论文部分内容阅读
激光成像雷达距离像与目标表面物理结构特性密切相关,体现目标的本质特征,是目标识别的主要研究方向。采用组合矩的神经网络方法进行了相干激光雷达距离像目标识别仿真研究。用Hu不变矩和仿射不变矩两者的低阶矩组合表示距离像目标区域特征,利用反向传播(BP)神经网络识别不同方位角的车辆。当视场角不变时,训练10个目标,每个目标取3～19个样本,在不同载噪比(CNR)情况下,分析Hu不变矩、仿射不变矩和两者组合矩的识别率。理论分析和仿真实验表明利用组合不变矩进行距离像目标识别性能优于单独利用其中一种不变矩。
Laser imaging radar distance image is closely related to the physical structure of the target surface, reflecting the essential characteristics of the target, which is the main research direction of target recognition. The method of combined moments of neural network was used to simulate the distance target imaging of coherent laser radar. The combination of lower moment of Hu moment invariants and affine moment invariants is used to represent the characteristics of the target image in range and the backpropagation (BP) neural network is used to identify vehicles of different azimuths. When the angle of view is invariant, 10 targets are trained and 3 ~ 19 samples are taken for each target. Under different CNR conditions, Hu moments, affine moment invariants and their combined moment Recognition rate. Theoretical analysis and simulation results show that the performance of distance target recognition using combination moment invariants is better than using one kind of moment invariants alone.