Article
Computer Science, Artificial Intelligence
Krishan Sharma, Renu Rameshan
Summary: Modeling image sets or videos as linear subspaces is common for classification in machine learning, but affine subspace modeling is less explored. This article introduces a novel approach to address the image sets classification problem by modeling them as affine subspaces and using a kernel-based method to map them to a finite-dimensional Hilbert space. Experimental results show promising performance in gait, object, hand, and body gesture recognition compared to state-of-the-art techniques.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2021)
Article
Computer Science, Information Systems
Krishan Sharma, Renu Rameshan
Summary: This paper explores the inherent geometry of video tensors by modeling them as points in product of Riemannian matrix manifolds and proposes positive definite kernels for feature mapping and classification. Experimental results on publicly available datasets show that the proposed methodology outperforms state-of-the-art methods in terms of classification accuracy.
JOURNAL OF VISUAL COMMUNICATION AND IMAGE REPRESENTATION
(2021)
Article
Computer Science, Information Systems
Dong Wei, Xiaobo Shen, Quansen Sun, Xizhan Gao, Zhenwen Ren
Summary: This study presents two metric learning algorithms based on Grassmann manifold for image set classification and exploring intrinsic geometry distance. The proposed algorithms perform favorably against the state-of-the-art methods in extensive experiments.
IEEE TRANSACTIONS ON MULTIMEDIA
(2023)
Article
Engineering, Electrical & Electronic
Arnaud Breloy, Sandeep Kumar, Ying Sun, Daniel P. Palomar
Summary: This paper introduces a framework for optimizing cost functions of orthonormal basis learning problems, using majorization-minimization framework with orthogonal projection reformulations to handle the orthogonality constraint systematically. Surrogate functions for various standard objectives are derived and utilized for constructing algorithms, with a new sparse PCA algorithm proposed as an example. Simulations and experiments demonstrate the effectiveness of the approach in terms of performance and computational complexity.
IEEE TRANSACTIONS ON SIGNAL PROCESSING
(2021)
Article
Mathematics, Applied
Shinya Moritoh, Nao Takemoto
Summary: An alternative wavelet inversion formula proposed by Lebedeva and Postnikov in 2014 is explored in this paper. The main objective is to provide a multidimensional version of their formula, expressing the Hilbert and Riesz transforms of functions in terms of wavelet transforms.
INTEGRAL TRANSFORMS AND SPECIAL FUNCTIONS
(2023)
Article
Computer Science, Information Systems
Peiguang Jing, Yuting Su, Zhengnan Li, Liqiang Nie
Summary: This paper proposes a robust multi-view affinity graph learning framework that can effectively handle data from multiple sources through improved feature selection algorithm and Consistent Affinity Graph Learning algorithm. Experimental results demonstrate that the proposed method achieves promising results on publicly available datasets.
INFORMATION SCIENCES
(2021)
Article
Computer Science, Artificial Intelligence
Liyu Su, Jing Liu, Jianting Zhang, Xiaoqing Tian, Hailang Zhang, Chaoqun Ma
Summary: In this paper, a smooth low-rank representation with a Grassmann manifold (SLRR-GM) model is proposed to tackle the identifiability issue in tensor completion. The proposed method effectively removes the preserved outliers of the sparse tensor and precisely estimates the low-rank tensor. Experiments demonstrate that it outperforms several state-of-the-art methods in terms of quantitative and visual aspects.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Engineering, Electrical & Electronic
Jingyu Ji, Yuhua Zhang, Zhilong Lin, Yongke Li, Changlong Wang, Yongjiang Hu, Jiangyi Yao
Summary: This article proposes a fusion algorithm based on iterative control of anisotropic diffusion and regional gradient structure to improve the fusion performance of infrared and visible images. The algorithm effectively controls the number of iterations and utilizes different fusion schemes for different layers. The fusion image is obtained by reconstructing the structure layer and the energy layer.
JOURNAL OF SENSORS
(2022)
Article
Engineering, Biomedical
Xiaoyan Li, Yuanhua Qiao, Lijuan Duan, Jun Miao
Summary: This paper proposes a new EEG classification method by representing EEG signals on the Grassmann manifold. The method achieves dimension reduction of EEG data through low rank representation and deep neural network, and achieves good classification results on multiple datasets.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2024)
Article
Computer Science, Information Systems
Freddy Alejandro Chaurra-Gutierrez, Claudia Feregrino-Uribe, Julio Cesar Perez-Sansalvador, Gustavo Rodriguez-Gomez
Summary: We propose the first quantum version of the one-dimensional integer wavelet S-Transform (QIST). The QIST is based on a new Quantum Block Representation by Basis States (QBRBS) and a quantum rounding operator that avoids nonlinearities. Complete quantum circuits for addition, subtraction, and halving operations are developed with polynomial quantum complexity and constant time complexity. The experimental results on MATLAB and Qiskit coincide with the theoretical results of the classical version. This new quantum transform enables new developments in information hiding, compression, and IoT applications.
