A Comparative Study of Neural Network Models for Mangrove Species Classification

Abstract

Mangrove ecosystems play a critical role in coastal protection, biodiversity conservation and carbon sequestration. Accurate classification of mangrove species is essential for the conservation and management of coastal ecosystems, which offer substantial ecological benefits, such as protecting shorelines and supporting diverse species. This study provides a comparative analysis of two neural network models, Learning Vector Quantization (LVQ) and Backpropagation, for the classification of mangrove species on Bintan Island. The methods used in this research starting from the acquisition of leaf and fruit images of mangroves, data preprocessing to reduce noise, feature extraction of shape and texture, to the application of Backpropagation and LVQ algorithms for classification. The findings revealed that the LVQ model achieved an accuracy of 67%, precision of 70%, recall of 67%, and an F1-score of 66%. Conversely, the Backpropagation model demonstrated superior performance with an accuracy of 88%, precision of 89%, recall of 88%, and an F1-score of 88%. These results suggest that the Backpropagation algorithm, with its multiple hidden layers and activation functions, is more adept at handling non-linear and complex data compared to LVQ.