A feature selection approach for terrestrial hyperspectral image analysis

Feature selection techniques are often employed for reducing data dimensionality, improving computational efficiency, and most importantly selecting a subset of the important features model building. The present study explored utility Filter-Wrapper (FW) approach feature using terrestrial hyperspectral remote sensing imagery. efficacy FW was evaluated in conjunction with Random Forest (RF) Extreme Gradient Boosting (XGBoost) classifiers, to discriminate between water-stressed non-stressed Shiraz vines. proposed yielded test accuracy 80.0% (KHAT = 0.6) both RF XGBoost, outperforming more traditional Kruskal-Wallis (KW) filter by than 20%. also less computationally expensive when compared commonly used Sequential Floating Forward Selection (SFFS) wrapper. Additionally, we examined effect hyperparameter optimisation on classification expense. results showed that marginally outperformed XGBoost all wavebands (p 176) optimised values. 83.3% 0.67), whereas 81.7% 0.63). Our further show optimising values an overall increase accuracy, ranging from 0.8% 5.0%, XGBoost. Overall, highlight performance machine learning ensembles modelling vineyard water stress.