New open machine learning models for geospatial applications

[Louisa Nakanuku-Diggs]

Dear all,

I am excited to share our latest open machine learning models for geospatial applications.

The first one is the AgriFieldNet Model for Crop Detection from Satellite Imagery. This model consists of 1 Unet + 8 Gradient Boosting Trees and is the first place solution for the AgriFieldNet India Challenge Crop Types Detection competition that aimed to classify crop types in agricultural fields across Northern India using multispectral observations.

The next is the Weighted Tree-based Crop Classification Models for Imbalanced Datasets that took second place in the same competition mentioned above. Ensembled weighted tree-based models "LGBM, CATBOOST, XGBOOST" with stratified k-fold cross validation, taking advantage of spatial variability around each field within different distances.

Both models were trained on the AgriFieldNet Competition Dataset, which is also freely available on Radiant MLHub. 

Best wishes for the new year,

--

Louisa Diggs

Marketing and Communications Manager

Radiant Earth Foundation

www.radiant.earth 

www.mlhub.earth --> an open library of geospatial training data and models