- Video Surveillance and Tracking Methods
- Advanced Measurement and Detection Methods
- Adversarial Robustness in Machine Learning
- Advanced Vision and Imaging
- Image and Object Detection Techniques
- Anomaly Detection Techniques and Applications
- Stroke Rehabilitation and Recovery
- Advanced Neural Network Applications
- Target Tracking and Data Fusion in Sensor Networks
- EEG and Brain-Computer Interfaces
- Bacillus and Francisella bacterial research
- Industrial Vision Systems and Defect Detection
- Medical Image Segmentation Techniques
- Digital Media Forensic Detection
- Medical Imaging and Analysis
- Image Enhancement Techniques
- Retinal Imaging and Analysis
- Gaze Tracking and Assistive Technology
- Infrared Target Detection Methodologies
- Human Pose and Action Recognition
- Robotics and Sensor-Based Localization
- Generative Adversarial Networks and Image Synthesis
- Brain Tumor Detection and Classification
- Remote Sensing and Land Use
- Image Processing Techniques and Applications
Shanghai University
2015-2025
Wenzhou University
2022-2024
Beijing University of Posts and Telecommunications
2023
Shandong University of Arts
2022
Shandong University of Art and Design
2022
Shanghai Jiao Tong University
2010-2015
Nihon University
2014
Ningbo University
2006
This exploration aims to study the emotion recognition of speech and graphic visualization expressions learners under intelligent learning environment Internet. After comparing performance several neural network algorithms related deep learning, an improved convolution network-Bi-directional Long Short-Term Memory (CNN-BiLSTM) algorithm is proposed, a simulation experiment conducted verify this algorithm. The experimental results indicate that Accuracy CNN-BiLSTM reported here reaches...
Hough Transform has been widely used for straight line detection in low-definition and still images, but it suffers from execution time resource requirements. Field Programmable Gate Arrays (FPGA) provide a competitive alternative hardware acceleration to reap tremendous computing performance. In this paper, we propose novel parallel (PHT) FPGA architecture-associated framework real-time high-definition videos. A resource-optimized Canny edge method with enhanced non-maximum suppression...
Class-Incremental Semantic Segmentation (CISS) addresses the challenge of catastrophic forgetting in semantic segmentation models. In autonomous driving scenarios, model can learn background class information from new data due to repetition many structural classes data. Traditional replay-based methods store original pixels these old data, resulting low memory efficiency. To enhance efficiency, we propose Spatial–Adaptive replay for Foreground objects (SAF), a method that stores only...
Universal Adversarial Perturbations (UAPs), which is type of image-agnostic adversarial attack, has been deeply investigated for Convolutional Neural Networks due to its high efficiency. On the other hand, as an architecture based on self-attention mechanism, Vision Transformers (ViTs) have boomed and widely applied solve various computer vision problems since most recent years. In this letter, we delve into robustness ViTs against universal attack propose inheritance attention matrix-based...
Background subtraction is a method typically used to segment moving regions in image sequences taken from static camera by comparing each new frame with model of the background scene. This paper proposes novel fuzzy algorithm for vehicle detection which achieves high rates, and reduces influence illumination changes shadows traffic The proposed adopts Choquet integral fusion similarity measures three color components YCbCr space uniform local binary pattern texture. Otherwise, an adaptive...
This paper proposes an embedded vision system for real-time moving object tracking using modified mean-shift algorithm mobile robot application. design of fully utilizing the advanced parallelism Field Programmable Gate Arrays (FPGA) is capable processing PAL video 720*576 at 25 fps. hardware implementation realizes time-consumed color space transformation pipeline operations, which completely removes dependence off-chip RAM memory. In addition, this incorporates adaptive kernel based mass...
Although Deep Neural Networks (DNNs) have achieved great success on various applications, investigations increasingly shown DNNs to be highly vulnerable when adversarial examples are used as input. Here, we present a comprehensive defense framework protect against examples. First, statistical and minor alteration detectors filter out contaminated by noticeable unnoticeable perturbations, respectively. Then, ensemble the detectors, deep Residual Generative Network (ResGN), an adversarially...
