Automatic detection of tree health using multispectral imagery and lidar data acquired by unmanned aerial vehicles

In this study, tree health was automatically detected by using multispectral images and LiDAR point cloud data together. A new window-based method, which can be adapted to the morphological characteristics of the target objects in classification studies, was proposed. As the dataset, airborne LiDAR data and multispectral images of a Scots pine forest area located in Kozağaç village, Şiran district, Gümüşhane province, were used. The proposed approach performs classification by simultaneously utilizing spectral, spatial, and morphological features within dynamically generated analysis windows for each tree crown area. In this context, spectral analysis methods such as the Normalized Difference Vegetation Index (NDVI) and the Normalized Difference Red Edge Index (NDRE), as well as tree crown width, crown area metrics, and spatial attributes were evaluated together. Within the proposed workflow, a fully automated process in the MATLAB environment was used, where individual trees were detected using the Canopy Height Model (CHM), and appropriate regions of interest (ROI) windows were generated by determining the relevant morphological metrics for each tree. Spectral analyses were applied within the obtained ROI windows, and object-focused classification was carried out. According to the object-based accuracy analysis, the highest classification accuracy was calculated as an average of 80% for the NDVI analysis and 65% for the NDRE analysis.