Qing Fei

Vision based ground target tracking for rotor UAV

This paper studies an efficient ground target tracking algorithm for rotor Unmanned Aerial Vehicle (UAV) to overcome the contradiction among the target tracking rapidity, precision and robustness for aerial vehicle. Firstly, Scale Invariant Feature Transform (SIFT) algorithm, which has a better robust performance during rotation, scaling and changes of illumination, is utilized to extract and match the feature points in order to realize target recognition and positioning. Secondly, using top-down tracking method, Kalman filter is combined to estimate the target position in the next frame and search target in the predicted area, it can avoid blind matching, improve tracking rapidity and reduce the ratio of losing target. Finally, an experimental platform of rotor UAV visual tracking is set up and the ground target tracking algorithm is tested. The experiment results show that the algorithm can achieve ground target tracking effectively and has good real-time performance and robustness.

Convolutional Neural Network Based on Extreme Learning Machine for Maritime Ships Recognition in Infrared Images

The success of Deep Learning models, notably convolutional neural networks (CNNs), makes them the favorable solution for object recognition systems in both visible and infrared domains. However, the lack of training data in the case of maritime ships research leads to poor performance due to the problem of overfitting. In addition, the back-propagation algorithm used to train CNN is very slow and requires tuning many hyperparameters. To overcome these weaknesses, we introduce a new approach fully based on Extreme Learning Machine (ELM) to learn useful CNN features and perform a fast and accurate classification, which is suitable for infrared-based recognition systems. The proposed approach combines an ELM based learning algorithm to train CNN for discriminative features extraction and an ELM based ensemble for classification. The experimental results on VAIS dataset, which is the largest dataset of maritime ships, confirm that the proposed approach outperforms the state-of-the-art models in term of generalization performance and training speed. For instance, the proposed model is up to 950 times faster than the traditional back-propagation based training of convolutional neural networks, primarily for low-level features extraction.

Modeling and control for longitudinal attitude of a twin-rotor tail-sitter unmanned aerial vehicle

In this paper, the modeling and model reference adaptive control (MRAC) for longitudinal attitude of a twin-rotor tail-sitter unmanned aerial vehicle (UAV), which is highly unstable during flight, are presented. First, the attitude dynamic models are established. Linearized model for longitudinal attitude in vertical flight mode is given that is used in later derivation of controller as well as for testing the algorithms in simulation. Then, a control law based on the MRAC technique is utilized to stabilize the longitudinal attitude control system with uncertainty. Simulation results show that the MRAC of pitch angle has good trajectory tracking and the designed control law has strong adaptive ability and anti-jamming ability.

Design of control system for an unmanned semi-submersible vehicle

Given the fact that ocean is vast and its condition is always changing, it is difficult to collect meteorological and oceanographic data. In this paper, a semi-submersible un-manned surface vehicle is introduced to decrease the difficulties. The hardware and software design of the USV, which is a rigid hull with tow screw propellers, is discussed at the beginning. Then, the mathematics model and control algorithm related to path following control of the USV are presented. Finally, several experiments are conducted to verify the feasibility and control performance of the unmanned vehicle.

Tailsitter VTOL flying wing aircraft attitude control

This paper introduces a flying-wing tail-sitter aircraft which can switch between vertical flight mode (rotary-wing mode) and level flight mode (fixed-wing mode). Equipped with two propellers and two elevons, the aircraft can fly by controlling these four actuators. The aircraft uses a microcomputer and various sensors to stabilize the attitude and to switch modes on command. Using the transition logic, it can receive and act on the signal sent by operator at any time, whether it is to switch modes or to retract the command. Using the target angle calculation algorithm, the aircraft can adjust its pitch angle during transition. PID feedback is used for attitude control both in vertical mode and during the transition. Test results show that the aircraft have the advantages of helicopters and fix-wing airplanes. It can hover midair and does not need runway to take off, and it can fly in high speed in fixed-wing mode.

Switching Control System Based on Robust Model Reference Adaptive Control

For conventional adaptive control, time-varying parametric uncertainty and unmodeled dynamics are ticklish problems, which will lead to undesirable performance or even instability and nonrobust behavior, respectively. In this study, a class of discrete-time switched systems with unmodeled dynamics is taken into consideration. Moreover, nonlinear systems are here supposed to be approximated with the class of switched systems considered in this paper, and thereby switching control design is investigated for both switched systems and nonlinear systems to assure stability and performance. For robustness against unmodeled dynamics and uncertainty, robust model reference adaptive control (RMRAC) law is developed as the basis of controller design for each individual subsystem in the switched systems or nonlinear systems. Meanwhile, two different switching laws are presented for switched systems and nonlinear systems, respectively. Thereby, the authors incorporate the corresponding switching law into the RMRAC law to construct two schemes of switching control respectively for the two kinds of controlled systems. Both closed-loop analyses and simulation examples are provided to illustrate the validity of the two proposed switching control schemes. Furthermore, as to the proposed scheme for nonlinear systems, its potential for practical application is demonstrated through simulations of longitudinal control for F-16 aircraft.

Composite μ-mod model reference adaptive control with input constraints

Based on the widely used model reference adaptive control (MRAC), a new controller is proposed to tackle a class of single-input(SI) systems in the presence of matched uncertainties and input constraints. The core idea is to substitute conventional simple direct adaptive control for composite model adaptive control (CMRAC). The main control structure that results in tractable responsibility is discussed in tail, in which the performance is adaptive to address the constraints of input in addition to retaining the usual transient properties and dynamic stability. To this end, a typical positive μ-mod method is developed on the basis of CMRAC frame to aim at a correct control. The formulated method supplies a convenient and intuitive interpretation to solve the control design problem, meanwhile utilizing the basic MRAC ideas. Moreover, Simulation is conducted to evaluate the control performance of longitudinal dynamics of an aerial vehicle.

On control design for a class of parametric uncertain nonlinear systems

In this paper, a class of nonlinear systems with parametric uncertainties in the control inputs are taken into consideration and several control design approaches are investigated and compared by simulation studies. As for nonlinearity, one popular method is gain-scheduling, of which the main idea is to linearize the system at many operation points and then adopt linear control design. However, it could not achieve good performance in the presence of uncertainties. Augmenting the gain-scheduling controller with adaptive control law may improve the closed-loop dynamics, but another disadvantage is the large quantities of data processing in advance resulting from linearization which would be impossibly addressed once we take the gain-scheduling as control strategy. Therefore, nonlinear control technique of less dependence on the mathematical model is our best choice. ADRC (active disturbance rejection control) benefits from its ESO (extended state observer) to cope with the disturbance and uncertainties. Moreover, the nonlinear feedback control based on ESO upgrades the performance of the closed-loop system. Simulations are conducted to validate the effectiveness of ADRC, and comparison is carried out to figure out advantages and disadvantages for each control law.