基于快速流式算法的局部余弦相似属性

Attribute of local cosine similarity based on a fast-streaming algorithm

相似性是一种常用的衡量不同图像之间差异程度的属性,广泛应用于地震数据处理环节.由于地震数据本质上是非平稳的,局部相似性比全局相似性更适用于刻画地震数据的时-空变化特征.现有的局部相似属性可以通过正则化最小二乘问题进行计算,但是其计算过程需要大量的计算时间和数据存储空间,难以适应当前的海量数据处理任务.本文提出了一种基于快速流式算法的局部余弦相似性计算方法,其采取局部数据递推的模式,避免迭代算法带来的计算负担,在保证计算精度的前提下能够快速地表征不同数据之间的差异.流式局部余弦相似性可以用于解决不同的地震数据处理问题,包括叠前地震数据加权叠加、多波多分量数据纵横波速度比估计以及基于构造预测的断层检测,更加适用于现阶段的宽方位角和高密度采集数据的处理流程.理论模型和实际数据测试结果可以验证流式局部余弦相似性算法的效率优势和解决不同地震数据处理问题的有效性.

Similarity is a common attribute to measure the difference between images,which is widely used in seismic data processing.As seismic data is naturally nonstationary,local similarity is more suitable than global similarity in characterizing time-space variation of seismic data.Existing local similarity can be calculated by solving regularized least-squares problems,but this process requires much computational time and storage,thus difficult to adapt to current large-scale data processing tasks.To solve this problem,this paper proposes a calculation method for the local cosine similarity attribute based on a fast-streaming algorithm.Instead of iteration,this method computes local similarity around each point recursively and quickly characterizes the difference between two datasets under the premise of accuracy.The streaming local cosine similarity can be applied to varied kinds of seismic data processing problems,adapted to the current wide-azimuth and high-density seismic workflow.The applications include weighted stacking,multi-component data registration with V_(P)/V_(S) estimation,and fault detection based on structure prediction.Tests on synthetic and field data show that the method based on streaming local similarity is effective and efficient when solving several seismic data processing problems.