时间序列植被指数频域滤波去噪算法的优化研究

Optimization of Frequency Domain Denoising Algorithms for Time-series Vegetation Index

频域低通滤波是一种比较常用的去噪方法,但在对时间序列植被指数去噪时易受异常值的影响,所以需要对传统的频率域滤波算法进行优化。本研究采用的优化思路是:首先根据植被的生长规律识别出异常值,再通过滤波后的数值代替异常值,然后迭代执行频率域滤波进行去噪,直到结果中不存在异常值为止。分别对MODIS-NDVI进行本文优化算法和传统傅里叶与小波低通滤波去噪,结果表明优化的算法能够在不借助已知图像质量评价信息的条件下,有效地避免了异常值的干扰,从而大幅度地提高了图像的质量和应用效果。

Traditional frequency domain filtering is a commonly used denoising method,which is apt to be affected by the abnormal values.This paper proposed an optimized approach：firstly,the abnormal values were detected according to the growth regulation of the plants,and the abnormal values were substitute with the filtered results;then the denoising process was implemented iteratively until there were no abnormal values in the result.The MODIS-NDVI was filtered by the optimized and the traditional Fourier and Wavelet low-pass algorithms.Result showed that the optimized algorithms could avoid the influence of abnormal values without the help of known image quality assessment information,which proved that the optimized algorithms could greatly improve the image quality and the application effects.