云边端协作的架空线路鸟巢检测与定位算法研究

针对云端单一集中数据处理时效性低、架空线路上鸟巢检测精度不高、模型对边缘计算设备算力高消耗以及目标定位不准确的问题,提出了一种基于云边端协作的架空线路鸟巢检测与定位算法。该算法通过云、终、边缘3端的协作,解决了云端集中处理效率低的问题,并通过云边数据可视化协作解决由于角度及光线引起的图像不清晰问题。为了提高架空线路鸟巢检测的精度,该算法在YOLOv5x模型基础上进行了优化。首先,通过将主干特征提取网络中的C3模块替换为C2f模块,并在最后一层加入SE(squeeze and excitation)注意力模块,以提升模型对小目标的检测能力。其次,将激活函数替换为Mish函数,解决训练梯度饱和导致神经元停止学习的问题。为了降低模型对边缘计算设备算力的消耗,对改进后的模型进行剪枝微调以降低模型参数规模。基于此优化模型,提出了三维目标定位算法,结合GIS(geographic information system)系统对定位结果进行修正,实现了对检测目标的精准定位。实验数据显示,改进后的模型平均精度均值达到93.25%,比原YOLOv5x模型提升了3.44%,优化后的模型剪枝率达到45%。检测目标经过三维空间建模计算并通过位置修正能够定位到相应的杆塔,有效指导工作人员快速准确排除隐患。

飞行模拟器线性最优洗出算法限幅控制

为在有限的运动平台空间内提高飞行模拟器的动感逼真度,结合人体前庭系统和线性最优洗出算法,对洗出算法的缩放限幅环节进行改进。对输入信号的低幅值段采用三阶多项式法,尽可能减小人体感觉加速度的触发时延;在高幅值段利用运动感觉作为BP神经网络控制器的输入变量,在有限运动空间内自适应调节输入信号的缩放比例系数。对飞行模拟器纵向/俯仰方向进行仿真验证,仿真结果表明,在低幅值段采用三阶多项式法可以有效避免局部极值信号的畸变,同时保留三阶多项式法克服感觉加速度触发时延的优点;在高幅值段采用BP神经网络自适应调节缩比系数法能将平台限制在运动空间内,最大程度减小飞行模拟器虚假暗示的发生,且提高飞行模拟器的安全性和可靠性,证明采用分段缩放限幅策略对飞行模拟器的动感逼真度提升有促进作用。

基于空-谱特征K-means的长波红外高光谱图像分类

高光谱图像(hyper spectral imagery,HSI)分类已成为探测技术的重要研究方向之一,同时也在军事和民用领域得到广泛运用。然而,波段数目巨大、数据冗余、空间特征利用率低等因素已成为高光谱图像分类的挑战,且现有的高光谱分类大多利用可见光或短波红外高光谱数据分类。针对这些问题,本文提出了一种基于光谱和空间特征的K-means分类方法。首先提取空间特征,然后将光谱与空间特征相结合并降维,最后引入K-means算法得到较普通K-means更佳的分类结果。并将此算法运用在长波红外的高光谱图像分类中。

Development of Cueing Algorithm Based on “Closed-Loop” Control for Flight Simulator Motion System

The classical washout algorithm had fixed gains and manually constructed filters, so that it led to poor adaptability. Furthermore, it lost the sustained acceleration cues of high-and mid-frequency in cross-over(tilt-coordination) channel, and the acceleration of cross-over frequency was also limited by angular velocity limiter, so the false cues in flight simulation process were clearly perceived by pilots. The paper studied the characteristics of the classical washout algorithm and flight simulator motion platform, tried to redesign the source of cross-over acceleration channel and translation acceleration channel, and transferred the part of cross-over acceleration that was unsimulated sustained acceleration to translation acceleration channel. Comparisons were mainly made between classical washout algorithm and revised algorithm in a longitudinal/pitch direction. The evaluation was based on the implementation of human vestibular perception system. The results demonstrated that the revised algorithm could significantly reduce the phase lag, and improved the spikes tracking performance. Furthermore, sensory angular velocity and the error of sensory acceleration were strictly controlled within the threshold of human perception system, and the displacement was a little broader than the classical washout algorithm. Therefore, it was proved that the new algorithm could diminish the filters parameters and heighten the self-adaptability for the washout algorithm. In addition, the magnitude of false cues was remarkably reduced during flight simulator, and the workspace utilization of the motion platform was developed by "closed-loop" control system.

A Novel Nonlinear Scaling Method for Optimal Motion Cueing Algorithm in Flight Simulator

Motion cueing algorithm plays a key role in simulator motion reproduction and improves the realism of movement sensation by combining with the human vestibular system.It is well established that scaling&limiting should be used to decrease the amplitude of the acceleration and angular velocity signals for making full use of limited workspace of motion platform.A novel nonlinear scaling method based on a third-order polynomial and back propagation(BP)neural networks for the motion cueing algorithm is proposed in this paper.The third-order polynomial method is applied to the low amplitude segment of the input signal to minimize the trigger delay of the sensation acceleration;in the high amplitude segment,the BP neural network is used to adaptively adjust the scaling factor of the input signal,to avoid washout displacement and angular displacement beyond the boundary of the workspace.The simulation experiment is verified in the longitudinal/pitch direction for flight simulator,and the result implies that the proposed method not only can overcome the problem of constant scaling parameter and improve motion platform workspace utilization,but also reduce the false cues during the motion simulation process.

A Design of Nonlinear Scaling and Nonlinear Optimal Motion Cueing Algorithm for Pilot’s Station

Motion cueing algorithms(MCA)are often applied in the motion simulators.In this paper,a nonlinear optimal MCA,taking into account translational and rotational motions of a simulator within its physical limitation,is designed for the motion platform aiming to minimize human’s perception error in order to provide a high degree of fidelity.Indeed,the movement sensation center of most MCA is placed at the center of the upper platform,which may cause a certain error.Pilot’s station should be paid full attention to in the MCA.Apart from this,the scaling and limiting module plays an important role in optimizing the motion platform workspace and reducing false cues during motion reproduction.It should be used along within the washout filter to decrease the amplitude of the translational and rotational motion signals uniformly across all frequencies through the MCA.A nonlinear scaling method is designed to accurately duplicate motions with high realistic behavior and use the platform more efficiently without violating its physical limitations.The simulation experiment is verified in the longitudinal/pitch direction for motion simulator.The result implies that the proposed method can not only overcome the problem of the workspace limitations in the simulator motion reproduction and improve the realism of movement sensation,but also reduce the false cues to improve dynamic fidelity during the motion simulation process.