Predicting Soil Properties from Hyperspectral Satellite Images

Abstract

The AI4EO HYPERVIEW challenge seeks machine learning methods that predict agriculturally relevant soil parameters (K, Mg, P 2 O 5 , pH) from airborne hyperspectral images. We present a hybrid model fusing Random Forest and Knearest 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 21.9%, with the first place on the public leaderboard. In addition, we discuss neural network architectures and potential future improvements.

Hyperspectral imaging – a short review of methods and applications

“…Research Foundation [48], -Pavia Centre and University data acquired by an airborne ROSIS sensor and provided by the Telecommunications and Remote Sensing Laboratory [49], -Kennedy Space Center data from AVRIS and opened by NASA [49], -HYPER-VIEW soil dataset acquired by an UAV platform using a XIMEA camera, opened by KP Labs from Poland within the Hyperview Challenge [50][51][52]. Despite many selective datasets and more than a dozen orbital hyperspectral missions, currently hyperspectral data from satellites have not been yet easily accessible to downstream companies.…”