Non-destructive chlorophyll prediction by machine learning techniques using RGB-based vegetation indices in wheat

Non-destructive RGB sensor-based estimation of chlorophyll content has significant uses in crop demand-based nitrogen fertilizer application in precision agriculture and plant phenotyping for crop improvement. In this work, we estimated the chlorophyll content of wheat rapidly and no-destructively with high-throughput phenotyping by using RGB-based vegetation indices. The RGB images were captured using the automated phenotyping and imaging platform at Nanaji Deshmukh Plant Phenomics Centre, New Delhi, India. Then the RGB images were analysed and used to calculate 39 RGB vegetative indices reported in the literature. We examined the RGB vegetative indices from RGB images and 16 machine-learning models were used to measure total chlorophyll content and the models were evaluated by coefficient of determination (R²), correlation coefficient (r), root means square error (RMSE) to select the best estimation model. The results showed that several RGB indices and total chlorophyll content values showed a highly significant correlation, with a high correlation reaching 0.84. In the prediction model, the highest precision was obtained with the BRNN model (R² = 0.71, r = 0.84, RMSE = 2.52). These indicated that the best model (BRNN) can precisely estimate whole plant total chlorophyll content in wheat from digital RGB images, which implies RGB indices have potential for low-cost wheat leaf chlorophyll estimation. Since RGB sensor is low-cost sensor, this model can be used drone-based and mobile based applications in precision agriculture and plant phenotyping. This non-invasive method can uncover potential genes for chlorophyll accumulation by estimating the pigment status of plants at various phenological stages.