Análisis de imágenes hiperespectrales capturadas por aeronaves remotamente pilotadas para la clasificación y caracterización de la vegetación en un ecosistema de páramo: estudio de caso parque ecológico Matarredonda (noviembre 2018)

Abstract

The study on the analysis of hyperspectral images captured by remotely piloted aircraft in the Matarredonda Ecological Park focused on developing a methodology for the classification and characterization of vegetation in a páramo ecosystem using advanced remote sensing. By acquiring hyperspectral images with a Headwall Nano-Hyperspec sensor mounted on an RPAS, high spectral and spatial resolution data were obtained, allowing for precise identification of the vegetation cover present in the study area. In image processing, dimensionality reduction techniques such as PCA (Principal Component Analysis), MNF (Minimum Noise Fraction), and ICA (Independent Component Analysis) were applied to optimize spectral information and reduce data redundancy. For classification, different algorithms were compared: Maximum Likelihood (ML), Support Vector Machines (SVM), and Random Forest (RF), evaluating their performance in terms of accuracy and reliability. The results showed that Random Forest achieved the highest overall accuracy, providing a detailed classification of vegetation cover with a high Kappa coefficient, while SVM exhibited high precision but with significant computational costs. Dimensionality reduction through MNF optimized classification by preserving key information without compromising accuracy. Species distribution maps were generated, highlighting the effective identification of frailejones and other typical páramo vegetation, validating the applicability of this methodology for monitoring and conserving these ecosystems. Additionally, the study demonstrated that hyperspectral imaging surpasses the limitations of multispectral sensors, enabling a more detailed analysis of vegetation. In conclusion, the combination of RPAS, hyperspectral sensors, and advanced classification algorithms represents an efficient strategy for analyzing fragile ecosystems such as the páramos, providing crucial information for environmental management and biodiversity conservation.