A5100614103- Speech and Audio Processing
- Advanced Adaptive Filtering Techniques
- Advanced Neural Network Applications
- COVID-19 diagnosis using AI
- Digital Imaging for Blood Diseases
- AI in cancer detection
- Radiomics and Machine Learning in Medical Imaging
- Network Security and Intrusion Detection
- Medical Imaging Techniques and Applications
- Music and Audio Processing
- Speech Recognition and Synthesis
- Medical Image Segmentation Techniques
- Industrial Vision Systems and Defect Detection
- Advanced Vision and Imaging
- Real-Time Systems Scheduling
- Image and Object Detection Techniques
- MRI in cancer diagnosis
- Video Surveillance and Tracking Methods
- Advanced Computational Techniques and Applications
- Advanced Image and Video Retrieval Techniques
- Software System Performance and Reliability
- Petri Nets in System Modeling
- Vehicle License Plate Recognition
- Artificial Intelligence in Healthcare
- Higher Education and Teaching Methods
South China Normal University
2021-2024
Harbin University of Science and Technology
2022-2023
Hebei University of Science and Technology
2023
Kuaishou (China)
2022
Xi'an Shiyou University
2022
Chongqing University of Posts and Telecommunications
2019
Anhui Science and Technology University
2010
Anhui University of Science and Technology
2010
State Key Laboratory of Remote Sensing Science
2010
Northeast Electric Power University
2007
Multi-channel speech enhancement is gaining increasing interest in recent years. By combining the beamforming framework with deep neural network, significant improvement on performance has been achieved. While beamformers designed for distributed microphone arrays are deployed practical applications such as teleconferencing and surveillance, there less approach co-located soundfield microphones. In this work, a new network proposed B-format 3D multi-channel recognition. The method...
<abstract> <sec><title>Background</title><p>Automatic liver segmentation is a prerequisite for hepatoma treatment; however, the low accuracy and stability hinder its clinical application. To alleviate this limitation, we deeply mine context information of different scales combine it with deep supervision to improve in paper.</p> </sec> <sec><title>Methods</title><p>We proposed new network called MAD-UNet automatic from CT. It...
<abstract> <p>The existing 2D/3D strategies still have limitations in human liver and tumor segmentation efficiency. Therefore, this paper proposes a 2.5D network combing cascaded context module (CCM) Ladder Atrous Spatial Pyramid Pooling (L-ASPP), named CCLNet, for automatic from CT. First, we utilize the mode to improve training efficiency; Second, employ ResNet-34 as encoder enhance accuracy. Third, L-ASPP is used enlarge receptive field. Finally, CCM captures more local...
As PET imaging is accompanied by radiation exposure and potentially increased cancer risk, reducing dose in scans without compromising the image quality an important topic. Deep learning (DL) techniques have been investigated for low-dose imaging. However, existing models often resulted compromised when achieving limited generalizability to different noise-levels, acquisition protocols, patient populations, hospitals. Recently, diffusion emerged as new state-of-the-art generative model...
Speech enhancement methods based on deep learning have surpassed traditional methods. While many of these new approaches are operating the wideband (16kHz) sample rate, a fullband (48kHz) speech system is proposed in this paper. Compared to existing full-band systems that utilize perceptually motivated features train with single network structure, two-step ensuring good quality while backward compatible systems.
Bone marrow cell morphology has always been an important tool for the diagnosis of blood diseases. Still, it requires years experience from a suitable person. Furthermore, outcomes their recognition are subjective and there is no objective quantitative standard. As result, developing deep learning automatic classification system bone cells extremely important. However, typical machine systems only produce answers, will not refuse to generate predictions when prediction reliability low. It...
Deep learning-based wideband (16kHz) speech enhancement approaches have surpassed traditional methods. This work further extends the existing systems to enable full-band (48kHz) while simultaneously ensuring automatic recognition compatibility and optionally, personalized enhancement. As shown in evaluation results, this is achieved by employing a multi-stage multi-loss training architecture that incorporates recently proposed two-step structure, ASR loss produced back-end encoder, speaker...
Scene text detection plays an important role in computer vision and pattern recognition field recent years due to extract the accurate rich information. At present, component-based methods have become trend, there are still some challenges because of illumination, blur difficult background. In this paper, orientation-correction method for scene based on Spatial Pyramid Pooling Convolutional Neural Networks, SPP-CNN , is proposed. Firstly, enhanced multi-channel MSER model, which constructed...
UAV remote sensing is low cost and convenient, but traditional methods for image registration mosaic are efficient, since it hard to find ground control points large quantity of images with small scale. In this study, adjacency transfer algorithm used between every two adjacent automate the work flow: First SIFT(Scale-invariant feature transform) employed detect correspondence points, then polynomial geometric correction, finally all digital orthophoto mosaiced into one image. Developed in...
In order to solve the high rate of wrong alarm in IDS, we designed control model by analyzing alarming information. This which using human olfaction passivation aims at sustaining frequency no-action information reduce achieve low IDS and make it convenient administrators.
Aiming at the current problem of low detection efficiency and high error when manually inspecting casing joint for calibration depth in visual inspection oil gas wells, a intelligent recognition based on YOLOv5 algorithm is proposed, which can realize joint. Firstly, large number pictures well joints were collected dataset was made by data enhancement method. Then, enhanced annotated with Labelimg tool sent to network training. Finally, use best trained weights testing result. The test...
Road intelligence monitoring is an inevitable trend of urban intelligence, and clothing information the main factor to identify pedestrians. Therefore, this paper establishes a multi-information recognition model proposes data augmentation method based on road monitoring. First, we use Mask R-CNN detect category in monitoring; then, transfer mask k-means cluster obtain color finally category. However, scene dataset are quite different, so suitable for designed improve ability small targets...
As PET imaging is accompanied by substantial radiation exposure and cancer risk, reducing dose in scans an important topic. Recently, diffusion models have emerged as the new state-of-the-art generative model to generate high-quality samples demonstrated strong potential for various tasks medical imaging. However, it difficult extend 3D image reconstructions due memory burden. Directly stacking 2D slices together create volumes would results severe inconsistencies between slices. Previous...