A5038291633- Advanced Chemical Sensor Technologies
- Remote-Sensing Image Classification
- Smart Agriculture and AI
- Advanced Vision and Imaging
- Spectroscopy and Chemometric Analyses
- Advanced Image Processing Techniques
- Advanced Neural Network Applications
- Face recognition and analysis
- Domain Adaptation and Few-Shot Learning
- Face and Expression Recognition
- Robotic Path Planning Algorithms
- Video Surveillance and Tracking Methods
- Emotion and Mood Recognition
- Image Processing Techniques and Applications
- Advanced Image and Video Retrieval Techniques
- Neural Networks and Applications
- Image and Signal Denoising Methods
- Modular Robots and Swarm Intelligence
- Interactive and Immersive Displays
- Gaze Tracking and Assistive Technology
- Data Mining Algorithms and Applications
- Medical Imaging and Analysis
- Neural dynamics and brain function
- Analytical Chemistry and Sensors
- EEG and Brain-Computer Interfaces
Kochi University of Technology
2018-2025
Shanghai Jiao Tong University
2024
Hangzhou Dianzi University
2017-2024
Hunan Institute of Engineering
2024
University of Science and Technology of China
2018-2023
Institute of Software
2023
Chinese Academy of Sciences
2023
Dalian University of Technology
2010-2020
Hong Kong Polytechnic University
2019
University of North Carolina at Charlotte
2017
This paper reviewed the 3rd NTIRE challenge on single-image super-resolution (restoration of rich details in a low-resolution image) with focus proposed solutions and results. The had 1 track, which was aimed at real-world single image problem an unknown scaling factor. Participants were mapping images captured by DSLR camera shorter focal length to their high-resolution longer length. With this challenge, we introduced novel dataset (RealSR). track 403 registered participants, 36 teams...
In this paper, a hierarchical deep multi-task learning (HD-MTL) algorithm is developed to support large-scale visual recognition (e.g., recognizing thousands or even tens of atomic object classes automatically). To achieve more effective accomplishment the coarse-to-fine tasks for recognition, multiple sets features are first extracted from different layers convolutional neural networks (deep CNNs). A tree then learned by assigning visually-similar with similar complexities into same group,...
In learning with noisy labels, the sample selection approach is very popular, which regards small-loss data as correctly labeled during training. However, losses are generated on-the-fly based on model being trained and thus large-loss likely but not certainly to be incorrect. There actually two possibilities of a point: (a) it mislabeled, then its loss decreases slower than other data, since deep neural networks "learn patterns first"; (b) belongs an underrepresented group has been selected...
Facial micro-expression refers to split-second muscle changes in the face, indicating that a person is either consciously or unconsciously suppressing their true emotions. Although these expressions are constantly occurring on people faces, they were easily ignored by with eye blinking. That say, most don't notice them and it representation of emotions mental health. Accordingly, both psychologists computer scientists (in fields vision machine learning particular) pay attention owing...
Discriminative dictionary learning has been successfully applied in pattern recognition field. In most of methods, ℓ <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sub> -norm or xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> is used to regularize the sparse representation coefficients, which makes computing time consuming. this paper, we present a novel method improve gas identification performance electronic nose. It significantly less...
In this paper, a level-wise mixture model (LMM) is developed by embedding visual hierarchy with deep networks to support large-scale recognition (i.e., recognizing thousands or even tens of object classes), and Bayesian approach used adapt pre-trained automatically the improvements features (that are for image class representation) when more representative learned along time. Our LMM can provide an end-to-end jointly learning: (a) extract discriminative representation; (b) tree classifier...
Micro-expression is a kind of brief facial movements which could not be controlled by the nervous system. indicates that person hiding his true emotion consciously. recognition has various potential applications in public security and clinical medicine. Researches are focused on automatic micro-expression recognition, because it hard to recognize people themselves. This research proposed novel algorithm for combined deep multi-task convolutional network detecting landmarks fused estimating...
Micro-expression is a kind of brief facial movements which could not be controlled by nervous system. indicates that person hiding his truly emotion consciously. analysis has various potential applications in public security and clinical medicine. Researches are focused on the automatic micro-expression recognition, because it hard to recognize naked eye. This research proposes novel algorithm for combines deep multi-task convolutional network detecting landmarks fused estimating optical...
We derive ensembles of decision trees through a nonparametric Bayesian model, allowing us to view random forests as samples from posterior distribution. This insight provides large gains in interpretability, and motivates class forest (BF) algorithms that yield small but reliable performance gains. Based on the BF framework, we are able show high-level tree hierarchy is stable samples. leads an empirical (EBF) algorithm for building approximate BFs massive distributed datasets EBFs...
Face detection has been well studied for many years. One remaining challenge is to detect faces from low-light images. The brightness of the image captured under extremely conditions could be very low and contrast will severely reduced. It easy cause confusion during feature extraction affects performance face detection. In this paper, we propose a single-stage method. First, design an improved MSRCR method increase quality condition ensuring that colors are not distorted. shows better...
Single-frame optical flow estimation is a more challenging task than predicting the between adjacent frames in video. This paper presents two-stream network that combines motion information from correlation images with appearance intensity to estimate single-frame manner. The image generated by three-phase sensor (3PCIS) records changes incident during exposure time and conveys about moving objects. crucial assist motion-blurred estimating state of Due lack directly available datasets train...
It seems difficult to recognize an object from its background with similar color using conventional segmentation methods. An efficient way is utilize hyperspectral images that contain more wave bands and richer information than only RGB components. Particularly in our task, we aim separate a pepper densely packed green leaves for automatic picking agriculture. Given imaging can be regarded as kind of propagation process, make novel attempt introducing complex neural network tailored...
A novel sparse representation classification method (SRC), namly SRC based on Method of Optimal Directions (SRC_MOD), is proposed for electronic nose system in this paper. By finding both a synthesis dictionary and corresponding coefficient vector, the i-th class training samples are approximated as linear combination few atoms. The optimal solutions vector found by MOD. Finally, testing identified evaluating which causes least reconstruction error. algorithm evaluated analysis hydrogen,...
Accurate and efficient segmentation of prostate image plays an important role in the diagnosis cancer. Since convolutional neural network demonstrates superior performance computer vision applications, we present a multi-layer deeply supervised deconvolution (DSDN) which completes end-to-end training to automatically segment magnetic resonance (MR) images. We put additional layers supervise hidden layers. During training, backpropagation process gradient information accelerates parameters...
Detecting and identifying objects of similar color is a challenging task in computer vision. Green peppers natural environment can be found using the abundant information provided by hyperspectral camera spectral domain, but an expensive device. Therefore, we propose novel framework called Optical Filter Net, which enables design optical filter that improves performance green pepper segmentation specific red-green-blue (RGB) system. When installed with filter, system efficiently utilize...
With the increasing number and variety of robots, there are still many problems limitations in offline communication cooperation platform. To address this issue, we present a universal ROS robot live interaction platform named TeleRobot that offers features such as multi-angle streaming, remote with online discussions. The implementation is based on microservice architecture WebRTC technology, utilizing Kurento Media Server (KMS) for streaming media transmission, rosbridge real-time robots....
Because of the challenge collecting labelled training data, zero-shot learning (ZSL) which transfers semantic knowledge represented by category attributes from seen classes to recognize unseen has received a lot attention recently. Existing methods assume that source are completely correct in learning. However, practice may contain noise due assignment mistake. In this paper, we develop novel robust method (RZSL) for ZSL. It seeks lowest rank representation among all candidates and extracts...