Article
Computer Science, Artificial Intelligence
Xiao Tan, Huaian Chen, Rui Zhang, Qihan Wang, Yan Kan, Jinjin Zheng, Yi Jin, Enhong Chen
Summary: In this paper, a deep dynamic multi-exposure fusion (DDMEF) framework is proposed to reconstruct a ghost-free high-quality image from two differently exposed images of a dynamic scene. The framework includes pre-enhancement-based alignment and privilege-information-guided fusion steps, which effectively improve the deghosting ability of the fusion network. Extensive experimental results show that the proposed method outperforms state-of-the-art dynamic MEF methods.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2023)
Article
Computer Science, Artificial Intelligence
Yuming Fang, Yan Zeng, Wenhui Jiang, Hanwei Zhu, Jiebin Yan
Summary: This study introduces a novel and efficient objective image quality assessment model for MEF images based on superpixels and an information theory adaptive pooling strategy for both static and dynamic scenes. Experimental results demonstrate promising performance with relatively low computational complexity, and show potential application for parameter tuning of MEF algorithms.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2021)
Article
Engineering, Electrical & Electronic
V. S. V. Unni, Pravin Nair, Kunal N. Chaudhury
Summary: In multiband fusion, an image with high spatial and low spectral resolution is combined with an image of low spatial but high spectral resolution to generate a single multiband image with high spatial and spectral resolutions. This study proposes a novel convex regularizer that considers texture information, long-distance correlations, and weighting using the higher spatial resolution image as a guide. The algorithm utilizes FFT-based convolution and soft-thresholding to efficiently solve the ADMM subproblems, outperforming state-of-the-art variational and deep learning techniques in terms of reconstruction quality.
IEEE TRANSACTIONS ON COMPUTATIONAL IMAGING
(2022)
Article
Chemistry, Analytical
Jun Dai, Songlin Liu, Xiangyang Hao, Zongbin Ren, Xiao Yang
Summary: This paper presents a factor graph optimization algorithm based on multi-source sensors for UAV localization state estimation. Mathematical simulations and validations demonstrate the advantages of this algorithm in terms of real-time performance, accuracy, and fault tolerance, as well as its ability to improve positioning accuracy in complex scenarios. Moreover, the algorithm is applicable to other scenarios and sensor combinations.
Article
Environmental Sciences
Ziyang Wang, Lin Yang, Yehua Sheng, Mi Shen
Summary: This paper proposes a method for segmenting pole-like objects under geometric structural constraints and combines classification results at different scales to effectively extract pole-like objects from point clouds in road scenes, achieving high-precision classification and identification.
Article
Remote Sensing
Qipeng Li, Yuan Zhuang, Jianzhu Huai
Summary: Accurate and robust localization is crucial for autonomous driving and multi-sensor robotic systems in complex urban scenes. This paper proposes a LiDAR-based dynamic scene perception and multi-sensor localization framework to address the limitations of existing methods. The framework includes a moving object segmentation module, optimized ground segmentation, and a lightweight LiDAR-IMU-GNSS odometry framework for real-time position estimation. Experimental results demonstrate that the proposed framework outperforms existing methods in terms of accuracy and robustness.
INTERNATIONAL JOURNAL OF APPLIED EARTH OBSERVATION AND GEOINFORMATION
(2023)
Article
Computer Science, Information Systems
Bouchra Honnit, Khaoula Belhaj Soulam, Mohamed Nabil Said, Ahmed Tamtaoui
Summary: This paper presents a new model for multi-classification in video surveillance based on data fusion. The model extracts features, conducts pre-classification for each feature separately, combines the obtained posterior probabilities using the T-conorm operator, and uses the maximum value to determine the label of each detected object. The evaluation on two public datasets demonstrates that the model improves the classification accuracy to an average of 99% using SVM and outperforms other methods.
JOURNAL OF KING SAUD UNIVERSITY-COMPUTER AND INFORMATION SCIENCES
(2022)
Article
Engineering, Civil
Thomas Golecki, Fernando Gomez, Juan Carrion, Billie F. Spencer Jr
Summary: Topology optimization of bridge structures under random moving traffic loading is challenging. Existing approaches are limited in addressing this problem. This study proposes a compact representation of random moving traffic loading and a corresponding optimization method. The results show significant improvements in bridge response and indicate a robust solution for different traffic parameters. This approach enables efficient optimization of bridge topology under loading uncertainties.
