A5009318707- Robotics and Sensor-Based Localization
- Advanced Image and Video Retrieval Techniques
- Advanced Vision and Imaging
- Robotic Path Planning Algorithms
- Indoor and Outdoor Localization Technologies
- Advanced Steganography and Watermarking Techniques
- 3D Surveying and Cultural Heritage
- Image Retrieval and Classification Techniques
- Digital Media Forensic Detection
- Inertial Sensor and Navigation
- Video Analysis and Summarization
- Advanced Data Compression Techniques
- Chaos-based Image/Signal Encryption
- Medical Image Segmentation Techniques
- Advanced Neural Network Applications
- Topic Modeling
- Advanced Algorithms and Applications
- Guidance and Control Systems
- Image Processing and 3D Reconstruction
- Optical measurement and interference techniques
- Visual Attention and Saliency Detection
- Computer Graphics and Visualization Techniques
- Generative Adversarial Networks and Image Synthesis
- Space exploration and regulation
- Fault Detection and Control Systems
Beijing Jiaotong University
2002-2025
Nanjing University of Aeronautics and Astronautics
2013-2024
Shanghai University of Electric Power
2024
Nanyang Technological University
2024
University of Electronic Science and Technology of China
2024
Peking University First Hospital
2024
Peking University
2024
Dalian University of Technology
2010-2023
Beijing Aerospace Flight Control Center
2020
Shanghai Ocean University
2019
Stitched images provide a wide field-of-view (FoV) but suffer from unpleasant irregular boundaries. To deal with this problem, existing image rectangling methods devote to searching an initial mesh and optimizing target form the deformation in two stages. Then rectangular can be generated by warping stitched images. However, these solutions only work for rich linear structures, leading noticeable distortions portraits landscapes non-linear objects. In paper, we address issues proposing first...
The rapid development of open-source large language models (LLMs) has been truly remarkable. However, the scaling law described in previous literature presents varying conclusions, which casts a dark cloud over LLMs. We delve into study laws and present our distinctive findings that facilitate scale two commonly used configurations, 7B 67B. Guided by laws, we introduce DeepSeek LLM, project dedicated to advancing with long-term perspective. To support pre-training phase, have developed...
Although diffusion models have achieved remarkable success in the field of image generation, their latent space remains under-explored. Current methods for identifying semantics within often rely on external supervision, such as textual information and segmentation masks. In this paper, we propose a method to identify semantic attributes pre-trained without any further training. By projecting Jacobian targeted region into low-dimensional subspace which is orthogonal non-masked regions, our...
Multiple description (MD) coding is a promising alternative for robust transmission of information over non-prioritized and unpredictable networks. In this paper, an effective MD image scheme introduced based on the lattice vector quantization (MDLVQ) wavelet transformed images. view characteristics coefficients in different frequency subbands, MDLVQ applied optimized way, including appropriate construction coefficient vectors, optimization encoding parameters such as choice sublattice index...
This paper presents a perception-aware path planner for active SLAM in dynamic environments using micro-aerial vehicles (MAV). The “Next-Best-View” (NBVP planner) is combined with an loop closing, which called the Active Loop Closing Planner (ALCP planner). proposed to avoid both static and obstacles unknown while reducing uncertainty of system further improving accuracy localization. First, receding horizon strategy adopted find next waypoint. cost function that combines exploration gain...
Objects detection and tracking using 3-D LiDAR has gained momentum lately while it not been extensively applied, the main challenge is that conventional mechanical expensive difficult for single sensor to obtain good performance over a long period of time. In this article, we propose multisensor fusion objects framework solid-state RGB camera. We use low-cost with irregular scan pattern universal clustering method which determines searching radius by laser density range. To improve overall...
Ultrasound guidance is used for many surgical applications such as biopsy and electrode insertion. This paper presents an improved method tracking micro tools inserted in human tissues. The RANSAC algorithm [1] has been implemented to detect the exact position of needle a stationary situation. In this paper, Kalman filter added estimate dynamical simulation, get tip needle, speckle inserting speed needle. uses results given by two methods above measurement make estimation tip. simulated show...
In the classification of remote-sensing sea ice images, labeled samples are difficult to acquire. To adequately utilize massive number unlabeled samples, which contain abundant information, we propose a cooperative framework based on active learning (AL) and semi-supervised (SSL) for image classification. We acquire most valuable using AL make full use information contained in SSL, then conduct label consistency verification procedure further ensure quality pseudo-labeled obtained through...
