Potential of Hybrid CNN-RF Model for Early Crop Mapping with Limited Input Data

When sufficient time-series images and training data are unavailable for crop classification, features extracted from convolutional neural network (CNN)-based representative learning may not provide useful information to discriminate crops with similar spectral characteristics, leading poor classification accuracy. In particular, limited input the main obstacles obtain reliable results early mapping. This study investigates potential of a hybrid approach, i.e., CNN-random forest (CNN-RF), in context mapping, that combines automatic feature extraction capability CNN superior discrimination an RF classifier. Two experiments on incremental unmanned aerial vehicle were conducted compare performance CNN-RF respect length sizes. used accuracy was slightly higher or comparable CNN. contrast, when fewer smallest at growth stage, substantially beneficial overall increased by maximum 6.7%p 4.6%p two areas, respectively, compared is attributed its ability insufficient using more sophisticated The experimental demonstrate effective classifier mapping only samples available.