INFORMATION SCIENCES
(2023)
Article
Computer Science, Artificial Intelligence
Jian Zou, Yue Zhang, Hongjian Liu, Lifeng Ma
Summary: This paper presents a novel method for single sample face recognition using grayscale monogenic features and kernel sparse representation on multiple Riemannian manifolds. The approach involves extracting local features from different regions of the face images, modeling the corresponding feature vectors as points on a Grassmann manifold, extracting co-occurrence distributions of feature images, and training a kernel sparse representation classifier using multiple kernel fusion. Experimental results demonstrate the superiority of the proposed method.
Article
Computer Science, Artificial Intelligence
Lele Fu, Jieling Li, Chuan Chen
Summary: Multi-view clustering aims to achieve higher accuracy in data clustering by leveraging complementary information embedded in multi-view data. This paper proposes a consistent affinity representation learning method with dual low-rank constraints, which learns a consistent affinity matrix by fusing multiple subspace representations and enhances global cluster structure using graph regularization.
Article
Computer Science, Artificial Intelligence
Yuping Wang, Junfei Zhang
Summary: It is challenging to classify multi-dimensional data with complex intrinsic geometry, such as human gesture recognition based on videos. This paper proposes a method using Grassmann manifold and data tensor to characterize the features of human gesture videos. Experimental results show that the proposed method is competitive to some relevant excellent methods.
PEERJ COMPUTER SCIENCE
(2022)
Article
Mathematics
Jinghao Huang, Fedor Sukochev, Dmitriy Zanin
Summary: This article investigates the distribution function of martingale transforms in a probability space and provides a sharp estimate, complementing and extending classical results from previous research.
JOURNAL OF FUNCTIONAL ANALYSIS
(2022)
Article
Engineering, Electrical & Electronic
Xu Guanlei, Xu Xiaogang, Wang Xiaotong
Summary: This paper presents a detailed introduction of a novel Bedrosian principle (BP) based on two-dimensional Hilbert transform (HT). The paper revisits several 2D BPs and provides new interpretations for two-dimensional time-frequency analysis. A new generalized BP (GBP) is derived with novel features using the Riesz transform (RT). The paper demonstrates the effectiveness of the proposed methods through theoretical analysis and comparison experiments on composite and natural images.
DIGITAL SIGNAL PROCESSING
(2022)
Article
Geochemistry & Geophysics
Ganggang Dong, Hongwei Liu, Jocelyn Chanussot
Summary: Despite extensive studies, radar target recognition remains a challenging issue. This article tackles the problem by utilizing local descriptors around keypoints instead of relying on holistic features or raw intensity values that are sensitive to real-world sources of variability.
IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE
(2021)
Article
Automation & Control Systems
Ganggang Dong, Hongwei Liu
Summary: The article introduces a new approach by building a neural network hierarchy for target recognition in a small sample environment, through feature generation and refinement modules. The strategy aims to achieve comparable or better performance with limited training resources.
IEEE TRANSACTIONS ON CYBERNETICS
(2021)
Article
Engineering, Electrical & Electronic
Rui Tang, Ganggang Dong
Summary: Bridge-over-water detection is crucial in urban surveillance and military reconnaissance. However, the arbitrary orientations and extreme aspect ratios of bridges in remote sensing images pose challenges for accurately extracting bridge-related features. This paper introduces modulated deformable convolution and attention mechanisms to address these issues, and demonstrates the effectiveness of the proposed methods through multiple experiments.
EURASIP JOURNAL ON ADVANCES IN SIGNAL PROCESSING
(2022)
Article
Engineering, Electrical & Electronic
Xiaojing Geng, Ganggang Dong, Ziheng Xia, Hongwei Liu
Summary: This research proposes a new method for open set target recognition, which involves constructing targets indistinguishable from known classes to detect and recognize unknown targets. The original open-world environment is transformed into a closed-world environment containing the unknown class.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2023)
Article
Engineering, Electrical & Electronic
Ganggang Dong, Hongwei Liu
Summary: In recent years, there has been a resurgence in neural networks with the use of hierarchical stacked hidden layers to learn high-level representations. These learned representations have achieved great performances. However, these learning models heavily rely on large quantities of labeled signal data, which is difficult and costly to obtain in realistic scenarios. To address this issue, a new family of signal augmentation strategies, segment-wise generation and signal-wise generation, are proposed to simulate unforeseen disturbances during signal sampling and improve recognition performance in realistic scenarios. Multiple comparative studies demonstrate the effectiveness of these strategies compared to classical methods and deep learning algorithms.
