IEEE Geoscience and Remote Sensing Letters | 2019
Histogram-Based Autoadaptive Filter for Destriping NDVI Imagery Acquired by UAV-Loaded Multispectral Camera
Abstract
Combinations of unmanned aerial vehicles (UAVs) and multispectral sensors provide low-cost approaches for detailed spatiotemporal vegetation studies. However, the resulting vegetation index images such as normalized difference vegetation index (NDVI) have unignorable stripe noise and seriously disturb the extraction of vegetation information. In this letter, the similar frequency phenomena caused by stripe noise were observed in the gray-scale histogram and the Fourier spectrum of a striped NDVI image. Thus, we tried establishing the empirical quantitative relationship between them, and then designed an autoadaptive filter for stripe noise removal in Fourier spectrum according to the characteristics including the dominant peak and troughs of the histogram curve of the raw NDVI image. Applying this autoadaptive filter to the corresponding Fourier spectrum image, we achieved automatic and effective stripe noise removal without any manual interference. Based on visual judgment and quantitative evaluation, the proposed autoadaptive filter demonstrated by far the better performance in retaining image fidelity and intelligibility than the improved high-pass filter and 2-D Weiner filter.