Home / Works / Drone-Imagery Analysis Toolbox

Drone-Imagery Analysis Toolbox

Consultancy · Remote Sensing
Python · NumPy · GDAL · SciPy · scikit-learn · QGIS

Overview

Random Forest land-cover classification on an engineered 11-feature stack, bare-land detection auto-calibrated by IoU grid search against ground-truth samples, per-square-metre vegetation grid analysis, and boundary refinement that snaps rough polygons to spectral edges.

How it works

Four tools share one RGB image; calibration feeds detection and refinement.

Training polygons Sample polygons(vegetated / bare) Rough non-vegetatedpolygon Drone RGB image Feature stack (11)RGB, indices, HSV, texture Random Forest(block-wise predict) Two-phase IoU search:thresholds, morphology Buffer, keep largestblob, smooth edge Class polygons(majority filter, area) 3-index vote +morphological cleanup Clean polygon Calibrate from samples(percentile midpoint) Aggregate to 1 m² cells(share vegetated) Grid polygons(non-vegetated cells) Bare-land polygons+ statistics thresholds thresholds

InputStepDecision / optionData storeOutput