IEEE TRANSACTIONS ON COMMUNICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Ziheng Xia, Penghui Wang, Ganggang Dong, Hongwei Liu
Summary: Open set recognition is more practical than closed set recognition due to the complexity of real-world applications. We propose three novel frameworks with kinetic pattern, namely kinetic prototype framework (KPF), adversarial KPF (AKPF), and an upgraded version AKPF++. KPF introduces a novel kinetic margin constraint radius to improve the compactness of known features and increase robustness for unknowns. AKPF generates adversarial samples and adds them to the training phase, enhancing performance with adversarial motion. AKPF++ further improves performance by adding more generated data into the training phase. Extensive experimental results demonstrate that the proposed frameworks are superior and achieve state-of-the-art performance.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Remote Sensing
Tingxuan Zhong, Ganggang Dong
Summary: SAR clutter statistical modelling research is important for interpreting SAR images. Previous studies used statistical modelling or electromagnetic calculation, but faced challenges in accurately characterizing ground clutter. To address this, attention has shifted to non-parametric models. However, there is limited research on complex-valued radar images. This paper proposes a dual-channel generative adversarial network and introduces a new evaluation metric called MCC to assess simulated clutter.
REMOTE SENSING LETTERS
(2023)
Article
Environmental Sciences
Ziyi Yu, Ganggang Dong, Hongwei Liu
Summary: Target recognition is a core application of radar image interpretation, and deep learning has become the mainstream solution. However, limited training samples may lead to underfitting and poor robustness. To address this issue, generative models have been proposed, and the quality of simulated images needs to be assessed. This paper presents a new evaluation strategy, including sample-wise and class-wise assessments, to evaluate simulated images from different perspectives.
Article
Engineering, Electrical & Electronic
Yao Wang, Ganggang Dong, Shuai Shao, Hongwei Liu
Summary: Data-driven ship detection methods using deep learning algorithms have become a popular research topic. Label assignment is an essential step in these models, and a new technique based on prediction and prior information guidance has been proposed to improve performance. The proposed ship detection network focuses on small ships and combines shallow texture and deep semantic information to achieve multiscale ship detection. Experimental results on SSDD and HRSID datasets demonstrate the superiority of the proposed method compared to advanced detectors.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Ganggang Dong, Hongwei Liu
Summary: Target recognition via deep learning has achieved great performances in previous works, but it requires large amounts of labeled training data. However, collecting labeled data for radar sensors is difficult due to the absence of imaging truth. This article proposes a new radar image generation method using target reimaging. The method transforms the original image into the frequency-aspect domain and applies inverse operations to generate new radar images. Multiple comparative studies are conducted to demonstrate the advantages of the proposed method.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Engineering, Electrical & Electronic
Ganggang Dong, Hongwei Liu
Summary: This article proposes a new method that combines physical modeling and deep learning to overcome the overfitting problem caused by limited training samples and poor quality radar images. The method includes radar measurement simulation, target reconstruction, residual calculation, data masking, and reimaging process to generate new radar measurements. Comparative studies demonstrate the advantages of this method in improving the learning efficiency of deep models under limited sample environments.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2023)
Article
Geochemistry & Geophysics
Yafei Song, Ganggang Dong
Summary: A single sensor radar is no longer sufficient in complex electromagnetic environments, leading to more attention on radar sensor networks for obtaining more information and achieving better detection and tracking performance through resource sharing and joint learning. This article presents a distributed learning model with three phases to improve utilization efficiency of multiple radar sensors with limited resources, addressing scenarios like limited training data or imbalanced samples. The proposed method includes self-reweighting loss, image generation, and federated learning framework with virtual representations for adjusting the classifier to enhance learning efficiency. Comparative studies on the MSTAR dataset demonstrate the advantages of the proposed method.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Ziheng Xia, Penghui Wang, Ganggang Dong, Hongwei Liu
Summary: Radar automatic target recognition (RATR) using high-resolution range profiles (HRRP) has gained attention, but previous works primarily focus on closed set recognition and may lead to errors in open set environments. This article proposes open set recognition to address this issue by establishing a closed classification boundary. The proposed extreme value boundary theorem demonstrates that the maximum distance from known features to the cluster center follows a generalized extreme value distribution, enabling the determination of a closed classification boundary to distinguish between known and unknown classes. Extensive experiments on measured HRRP data validate the proposed theorem and method.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Ziheng Xia, Penghui Wang, Ganggang Dong, Hongwei Liu
Summary: The open set recognition (OSR) model can identify both known and unknown samples simultaneously, making it more practical for radar automatic target recognition (RATR) than closed set recognition (CSR). By controlling the distribution of features in the space, the OSR performance can be improved.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2022)
Article
Engineering, Electrical & Electronic
Ganggang Dong, Hongwei Liu
Summary: This article presents a model-data co-driven ship detection strategy using deep neural networks. By simulating sea clutter and embedding targets, the proposed strategy solves the problem of expensive data labeling for radar sensors.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2022)