Multi-crop recognition using UAV-based high-resolution NDVI time-series

Multi-crop recognition is a highly nonlinear task in nature as it involves many dynamic factors to address. In this paper, a decision tree based approach is presented to classify and recognize 17 different crops. High spatial and temporal normalized difference vegetation index (NDVI) signatures were extracted from multispectral imagery using a multispectral sensor onboard the quadrotor. Detailed datasets were prepared through sampling based on normal distribution with different standard deviations. The impact of reduced dimensions was also tested using principal component analysis. A very high degree of accuracy was achieved for classification. The results also indicate that NDVIs pertaining to early-to-mid season have much more weight in the classification process for multiple crops.