Tree Species Identification using RGB Time Series and Multispectral Images Obtained from UAV
Mojdeh Miraki
1
,
هرمز سهرابی
2
1 - Ph.D. Graduated of Forest Management, Faculty of Natural Resources, Tarbiat Modares University, Tehran, Iran.
2 - Associate Professor of Department of Forestry, Faculty of Natural Resources, Tarbiat Modares University, Tehran, Iran.
Keywords: RGB Images, Classification, Random Forest Algorithm, Phenology, multispectral images,
Abstract :
Detailed information on forest combination is required for many environmental, monitoring, and forest protection purposes. The link between ecology and remote sensing provides valuable information for the study of forest trees to facilitate the study of ecosystem performance and to measure the spatial distribution of vegetation. In recent years, the use of modern remote sensing methods and techniques based on UAVs have been used for regular updating of forest inventory. In this research, different data sources including multi-spectral and RGB images with very high spatial resolution, were used for tree species recognition in plain forests of Noor City located in Mazandaran province. Also, taking images was performed in the growing season to prepare a time series of UAV-RGB images for investigating the effect of tree crown phonological changes on classification accuracy. Following orthomosaic generation, RGB (NGB, NRB) and multi-spectral (NDVI, CIgreen) indices were calculated and the random forest classification method was used for forest species classification. Based on single-time images, late April images provided the highest overall accuracy (75%). However, the results of the time series obtained from RGB images showed an increase in accuracy of up to 86%. Species identification based on multispectral images obtained from the Sequoia sensor also provided 85% accuracy. The results showed that the single-time image at the appropriate time using a UAV-RGB, compared to taking a time series and using a UAV equipped with multispectral sensors, has acceptable and less expensive results for tree recognition in the study area.
_||_
