model_ecaas_agrifieldnet_silver
Weighted Tree-based Crop Classification Models for Imbalanced Datasets
Second place solution in the Zindi AgriFieldNet India Challenge to classify crop types in agricultural fields across Northern India using multispectral observations from Sentinel-2 satellite. 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.
MLHub model id: model_ecaas_agrifieldnet_silver_v1. Browse on Radiant MLHub.
Training Data
Related MLHub Dataset
AgriFieldNet Competition Dataset
Citation
Alasawdah, M. “Weighted Tree-based Crop Classification Models for Imbalanced Datasets”, Version 1.0, Radiant MLHub. [Date Accessed] Radiant MLHub https://doi.org/10.34911/rdnt.qiuwp5
License
Creator
Mohammad Alasawdah - Earth Observation and Climate Data Science https://www.linkedin.com/in/mohammad-alasawdah-b3b541a5/
Contact
masawdah@gmail.com
Applicable Spatial Extent
The applicable spatial extent, for new inferencing.
{
"type": "FeatureCollection",
"features": [
{
"properties": {
"id": "ref_agrifieldnet_competition_v1"
},
"type": "Feature",
"geometry": {
"type": "MultiPolygon",
"bbox": [
76.2448,
18.9414,
88.046,
28.327
],
"coordinates": [
[
[
[
88.046,
18.9414
],
[
88.046,
28.327
],
[
76.2448,
28.327
],
[
76.2448,
18.9414
],
[
88.046,
18.9414
]
]
]
]
}
}
]
}
Applicable Temporal Extent
The recommended start/end date of imagery for new inferencing.
| Start | End |
|---|---|
| 2022-01-01 | present |
Learning Approach
- Supervised
Prediction Type
- Classification
Training Operating System
- Linux
Training Processor Type
- CPU
Model Inferencing
Review the GitHub repository README to get started running this model for new inferencing.
Training
Prepare the data for tree models by computing the average values of the pixels within each field, then feature engineering by computing spatial variability, more vegetation, and flowering phenology indices.
Zonal statistics (mean , min, max, std) within different radiuses (0.50, 1.00, 1.50, 2.50, 3.50, 5.00) Km around each field
Model
Weighted average tree-based models: lightgbm. catboost, xgboost classifers.
Structure of Output Data
- Predictions.csv: Final predictions text file, with 13 crops classes as following
Wheat, Mustard, Lentil, No Crop, Sugarcane, Garlic, Potato, Green pea, Bersem, Coriander, Gram, Maize, Rice - veg_indices.csv: Extracted vegetation indices for each field.
- Field_stats_indices.csv: Extracted statistics for each field.