小光斑ALS全波形数据处理技术研究
发布时间：2018-08-12 18:07
【摘要】：机载激光雷达(Airborne Laser Scanning,ALS),是综合了多个子系统于一体的主动式遥感器,具有实时快速、精度高、穿透性强等优点。其中小光斑全波形系统照射到地面的光斑较小(光斑直径在1m以内),相比于传统的大光斑全波形系统采集地物信息更为精细、精度更高,在农业、林业、电力、军事等领域具有较大应用价值。由于其较为精细的采集特性,小光斑ALS全波形数据中存在较多地物回波叠加,同时又存在多种复杂噪声的影响,因此对小光斑ALS全波形数据处理方法的要求较高,通常包括预处理、波形分解、组分信息(三维点云、强度、波宽等)解算等步骤,其中难点在于预处理和波形分解。在目前广泛使用的小光斑ALS全波形数据处理方法(高斯分解法和反卷积法)中,都存在预处理时去噪效果与波形特征保留平衡度不高,波形分解效果不佳的问题。本文通过对小光斑ALS全波形数据处理技术的研究,在波形数据处理中选用波形分解能力较强的反卷积法。其中,在稳定性、边缘探测能力和对低信噪比数据的处理能力方面表现较好的是基于RL算法的反卷积方法(后文简称RL反卷积法),但这种方法仍存在着诸如收敛速度慢、噪声放大等问题;并且随着迭代次数的增加,波宽增大的波形在分解过程中容易出现虚假波峰,从而影响波形分解准确度。针对存在的问题,本文在预处理和波形分解方面进行了一些改进:(1)在预处理中,使用小波阈值去噪,并针对小光斑ALS全波形数据处理实际,对去噪参数进行优选,实现了在取得较好去噪效果,降低噪声对后期波形分解影响的同时,更多地保留波形特征;(2)为降低波宽对RL算法分解波形数据准确度的影响,本文通过在收敛曲线上设置特定截止变化率的方法,实现了对RL反卷积法迭代次数的有效控制;(3)为提高RL算法的收敛速度,本文在提高点扩散函数构造质量的同时,引入基于二阶矢量外推的加速RL算法,使RL算法分解小光斑ALS全波形数据的速度提高了近六倍。最后,从小光斑ALS全波形数据处理效果和目标提取应用两个方面,验证了本文对RL反卷积法的改进效果。
[Abstract]:Airborne lidar (Airborne Laser Scannings-ALS) is an active remote sensor which integrates many subsystems. It has the advantages of fast real-time, high precision, strong penetration and so on. The whole waveform system of small spot is smaller (the diameter of light spot is less than 1 m). Compared with the traditional full-waveform system of large light spot, it is more precise and accurate. In agriculture, forestry, electric power, Military and other fields have great application value. Because of its fine collection characteristics, there are many ground objects echo superposition in the ALS full waveform data of small spot, and at the same time, there are many kinds of complex noise, so the method of processing the whole waveform data of the small spot ALS is very high. It usually includes pretreatment, waveform decomposition, component information (3D point cloud, intensity, wave width, etc.), and so on. The difficulty lies in pretreatment and waveform decomposition. In the widely used ALS full waveform data processing methods (Gao Si decomposition method and deconvolution method), there are some problems such as the poor balance of denoising effect and waveform characteristic retention, and the poor waveform decomposition effect. In this paper, the ALS full waveform data processing technique with small spot is studied, and the deconvolution method with strong waveform decomposition ability is chosen in waveform data processing. Among them, in terms of stability, edge detection ability and processing ability of low signal-to-noise ratio data, the method based on RL algorithm (RL deconvolution method) is better, but this method still has some problems such as slow convergence rate. Noise amplification and so on, and with the increase of iteration times, the wave with the increase of wave width is prone to appear false wave peaks in the decomposition process, thus affecting the waveform decomposition accuracy. Aiming at the existing problems, some improvements are made in the aspects of pretreatment and waveform decomposition. (1) in the pretreatment, wavelet threshold is used to de-noise, and the denoising parameters are optimized according to the actual data processing of ALS with small spot. In order to reduce the effect of RL algorithm on the accuracy of waveform decomposition, we can achieve better denoising effect and reduce the effect of noise on waveform decomposition, while retaining more waveform characteristics. (2) in order to reduce the influence of wave width on the accuracy of RL algorithm to decompose waveform data, In this paper, the effective control of the iteration times of RL deconvolution method is realized by setting the specific cutoff rate on the convergence curve. (3) in order to improve the convergence speed of the RL algorithm, the construction quality of the point diffusion function is improved. An accelerated RL algorithm based on second-order vector extrapolation is introduced, which improves the speed of decomposition of ALS full waveform data of small spot by nearly six times. Finally, the improvement of RL deconvolution method is verified from two aspects: the processing effect of ALS full waveform data and the application of target extraction.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN958.98
