The tracking problem of nonholonomic mobile robots with uncertainties is investigated in this paper. An uncertain model of the nonholonomic kinematic system is presented based on the visual feedback and the state and input transformations for a kind of mobile robots in chained form with uncertainties. Two transformations are exploited based on the idea of backstepping and the structure of tracking error system. Then, both an adaptive control law and a dynamic feedback robust controller are designed to track the desired trajectory by using Lyapunov direct method and the extended Barbalat Lemma. The asymptotic convergence of a closed-loop error system is proved rigorously. Finally, simulation results demonstrate the effectiveness of the proposed strategies.
近年来有不少基于深度学习的单幅遥感图像超分辨率重建方法,虽然在超分辨率的各项指标上都有所提升,但通过人眼观察的超分辨率效果却仍不明显.此前方法在制作低分辨率图像过程中采用的降采样会带来部分信息损失,为了避免此类问题,本文以在数据集部分通过用不同比例尺得到清晰和模糊两种分辨率的遥感图像作为训练数据集对,有效避免了由于降采样带来的原图信息损失.在主网络部分运用了一种基于深度残差块的generative adversarial network(GAN)图像超分辨率模型,生成网络部分生成超分辨率图像,对抗网络部分对生成网络生成图像的逼真程度进行判断并反馈到生成网络中.在损失函数部分,为了减少由于特征匹配错误产生的不良影响,本文在上下文损失函数(contextual loss function)的基础上添加了图像特征之间的空间位置信息,并采用相对判别器来判断图片相对真实性,优化了超分辨率效果.通过在NWPU-RESISC45数据集上进行实验验证,本文提出的方法在峰值信噪比(peak signal to noise ratio,PSNR)、结构相似性(structural similarity,SSIM)和平均梯度(average gradient,AG)三项指标上有了较明显的提升,并可通过人眼观测到网络输出良好的超分辨率效果图.