ENGINEERING STRUCTURES
(2023)
Article
Computer Science, Artificial Intelligence
Rui She, Qiyu Kang, Sijie Wang, Wee Peng Tay, Yong Liang Guan, Diego Navarro Navarro, Andreas Hartmannsgruber
Summary: This paper investigates the matching of real-time images captured by on-vehicle cameras with landmark patches in an image database, which plays a crucial role in various computer perception tasks for autonomous driving. Current methods focus on local matching for regions of interest without considering the spatial relationships among image patches, which correspond to objects in the environment. In this study, a spatial graph is constructed to capture the spatial neighborhood information, and a joint feature and metric learning model with graph-based learning is proposed. The evaluation using several street-scene datasets demonstrates that the approach achieves state-of-the-art matching results.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2023)
Article
Environmental Sciences
Yao Wang, Lihua Cao, Keke Su, Deen Dai, Ning Li, Di Wu
Summary: The accurate detection and localization of small targets in infrared images with complex backgrounds is crucial in applications involving infrared search and tracking. In this paper, the authors propose a method that integrates spatio-temporal information to improve detection performance. Experimental results show that the proposed method outperforms existing methods in terms of background suppression and false alarm rates.
Article
Multidisciplinary Sciences
Jingxia Guo, Nan Jia, Jinniu Bai
Summary: In this study, we propose an effective framework for high-resolution remote sensing (HRRS) image scene classification by combining the channel-spatial attention (CSA) mechanism with the Vision Transformer method. The proposed model extracts channel and spatial features using CSA and the Multi-head Self-Attention mechanism, and solves the gradient disappearance problem and avoids overfitting using the residual network structure. Experimental results show that the proposed network outperforms state-of-the-art methods in scene classification.
SCIENTIFIC REPORTS
(2022)
Article
Environmental Sciences
Chongdi Duan, Yu Li, Weiwei Wang, Jianguo Li
Summary: This paper presents a novel multi-source fusion detection method based on LEO satellites, which solves the problems of Doppler bandwidth expansion and offset by constructing a pre-compensation function and designing theoretical upper and lower limits of Doppler frequency. The CFAR detection threshold based on the exponential weighted likelihood ratio is used for target detection.
Article
Engineering, Electrical & Electronic
Oguzhan Ulucan, Diclehan Ulucan, Mehmet Turkan
Summary: This paper proposes an image fusion algorithm based on weight characterization and guided filter, which can effectively extract weight maps and obtain blended output images through pyramidal decomposition. Comprehensive experiments and comparisons demonstrate that the algorithm performs well in both static and dynamic scenes, and it is also successfully applied to the visible-infrared image fusion problem without further optimization.
Article
Geochemistry & Geophysics
Pengcheng Wang, Wenlong Niu, Weihua Gao, Yingyi Guo, Xiaodong Peng
Summary: This letter proposes an approach for detecting dim moving point targets in cloud clutter scenes based on temporal profile learning. A novel signal-to-signal network is introduced to learn the temporal characteristics of background and clutter, with the extraction of transient disturbance through residual calculation. The introduced ConvBlock-1-D structure enhances feature flow and propagation, and a proposed loss function addresses the imbalance problem. Experimental results validate the superior performance of the proposed method in qualitative and quantitative assessments.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2023)
Article
Computer Science, Artificial Intelligence
Luming Zhang, Junjie Peng, Wenfu Liu, Haochen Yuan, Shuhua Tan, Lu Wang, Fen Yi
Summary: With the explosive growth of the express logistics industry, the recognition of express bills in automated express transportation is challenging due to poor image quality and complex scenes. In order to address this problem, we propose the Semantic Fusion Rotated Object Detector (SFRDet) method which uses semantic fusion mechanism and the Semantic Reinforcement Feature Pyramid Network (SRFPN) to enhance feature extraction capability and improve efficiency. Extensive experiments show that our method outperforms other state-of-the-art methods in precision and efficiency, making it highly promising in the industry.