本文编号：2179875
[Abstract]:Airborne lidar (Airborne Laser Scannings-ALS) is an active remote sensor which integrates many subsystems. It has the advantages of fast real-time, high precision, strong penetration and so on. The whole waveform system of small spot is smaller (the diameter of light spot is less than 1 m). Compared with the traditional full-waveform system of large light spot, it is more precise and accurate. In agriculture, forestry, electric power, Military and other fields have great application value. Because of its fine collection characteristics, there are many ground objects echo superposition in the ALS full waveform data of small spot, and at the same time, there are many kinds of complex noise, so the method of processing the whole waveform data of the small spot ALS is very high. It usually includes pretreatment, waveform decomposition, component information (3D point cloud, intensity, wave width, etc.), and so on. The difficulty lies in pretreatment and waveform decomposition. In the widely used ALS full waveform data processing methods (Gao Si decomposition method and deconvolution method), there are some problems such as the poor balance of denoising effect and waveform characteristic retention, and the poor waveform decomposition effect. In this paper, the ALS full waveform data processing technique with small spot is studied, and the deconvolution method with strong waveform decomposition ability is chosen in waveform data processing. Among them, in terms of stability, edge detection ability and processing ability of low signal-to-noise ratio data, the method based on RL algorithm (RL deconvolution method) is better, but this method still has some problems such as slow convergence rate. Noise amplification and so on, and with the increase of iteration times, the wave with the increase of wave width is prone to appear false wave peaks in the decomposition process, thus affecting the waveform decomposition accuracy. Aiming at the existing problems, some improvements are made in the aspects of pretreatment and waveform decomposition. (1) in the pretreatment, wavelet threshold is used to de-noise, and the denoising parameters are optimized according to the actual data processing of ALS with small spot. In order to reduce the effect of RL algorithm on the accuracy of waveform decomposition, we can achieve better denoising effect and reduce the effect of noise on waveform decomposition, while retaining more waveform characteristics. (2) in order to reduce the influence of wave width on the accuracy of RL algorithm to decompose waveform data, In this paper, the effective control of the iteration times of RL deconvolution method is realized by setting the specific cutoff rate on the convergence curve. (3) in order to improve the convergence speed of the RL algorithm, the construction quality of the point diffusion function is improved. An accelerated RL algorithm based on second-order vector extrapolation is introduced, which improves the speed of decomposition of ALS full waveform data of small spot by nearly six times. Finally, the improvement of RL deconvolution method is verified from two aspects: the processing effect of ALS full waveform data and the application of target extraction.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN958.98
【参考文献】
相关期刊论文 前2条
1 黄涛;胡以华;;遮蔽目标的激光雷达回波信息处理[J];光电技术应用;2011年01期
2 赖旭东;秦楠楠;韩晓爽;王俊宏;侯文广;;一种迭代的小光斑LiDAR波形分解方法[J];红外与毫米波学报;2013年04期
，本文编号：2179875
本文链接：https://www.wllwen.com/kejilunwen/wltx/2179875.html
