- 标题
- 摘要
- 关键词
- 实验方案
- 产品
-
[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 - Desnet: Deep Residual Networks for Descalloping of Scansar Images
摘要： Scalloping is one of the critical problems in ScanSAR images. It not only affects image visualization, but also influences the quantitative applications such as surface wind and wave retrievals in the ocean area. The existing method of descalloping needs artificial parameter setting and lacks generality in the image domain. A novel deep neural network based on residual learning for descalloping of ScanSAR images is proposed in this paper. The proposed method can eliminate scalloping patterns and has strong adaptive ability, which can handle inhomogeneous scalloping patterns and different scenarios. Experiments on GF-3 ScanSAR images verify the good performance of this method. The code for our models is available online.
关键词： synthetic aperture radar (SAR),deep neural network,scalloping patterns,ScanSAR,Residual learning
更新于2025-09-23 15:22:29
-
[IEEE 2018 China International SAR Symposium (CISS) - Shanghai (2018.10.10-2018.10.12)] 2018 China International SAR Symposium (CISS) - Reconstruction Full-Pol SAR Data from Single-Pol SAR Image Using Deep Neural Network
摘要： Compared with single channel polarimetric (single-pol) SAR image, full polarimetric (full-pol) data convey richer information, but with compromises on higher system complexity and lower resolution or swath. In order to balance these factors, a deep neural networks based method is proposed to recover full-pol data from single-pol data in this paper. It consists of two parts: a feature extractor network is applied first to extract hierarchical multi-scale spatial features, followed by a feature translator network to predict polarimetric features with which full-pol SAR data can be recovered. Both qualitative and quantitative results show that the recovered full-pol SAR data agrees well with the real full-pol data. No prior information is assumed for scatterer media, and the framework can be easily expanded to recovery full-pol data from non-full-pol data. Traditional PolSAR applications such as model-based decomposition and unsupervised classification can now be applied directly to recovered full-pol SAR image to interpret the physical scattering mechanism.
关键词： synthetic aperture radar (SAR),deep neural network (DNN),polarimetric reconstruction
更新于2025-09-23 15:22:29
-
[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 - Reconstruction of Full-Pol SAR Data from Partialpol Data Using Deep Neural Networks
摘要： We propose a deep neural networks based method to reconstruct full polarimetric (full-pol) information from single polarimetric (single-pol) SAR data. It consists of two parts: feature extractor which is used to obtain multi-scale multi-layer features of targets in single-pol gray image, and feature translator that converts the geometric features to defined polarimetric feature space. The proposed method is demonstrated on L-band UAVSAR of NASA/JPL images over San Diego, CA, and New Orleans LA, USA. Both qualitative and quantitative results show the reconstructed full-pol images agree well with true full-pol images, the proposed networks have a good spatial robustness. Model-based target decomposition and unsupervised classification can be used directly on constructed full-pol images.
关键词： Deep Neural Network,unsupervised classification,Polarimetric Synthetic Aperture Radar (PolSAR),SAR image colorization
更新于2025-09-23 15:22:29
-
[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 Multi-Direction Subbands and Deep Neural Networks Bassed Pan-Sharpening Method
摘要： This paper proposes a pan-sharpening method based on multi-direction subbands and deep neural networks. First, by utilizing the multi-scale and multi-direction properties of the nonsubsampled contourlet transform (NSCT), panchromatic (PAN) image is decomposed into the low frequency subbands in different resolutions and the high frequency subbands in different directions. Pan-sharpening method aims to fuse the high frequency subband coefficients of PAN image and the low frequency subband coefficients of multispectral (MS) image. Second, in order to better extract the feature of the high frequency subbands in different directions of PAN image, the deep neural network (DNN) is trained using the image patches of high frequency subbands of PAN image. Third, in the fusion stage, we exploit NSCT on the principal component of resampled low resolution (LR) MS image. The high frequency subbands of output high resolution (HR) MS image is obtained by forward propagation of the trained DNN, which input is the high frequency subbands of LR MS image. Finally, a new subband set is obtained by fusing the reconstructed high frequency subband and the original low frequency subband of LR MS image. The HR MS image is produced by executing the inverse transform of NSCT and adaptive PCA (A-PCA) on the new subband set. The experimental results show the proposed method outperforms other well-known methods in terms of both objective measurements and visual evaluation.
