Alpine Wetlands Information Extraction Using Optimized Multifeatures and Random Forest Algorithm

Abstract

Alpine wetlands are a special ecosystem that is extremely sensitive to global climate change. The unique geographical location and climatic conditions of the Yellow-River-Source National Park give rise to diverse wetland types at the land-water interface, which exhibit phenomena such as “same object with different spectra” and “different objects with the same spectrum.” These complexities pose significant challenges in accurately extracting information from alpine wetlands in the study area. To address these challenges, this study proposes a novel integrated multialgorithm feature optimization model that combines three filtering algorithms with a random forest (RF) algorithm for classifying alpine wetland information. First, the rich feature information in the image is initially filtered using a fusion of the three algorithms. Then, the RF algorithm is applied to optimize the filtered features. Finally, the RF classification model is used to refine wetland extraction based on this optimized feature set. The results show that 1) the fused filtering algorithm demonstrates higher stability than each individual algorithm and takes into consideration the strengths and weaknesses of each individual algorithm; 2) the classification accuracy of the RF algorithm reaches its highest value when the number of feature variables is 21; 3) the optimal classification of alpine wetlands is achieved using the RF classification model based on the best set of feature variables, resulting in an overall accuracy of 93.32% and a Kappa coefficient of 91.65% . Compared to existing land cover datasets, the proposed method provides a more detailed classification of alpine wetlands.