This repository contains the contribution of the team EagleEyes to the AI4EO Hyperview Machine Learning Challenge
Overview
The objective of the AI4EO HYPERVIEW challenge is to predict agriculturally relevant soil pa- rameters (K, Mg, P2O5, pH) from airborne hyperspectral images. We present a hybrid model fusing Random Forest and K-nearest neighbor regressors that exploit the average spectral reflectance, as well as derived features such as gradients, wavelet coefficients, and Fourier transforms. The solution is computationally lightweight and improves upon the challenge baseline by 21%.
This Repository contains the following:
- A jupyter notebook containing the final solution can be found in final-submission.
- Conference paper: Kuzu, R.S., Albrecht, F., Arnold, C., Kamath, R.,Konen, K. (2022), Predicting Soil Properties from Hyperspectral Satellite Images, in 29th IEEE International Conference on Image Processing (IEEE ICIP 2022), Bordeaux, France.
- The folder notebooks contains some jupyter notebooks for data exploration and first simple models
- We explored several other approaches, which can be found in hyperview
- Different Neural Network architectures (based on keras) NN keras
- PSELTAE model (based on pytorch-lightning) PSELTAE
- Random Forest and XGBoost models
A more detailed README can be found here