- 标题
- 摘要
- 关键词
- 实验方案
- 产品
-
[IEEE 2018 24th International Conference on Pattern Recognition (ICPR) - Beijing, China (2018.8.20-2018.8.24)] 2018 24th International Conference on Pattern Recognition (ICPR) - Wasserstein Generative Recurrent Adversarial Networks for Image Generating
摘要： Most generative models are generating images at a time, but in fact, painting is usually done iteratively and repeatedly. Generative Adversarial Networks (GAN) are well known for generating images, however, it is hard to train stably. To tackle this problem, we propose a framework named the Wasserstein generative recurrent adversarial networks (WGRAN), which merges Wasserstein distance with recurrent neural networks to iteratively generate realistic looking images and trains our model in an adversarial way. Therefore, our generative model is gradually generates images using the feedback of discriminate model. And our approach allows us to control the number of iterations of generation. We train our model on various image datasets and compare our model with the recurrent generative adversarial networks (GRAN) and other state-of-the-art generative models using Generative Adversarial Metric. From these experiments, we show evidence that our model has the ability to generate high quantity images.
关键词： recurrent nerual netwoks,image generating,Generative Adversarial networks,Wasserstein distance
更新于2025-09-09 09:28:46
-
[IEEE 2018 20th International Conference on Transparent Optical Networks (ICTON) - Bucharest (2018.7.1-2018.7.5)] 2018 20th International Conference on Transparent Optical Networks (ICTON) - Medium-Transparent Packet-Based Fronthauling for 5G Hot-Spot Networks
摘要： In order to meet the traffic density requirement and cope with the technical challenges of hot-spot scenarios exhibiting extreme user density in very confined geographical areas, the fifth generation (5G) networks must undergo radical technological and architectural innovations that support a broad solution portfolio. This solution mix includes spectrum expansion to the millimeter Wave (mmWave) band, massive Multiple-Input Multiple- Output antennas and network densification through remote radio head deployment. Since the current Common Public Radio Interface-based mobile fronthaul cannot cope with massive mmWave multi-Gbps traffic streams, it is imperative to introduce a novel converged 5G architecture, specifically designed to facilitate mmWave access to massive amounts of users. To this end, a Medium Transparent Medium Access Control (MT-MAC) protocol has been proposed, designed to operate over a converged mmWave analog Fiber-Wireless (FiWi) fronthaul. MT-MAC allows for fast and direct negotiation of wavelength, radio frequency and time resources between the base band unit and the mmWave users, while offering fast on-demand link formation. However, its performance has not yet been evaluated under 5G heavy-traffic scenarios, as the ones previously described. Hence, in this paper, we investigate the performance of an MT-MAC-enabled fronthaul and report on its suitability for mmWave hot-spot 5G access networks.
关键词： fronthaul,optical-wireless,millimetre wave (mmWave),C-RAN,passive-optical-networks (PON),radio-over-fiber (RoF)
更新于2025-09-09 09:28:46
-
[IEEE IGARSS 2018 - 2018 IEEE International Geoscience and Remote Sensing Symposium - Valencia (2018.7.22-2018.7.27)] IGARSS 2018 - 2018 IEEE International Geoscience and Remote Sensing Symposium - A Neural Network Method for Nonlinear Hyperspectral Unmixing
摘要： Because of the complex interaction of light with the Earth surface, a hyperspectral pixel can be composed of a highly nonlinear mixture of the re?ectances of the materials on the ground. When nonlinear mixing models are applied, the estimated model parameters are usually hard to interpret and to link to the actual fractional abundances. Moreover, not all spectral re?ectances in a real scene follow the same particular mixing model. In this paper, we present a supervised learning method for nonlinear spectral unmixing. In this method, a neural network is applied to learn mappings of the true spectral re?ectances to the re?ectances that would be obtained if the mixture was linear. A simple linear unmixing then reveals the actual abundance fractions. This technique is model-independent and allows for an easy interpretation of the obtained abundance fractions. We validate this method on several arti?cial datasets, a dataset obtained by ray tracing, and a real dataset.
