Machine-vision-based nitrogen management models for rice

Machine-vision-based yield prediction models were developed for mid-season nitrogen (N) management for two rice cultivars: Oryza sativa 'Millie' and Oryza sativa 'Lemont'. Field images of rice plants were acquired using a camcorder mounted on an image acquisition unit (IAU) designed for flooded rice fields. The acquired images were digitized and then segmented into plant and background pixels using a segmentation algorithm based on spatially varying mean intensity values and mathematical morphology. Segmented images were used to extract features related to plant health. Several models were developed to predict yield as a function of mid-season N application rate and mid-season plant measurements; the measurements included features extracted from the rice plant images, manual size measurements and Y-leaf chlorophyll readings. The best models (R(2) = 0.846 and 0.828 for Millie and Lemont, respectively) included 20 variables comprised of combinations of machine vision based measurements and leaf-chlorophyll readings. The models were superior to models based on manual measurements alone. The machine vision based N management system may provide an objective method for performing mid-season N assessments and making N recommendations that maximize yield or profit.