A5100352614- Advanced Image and Video Retrieval Techniques
- Image Retrieval and Classification Techniques
- Image Enhancement Techniques
- Advanced Neural Network Applications
- Infrared Target Detection Methodologies
- Advanced Vision and Imaging
- Video Surveillance and Tracking Methods
- Generative Adversarial Networks and Image Synthesis
- Robotics and Sensor-Based Localization
- Medical Image Segmentation Techniques
- Advanced Image Processing Techniques
- Advanced Image Fusion Techniques
- Radiomics and Machine Learning in Medical Imaging
- Digital Media Forensic Detection
- Robotic Path Planning Algorithms
- Face recognition and analysis
- Advanced Measurement and Detection Methods
- Medical Imaging and Analysis
- Image Processing Techniques and Applications
- Anomaly Detection Techniques and Applications
- Image and Video Quality Assessment
- Human Pose and Action Recognition
- Video Analysis and Summarization
- Advanced X-ray and CT Imaging
- Face and Expression Recognition
Sun Yat-sen University
2017-2025
Northeastern University
2024
Northwest A&F University
2016-2023
Bridge University
2021
Shanghai Public Security Bureau
2017
Tsinghua University
2014-2015
National University of Defense Technology
2015
Hebei Agricultural University
2014
Zhejiang Gongshang University
2010-2012
University of Louisville
2008-2010
Abstract Deep neural networks (DNNs) have gained remarkable success in speech recognition, partially attributed to the flexibility of DNN models learning complex patterns signals. This flexibility, however, may lead serious over-fitting and hence miserable performance degradation adverse acoustic conditions such as those with high ambient noises. We propose a noisy training approach tackle this problem: by injecting moderate noises into data intentionally randomly, more generalizable can be...
Exploring the relationships between plant phenotypes and genetic information requires advanced phenotypic analysis techniques for precise characterization. However, diversity variability of morphology challenge existing methods, which often fail to generalize across species require extensive annotated data, especially 3D datasets. This paper proposes a zero-shot leaf instance segmentation method using RGB sensors. It extends 2D model SAM (Segment Anything Model) through multi-view strategy....
Deep-learning-based technologies such as deepfakes ones have been attracting widespread attention in both society and academia, particularly used to synthesize forged face images. These automatic professional-skill-free manipulation can be replace the an original image or video with any target object while maintaining expression demeanor. Since human faces are closely related identity characteristics, maliciously disseminated manipulated videos could trigger a crisis of public trust media...
The latest medical image segmentation methods uses UNet and transformer structures with great success. Multiscale feature fusion is one of the important factors affecting accuracy segmentation. Existing transformer-based do not comprehensively explore multiscale fusion, there still much room for improvement. In this paper, we propose a novel multiresolution aggregation (MRA-TUNet) based on input coordinate attention It realizes from following two aspects: (1) On side, module used to fuse...
Multitarget tracking (MTT) in surveillance system is extremely challenging, due to uncertain data association, maneuverable target motion, dense clutter disturbance, and real-time processing requirements.A good many methods have been proposed cope with these challenges.However, no up-to-date survey available the literature that can help select suitable algorithm for practical problem.This paper provides a comprehensive review of state-of-the-art motion-based MTT techniques, classifies...
Postmortem investigation of methamphetamine (MA) abuse is an important task in forensic pathology. The present study investigated morphological changes the astrocytes parietal cerebral cortex MA abusers. Glial fibrillary acidic protein immunoreactivity was examined autopsy cases for MA-detected group and control group. Clasmatodendrotic (including those with swollen cell bodies disintegrating distal processes) were frequently observed Quantitative analysis using a colour image processor...
This paper presented a new zero-watermarking algorithm for vector digital maps based on statistical characteristics. The watermark information is constructed by utilizing the original data's We divide map into rings using concentric circles and count number of vertices in each ring, which feature information. A zero image copyright Experiments show that watermarks are resilient to translation, scaling, vertex deletion growth, rotation, random noise, objects scrambling cropping, making it...
Task planning involving multiple unmanned aerial vehicles (UAVs) is one of the main research topics in field cooperative vehicle control systems. This a complex optimization problem where task allocation and path are dealt with separately. However, recalculation optimal results too slow for real-time operations dynamic environments due to large amount computation required, traditional algorithms difficult handle scenarios varying scales. Meanwhile, approach confines 2D environment, which...
Infrared ship target detection is crucial technology in marine scenarios. Ship targets vary scale throughout navigation because the distance between and infrared camera constantly changing. Furthermore, complex backgrounds, such as sea clutter, can cause significant interference during tasks. In this paper, multiscale morphological reconstruction-based saliency mapping, combined with a two-branch compensation strategy (MMRSM-TBC) algorithm, proposed for of various sizes against backgrounds....
Abstract The three successive coronal mass ejections (CMEs) that erupted from 2023 November 27–28, provide the first opportunity to shed light on entire process of a shock propagating through, sequentially compressing, and modifying two preceding CMEs using in situ data Solar Orbiter, Wind, STEREO-A. We describe interaction as follows: CME-1 CME-2 interacted with each other at distances close Sun. Subsequently, (S3) driven by CME-3 caught up compressed ICME-2 before 0.83 au, forming typical...
Camera calibration is an important step for vision-based measurement applications. A well-known flexible camera method analyzed that uses the checkerboard pattern plane and in which can be moved freely. When using a perspective projection model, characteristics of both objective image are utilized accurate results obtained. However, method's may fail when rotation angles planar small, distortion coefficients obtained under model not used real-time vision application. We solve ill-conditioned...
针对光学成像制导武器系统对图像处理的实时性要求,该文提出了一种基于硬件加速的2次扫描连通域标记算法。算法结合基于像素和基于游程扫描算法的优点,以像素为基本的扫描单元,以线段为基本的标号单元,在第1次扫描过程中建立临时标号的树形拓扑结构,并输出线段作为结果。第2次扫描对线段进行标号替换完成连通域标记。通过在FPGA+DSP平台中进行实验证明,该文算法的硬件加速实现占用资源少,能够达到较高的性能和执行效率,保证了系统的实时性,具有较高的实用价值。
Deep neural networks (DNN) have gained remarkable success in speech recognition, partially attributed to its flexibility learning complex patterns of signals. This flexibility, however, may lead serious over-fitting and hence miserable performance degradation adverse environments such as those with high ambient noises. We propose a noisy training approach tackle this problem: by injecting noises into the intentionally randomly, more generalizable DNN models can be learned. `noise injection'...
This paper detects violations of smoking in non-smoking areas by construction workers, uses the YOLO object detection algorithm combined with Kalman filter to track human body, then Alphapose's pose estimation obtain key points body. We input into Spatial-Temporal Graph Convolutional Networks for preliminary identification workers' behavior. However, this will lose texture feature information image, resulting drinking, scratching, etc. also be recognized as smoking. Therefore, based on...