Application of deep learning techniques for determining the spatial extent and classification of seagrass beds, Trang, Thailand

Abstract

Abstract Few studies have investigated the long-term temporal dynamics of seagrass beds, especially in Southeast Asia. Remote sensing is one of the best methods for observing these dynamic patterns, and the advent of deep learning technology has led to recent advances in this method. This study examined the feasibility of applying image classification methods to supervised classification and deep learning methods for monitoring seagrass beds. The study site was a relatively natural seagrass bed in Hat Chao Mai National Park, Trang Province, Thailand, for which aerial photographs from the 1970s were available. Although we achieved low accuracy in differentiating among various densities of vegetation coverage, classification related to the presence of seagrass was possible with an accuracy of 80% or more using both classification methods. Automatic classification of benthic cover using deep learning provided similar or better accuracy than that of the other methods even when grayscale images were used. The results also demonstrate that it is possible to monitor the temporal dynamics of an entire seagrass area, as well as variations within sub-regions, located in close proximity to a river mouth.