Peanut crop identification based on UAV remote sensing image and random forest

Peanut is considered a crucial economic crop, and accurately identifying its planting areas is regarded as vital for ensuring food and oil security. This study utilizes ultra-high spatial resolution unmanned aerial vehicle (UAV) imagery and the random forest (RF) method to classify multispectral imagery. Three types of features are combined to identify peanut crop and verify accuracy. Results indicate that using only multispectral features yields unsatisfactory accuracy. Introducing texture features significantly improves classification accuracy, particularly for spectral features alone. However, accurately identifying crop pixels at field edges remains challenging due to noise influence. When UAV image spectral and texture features are employed for classification, the principal component analysis (PCA) method, combined with post-classification patch removal, proves to be the best strategy, achieving an overall classification accuracy of 98.17% and a Kappa coefficient of 0.97. The comprehensive peanut recognition accuracy reaches 90.39%.