A5074234291- Advanced Neural Network Applications
- Advanced Image and Video Retrieval Techniques
- Visual Attention and Saliency Detection
- Video Surveillance and Tracking Methods
- Image and Video Quality Assessment
- Olfactory and Sensory Function Studies
- Human Pose and Action Recognition
- Domain Adaptation and Few-Shot Learning
- Graph Theory and Algorithms
- Advanced Graph Neural Networks
- Ear Surgery and Otitis Media
- Gait Recognition and Analysis
- Remote-Sensing Image Classification
- Radiomics and Machine Learning in Medical Imaging
- Data Management and Algorithms
- Infrastructure Maintenance and Monitoring
- Fire Detection and Safety Systems
- AI in cancer detection
- Image Retrieval and Classification Techniques
- Remote Sensing and Land Use
- Breast Cancer Treatment Studies
- Medical Image Segmentation Techniques
- Robotics and Sensor-Based Localization
- Face Recognition and Perception
- Smart Agriculture and AI
South China University of Technology
2018-2024
Guangzhou Experimental Station
2021-2023
Shenyang Institute of Computing Technology (China)
2015-2023
Guangdong 999 Brain Hospital
2021
Wuhan University
2015-2019
State Key Laboratory of Information Engineering in Surveying Mapping and Remote Sensing
2015-2019
Huazhong University of Science and Technology
2012-2016
University of Nebraska–Lincoln
2014-2016
Chinese Academy of Sciences
2015
Wuhan University of Technology
2013
To develop a deep convolutional neural network (DCNN) that can automatically detect laryngeal cancer (LCA) in laryngoscopic images.A DCNN-based diagnostic system was constructed and trained using 13,721 images of LCA, precancerous lesions (PRELCA), benign tumors (BLT) normal tissues (NORM) from 2 tertiary hospitals China, including 2293 206 LCA subjects, 1807 203 PRELCA 6448 774 BLT subjects 3191 633 NORM subjects. An independent test set 1176 other 3 132 44 129 43 504 168 411 137 applied to...
Due to its wide applications, remote sensing (RS) image scene classification has attracted increasing research interest. When each category a sufficient number of labeled samples, RS can be well addressed by deep learning. However, in the big data era, it is extremely difficult or even impossible annotate samples for all categories one time as often needs extended along with emergence new applications that inevitably involve class images. Hence, era fairly requires zero-shot (ZSRSSC)...
Bridge detection in remote sensing images (RSIs) plays a crucial role various applications, but it poses unique challenges compared to the of other objects. In RSIs, bridges exhibit considerable variations terms their spatial scales and aspect ratios. Therefore, ensure visibility integrity bridges, is essential perform holistic bridge large-size very-high-resolution (VHR) RSIs. However, lack datasets with VHR RSIs limits deep learning algorithms' performance on detection. Due limitation GPU...
Few-shot instance segmentation (FSIS) conjoins the few-shot learning paradigm with general segmentation, which provides a possible way of tackling in lack abundant labeled data for training. This paper presents Fully Guided Network (FGN) segmentation. FGN perceives FSIS as guided model where so-called support set is encoded and utilized to guide predictions base network (i.e., Mask R-CNN), critical guidance mechanism. In this view, introduces different mechanisms into various key components...
Visual saliency is attracting more and research attention since it beneficial to many computer vision applications. In this paper, we propose a novel bottom-up model for detecting salient objects in natural images. First, inspired by the recent advance realm of statistical thermodynamics, adopt mathematical model, namely, maximal entropy random walk (MERW) measure saliency. We analyze rationality superiority MERW modeling visual Then, based on establish generic framework detection. Different...
This paper proposes an effective Temporally Aligned Pooling Representation (TAPR) for video-based person re-identification. To extract the motion information from a sequence, we propose to track superpixels of lowest portions human. perform temporal alignment videos, select "best" walking cycle noisy according intrinsic periodicity property persons, that is fitted sinusoid in our implementation. describe video data selected cycle, first divide into several segments sinusoid, and then each...
