In the realm of contemporary forestry, the need for precise forest management has
led to a
demand for detailed digital forest information. This research focuses on the methodological
advancements
achieved through a segmentation procedure integrated into post-processing. The approach
involves
image segmentation using the MRS algorithm on a pre-classified image, progressively
building from
the bottom up at different intervals (10 and 100 levels). Across three diverse sample
areas, this procedure
yielded substantial enhancements in overall classification accuracy and accuracy values
per class. The
pixel-based results demonstrated an average accuracy of 87.6%, with further improvements
to 89.93%
and 89.98% for the 10-level and 100-level versions, respectively. Notably, the 10-level
version exhibited
the highest improvement on average at the L3 level, reaching 102.51% of pixel-based
accuracy, while
the 100-level version achieved 102.71%.
