Structure:

• All relevant constants are stored in config/, and are accessed via the Config class.
• Each file must be runnable on its own, and hence must have an if __name__=="__main__" section.

Installation flow: always at root

1. Create a venv using python3 -m venv venv
2. Activate venv usin source venv/bin/activate
3. pip3 install -e . to install everything in requirements.txt

Running the pipeline:

• Configuration
  • Open config/
  • Configure tables.json (The codebase runs only on the first table in the list)
  • Configure psql.json (postgres database configuration)
• Satellite data
  • All satellite data (quads or otherwise) must be store in quads/ in the root
  • Run python3 src/planet_labs/planet_labs.py to get monthly data of a quad that intersects the table's geometry - Good for small tables
  • Run python3 src/planet_labs/planet_labs_quadpy to get specific quads - Better when dealing with a large table (say a taluka)
• Clipping
  • Run python3 src/crop_presence_basic/clip_automatic.py to clip satellite quads and store average RGB values in the database as extra columns
• Crop Presence
  • Run python3 src/crop_presence_basic/crop_presence_inference.py to infer monthly crop presence probabilities based on the stored average RGB values
  • python3 src/crop_presence_basic/train_crop_presence.py reads annotations.csv that must correspond to the existing table being processed (the first table in tables.json) and trains the crop presence model
• Crop Cycle
  • Run python3 src/crop_cycle_monthly/crop_cycle_inference.py to infer cropping patterns based on crop probabilities and store the results in the database
  • python3 src/crop_cycle_monthly/train_crop_cycle.py trains the crop cycle model based on cycle_annotations.py
  • python3 src/crop_cycle_monthly/annotate_cycle.py is a tool to annotate crop cycles
• Analysis
  • Run python3 src/analysis/study_plot.py -k <some_key> to study the farm plot with that specific key
  • Run python3 src/analysis/study_random_plot.py to study a random plot
• CNN Crop Presence
  • Run python3 src/crop_presence_cnn/train_crop_presence_cnn.py to train CNN model (Refer to src/crop_presence_cnn/model.py for CNN architecture)
  • Run python3 src/crop_presence_cnn/crop_presence_inference_cnn.py -m <hsv/rgb/hist> to infer monthly crop presence probabilities from histogram model
  • Run python3 src/crop_presence_cnn/annotate_cnn.py -s <start-key> -e <end-key> -d <train/test> to annotate farm plots

Work in Progress:

1. Switching to a CNN based crop presence predictor
   1. Trying HSV based models instead of RGB
   2. Using HSV histogram bins as inputs to a neural network
2. Shifting to weekly data from planet labs to improve rabi sowing period detection
3. Planet dependencies:
   1. pip3 install click --upgrade and pip3 install planet --upgrade makes planet go to version 2 which is required for its api