Background UAV remote sensing technology has shown unique advantages in vegetation coverage extraction. However, in arid desert areas with low vegetation density, the spectral characteristics of vegetation are relatively weak. The application of UAV images to extract vegetation coverage information in arid desert areas has issues such as difficulty in selecting thresholds and unsatisfactory extraction results. This work aims to quickly and accurately obtain vegetation coverage information in arid desert areas.

Methods Haibowan district, Wuhai city, Inner Mongolia was selected as the research area. Based on the visible light images of different site conditions near Table Mountain collected by UAV (Phantom 4 Advanced), the segmentation threshold was determined by the combination of supervised classification and vegetation index statistical histogram, and the vegetation coverage was extracted. The vegetation information obtained by manual visual interpretation was used as the true value for accuracy verification.

Results 1) When the visible light vegetation index was used to extract the vegetation information in the study area, the pixel statistical histogram showed a single peak feature. There was a large overlap between the vegetation and the bare land pixel values in the grayscale images of the normalized green-red difference index, the red-green ratio index, and the vegetation index. 2) The combination of supervised classification and vegetation index statistical histogram could improve the problem that the vegetation index histogram method was difficult to present bimodal characteristics when applied to vegetation extraction in arid areas. 3) Through the verification of the extraction accuracy of each vegetation index, the difference enhanced vegetation index was better than other vegetation indexes in the study area, the average accuracy could reach 93.09%, and the threshold stability was better. The results showed that the segmentation threshold determined by this method had good stability and could be applied to the extraction of vegetation information in arid desert areas. 4) The applicability test of the method of combining supervised classification with vegetation index statistical histogram to determine the segmentation threshold showed that the segmentation threshold determined by this method had good stability and could be applied to the extraction of vegetation information in arid desert areas.

Conclusions The method of combining supervised classification with vegetation index histogram to determine the threshold has a good effect on extracting vegetation information in arid desert area. It can quickly and accurately obtain vegetation coverage information in arid desert area and has practical significance for the application of visible light image vegetation information extraction in arid desert area.