AUTOMATIC SEGMENTATION OF TREE CROWNS IN PINE FORESTS USING MASK R-CNN ON RGB IMAGERY FROM UAVS

Abstract

The article presents the results of applying an improved method for automatic segmentation of RGB imagery obtained using consumer-grade UAVs, based on the Mask R-CNN neural network architecture. Blocks for the preparation and post-processing of raster and vector files have been developed for working with geospatial data. The model was trained on 7000 crowns identified in pine forest of automorphic habitats in the mixed coniferous-broadleaf forest subzone. Training was carried out using cross-validation. Additional data of 1337 crowns were used for verification. During the sequential filtering by area, confidence level, and duplicate segments, the quality of the final segmentation results improved for all age groups of pine forests. The final average precision is 0.87, recall – 0.81, F1-score – 0.83. The results demonstrate the high efficiency of the filtering algorithm in reducing segment redundancy and increasing data reliability. The Mask R-CNN automatic segmentation method is an effective tool for analyzing the characteristics of pine canopies using RGB imagery from UAV surveys. It is capable of replicating the results of visual interpretation with high accuracy. This method particularly advantageous for scaling studies to large areas where manual delineation becomes labor-intensive.

Keywords

• mask r-cnn
• automatic segmentation
• detection trees
• pine forests
• rgb imagery
• uavs
• ecological monitoring
• remote sensing