Enhancing Agricultural Crop Classification with Sentinel-2 Imagery: A Comparative Approach of Random Forest and CNN Algorithms

The primary limitations of surface irrigation network operations lie in the dynamics of the supply and needs of crop water, particularly with climate change and the freedom of farmers to decide on crop-type cultivation. Accurate spatial information is important for crop water demand as a basis for irrigation network operation. This information is extremely significant and depends on the crop type. The purpose of this study was to test two machine-learning algorithms (i.e., random forest RF and convolutional neural network CNN), to appropriately forecast crop types on an irrigation area. Validation experiments were performed in three different irrigation areas. The overall accuracies of the CNN and RF algorithms were 70.1% and 75.1%, respectively. These results provide an alternative that is less time-consuming and more effective for the landscape size of crop water requirements in surface irrigation management. ©The Korean Institute of Intelligent Systems