Deep Learning Approach for Seagrass Species Classification

Integrating deep learning algorithms with image-based data presents considerable potential for improving classification tasks, specifically in the context of seagrass monitoring and conservation initiatives. As exceptionally productive ecosystems, seagrasses encounter significant risks from climatic fluctuations and human activities. Furthermore, careful monitoring and conservation approaches are imperative. However, the intricate characteristics of seagrass species present obstacles to developing robust classification models. This research presents a novel methodology for enhancing the identification of seagrass species by curating an extensive image dataset that includes temporal variations, a wide range of geographic locations, water depths, and turbidity levels. A deep learning approach based on differentiable architecture search with early stopping was used to train a neural network. The results showed an impressive 93.72% in classifying five seagrass species sampled in Calatagan, Batangas and Puerto Galera, Oriental Mindoro, Philippines namely, Cymodocea rotundata, Enhalus acoroides, Halophila ovalis, Syringodium isoetifolium, and Thalassia hemprichii.