Greenhouse image classification

This repo contains code on image classification for the paper Global area boom for greenhouse cultivation revealed by satellite mapping

The purpose is to find the presence of greenhouses globally tile by tile (a region of approximately 1 degree cell). The tiles with a positive prediction of greenhouses will be served to the image segmentation model

Key features

This code splits the image chip labels with the target 'greenhouse' and background 'non-greenhouse', which were saved as csv, into training, validation and testing data.
Train model using EfficientNet backbones
Predict at 1km grid for a large area

Code structure:

Prepare labels

python data_prepare_classification.py

--- 🔖 set configs ---

config/config_classification.yaml

Train 1st model: Greenhouse image classification:

python main_classification.py

Test 1st model: Predict at 1km grid for large area using satellite images (e.g. PlanetScope):

python inference_run_classification.py