关键词： endmembers,Hyperspectral,abundances,neural networks
更新于2025-09-09 09:28:46
-
Automatic Graph-based Modeling of Brain Microvessels Captured with Two-Photon Microscopy
摘要： Graph models of cerebral vasculature derived from 2-photon microscopy have shown to be relevant to study brain microphysiology. Automatic graphing of these microvessels remain problematic due to the vascular network complexity and 2-photon sensitivity limitations with depth. In this work, we propose a fully automatic processing pipeline to address this issue. The modeling scheme consists of a fully-convolution neural network to segment microvessels, a 3D surface model generator and a geometry contraction algorithm to produce graphical models with a single connected component. Quantitative assessment using NetMets metrics, at a tolerance of 60 μm, false negative and false positive geometric error rates are 3.8% and 4.2%, respectively, whereas false negative and false positive topological error rates are 6.1% and 4.5%, respectively. Our qualitative evaluation confirms the efficiency of our scheme in generating useful and accurate graphical models.
关键词： segmentation,graph,Cerebral microvasculature,deep learning,convolution neural networks,two-photon microscopy
更新于2025-09-09 09:28:46
-
[IEEE 2018 37th Chinese Control Conference (CCC) - Wuhan (2018.7.25-2018.7.27)] 2018 37th Chinese Control Conference (CCC) - Real-Time Fire Detection Based on Deep Convolutional Long-Recurrent Networks and Optical Flow Method
摘要： A new ?re monitoring method was proposed in this paper, which proposed a Neural Network of Deep Convolutional Long-Recurrent Networks (DCLRN), and combining DCLRN network and optical ?ow method for ?re monitoring in open space environment in real time. This is achieved by utilizing the static and dynamic characteristics of the ?re, converting ?re RGB images to optical ?ow images in real-time, and use convolutional neural network for spatial learning, a class of recurrent convolutional architectures for sequence learning, eventually achieve the purpose of ?re monitoring. Which is end-to-end trainable and suitable for large-scale visual understanding tasks of ?re monitoring. Our novelties are: ?rstly, our method is the ?rst to our knowledge to put forward DCLRN, and combining DCLRN with optical ?ow method for ?re monitoring. Secondly, our method has the ability that can detect the smoke as well as the ?ames. Finally, in this way the ?re can be detected as soon as it occurs, achieved early detection of ?re. The experiments have proved that DCLRN combined with optical ?ow images have good accuracy and reliability in the detection and recognition of ?re monitoring videos, and give good performance on a more challenging dataset.
关键词： sequence learning,optical ?ow method,Fire monitoring,Deep Convolutional Long-Recurrent Networks
更新于2025-09-09 09:28:46
-
[IEEE 2018 IEEE 20th International Workshop on Multimedia Signal Processing (MMSP) - Vancouver, BC, Canada (2018.8.29-2018.8.31)] 2018 IEEE 20th International Workshop on Multimedia Signal Processing (MMSP) - Fast, Robust, and Accurate Image Denoising via Very Deeply Cascaded Residual Networks
摘要： Patch based image modelings have shown great potential in image denoising. They mainly exploit the nonlocal self-similarity (NSS) of either input degraded images or clean natural ones when training models, while failing to learn the mappings between them. More seriously, these algorithms have very high time complexity and poor robustness when handling images with different noise variances and resolutions. To address these problems, in this paper, we propose very deeply cascaded residual networks (VDCRN) to build the precise relationships between the noisy images and their corresponding noise-free ones. It adopts a new residual unit with an identity skip connection (shortcut) to make training easy and improve generalization. The introduction of shortcut is helpful to avoid the problem of gradient vanishing and preserve more image details. By cascading three such residual units, we build the VDCRN to deploy deeper and larger convolutional networks. Based on such a residual network, our VDCRN achieves very fast speed and good robustness. Experimental results demonstrate that our model outperforms a lot of state-of-the-art denoising algorithms quantitively and qualitively.
关键词： image denoising,nonlocal self-similarity,cascaded residual networks
更新于2025-09-09 09:28:46
-
Hyperspectral Imaging Classification based on a Convolutional Neural Network with Adaptive Windows and Filters Sizes
摘要： Image classification by the Convolutional Neural Networks (CNN) has shown its great performances in recent years, in several areas, such as image processing and pattern recognition; However, there is still some improvement to do. The main problem in CNN is the initialization of the number and size of the filters, which can obviously change the results. In this article, we assign three major contributions, based on the CNN model; (1) adaptive selection of the number of filters. (2) using an adaptive size of the windows. (3) using an adaptive size of the filters. The tests results, applied to different hyperspectral datasets (SalinasA, Pavia University, and Indian Pines), have proven that this framework is able to improve the accuracy of the HSI classification.
