一种粮粒图像快速重构方法  

A FAST METHOD FOR IMAGE RECONSTRUCTION OF GRAIN KERNELS

作　　者：许德刚 廉飞宇[1,2,3] 

机构地区：[1]河南工业大学粮食信息处理与控制教育部重点实验室,河南郑州450001 [2]河南工业大学粮食光电探测与控制河南省重点实验室,河南郑州450001 [3]河南工业大学信息科学与工程学院,河南郑州450001

出　　处：《河南工业大学学报（自然科学版）》2017年第6期74-79,共6页Journal of Henan University of Technology：Natural Science Edition

基　　金：河南省科技发展计划项目(162102310405);河南省基础与前沿计划项目(152300410079);国家重点研发计划项目(2017YFD0401003)

摘　　要：针对粮情实时检测过程中图像清晰度下降,给后续分析造成困难的问题,提出一种新的图像重建记忆梯度追踪(MGP)算法,用于减少压缩感知方向追踪算法的重构时间并提高重构精度。该算法结合正则化正交匹配追踪(ROMP)的元胞生成方法,利用非单调非精确阿米霍线搜索方法确定迭代步长,利用MGP算法锁定搜索方向,可以在保证图像重构精度的同时,缩短重构时间。对原有MGP算法的方向参数公式进行了推导和改进,得到了效率更高的计算公式,使得算法的运行时间较共轭梯度追踪算法节省30%,并可精确重构二维粮粒图像信号。本算法的运行结果表明,在相同硬件平台下,其二维粮粒图像信号的重构性能优于其他的同类重构算法。For the problem that charity of images decreasing in the real-time monitoring of grain kernels situation,which causing difficulties for the subsequent analysis,a new image reconstruction method based on memory gradient pursuit(MGP) algorithm was proposed used to reduce the reconstruction time and improve the reconstruction precision of the compressed sensing direction tracking algorithm. The algorithm combined the cell generation method of regularized orthogonal matching pursuit,utilizing non-monotonic and un-precise Armijo linear search to determine step size,and using MGP to lock search direction,which could reduce reconstruction time under satisfactory reconstruction precision. In this study,the former formula of direction parameters for MGP was inferred and improved,and the formula with higher efficiency was afforded,which made the proposed algorithm to save time 30% compared with Conjugate gradient tracking algorithm,and could also reconstruct two-dimensional images of grian precisely. Experimental results indicated that the method proposed in the present study had better reconstruction performance of grain images than other algorithms under the same test conditions.

关 键 词：粮粒检测 压缩感知 方向追踪 记忆梯度 图像重构 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]