关键词： adaptive Principal Component Analysis (A-PCA),deep neural network (DNN),pan-sharpening,nonsubsampled contourlet transform (NSCT)
更新于2025-09-23 15:22:29
-
[IEEE 2018 25th IEEE International Conference on Image Processing (ICIP) - Athens, Greece (2018.10.7-2018.10.10)] 2018 25th IEEE International Conference on Image Processing (ICIP) - Sketchpointnet: A Compact Network for Robust Sketch Recognition
摘要： Sketch recognition is a challenging image processing task. In this paper, we propose a novel point-based network with a compact architecture, named SketchPointNet, for robust sketch recognition. Sketch features are hierarchically learned from three miniPointNets, by successively sampling and grouping 2D points in a bottom-up fashion. SketchPointNet exploits both temporal and spatial context in strokes during point sampling and grouping. By directly consuming the sparse points, SketchPointNet is very compact and efficient. Compared with state-of-the-art techniques, SketchPointNet achieves comparable performance on the challenging TU-Berlin dataset while it significantly reduces the network size.
关键词： point set,stroke pattern,Sketch recognition,deep neural network
更新于2025-09-23 15:22:29
-
[IEEE 2018 OCEANS - MTS/IEEE Kobe Techno-Ocean (OTO) - Kobe, Japan (2018.5.28-2018.5.31)] 2018 OCEANS - MTS/IEEE Kobe Techno-Oceans (OTO) - Deep Neural Network for Source Localization Using Underwater Horizontal Circular Array
摘要： This paper applies deep neural network (DNN) to source localization in a shallow water environment using underwater horizontal circular array. The proposed method can discriminate source locations in a three-dimension space. The proposed method adopts a two-stage scheme, incorporating feature extraction and DNN analysis. In feature extraction step, the eigenvectors corresponding to the modal signal space, which are shown to be able to represent the propagating modes of the sound source, are extracted as the input feature of DNN. The eigenvectors are obtained by applying eigenvalue decomposition (EVD) of the covariance matrix of the received multi-channel signal. In DNN analysis step, time delay neural network (TDNN) is used to construct the mapping relationship between the eigenvectors and the source locations, because it is capable of making use of sequential information of the source signal. The output of the network is the source location estimates. Several experiments are conducted to demonstrate the effectiveness of the proposed method.
关键词： shallow water environment,modal signal space,Deep neural network,horizontal circular array,source localization
更新于2025-09-23 15:22:29
-
[IEEE 2019 Conference on Lasers and Electro-Optics Europe & European Quantum Electronics Conference (CLEO/Europe-EQEC) - Munich, Germany (2019.6.23-2019.6.27)] 2019 Conference on Lasers and Electro-Optics Europe & European Quantum Electronics Conference (CLEO/Europe-EQEC) - A Multi-Copy Approach to Quantum Entanglement Characterization
摘要： Automatic speech recognition (ASR) systems are used daily by millions of people worldwide to dictate messages, control devices, initiate searches or to facilitate data input in small devices. The user experience in these scenarios depends on the quality of the speech transcriptions and on the responsiveness of the system. For multilingual users, a further obstacle to natural interaction is the monolingual character of many ASR systems, in which users are constrained to a single preset language. In this work, we present an end-to-end multi-language ASR architecture, developed and deployed at Google, that allows users to select arbitrary combinations of spoken languages. We leverage recent advances in language identification and a novel method of real-time language selection to achieve similar recognition accuracy and nearly-identical latency characteristics as a monolingual system.
