基于多核计算机的遥感影像多尺度分割

Multi-scale Segmentation of Remote Sensing Image based on Multi-core Computer

多尺度分割作为一种成熟的影像分割方法,在遥感影像信息提取中得到广泛应用,但算法整体效率较低。利用多核计算机实现了基于数据并行的遥感影像多尺度分割。传统的影像IO(In-put and Output)方法在影像数据量较大的情况下无法满足多核计算机并行处理的需要,设计了一种新的影像IO策略消除了这种缺陷;此外,在遥感影像多尺度并行分割的过程中,普遍存在分割结果无法直接进行合并的问题,利用对特定区域重分割的方法在保证效率的前提下解决了这个问题。结果表明:针对各种数据量与尺寸的遥感影像,并行分割效率有了较大提升,并且分割算法具备了处理大数据量影像的能力,极大地增强了通用性。利用多核计算机提升影像分割效率取得了显著成效。

As a mature image segmentation method, Multi-scale segmentation is used widely in remote sensing image information extraction, but the whole algorithm efficiency is low. In this paper, data paralleled segmentation of remote sensing image is achieved based on multi-core computer. The traditional image IO method can＇t satisfy the needs of parallel computing based on multi-core computer. A new image IO strategy is devised to eliminate this drawbaek; In addition,in the process of parallel segmentation,it universally exists a problem that the segmentation results can not be merged directly. The method of re-segmentation is used to solve the problem to the special areas. The experimental results show that parallel segmentation efficiency has improved significantly on images of various quantity and size. Besides, the segmentation algo- rithm has been provided with the ability of processing image of large data quantity and the applicability is greately enhanced . Image segmentation based on multi-core computer acquires remarkable effect.