Application of deep learning with stratified K-fold for vegetation species discrimation in a protected mountainous region using Sentinel-2 image

Understanding the spatial distribution of vegetation species is essential to gain knowledge on the recovery process of an ecosystem. Few studies have used deep learning and machine learning models for image processing focusing on forest/crop classification. This study, therefore, makes use of a multi-layer perceptron (MLP) deep neural network to discriminate grass species in a mountainous region using Sentinel-2 images. Vegetation indices, Sentinel-1 and ASTER DEM were combined with Sentinel-2 images to improve classification accuracy. Stratified K-fold was used to ensure balanced training and test data. The results, when compared with other commonly used machine learning models, outperformed them all. It produced a better discriminate of the grass species when ASTER DEM was combined with Sentinel-2 images, with overall F1 score of 92%. The results of the species discrimination show a general increase in increaser II species such as Eragrostis curvula and a decrease in decreaser species like Phragmites australis.

Automated summary

Understanding the spatial distribution of vegetation species is crucial for ecosystem recovery. This study utilized deep learning and machine learning models to differentiate grass species in a mountainous region, achieving better results than other models by combining multiple data sources.