Extraction of Morphometric Features the Shape of Mangrove Leaves Based on Digital Images and Classification Using the Support Vector Machine

At present, several botanists still rely on manual estimating methods to assess the carbon content in mangroves. However, these methods have been reported to be extremely time-consuming, showing the need to develop a system for prediction. An effective solution lies in the creation of an artificial intelligence application that can provide rapid and cost-effective results. In constructing this application, careful consideration must be given to the selection of parameters or attributes. Species are essential parameters for the assessment of carbon content, but their determination has proven to be challenging due to the similarities between mangroves. The occurrence of errors in identifying species can lead to inaccurate prediction in a given tree. To address this challenge, the identification process can be greatly improved by leveraging plant morphology, particularly leaf. Previous reports have shown that leaf exhibits distinctive morphological features, and the application of geometric mathematics proved instrumental in extracting these characteristics. Therefore, this study aimed to extract the shape of mangrove leaf images using morphometric features. Based on the features obtained, a classification was performed to identify mangrove species using a machine learning algorithm, Support Vector Machine (SVM). SVM is able to solve high-dimensional problems, apply Structural Risk Minimization (SRM) strategies, and has a theoretical basis that can be analyzed clearly, so it is possible to use it as an innovative approach to identify mangrove species. The results showed that the geometric method was effective in extracting values for roundness, solidity, eccentricity, convexity, compactness, elongation, rectangularity, and aspect ratio. The analysis of each feature showed that the roundness feature could be used to distinguish the 4 mangrove species effectively. Furthermore, the classification results using SVM obtained the highest average accuracy of 91.26%