Machine Learning-Based Coastal Terrain Classification in Tropical Regions Using Multispectral UAV Imaging: A Comparative Study of Random Forest and SVM Models
Keywords: UAV, Multispectral Imaging, SVM, Random Forest
TL;DR: SVM and RF algorithms are used to label terrain in a coastal tropical area captured by a multispectral UAV
Abstract: Advances in various technologies and machine learning (ML) are transforming the field of remote sensing. This study proposes an ML-centered methodology for classifying coastal terrain in tropical coastal regions using multispectral unmanned aerial vehicle (UAV) image inputs. The objective is to identify suitable ML algorithms for analyzing multispectral images on limited hardware. Multispectral images of the study area were collected using a DJI Mavic 3M UAV in March 2023. K-means clustering was implemented to assist in coastal terrain identification, and the labeled data were used to train pixel-based Support Vector Machine (SVM) and Random Forest (RF) models utilizing a 5-fold block cross-validation scheme. The results showed that the optimized RF model outperformed the SVM model across most metrics. Despite this, the SVM model showed potential for live image classification due to its smaller size and quick classification speed. Additionally, the optimized models effectively classified images from areas set as an independent hold-out test set, demonstrating the applicability of ML in this type of remote sensing problem.
Submission Number: 34
Loading