Multi Modal Deep Learning Based Crop Classification Using Multispectral and Multitemporal Satellite Imagery

Summary

This repository contains the code for the paper Multi Modal Deep Learning Based Crop Classification Using Multispectral and Multitemporal Satellite Imagery that will be published as a poster paper in KDD Applied Data Science Track 2020.

Data

Data can be downloaded from here

Software Configuration

• python==3.7.4
• tensorflow-gpu==1.13.1
• keras==2.2.4
• sklearn==0.21.2
• numpy==1.16.4
• matplotlib==3.1.1
• pandas==0.25.1
• configparser

How to run the code:

The following gives the folder descriptions. Each folder is a separate set of experiment:

• purely_spatial: purely spatial 2D crop image classification using well-known neural networks.

• lstm, bi-lstm and 1dcnn: purely temporal crop time series classification that only uses the temporal part of our data.

• concatenate: concantenates purely spatial and purely temporal streams.

• avg-fusion: uses average to combine the spatial and temporal streams.

• svm-fusion: uses SVM classification to predict on a concatenation of spatial and temporal stream predicted probabilites.

To run the experiments, each folder has a readme file with instructions.