Mangrove Tree Species Classification Based on Leaf, Stem, and Seed Characteristics Using Convolutional Neural Networks with K-Folds Cross Validation Optimalization

Abstract

Mangrove classification plays a pivotal role in environmental monitoring and conservation efforts. In this study, our meticulously curated dataset comprised diverse mangrove tree images standardized to 250 x 250 pixels, capturing the nuances of various species. Employing advanced deep learning techniques, our models demonstrated exceptional accuracy, reaching 99.23% without K-Folds and a slightly enhanced 99.78% with K-Folds. These models exhibited outstanding consistency, showcasing recall, precision, and F1-Score metrics all surpassing 99%. Through rigorous testing in 10 experiments, both K-Folds and non-K-Folds methods consistently achieved 100% accuracy, evidenced by the presence of True Positives in every classification scenario. This remarkable performance underscores the robustness of our algorithms in precisely classifying mangrove species, offering a valuable tool for ecological research and conservation initiatives.