In 3D frequency domain seismic forward and inversion calculation,the huge amount of calculation and storage is one of the main factors that restrict the processing speed and calculation efficiency.The frequency domain...In 3D frequency domain seismic forward and inversion calculation,the huge amount of calculation and storage is one of the main factors that restrict the processing speed and calculation efficiency.The frequency domain finite-difference forward simulation algorithm based on the acoustic wave equation establishes a large bandwidth complex matrix according to the discretized acoustic wave equation,and then the frequency domain wave field value is obtained by solving the matrix equation.In this study,the predecessor's optimized five-point method is extended to a 3D seven-point finite-difference scheme,and then a perfectly matched layer absorbing boundary condition(PML)is added to establish the corresponding matrix equation.In order to solve the complex matrix,we transform it to the equivalent real number domain to expand the solvable range of the matrix,and establish two objective functions to transform the matrix solving problem into an optimization problem that can be solved using gradient methods,and then use conjugate gradient algorithm to solve the problem.Previous studies have shown that in the conjugate gradient algorithm,the product of the matrix and the vector is the main factor that affects the calculation efficiency.Therefore,this study proposes a method that transform bandwidth matrix and vector product problem into some equivalent vector and vector product algorithm,thereby reducing the amount of calculation and storage.展开更多
鉴于现有的M ean Sh ift跟踪方法都是使用单一半径参数来描述目标大小变化,且每个目标仅有位置和尺寸两个自由度,因而不能适应复杂的目标运动情况。针对该问题,首先提出了一种新的M ean Sh ift跟踪方法,由于该方法是通过引入带宽矩阵来...鉴于现有的M ean Sh ift跟踪方法都是使用单一半径参数来描述目标大小变化,且每个目标仅有位置和尺寸两个自由度,因而不能适应复杂的目标运动情况。针对该问题,首先提出了一种新的M ean Sh ift跟踪方法,由于该方法是通过引入带宽矩阵来描述目标尺寸,因此能够在水平和垂直两个方向上独立描述目标的大小变化,并通过加入目标倾角,使得目标旋转运动得以很好描述;然后借鉴了三步搜索的思想,提出了一种快速搜索策略,以解决目标遮挡问题。实验表明,该算法能够准确跟踪序列图像中的任意复杂运动,尤其对目标的缩放、旋转运动以及遮挡有良好的适应性。展开更多
基金supported by the National Natural Science Foundation of China(Project U1901602&41790465)Key Special Project for Introduced Talents Team of Southern Marine Science and Engineering Guangdong Laboratory(Guangzhou)(GML2019ZD0203)+2 种基金Shenzhen Key Laboratory of Deep Offshore Oil and Gas Exploration Technology(Grant No.ZDSYS20190902093007855)Shenzhen Science and Technology Program(Grant No.KQTD20170810111725321)the leading talents of Guangdong province program(Grant No.2016LJ06N652).
文摘In 3D frequency domain seismic forward and inversion calculation,the huge amount of calculation and storage is one of the main factors that restrict the processing speed and calculation efficiency.The frequency domain finite-difference forward simulation algorithm based on the acoustic wave equation establishes a large bandwidth complex matrix according to the discretized acoustic wave equation,and then the frequency domain wave field value is obtained by solving the matrix equation.In this study,the predecessor's optimized five-point method is extended to a 3D seven-point finite-difference scheme,and then a perfectly matched layer absorbing boundary condition(PML)is added to establish the corresponding matrix equation.In order to solve the complex matrix,we transform it to the equivalent real number domain to expand the solvable range of the matrix,and establish two objective functions to transform the matrix solving problem into an optimization problem that can be solved using gradient methods,and then use conjugate gradient algorithm to solve the problem.Previous studies have shown that in the conjugate gradient algorithm,the product of the matrix and the vector is the main factor that affects the calculation efficiency.Therefore,this study proposes a method that transform bandwidth matrix and vector product problem into some equivalent vector and vector product algorithm,thereby reducing the amount of calculation and storage.
文摘鉴于现有的M ean Sh ift跟踪方法都是使用单一半径参数来描述目标大小变化,且每个目标仅有位置和尺寸两个自由度,因而不能适应复杂的目标运动情况。针对该问题,首先提出了一种新的M ean Sh ift跟踪方法,由于该方法是通过引入带宽矩阵来描述目标尺寸,因此能够在水平和垂直两个方向上独立描述目标的大小变化,并通过加入目标倾角,使得目标旋转运动得以很好描述;然后借鉴了三步搜索的思想,提出了一种快速搜索策略,以解决目标遮挡问题。实验表明,该算法能够准确跟踪序列图像中的任意复杂运动,尤其对目标的缩放、旋转运动以及遮挡有良好的适应性。
