基于背景扣除的弱吸收信息精确提取方法研究

随着半导体器件的快速发展，基于分子振-转吸收光谱的研究日益深入，以可调谐半导体激光吸收光谱（Tunable Diode Laser Absorption Spectroscopy， TDLAS）为代表的检测技术取得巨大进步，应用领域进一步扩大，已有超过一千类仪器应用于连续监测以及过程控制，每年出售的气体检测仪器占据了红外气体传感检测仪器总数的5%以上，已实现不同领域组分浓度、温度、压力等参数的高精度探测。本文针对仪器开发领域目标吸收信息被完全覆盖的复杂应用环境，利用调制技术将吸收信息转移到高频部分，经过分通道背景扣除和谐波信号归一化处理提取吸收信息。以测量含硫天然气中微量H₂S为例，天然气中CH₄含量超过90%，目标组分H₂S的吸收信息被完全覆盖，将吸收信息转移到1 kHz频率，得到的谐波信号峰值与H₂S浓度呈正相关，线性参数达到0.999 9，实现了弱吸收信号的有效提取，从方法上验证了提取弱吸收信号的有效性，进一步扩展了光学气体传感的应用领域，为仪器开发提供技术保障。     

基于混合神经网络的无线电信号自动调制识别

随着无线电信号数据海量增加，复杂电磁环境下面临着未知威胁和目标侦察识别复杂度高的问题，本文针对未知无线电信号的特征提取任务，设计了一种混合神经网络以提高目标无线电信号的识别能力。先通过胶囊神经网络对未知信号的空间信息进行提取，再进一步运用门控循环单元提取信号在时间上的特征信息。设计混合网络模型将信号的时间和空间特征相结合，提高对目标信号的分类精度。通过RML2016.04C调制信号数据集，验证了混合神经网络的识别性能。结果表明:当信噪比为6 dB时，混合网络模型对多种不同调制信号的分类精度大于95%。因此，本文所设计的混合神经网络能够有效对不同调制信号进行准确分类。     

基于点云深度学习的对称结构空间目标相对位姿测量

针对对称结构空间目标相对位姿解算过程中点云误匹配带来的误差问题，提出一种基于点云深度学习的对称结构空间目标相对位姿测量方法。首先设计空间目标点云特征提取网络及关键点回归网络，将位姿测量问题转换为空间目标点云关键点回归问题，通过两个并行的回归网络分别输出空间目标平移向量和具有固定标签的目标点云三维边界框角点；其次利用具有连续稳定标签的角点求解目标姿态，可有效解决目标的对称结构导致的点云误配准问题；最后通过仿真数据集的实验表明，该方法相比于传统的点云配准方法有更高的准确率，能够精确求解具有对称结构的空间目标相对位姿。     

顾及动态物体感知的增强型视觉SLAM系统

传统的同步定位与制图（Simultaneous localization and mapping, SLAM）系统在复杂环境下工作时，无法分辨环境中的物体是否存在运动状态，图像中运动的物体可能导致特征关联错误，引起定位的不准确和地图构建的偏差。为了提高SLAM系统在动态环境下的鲁棒性和可靠性，本文提出了一种顾及动态物体感知的增强型视觉SLAM系统。首先，使用深度学习网络对每一帧图像的动态物体进行初始检测，然后使用多视图几何方法更加精细地判断目标检测无法确定的动态物体区域。通过剔除属于动态物体上的特征跟踪点，提高系统的鲁棒性。本文方法在公共数据集TUM和KITTI上进行了测试，结果表明在动态场景中定位结果的准确度有了明显提升，尤其在高动态序列中相对于原始算法的精度提升在92%以上。与其他顾及动态场景的SLAM系统相比，本文方法在保持精度优势的同时，提高了运行结果的稳定性和时间效率。     

Comparison of residual and dense neural network approaches for building extraction from high-resolution aerial images

