Multifractal detrended fluctuation analysis parallel optimization strategy based on openMP for image processing

Abstract

In the past few years, multifractal detrended fluctuation analysis (MF-DFA) method has been widely applied in the field of agricultural image processing. However, the agricultural image feature MF-DFA analyses involves a great deal of iterative processes and complex matrix operations, which require massive computation and processing time. In order to reduce processing time and improve analysis efficiency, we first develop a MF-DFA program that involves image preprocessing, image segmentation, local area accumulation matrix calculation, local area trend fitting, local area trend elimination, a global qth-order fluctuation function, and the Hurst index. Then, we analyze and compare MF-DFA each modules’ performance characteristics and explore its parallelism according to various segmentation scales s. Lastly, we propose a parallel optimization scheme based on OpenMP for the MF-DFA. The results of our rigorous performance evaluation clearly demonstrate that our proposed parallel optimization scheme can efficiently use multicore capability to extract rape leaf image texture characteristics.