IMAGE AND VISION COMPUTING
(2023)
Article
Automation & Control Systems
Yuwu Lu, Zhihui Lai, Xuelong Li, Wai Keung Wong, Chun Yuan, David Zhang
IEEE TRANSACTIONS ON CYBERNETICS
(2019)
Article
Automation & Control Systems
Jianbing Shen, Zhiyuan Liang, Jianhong Liu, Hanqiu Sun, Ling Shao, Dacheng Tao
IEEE TRANSACTIONS ON CYBERNETICS
(2019)
Article
Engineering, Electrical & Electronic
Hui Wang, Jianbing Shen, Junbo Yin, Xingping Dong, Hanqiu Sun, Ling Shao
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2020)
Article
Biochemistry & Molecular Biology
Jian-Bing Shen, Kiran S. Toti, Saibal Chakraborty, T. Santhosh Kumar, Chunxia Cronin, Bruce T. Liang, Kenneth A. Jacobson
PURINERGIC SIGNALLING
(2020)
Editorial Material
Computer Science, Artificial Intelligence
Shuo Shang, Jianbing Shen, Ji-Rong Wen, Panos Kalnis
Article
Ophthalmology
Huazhu Fu, Fei Li, Yanwu Xu, Jingan Liao, Jian Xiong, Jianbing Shen, Jiang Liu, Xiulan Zhang
TRANSLATIONAL VISION SCIENCE & TECHNOLOGY
(2020)
Article
Computer Science, Artificial Intelligence
Xingping Dong, Jianbing Shen, Wenguan Wang, Ling Shao, Haibin Ling, Fatih Porikli
Summary: This paper introduces a novel dynamical hyperparameter optimization method utilizing an action-prediction network based on continuous deep Q-learning, to adaptively optimize hyperparameters for a given sequence and improve visual object tracking performance.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2021)
Article
Optics
Timothy O'Connor, Jian-Bing Shen, Bruce T. Liang, Bahram Javidi
Summary: This study presents a rapid screening method for COVID-19 infection in red blood cells using a compact, field-portable 3D-printed shearing digital holographic microscope. By analyzing the spatiotemporal behavior of individual red blood cells, a bi-directional long short-term memory network is used to classify between healthy and COVID positive red blood cells. The proposed system may benefit under-resourced healthcare systems.
Article
Chemistry, Analytical
Hao Li, Sanyuan Zhao, Wenjun Zhao, Libin Zhang, Jianbing Shen
Summary: The proposed anchor-free 3D vehicle detection algorithm encodes object positions as keypoints in the bird's-eye view of LiDAR point clouds, mapping them to a single-channel 2D heatmap using voxel/pillar feature extractor and convolutional blocks. The method achieves high average orientation similarity without direction classification tricks, eliminating the need for anchor boxes in target assignment and bounding box decoding processes.
Article
Computer Science, Artificial Intelligence
Xiankai Lu, Wenguan Wang, Jianbing Shen, David Crandall, Jiebo Luo
Summary: We introduce a novel network called COSNet that addresses the zero-shot video object segmentation task by incorporating a global co-attention mechanism. COSNet outperforms current alternatives by capturing global correlations and scene context through joint computation and appending co-attention responses into a joint feature space.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Computer Science, Artificial Intelligence
Xiankai Lu, Chao Ma, Jianbing Shen, Xiaokang Yang, Ian Reid, Ming-Hsuan Yang
Summary: This paper addresses the problem of data imbalance in learning deep models for visual object tracking. The proposed shrinkage loss function helps balance the training data and improves the performance of both deep regression and classification trackers. Experimental results on six benchmark datasets demonstrate the effectiveness of the shrinkage loss.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Computer Science, Artificial Intelligence
Keren Fu, Deng-Ping Fan, Ge-Peng Ji, Qijun Zhao, Jianbing Shen, Ce Zhu
Summary: This paper proposes a novel RGB-D salient object detection model, which achieves effective feature learning from RGB and depth inputs through joint learning and densely cooperative fusion. Experimental results demonstrate significant improvements over state-of-the-art models on multiple datasets, as well as comparable or better performance on other multi-modal detection tasks and RGB-D semantic segmentation tasks.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Computer Science, Information Systems
Qiuxia Lai, Salman Khan, Yongwei Nie, Hanqiu Sun, Jianbing Shen, Ling Shao
Summary: Both human and machine attention play crucial roles in deep learning models, and understanding the relationship between the two is essential for interpreting and designing neural networks. Recent studies have shown that artificial attention does not always align with human intuition, highlighting the importance of better aligning machine and human attention for improved performance and explainability in neural network design.
IEEE TRANSACTIONS ON MULTIMEDIA
(2021)
Article
Computer Science, Artificial Intelligence
Fan Yang, Xin Li, Jianbing Shen
Summary: State-of-the-art object skeleton detection methods leveraging Convolutional Neural Networks have been enhanced by a new architecture called Multi-Scale Bidirectional Fully Convolutional Network (MSB-FCN), which aims to improve the accuracy by better gathering and enhancing multi-scale high-level contextual information. By solely utilizing deep features and a bidirectional structure, along with dense connections and an attention pyramid, MSB-FCN successfully learns semantic-level information and achieves significant improvements over existing algorithms in various benchmarks.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2021)
Proceedings Paper
Computer Science, Artificial Intelligence
ZheHui Wang, Sanyuan Zhao, Jianbing Shen, Zhengchao Lei
2020 IEEE INTERNATIONAL CONFERENCE ON VISUAL COMMUNICATIONS AND IMAGE PROCESSING (VCIP)
(2020)