Applications including change detection, disaster management, and urban planning require precise building information, and therefore automatic building extraction has become a significant research topic. With the improvements in sensor and satellite technologies, more data has become available, and with the increased computational power, deep learning methods have emerged as successful tools. In this study, U-Net and FPN architectures using four different backbones (ResNet-50, ResNeXt-50, SE-ResNext-50, and DenseNet-121), and an Attention Residual U-Net approach were used for building extraction from high-resolution aerial images. Two publicly available datasets, Inria Aerial Image Labeling Dataset and Massachusetts Buildings Dataset were used to train and test the models. According to the results, Attention Residual U-Net model has the highest F1 score with 0.8154, IoU score with 0.7102, and test accuracy with 94.51% on the Inria dataset. On the Massachusetts dataset, FPN Dense-Net-121 model has the highest F1 score with 0.7565 and IoU score with 0.6188, and Attention Residual U-Net model has the highest test accuracy with 92.43%. It has been observed that, FPN with DenseNet backbone can be a better choice when working with small size datasets. On the other hand, Attention Residual U-Net approach achieved higher success when a sufficiently large dataset is provided.     

A Robotic TeleoperationSystem for Precise Robotic Manipulation by Human-Machine Interaction

Teleoperation is of great importance in the area of robotics， especially when people are unavailable in the robot workshop. It provides a way for people to control robots remotely using human intelligence. In this paper， a robotic teleoperation system for precise robotic manipulation is established. The data glove and the 7-degrees of freedom (DOFs) force feedback controller are used for the remote control interaction. The control system and the monitor system are designed for the remote precise manipulation. The monitor system contains an image acquisition system and a human-machine interaction module， and aims to simulate and detect the robot running state. Besides，a visual object tracking algorithm is developed to estimate the states of the dynamic system from noisy observations. The established robotic teleoperation systemis applied to a series of experiments， and high-precision results are obtained， showing the effectiveness of the physical system.     

Orbital error propagation considering atmospheric density uncertainty

Due to the influence of various errors, the orbital uncertainty propagation of artificial celestial objects while orbit prediction is required, especially in some applications such as conjunction analysis. In the orbital error propagation of artificial celestial objects in low Earth orbits (LEOs), atmospheric density uncertainty is one of the important factors that require special attention. In this paper, on the basis of considering the uncertainties of position and velocity, the atmospheric density uncertainty is also taken into account to further investigate the orbital error propagation of artificial celestial objects in LEOs. Artificial intelligence algorithms are introduced, the MC Dropout neural network and the heteroscedastic loss function are used to realize the correction of the empirical atmospheric density model, as well as to provide the quantification of model uncertainty and input uncertainty for the corrected atmospheric densities. It is shown that the neural network we built achieves good results in atmospheric density correction, and the uncertainty quantization obtained from the neural network is also reasonable. Moreover, using the Gaussian mixture model - unscented transform (GMM-UT) method, the atmospheric density uncertainty is taken into account in the orbital uncertainty propagation, by adding a sampled random term to the corrected atmospheric density when calculating atmospheric density. The feasibility of the GMM-UT method considering atmospheric density uncertainty is proved by the further comparison of abundant sampling points and GMM-UT results (with and without considering atmospheric density uncertainty).     

“珠海一号”高光谱数据城市不透水面提取研究

不透水面是评估城市化进程和评价城市环境的重要参考指标，研究城市不透水面的空间覆盖程度对城市生态环境保护、宜居城市建设具有重要意义。本文以“珠海一号”高光谱影像为数据源、珠海市陆域为研究区域，选取支持向量机和随机森林机器学习算法，结合光谱特征，实现城市不透水面信息提取。结果表明：使用“珠海一号”高光谱数据提取不透水面可行性较强，且支持向量机算法更适用于“珠海一号”不透水面提取，总体精度和Kappa系数分别达到92.4%、0.78；基于“珠海一号”高光谱数据进行城市不透水面提取，可为城市规划建设、分析城市土地扩张、评估城市灾害风险提供可靠理论依据和数据支撑。