红外激光遥感图像散斑特征提取  

Speckle feature extraction from infrared laser remote sensing image

作　　者：周贵华 周伟昌 ZHOU Guihua;ZHOU Weichang(Songtian College,Guangzhou University.Guangdong 511370,China;School of Physics and Electronics,Hunan Normal University,Hunan 410000,China)

机构地区：[1]广州大学松田学院,广东511370 [2]湖南师范大学物理与电子科学学院,湖南410000

出　　处：《激光杂志》2020年第1期108-112,共5页Laser Journal

基　　金：国家自然科学基金(No.51102091)

摘　　要：当前方法对红外激光遥感图像进行散斑特征提取时,采用2D(二维)图像作为数据采集手段,存在散斑特征提取区域误差大的问题。据此,提出基于深度学习的红外激光遥感图像散斑特征提取方法。采用无人机配置自动旋转云台上搭配2D激光测距传感器对区域红外激光数据进行采集、标定和利用激光里程计算法对区域内特征值提取和匹配。结合线性插值法对应的周期内接收的激光数据进行区域转换,利用L-M算法解决红外激光遥感图像范围的预计问题,采用信赖区域法迭代出最优区域范围,将激光数据转化到世界坐标系下的点云中,通过点云叠加完成区域的散斑特征提取。仿真实验结果表明,所提方法在进行红外激光遥感图像散斑特征提取时水平和区域深度定位误差都能控制在±1 m之内,符合分割要求。与其他方法相比,该方法的提取准率95%~99%的范围内变化,平均耗时为0. 45 s,说明该方法提取精度更高,提取速度更快。In the current method of speckle feature extraction from infrared laser remote sensing image,2D( twodimensional) image is used as data acquisition method,which has the problem of large regional error in speckle feature extraction. Based on this,a speckle feature extraction method for infrared laser remote sensing image based on depth learning is proposed. The infrared laser data are collected and calibrated by UAV equipped with automatic rotating platform and 2D laser ranging sensor. Laser mileage calculation method is used to extract and match the eigenvalues in the region. Combining the laser data received in the corresponding period of the linear interpolation method,the region conversion is carried out. The L-M algorithm is used to solve the prediction problem of the infrared laser remote sensing image range. The trust region method is used to iterate optimal region range. Laser data is transformed into point clouds in the world coordinate system,and the speckle feature of the region is extracted by point cloud superposition. Simulation results show that both horizontal and regional depth positioning errors of the proposed method can be controlled within ± 1 m when extracting speckle features of infrared laser remote sensing images,which meets the requirements of segmentation. Compared with other methods,the extraction accuracy of this method changes within the range of 95-99%,and the average time is 0.45 s,indicating that the extraction accuracy of this method is higher and the extraction speed is faster.

关 键 词：红外激光遥感图像分割 散斑特征提取 深度计算 信息迭代 区域转换 精度 

分 类 号：TN751[电子电信—电路与系统]