In this paper, we propose a simple yet effective self-supervised method called spatio-temporal contrastive learning (ST-CL) for 3D skeleton-based action recognition. ST-CL acquires action-specific features by regarding the continuity of motion tendency as supervisory signal. To yield representations, first designs some novel proxy tasks providing different observation scenes same and pulling them together in embedding space. Second, three key components are devised encoding to efficiently...
Objectives This study investigated the usefulness and performance of a two-stage attention-aware convolutional neural network (CNN) for automated diagnosis otitis media from tympanic membrane (TM) images. Design A classification model development validation in ears with based on otoscopic TM Two commonly used CNNs were trained evaluated dataset. On basis Class Activation Map (CAM), pipeline was developed to improve accuracy reliability, simulate an expert reading Setting participants is...
Whole-slide image (WSI) classification is fundamental to computational pathology, which challenging in extra-high resolution, expensive manual annotation, data heterogeneity, etc. Multiple instance learning (MIL) provides a promising way towards WSI classification, nevertheless suffers from the memory bottleneck issue inherently, due gigapixel high resolution. To avoid this issue, overwhelming majority of existing approaches have decouple feature encoder and MIL aggregator networks, may...
The past few years have witnessed impressive progress on the research of salient object detection. Nevertheless , existing approaches still cannot perform satisfactorily in case complex scenes, particularly when objects non- uniform appearance or complicated shapes, and background is complexly structured. One important reason for such limitations may be that these commonly ignore factor perceptual grouping saliency modeling. To address this issue, paper presents a novel computational model...
A brain-computer interface (BCI) measures and analyzes brain activity converts this into computer commands to control external devices. In contrast traditional BCIs that require a subject-specific calibration process before being operated, subject-independent BCI learns model eliminates for new users. However, building remains difficult because electroencephalography (EEG) is highly noisy varies by subject. study, we propose an invariant pattern learning method based on convolutional neural...
Recently, deep learning based methods have demonstrated promising results on the graph matching problem, by relying descriptive capability of features extracted nodes. However, one main limitation with existing (DGM) lies in their ignorance explicit constraint structures, which may lead model to be trapped into local minimum training. In this paper, we propose explicitly formulate pairwise structures as a quadratic incorporated DGM framework. The minimizes structural discrepancy between...
An increasing number of people fail to properly regulate their emotions for various reasons. Although brain-computer interfaces (BCIs) have shown potential in neural regulation, few effective BCI systems been developed assist users emotion regulation. In this paper, we propose an electroencephalography (EEG)-based regulation with virtual reality (VR) neurofeedback. Specifically, music clips positive, neutral, or negative were first presented, based on which the participants asked emotions....
The smooth operation of autonomous underwater vehicles (AUVs) relies heavily on the accurate detection surrounding objects. Toward this end, letter presents a novel method for object based gravity gradient differential and ratio caused by relative motion between AUV object. Unlike existing techniques, proposed works in passive manner achieves invisibility without energy emission. In addition, method, no map or is required, which improves its practicality. Experimental results demonstrate...
Tracking by detection has become an attractive tracking technique, which treats as object problem and trains a detector to separate the target from background in each frame. While this strategy is effective some extent, we argue that task should be searching for specific instance instead of category. Based on viewpoint, novel framework based exemplar detectors proposed visual tracking. To build discriminative model background, method exemplar-based linear discriminant analysis (ELDA)...
Few-shot learning (FSL) usually assumes that the query is drawn from same label space as support set, while queries unknown classes may emerge unexpectedly in many open-world application scenarios. Such an open-set issue will limit practical deployment of FSL systems, which remains largely unexplored. In this paper, we investigate problem few-shot recognition (FSOR) and propose a novel solution, called Relative Feature Displacement Network (RFDNet), empowers systems to reject accurately...
Instance segmentation is a challenging computer vision problem which lies at the intersection of object detection and semantic segmentation. Motivated by plant image analysis in context phenotyping, recently emerging application field vision, this paper presents Exemplar-Based Recursive Segmentation (ERIS) framework. A three-layer probabilistic model firstly introduced to jointly represent hypotheses, voting elements, instance labels their connections. Afterwards, recursive optimization...