Semiarid mangrove species classification using machine learning algorithms and visible UAV data

Abstract

Remote sensing studies have emerged as a crucial tool for monitoring and managing mangrove forests, with Unmanned Aerial Vehicles (UAV) being a popular platform for data collection. UAV surveys offer a non-invasive and efficient means of gathering high-resolution data on mangrove forests, enabling detailed analysis of their structure and health. However, the cost of UAV platforms and software can be a major drawback, particularly for small-scale projects. Despite these difficulties, this research provides valuable insights as it aimed to assess seasonal variations in mangrove forest vegetation at a local scale during the winter, dry, and rainy seasons utilizing solely visible data from a conventional UAV camera and processed through three machine learning algorithms and an 11×11 Lee Sigma speckle filter. The Support Vector Machine (SVM) algorithm outperformed the Random Forest (RF) and Neural Network (NN) algorithms in terms of overall accuracy during the dry season. Specifically, the SVM algorithm achieved an accuracy of 85 % for the dry, 61 % for the winter, and 57 % for the rainy season. In comparison, the RF algorithm achieved accuracies of 80 % for the dry, 77 % for the rainy, and 63 % for the winter season, and the NN algorithm recorded accuracies of 71 % for the dry, 65 % for the rainy, and 57 % for the winter season. These findings indicate that the SVM algorithm is a more effective approach for the dry season. Importantly, this study demonstrates the potential of machine learning algorithms in classifying mangrove species using visible UAV data, which could substantially enhance the efficiency and cost-effectiveness of mangrove forest management.