Agricultural landscapes are constantly at risk due to erratic weather phenomena. Monitoring these landscapes, especially through crop-type maps produced from remote sensing imagery, becomes imperative. These maps serve as pivotal tools in forecasting and estimating crop yields, where accuracy is crucial. In this research, we conducted a thorough investigation using a series of multispectral images taken by an unmanned aerial vehicle (UAV). The focus was on three specific crops: soybean, barley, and wheat. The primary objective was to determine the comparative efficiency of crop classification models, either trained using Gray Level Co-occurrence Matrix (GLCM) features or those leveraging the UAV multispectral images directly. The results emerging from this study were as follows: the fusion of GLCM features derived from a time series of images and the ExtraTreesClassifier emerged as the standout performer, achieving an accuracy, precision, and recall of 0.87, 0.88, and 0.87, respectively. Following closely was the combination using the same GLCM features, but with the Random Forest classifier.

The outcomes of this research highlight the primacy of feature selection for accurate crop classification tasks, which often is more important than the choice of classification algorithm itself. Effective agricultural planning depends on accurate crop classification. It makes it possible for decision-makers to allocate resources wisely, modify the supply chain, and adopt adaptive farming techniques. Additionally, it provides a path for sustainable policies that consider market demands and climate conditions and subsequently contribute to the field of agricultural strategy and governance.