Identification of Selected Tree Species Using UAV Multispectral Image and Machine Learning Techniques

Tree species identification is crucial for maintaining biodiversity and ecosystem health, supporting agroforestry management, conservation and climate change studies. The present study evaluates the effectiveness of Object-Based Image Analysis (OBIA) and Support Vector Machine (SVM) for classifying selected tree plantations using very highresolution multispectral UAV imagery. Traditional pixelbased methods are not suitable for classification of very highresolution images due to significant pixel heterogeneity and spectral similarity. Hence, OBIA which incorporates spectral and spatial features was employed to address these challenges. Image objects generated through mean shift segmentation were utilized as training samples for classification of arecanut, bamboo, date palm and jackfruit using SVM algorithm. Addition of Normalized Difference Red Edge (NDRE) index layer enhanced classification accuracy by detecting subtle spectral responses. Overall classification accuracy was estimated as 86.8 percent with a kappa coefficient of 0.813, highlighting the potential of integrating these advanced techniques for accurate tree identification. Future scope includes utilization of LiDAR data and Deep Learning techniques for achieving better accuracy towards above ground biomass assessment.