关键词： Adaptive Filters,Convolutional Neural Networks,Image Classification,Hyperspectral Imaging
更新于2025-09-09 09:28:46
-
Space-Frequency Joint Contention Scheduling Algorithm based on AoD in SDM-EONs
摘要： Compared with elastic optical networks, spectrum contention problem may become more serious in space-division multiplexing elastic optical networks. Spectrum converters (SCs) can effectively resolve the problem of spectrum contention without additional delays, but SCs have the disadvantages of complicated technologies and high costs. To resolve the two problems aforesaid, taking into account the alleviated spectrum continuity constraint caused by the space dimension and the node flexibility introduced by architecture on demand (AoD), a space-frequency joint contention scheduling algorithm (SFJSA) based on AoD is proposed in this paper. SFJSA divides the contention scheduling process into two phases: scheduling in space domain and in frequency domain. It gives priority to resolve contention in the spatial domain, and then considers using spectrum converters in spectrum domain. In the stage of space switching, a weight formula is designed to balance the costs of spectrum selective switches and the loads of cores. In the stage of spectrum conversion, a concept of spectrum compactness is introduced to further optimize the utilization of spectrum resources and spectrum converters. The simulation results indicate that the proposed algorithm can reduce the number of spectrum converters required while improving the blocking performance, which means it achieves better blocking performance in a more cost-effective manner.
关键词： Space division multiplexing elastic optical networks,Bandwidth blocking probability,Architecture on demand,Spectrum converter,Space-Frequency Joint Scheduling
更新于2025-09-09 09:28:46
-
Deep-Learning-Assisted Network Orchestration for On-Demand and Cost-Effective vNF Service Chaining in Inter-DC Elastic Optical Networks
摘要： This work addresses the relatively long setup latency and complicated network control and management caused by on-demand virtual network function service chain (vNF-SC) provisioning in inter-datacenter elastic optical networks. We first design a provisioning framework with resource pre-deployment to resolve the aforementioned challenge. Specifically, the framework is designed as a discrete-time system, in which the operations are performed periodically in fixed time slots (TS). Each TS includes a pre-deployment phase followed by a provisioning phase. In the pre-deployment phase, a deep-learning (DL) model is designed to predict future vNF-SC requests, then lightpath establishment and vNF deployment are performed accordingly to pre-deploy resources for the predicted requests. Then, the system proceeds to the provisioning phase, which collects dynamic vNF-SC requests from clients and serves them in real-time by steering their traffic through the required vNFs in sequence. In order to forecast the high-dimensional data of future vNF-SC requests accurately, we design our DL model based on the long/short-term memory-based neural network and develop an effective training scheme for it. Then, the provisioning framework and DL model are optimized from several perspectives. We evaluate our proposed framework with simulations that leverage real traffic traces. The results indicate that our DL model achieves higher request prediction accuracy and lower blocking probability than two benchmarks that also predict vNF-SC requests and follow the principle of the proposed provisioning framework.
关键词： Long/short-term memory (LSTM),Elastic optical networks (EONs),Datacenter (DC),Service chaining,Network function virtualization (NFV),Deep learning
更新于2025-09-09 09:28:46
-
Handbook of Smart Cities (Software Services and Cyber Infrastructure) || Energy Harvesting in Smart Building Sensing: Overview and a Proof-of-Concept Study
摘要： Modern “smart” buildings require a plethora of sensors to be installed at various locations during the construction phase. Wiring costs and limited ?exibility of installation make wired installations less attractive. An alternative, ?exible, approach is to introduce wireless sensors and endow them with ways to harvest energy from the environment such that they attain the same “zero cost” of maintenance as their wired counterparts. The chapter reviews the sensing needs of smart buildings, and the related merits of energy harvesting to power embedded wireless sensor nodes. A proof-of-concept device exploiting thermoelectric harvesting is designed, built and tested to demonstrate how todays wireless sensing devices enable sustained continuous operation with minor energy harvesting requirements. In multi-hop environments, the underlying optimization problems are described and simple strategies that forego the solution of the hard computation problems but appear effective are outlined.
关键词： thermoelectric harvesting,smart buildings,energy harvesting,wireless sensor networks,multi-hop environments
更新于2025-09-09 09:28:46