Recent object detection models have achieved satisfactory performance by deep learning with large-scale annotated datasets. However, these models often perform poorly when the training examples are not sufficient enou...
详细信息
ISBN:
(纸本)9781728137933
Recent object detection models have achieved satisfactory performance by deep learning with large-scale annotated datasets. However, these models often perform poorly when the training examples are not sufficient enough, especially for one-shot object detection where there is only one training example of each target class. In this paper, we propose a network parameter generation framework for effective one-shot object detection. By leveraging the semantic relationship between the source-domain and the target-domain (new classes), our framework can generate network parameters for the new classes directly. Furthermore, we develop two implementation schemes for this framework: weighting-based parameter generation and regression-based parameter generation. These two schemes use different information to construct the relationship and transfer knowledge from the source-domain to the target-domain. Experiments on the PASCAL-OSCD benchmark show that our methods can significantly boost the performance of one-shot object detection.
The study shows that the leakage of the expansion joints in the underground heating primary pipe network will lead to the fluctuation of soil temperature and conductivity around the leakage point. Based on the online ...
详细信息
In this paper, we study two-player evolutionary prisoner's dilemma on regular graphs and identify the stochastically stable equilibria for infinite populations. We consider four different update rules: Birth-death...
详细信息
Cooperative monitoring targets of mobile robots is of great importance in military, civil, and medical applications. In order to achieve multi-robot coordinated monitoring, this paper proposes a new distributed path p...
详细信息
For distributed generation(DG) network, it is important to estimate the real-time states. The information-centric networking(ICN) is established to take charge of the communication of DG network. However, the assumpti...
详细信息
ISBN:
(纸本)9781538629185
For distributed generation(DG) network, it is important to estimate the real-time states. The information-centric networking(ICN) is established to take charge of the communication of DG network. However, the assumption of ideal communication between sensors and the estimation center cannot be guaranteed due to the communication constraints of ICN with the increasing DG network. A conventional algorithm, which reduces the communication burden in ICN, is to drop the observation of each smart grid in a random way. However, the accuracy of this algorithm recession decays rapidly with the increasing drop rate. To guarantee an appropriate estimation accuracy when the drop rate increases, this paper introduces the event-trigger strategy into the estimation algorithm. An event-trigger extended kalman filter(ET-EKF) is established in this paper to adapt the nonlinearity of DG system. ET-EKF reduces the communication burden and achieves an appropriate estimation accuracy at the same time. Finally, its feasibility and performance are demonstrated using the standard IEEE 39-bus system with phasor measurement units(PMUs).
There are usually large uncertainties in the aerodynamic parameters of Mars entry vehicles,which inevitably degrades the performance of entry guidance *** paper presents a backstepping sliding mode controller for drag...
详细信息
ISBN:
(纸本)9781538629185
There are usually large uncertainties in the aerodynamic parameters of Mars entry vehicles,which inevitably degrades the performance of entry guidance *** paper presents a backstepping sliding mode controller for drag tracking in Mars entry guidance,which can guarantee the tracking error with finite-time ***,the drag can track the reference trajectory under the virtual control law by introducing a sliding manifold,and the extended state observer(ESO) is used to estimate and counteract the ***,the finite-time convergence of drag tracking error can be achieved under the proposed backstepping sliding mode guidance *** addition,the rigorous proof of tracking error convergence is ***,numerical simulation results are presented to demonstrate the effectiveness of the proposed guidance law.
In this paper, an adaptive optimal control scheme based on parameter estimation is proposed for optimal control problem of servomechanisms with unknown parameters. Parameters of the servomechanism dynamics are identif...
详细信息
Wireless networked controlsystems have gained significant popularity due to commissioning and maintenance *** 802.15.4 standard is one of wireless communication protocol for low delay deterministic *** automation,not...
详细信息
ISBN:
(纸本)9781538629185
Wireless networked controlsystems have gained significant popularity due to commissioning and maintenance *** 802.15.4 standard is one of wireless communication protocol for low delay deterministic *** automation,not only needs low delay but also requires the network that should has the capacity of various delay *** this paper,we propose a scheme,satisfying the muli-delay network in the networked control *** are using TI SmartRF04 EB evaluation boards with CC2530 SoC and TIMAC software stack CC2530 boards,as well as dc servo motors for a speed tracking control problem.
Traditional forest fire detection methods based on flame are subject to detection delays due to the later appearrance of flame than smoke. When the flame is detected, the fire might have already been spread and out of...
详细信息
Traditional forest fire detection methods based on flame are subject to detection delays due to the later appearrance of flame than smoke. When the flame is detected, the fire might have already been spread and out of ***, the objective of this paper is to study new fire detection method based on smoke. First a quadrotor unmanned aerial vehicle(UAV) is adopted to capture the fire video;then the collected video sequence will be processed by a series of preprocessing including filtering and image enhancement;finally a new significance detection algorithm is investigated for early and efficient smoke detection. Experimental results show that the algorithm can detect smoke effectively.
To make robots walk smoothly and stably,it is necessary to solve ground impacts during dynamic *** motion control can help to reduce the impacts and improve the robot's adaptability to complex *** control is a use...
详细信息
ISBN:
(纸本)9781509046584
To make robots walk smoothly and stably,it is necessary to solve ground impacts during dynamic *** motion control can help to reduce the impacts and improve the robot's adaptability to complex *** control is a useful framework to provide robot's leg *** conventional impedance control employs second-order mass-damping-spring system as a force disturbance *** this paper,position-based impedance control is applied to legged locomotion of the wheel-legged robot and a novel impedance controller with fractional order form is put forward to increase the system degrees of freedom for *** simulation tests of electric cylinder actuator and leg systems are carried out to demonstrate the effectiveness of the proposed algorithm.
暂无评论