关键词： Automatic speech recognition (ASR),multilingual,deep neural network (DNN),language identification (LID)
更新于2025-09-23 15:21:01
-
[Lecture Notes in Computer Science] Pattern Recognition and Computer Vision Volume 11256 (First Chinese Conference, PRCV 2018, Guangzhou, China, November 23-26, 2018, Proceedings, Part I) || Automatic Measurement of Cup-to-Disc Ratio for Retinal Images
摘要： Glaucoma is a chronic eye disease which results in irreversible vision loss, and the optic cup-to-disc ratio (CDR) is an essential clinical indicator in diagnosing glaucoma, which means precise optic disc (OD) and optic cup (OC) segmentation become an important task. In this paper, we propose an automatic CDR measurement method. The method includes three stages: OD localization and ROI extraction, simultaneous segmentation of OD and OC, and CDR calculation. In the ?rst stage, the morphological operation and the sliding window are combined to ?nd the OD location and extract the ROI region. In the second stage, an improved deep neural network, named U-Net+CP+FL, which consists of U-shape convolutional architecture, a novel concatenating path and a multi-label fusion loss function, is adopted to simultaneously segment the OD and OC. Based on the segmentation results, the CDR value can be calculated in the last stage. Experimental results on the retinal images from public databases demonstrate that the proposed method can achieve comparable performance with ophthalmologist and superior performance when compared with other existing methods. Thus, our method can be a suitable tool for automated glaucoma analysis.
关键词： OD&OC segmentation,Glaucoma diagnosis,Deep neural network,OD localization,Cup-to-disc ratio (CDR)
更新于2025-09-23 15:21:01
-
Setting Up Surface-Enhanced Raman Scattering Database for Artificial Intelligence-Based Label-Free Discrimination of Tumor Suppressor Genes
摘要： The quality of input data in deep learning is tightly associated with the ultimate performance of machine learner. Taking advantages of unique merits of surface-enhanced Raman scattering (SERS) methodology in the collection and construction of database (e.g., abundant intrinsic fingerprint information, noninvasive data acquisition process, strong anti-interfering ability, etc.), herein we set up SERS-based database of deoxyribonucleic acid (DNA), suitable for artificial intelligence (AI)-based sensing applications. The database is collected and analyzed by silver nanoparticles (Ag NPs)-decorated silicon wafer (Ag NPs@Si) SERS chip, followed by training with a deep neural network (DNN). As proof-of-concept applications, three kinds of representative tumor suppressor genes, i.e., p16, p21 and p53 fragments, are readily discriminated in label-free manners. Prominent and reproducible SERS spectra of these DNA molecules are collected and employed as input data for DNN learning and training, which enables selective discrimination of DNA target(s). The accuracy rate for the recognition of specific DNA target reaches 90.28%.
关键词： surface-enhanced Raman scattering,label-free discrimination,deep neural network,tumor suppressor genes,artificial intelligence
更新于2025-09-23 15:21:01
-
Enhancing the Reliability of Protection Scheme for PV Integrated Microgrid by Discriminating between Array Faults and Symmetrical Line Faults using Sparse Auto Encoder
摘要： The ever increasing power demand and the stress on reducing carbon footprint have paved the way for widespread use of PV integrated microgrid. However, the development of a reliable protection scheme for PV integrated microgrid is challenging because of the similar voltage-current profile of PV array faults and symmetrical line faults. Conventional protection schemes based on pre-defined threshold setting are not able to distinguish between PV array and symmetrical faults, and hence fail to provide separate controlling actions for the two cases. In this regard, a protection scheme based on sparse auto-encoder and deep neural network (SAE-DNN) approach has been proposed to discriminate between array faults and symmetrical line faults in addition to performing the tasks of mode detection, fault detection, classification and section identification. The voltage and current signals retrieved from relaying buses are converted into grayscale image dataset, which is fed as input to the SAE to perform the unsupervised feature learning. The performance of proposed scheme has been evaluated through reliability analysis and compared with ANN, SVM and DT based techniques under both islanding and grid-connected mode of the microgrid. The scheme has been also validated for field applications by performing real-time simulations on OPAL-RT digital simulator.
关键词： sparse auto-encoder,classification,deep neural network,PV integrated microgrid,section identification,protection scheme,fault detection,OPAL-RT digital simulator
更新于2025-09-23 15:21:01