Early rehabilitation with the right intensity contributes to physical recovery of stroke survivors. In clinical practice, physicians determine whether training is suitable for based on patients' narratives, scores, and evaluation scales, which puts tremendous pressure medical resources. this study, a lightweight facial expression recognition algorithm proposed diagnose motivations automatically. First, properties convolution are introduced into Vision Transformer's structure, allowing model...
Traffic sign detection plays an important role in a traffic recognition system. It is also unsolved problem the intelligence transportation Two main technique routes are typically used current research of detection: using gray-level object image or that color image. The former has advantage related theories and methods more mature, but weakness short information, which results chances misdetection. latter abundant information can be provided same as human sense vision, still immaturity. This...
Background: Hand dysfunction is one of the main symptoms stroke patients, but there still a lack accurate hand function assessment systems. This study focused on application multi-dimensional intelligent visual quantitative system (MDIVQAS) in rehabilitation patients and evaluate patients. Methods: Eighty-two with unilateral were evaluated by MDIVQAS. Cronbach’s Alpha coefficient was used to assess internal consistency MDIVQAS; F-test differences MDIVQAS for multiple repeated measures....
Target tracking based only on color feature often leads to an inaccurate result when the background is complex. In this paper, a novel particle filter algorithm and contour features proposed. We first obtain information of moving target from its motion segmentation, then use Harris corner detector extract points contour. Afterwards, Hausdorff distance applied measure similarity between hypothetical candidate target. Experiments show that with fusion two features, can track robustly...
Deep neural networks (DNNs) have recently achieved impressive performances on various applications. However, recent researches show that DNNs are vulnerable to adversarial perturbations injected into input samples. In this paper, we investigate a defense method for face verification: deep residual generative network (ResGN) is learned clean perturbations. We propose novel training framework composed of ResGN, pre-trained VGG-Face and FaceNet network. The parameters ResGN optimized by...
As a self-supervised learning paradigm, contrastive has been widely used to pre-train powerful encoder as an effective feature extractor for various downstream tasks. This process requires numerous unlabeled training data and computational resources, which makes the pre-trained become valuable intellectual property of owner. However, lack priori knowledge tasks it non-trivial protect by applying conventional watermarking methods. To deal with this problem, in paper, we introduce AWEncoder,...
Nowadays, deep neural networks (DNN) have achieved significant success in computer vision. However, recent investigations shown that DNN models are highly vulnerable to an input adversarial example. How defense against examples is essential issue improve the robustness of models. In this paper, we present a hybrid framework integrates detecting and cleaning perturbations protect DNN. Specifically, part consists statistical detector Gaussian noise injection which adaptive perturbation...
This literature critically explains the intelligent method for detection of traffic signs. uses a particular color and shape signs, as an example, we used red down triangle sign, to explain this method. is mainly carried out in four steps, which are follows. First, convert RGB space HIS space, extract pixels with color. Then perform LOG mask operation on got from step 1, edges. By using neural network, determine angle pixels, at same time, also specific pixel is. And finally detect sign by...
In order to track the object effectively in real complex environment, paper proposes an algorithm which fuses color and edge feature establish observation model. Color histogram is not sensitive scale, rotation deformation a global description of target it suitable apply tracking shelter non-rigid object. However, particle filter based on cannot correctly when condition illumination changes similar clutters. Edge direction contains some information about structure space, can complement we...
Edge is an important feature of image and it very useful in machine vision application. In view the parallelism, logic operation pipelined Field Programmable Logic Array (FPGA), this paper proposes improved edge detection algorithm based on Canny operator for FPGA. Median Filter 3-way Parallel can complete preprocessing high-speed. Second Harmonic Variable Parameters (SHOVP) calculates gradient easily flexibly. 45 Degrees into Direction Gradient, Non Maximum Suppression Quarter Gradient will...
Occlusion is a challenge for tracking especially in dynamic scene. It adds the consideration background modeling. In condition, tracker will be influenced by both occlusions and background. this paper, we address problem proposing robust algorithm based on improved particle filter using discriminative model without Discriminative offers accurate templates occlusion detection alleviating influence from pixels. Since cannot carry out effective under heavy occlusion, blocking introduced to...