FerdiMT/Zindi_Crop_Detection_Satellite_Imagery
Domaine:
agriculturegeospatialType de record:
project
Créateur:
Fer
Hôte:
zindi.africa
# Zindi_Radiant Earth Computer Vision for Crop Detection from Satellite Imagery
Quick models to participate in the computer vision crop recognition challenge by Zindi.
zindi.africa
#### Changelog:
| ID | Version | Comments | Zindi score | Issues/Todo's |
| ------ | ----------------------------------- | ------------------------------------------------------------ | ----------- | ------------------------------------------------------------ |
| id_1 | ML-Baseline | Baseline | 1.2549 | Some fields in training set might be NA/0, the ones that are less than a pixel. Instea of fillna(0) maybe just remove them. Gridsearch parameters Generate algorithm to detect neighbourhood. |
| id_2 | ML-Neighbourhood Algorithm | Added the Neighbourhood Algorithm (own) | 1.2570 | Standardize results from neighbourhood algo (matrix_distances). PCA Check other models Gridsearch parameters |
| id_3 | ML-Neighbourhood Algorithm XGB | | 1.2554 | |
| id_3.1 | ML- XGB | No neighbourhood algo | 1.2800 | |
| id_4 | ML-XGB-Neighbourhood in percentages | Put the results of neighbourhood algorithm in percentages (by rows) | 1.2286 | |
| id_4.1 | ML-RandomForest | Same but RF number estimators 500 | 1.2276 | |
| i …