of the feature map. Experiments demonstrate that the OPU kernel and its RF approximation achieve competitive performance in applications using kernel ridge regression and transfer learning for image classification. Crucially, thanks to the use of the OPU, these results are obtained with time and energy savings.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Ship detection and classification with terrestrial hyperspectral data based on convolutional neural networks 25

Ship detection and classification with terrestrial hyperspec...

引用

conference on image and signal processing for Remote Sensing xxV

作者： Schenkel, Fabian Wohnhas, Benjamin Gross, Wolfgang Schreiner, Simon Bagov, Ilia Middelmann, Wolfgang Fraunhofer Inst Optron Syst Technol & Image Explo Karlsruhe Germany

ISBN: (数字)9781510630147

ISBN: (纸本)9781510630147

The detection and classification of maritime objects in a harbour environment or coastal areas using a terrestrial hyperspectral system in combination with a high-resolution RGB sensor is a challenging task since the large number of spectral channels requires a robust analytical method. Recently, deep learning methods have shown a good performance in many computer vision applications. In this paper, we present a general analysis workflow for ship detection and classification based on fused terrestrial RGB and hyperspectral images, which employs a deep learning network for the localization of ships in the high-resolution images and a following convolutional neural network based multi-input model for the classification of each detected object. During a measurement campaign, images of various ship types were collected under distinct weather conditions for the training and evaluation of the network model. In the first part of the workflow, ship candidates were located using the Mask R-CNN framework based on the RGB images. For the following classification process, which was trained to separate different ship type classes, we developed a multi-input convolutional neural network using the RGB and the hyperspectral images as input data. For the pre-processing procedure of the hyperspectral data a principal component analysis was applied to reduce the number of input channels for the network while still maintaining a large fraction of the initial information. For the architecture of the RGB classification branch, the structure and the weights of a pre-trained model were integrated and fine-tuned. Since only limited training data was available, regularization methods and data augmentation were employed. The detection and the multi-input classification network was finally evaluated and showed that the classification performance can be increased when integrating additional information from a hyperspectral sensor.

关键词： Hyperspectral Data Classification Object Detection Convolutional Neural Networks Principal Component Analysis Maritime Environment

来源：评论

学校读者我要写书评

暂无评论

The Design, by Physical Topology Optimizing, of a Passive UHF RFID Identification System: Suitable for applications with Various Constraints

The Design, by Physical Topology Optimizing, of a Passive UH...

引用

International conference on Advanced Technologies for signal and image processing (ATSIP)

作者： Rahma Zayoud Habib Hamam Faculty of Engineering Université de Moncton N-B CANADA

ISBN: (数字)9781728175133

ISBN: (纸本)9781728175140

RFID technology is booming. It catches the attention of several researchers. This technology uses the radio wave to identify objects on which an RFID tag is placed. RFID has a lot of advantages, but it also has limitations. Its limits are its sensitivity to liquids, metals and speed. We design an objects identification system by radio frequency that works in different RFID technology application environments, based on passive UHF RFID technology and through the physical topology optimization in real time too. This system is also adapted to various constraints, in order to solve overcome difficulties, related to various applications, at once, and operates, without any problem, in different fields. We use the simulated annealing algorithm to find the optimal physical topology that has the highest average reading rate. RFID antennas are installed on brackets, attached to a tripod system by servo motors. These servo motors are controlled by the middleware, where the simulated annealing optimization algorithm is already implemented, to vary the angles of the detection connectors in an automated way, in order to find the optimal topology. So the optimization process will be guided by an optimization algorithm and not by trial and error processes. Tripod mechanical systems are movable media for antennas to play on xYZ dimensions too. The design of the mechanical system and the choice of materials were made. The first results, such as the identity of each detected object and the identity of each antenna that will detect an object, are here. Following reached results will be shared in several next publications.

关键词： Radiofrequency identification Antennas Passive RFID tags Topology Optimization Object recognition Metals

来源：评论

学校读者我要写书评

暂无评论

Non-line-of-sight Imaging with Radio signals

Non-line-of-sight Imaging with Radio Signals

引用

Asia-Pacific signal and Information processing Association Annual Summit and conference (APSIPA)

作者： Ying He Dongheng Zhang Yang Hu Yan Chen School of Information and Communication Engineering University of Electronic Science and Technology of China School of Cyberspace Security University of Science and Technology of China

ISBN: (数字)9789881476883

ISBN: (纸本)9781728181301

Non-line-of-sight (NLOS) imaging reconstructs the shape and albedo of objects outside the field of view of an imaging system. Based on this technique, a wide range of applications such as autonomous vehicle navigation, robotic and machine vision, as well as medical imaging will possess unprecedented capabilities. Recent NLOS imaging approaches employ ultra-fast pulsed laser and time-resolved photon detector to collect light transients, which are however time-consuming due to the point-by-point grid scan on the wall. Moreover, the maximum imaging distance is only a few meters because of the limitation of the photon counter. In this paper, we introduce Radio Frequency (RF) NLOS imaging to reconstruct hidden objects with radio signals. By analyzing the image formation model, we transform the reconstruction problem into an optimization problem that can be solved with additional constraints. The simulation results validate that we can reconstruct letter-shaped objects using a 12×12 multipleinput multiple-output (MIMO) array and a stepped frequency signal with a bandwidth of 4GHz., Compared with the optical NLOS imaging, the proposed method can capture the hidden 3D geometry at longer ranges with shorter acquisition times.

关键词： Imaging Nonlinear optics image reconstruction Scattering Receivers Photonics Optical imaging

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：