Enhancing Semantic Forestry Segmentation Through Advanced Preprocessing With ML Models

This study explores the application of advanced artificial intelligence based preprocessing techniques to improve semantic forestry segmentation. The research investigates the performance of You Only Look Once version 8 (YOLOv8) from Ultralytics, Detectron2 from Meta, and the Segment Anything Model (SAM) from Meta, with the goal of enhancing segmentation accuracy and detail in forest environments. Goal is accomplished by combining models into a pipeline along with machine-larning-based preprocessing models. The integration of SAM and pre-processing steps significantly enhanced the quality and number of segmentation masks, resulting in more accurate and detailed representations of complex forest structures compared to the ground truth. The findings highlight the importance of combining AI models with effective preprocessing strategies to navigate the complexities of forest environments, offering valuable insights for future advancements in semantic forestry segmentation.