A5100749764- Adaptive Control of Nonlinear Systems
- Face and Expression Recognition
- Industrial Vision Systems and Defect Detection
- Magnetic Bearings and Levitation Dynamics
- Inertial Sensor and Navigation
- Advanced Algorithms and Applications
- Distributed Control Multi-Agent Systems
- Neural Networks Stability and Synchronization
- Guidance and Control Systems
- Advanced Image and Video Retrieval Techniques
- Biometric Identification and Security
- Face recognition and analysis
- Control and Dynamics of Mobile Robots
- Smart Agriculture and AI
- Video Surveillance and Tracking Methods
- Advanced Neural Network Applications
- Handwritten Text Recognition Techniques
- Data Stream Mining Techniques
- Domain Adaptation and Few-Shot Learning
- Vehicle License Plate Recognition
- Artificial Intelligence in Games
- Optical Systems and Laser Technology
- Multimodal Machine Learning Applications
- Advanced Graph Neural Networks
- Reinforcement Learning in Robotics
Nanjing Agricultural University
2023-2024
Changsha University of Science and Technology
2023
Hefei University of Technology
2017-2021
Huazhong University of Science and Technology
2018
Wuhan National Laboratory for Optoelectronics
2018
Liaoning University of Technology
2016
Shanghai Micro Satellite Engineering Center
2015
Shaanxi Xueqian Normal University
2014
The Optical Character Recognition (OCR) systems have been widely used in various of application scenarios, such as office automation (OA) systems, factory automations, online educations, map productions etc. However, OCR is still a challenging task due to the text appearances and demand computational efficiency. In this paper, we propose practical ultra lightweight system, i.e., PP-OCR. overall model size PP-OCR only 3.5M for recognizing 6622 Chinese characters 2.8M 63 alphanumeric symbols,...
Optical Character Recognition (OCR) systems have been widely used in various of application scenarios. Designing an OCR system is still a challenging task. In previous work, we proposed practical ultra lightweight (PP-OCR) to balance the accuracy against efficiency. order improve PP-OCR and keep high efficiency, this paper, propose more robust system, i.e. PP-OCRv2. We introduce bag tricks train better text detector recognizer, which include Collaborative Mutual Learning (CML), CopyPaste,...
In the reinforcement learning field, traditional Q-Learning method can make agent get a good grades on some simple games whose states are limited. Meanwhile, if model of game is relatively simple, continuous also be transformed into finite through discretization so that achieve results. However, will face many difficulties for video with image output, such as too complicated to store. this paper, an algorithm deep learning(DRL), called Deep Q-Network(DQN), which originated from and combining...
The topic about the consensus tracking algorithm design is investigated for nonlinear second-order leader-following multi-agent systems via event-triggered sampled-data control. By combing techniques of with time-triggered, an feedback control law designed, which can ensure that state reachable and time two consecutive transmissions bounded below by a positive constant. Simulation result presented to show efficiencies given algorithm.
Abstract Semantic segmentation is crucial to the autonomous driving, as an accurate recognition and location of surrounding scenes can be provided for street understanding task. Many existing networks usually fuse high‐level low‐level features boost performance. However, simple fusion may impose a limited performance improvement because gap between features. To alleviate this limitation, we respectively propose spatial aggregation channel bridge gap. Our implementation, inspired by attention...
<title>Abstract</title> In deep learning based vision tasks, improving multiscale representation by combining shallow and features has consistently led to performance gains across a wide range of applications. However, significant discrepancies in both scale semantic content often occur during the fusion features. Most existing approaches rely on standard convolutional structures for representing features, which may not fully capture complexity underlying data. To address this, we propose...
Abstract Morphing attacks (MAs) pose a substantial security threat to the Automatic Border Control (ABC) system. While few morphing attack detection (MAD) methods have been proposed, face accomplice's facial restoration has not received sufficient attention. Due inability foresee factor used for particular morphed image, selecting appropriate de‐morphing becomes challenging problem in of image. If cannot be chosen reasonably, achieving desired effect is difficult. Therefore, this paper...
This paper considers a method for the finite-time consensus control of second-order multi-agent systems in leaderless case by both velocity and input constraints. A new class distributed controllers are proposed, which shows that under appropriate initial speed conditions, constraints can be met process while achieving consensus. Then, some simulation results proposed to verify effectiveness controller.
The development of intelligent vehicle involves many key technologies. machine vision technology based on deep learning has become one the research hotspots because its good performance.In this paper, aiming at environment perception smart cars, YOLOv2 algorithm is improved by combining with experimental data set, and it applied to real-time detection traffic scene objects.First, PASCAL VOC, different network structures loss functions was trained tested.According test results, structure...
For position tracking problem for quad-rotor system with velocity constrains, a finite-time control method constrains is proposed in this paper. Firstly, based on backstepping design, controller constraints to ensure that the desired can be tracked finite time constrains. Finally, simulation results are given verify effectiveness and advantages of
This paper mainly investigates the attitude control problem with angular velocity constraints for a spacecraft. First of all, finite-time controller constrains is proposed. Under action this controller, quaternion can be stabilized to desired value and origin in finite time. Secondly, simulation section, effectiveness proposed demonstrated.
This paper mainly investigates the attitude control problem for a spacecraft. Not only angular velocity constraint but also external disturbances are considered in system model. Under action of proposed controller, posture can be converged to desired value, and asymptotically origin.
In many applications of modern world, moving object detection is an important work. Especially, the under complicated background has more difficult. On basis introduction conventional methods such as temporal difference (TDD) and subtraction detection, paper proposes a new mathematical morphology based algorithm in view background. The proposed utilizes opening closing operations to process differenced image. Then, it compares processed image with experimental results show that can be able...
Face morphing attacks have been demonstrated to present a significant risk current Automatic Border Control systems. Despite research on attack detection has made certain progress, the forensic analysis and restoration of accomplices' facial images remain in initial stage. Currently, face de-morphing methods based deep learning technology usually ignore role landmarks, which are crucial structure images. Therefore, paper proposes CALP-FD, novel method for using cross-attention-based...
This paper depended on that video monitoring and images in the coal mine were susceptible to dust, light miner's safety helmet other special environmental impact.In order realize real-time accurate face recognition rate, lay good foundation for miners' behavior characteristics subsequent research intelligent mine.The inherent degeneration 、stability Rotational invariance of singular value can reflect matrix fully.In image as miner is a very method.But only it doesn't work using miners...