[general]
name=Fazenda Classifier
description=Forest inventory tools: sampling plots, canopy heights, crown measurements, RGB species classification and CSV export.
about=Supports forest inventory with individual crown delineation, planar area and maximum diameter measurements, and pointwise DSM/DTM heights. Creates metric sampling plots, extracts DSM maxima in 1 x 10 m windows, and samples DTM at the same physical location, with horizontal CRS conversion. Exports point measurements and CRS provenance to CSV. Plot and height tools require QGIS and NumPy only. Optional CNN classification requires PyTorch, compatible torchvision, Pillow, and a user-supplied Fazenda LeNet state-dict model. Tested by the author with PyTorch 2.10.0 and torchvision 0.25.0 for CPU on Windows. No PyTorch version gate is imposed. The author's trained weights are private and not available for download. Species classification needs externally supplied compatible weights; plot, height and crown tools work without them. Measurements and exports can support external forest carbon, AGC and climate studies; this version does not calculate carbon or climate projections. Height sampling points are not identified individual trees and are not automatically converted to H5 or AGC. QGIS 4 support has not been validated.
version=1.1.0-rc4
qgisMinimumVersion=3.36
qgisMaximumVersion=3.99
author=Leonardo Heraki
email=leonardoheraki@outlook.com
homepage=https://github.com/leonardoheraki/fazenda-classifier#readme
repository=https://github.com/leonardoheraki/fazenda-classifier
tracker=https://github.com/leonardoheraki/fazenda-classifier/issues
license=GPL-2.0-or-later
category=Raster
tags=forestry,canopy,DSM,DTM,forest inventory,remote sensing
icon=icon.svg
experimental=True
deprecated=False
hasProcessingProvider=False
changelog=1.1.0-rc4: Preserve the RC3 inference path and weights_only=True; tested by the author with PyTorch 2.10.0 and torchvision 0.25.0; unique temporary workspaces independent of UA; write only to marked memory outputs; stage complete classifications before replacing previous results; 12 native QGIS checks passed and real outputs reviewed; display tree confidence as a percentage while keeping CSV precision and the 0.60 threshold.