Photovoltaic (PV) power is greatly uncertain due to the random meteorological parameters. Therefore, accurate PV forecasting results are significant for dispatching of and improving system stability. This paper proposes a hybrid model one-day-ahead under different cloud amount conditions. The proposed consists an improved artificial neural network (ANN) algorithm conversion model. First, ANN designed forecast plane array (POA) irradiance ambient temperature. Backpropagation, gradient...
In this paper, we explore a principal way to enhance the quality of object masks produced by different segmentation models. We propose model-agnostic solution called SegRefiner, which offers novel perspective on problem interpreting refinement as data generation process. As result, process can be smoothly implemented through series denoising diffusion steps. Specifically, SegRefiner takes coarse inputs and refines them using discrete By predicting label corresponding states-transition...
In open-loop starting process of Brushless DC motor without position sensors, pulsations torque and speed are large cannot even start successfully. order to solve those problems, Space Vector Pulse Width Modulation control schema three-phase inverter employing 180° switch-on mode is analyzed. Moreover, relationship between the composed stator magnetomotive force current maximum phase's derived with windings in Y connection. Hence, this paper proposes a new strategy combing regulation SVPWM....
Artificial Intelligence (AI) has achieved significant advancements in technology and research with the development over several decades, is widely used many areas including computing vision, natural language processing, time-series analysis, speech synthesis, etc. During age of deep learning, especially arise Large Language Models, a large majority researchers' attention paid on pursuing new state-of-the-art (SOTA) results, resulting ever increasing model size computational complexity. The...
In inertial/visual integrated navigation system, image registration is the key to achieve positioning. practical application of template matching, there are inevitably changes in rotation and scale between source image. selection matching search strategy, traversal method very time-consuming inefficient, which makes it difficult meet requirements real-time positioning navigation. To address problem above, an acceleration algorithm based on improved ORB proposed this paper. Image pyramid...
Classical relevance vector machine is sensitive to outliers during training and has weak robustness. In order solve this problem, a novel robust presented in paper. The key idea of the proposed method introduce individual noise variance coefficient for each sample. process model training, coefficients gradually decrease so as automatically detect eliminate outliers. addition, iterative formulae optimization hyperparameters are derived according Bayesian evidence framework. Simulation results...
Aiming at the low prediction accuracy of current lithium-ion battery cycle, this paper proposes a model based on differential evolution algorithm (DE) and BP neural network fusion. is used to predict cycle life battery. The DE optimize initial weight threshold network, which reduces number iterations accelerates convergence speed. results show that has higher accuracy, effectively improves speed meets characteristics operation, great significance for improving timeliness assessment.
Automatic image annotation (AIA) is a promising way to improve the performance of retrieval. In this paper, we propose novel AIA scheme based on multiple-instance learning (MIL). By introducing minimum reference set (MRS) into MIL (denoted by MRS-MIL), positive instances (i.e. regions in images) embedded bags can be picked out via reliable inferring for concept. Generated through 1-NN classifier, MRS denotes number that correctly classify all labeled bags. Following principle structure risk...
With the rapid development of Internet, P2P has become main network application in which consumes most resources.Accurately identifying and making control traffic is great significance.As a mature classification theory, support vector machine (SVM) algorithm suitable for identification.This paper proposes SVM based flow identification method, adopting multidimensional properties as input vector, can improve accuracy.Analysis shows this method many advantages over other methods.
Learned B-frame video compression aims to adopt bi-directional motion estimation and compensation (MEMC) coding for middle frame reconstruction. However, previous learned approaches often directly extend neural P-frame codecs relying on optical-flow or interpolation. They suffer from inaccurate quantized motions inefficient compensation. To address these issues, we propose a simple yet effective structure called Interpolation-driven Video Compression (IBVC). Our approach only involves two...
Thin-plate spline (TPS) is a principal warp that allows for representing elastic, nonlinear transformation with control point motions. With the increase of points, becomes increasingly flexible but usually encounters bottleneck caused by undesired issues, e.g., content distortion. In this paper, we explore generic applications TPS in single-image-based warping tasks, such as rotation correction, rectangling, and portrait correction. To break bottleneck, propose coupled thin-plate model...