Detecção de estradas florestais usando dados LiDAR
Nenhuma Miniatura disponível
Data
2025-07-30
Autores
Brisson, Estefany Vaz
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Federal do Espírito Santo
Resumo
Knowledge of forest road networks is essential for sustainable forest management, including silvicultural operations, harvesting, transportation, and firefighting. Mapping roads in dense forest areas using costly field methods or low-spatial-resolution sensing can be inaccurate, especially on narrow roads or under canopy cover. Light Detection and Ranging (LiDAR) technology has a high penetration capacity in difficult-to-access forest environments with dense vegetation, thus presenting itself as a promising technology to support forest road detection. Therefore, the objective of this study was to detect different types of forest roads in commercial pine plantations using data from airborne LiDAR sensors. The research was structured in two sections: in the first section, an accurate and representative Digital Terrain Model for dense forests was developed based on LiDAR sensor data; in the second section, a classification of covered and uncovered roads was conducted using the automatic image classification method Random Forest in the areas of interest. The area comprises approximately 800 ha of forest aged 7 to 16 years and approximately 4,143 km of roads located in northern Spain. Image segmentation techniques and the Random Forest automatic classifier were used to map the covered and uncovered forest roads. The input variables for the classification were based on height and intensity values generated by the LiDAR point cloud. The classification process achieved an overall accuracy of 97%, resulting in LiDAR-identified forest roads for 86% of the reference roads (field survey). The height variable stood out in the identification of Exposed Roads (ER), and the combination of height and intensity variables stood out in the identification of Covered Roads (CR). The quality metrics composed of completeness, correctness, and quality individually obtained values for ER (74%, 76%, and 60%) and for CR (86%, 66%, and 60%). Although the highest integrity (completeness) was observed for covered roads, the lowest errors were observed in identifying exposed roads. However, due to the road's length being 1,903 km longer than that of the covered road, the error distribution was disproportionate. The proposed method can provide accurate road mapping to support forest management, although improvements are needed in identifying roads below the canopy, which can be improved by adding more detailed site features
Descrição
Palavras-chave
Sensoriamento remoto , Estradas florestais , Segmentação de imagem , LiDAR , Remote sensing , Forest roads , Image segmentation