Chlorophyll estimation using multi-spectral unmanned aerial system based on machine learning techniques

Remote sensing for precision agriculture is a proven tool for efficient management of crop inputs. High spatial and temporal resolutions are requisite for accurate and timely estimate of crop parameters. We made an attempt to estimatetheleaf chlorophyll concentration of standing maize plant from high resolution (5 cm) multi-spectral Unmanned Aerial Vehicle (UAV) images. The UAV images in Green, Red, Red-Edge and NIR bands of standing maize fields under various nutrient induced abiotic stresses were acquired by flying a hexacopter at an elevation of 60 mat ICAR Research Complex for NEH Region, Meghalaya. A handheld spectroradiometer was also used to record the spectra in 1024 bands, ranging from 350 nm to 2500 nm to support the UAV study. We evaluated advanced machine learning algorithms combined with spectral data and ground truth chlorophyll to model the chlorophyll estimates. Algorithms included Support vector regression, Relevance vector regression, Gaussian process regression, Kernel ridge regression and Random forest with K-fold cross validation. The multivariate analysis applied on spectroradiometer and UAV data showed the dominance of Red-band for chlorophyll prediction with R-2 values greater than 0.80. Among the machine learning algorithms, we found the Kernel-Ridge regression was most robust method for developing chlorophyll estimation model with minimal RMSE (0.057 mg/gm) and regression coefficient of determination (R-2= 0.904). The relevance vector machine also predicted chlorophyll concentration satisfactorily (R-2= 0.87 with RMSE of 0.06 mg/gm), but took larger training time. The optimization and hybridisation of kernel based algorithms is further needed to enhance the reliability of models for prediction of leaf chlorophyll concentrations.