Early Detection of Nitrogen Deficiency in Corn Using High Resolution Remote Sensing and Computer Vision.

There is significant interest in early detection of nitrogen deficiency in corn fields using remote sensing.  Early detection can be used to guide variable rate nitrogen (VRT N) applications that are customized according to location in the field based on extent and severity of the deficiency.  We have developed a  semi-automated methodology for the early detection of N deficiencies utilizing low altitude remote sensing with low cost high resolution RGB cameras. We gathered 39 high resolution (0.01 m) RGB images from corn plots that were treated with between 0 to 252 kg/ha of N fertilizer. Each image covers 3 rows of corn plants at the R1 stage of development. The goals are to accurately identify signs of N deficiencies on the plant leaves and associate the number of those deficient leaves with the amount of N fertilizer needed to correct deficiencies. Our approach is based on image processing and machine learning techniques. We developed a 2-step methodology that starts with the localization of rectangular regions in the image that exhibit potential N deficiencies, and continues with the delineation of individual leaves in those regions that exhibit symptoms associated with N deficiency.  This approach had a 96.4% accuracy in detecting colors of the images associated with potential N deficiency, and an 84.2% accuracy in identifying leaves with N deficiency in the localized rectangular regions. Accuracy of detecting N deficiency in the same experiment with a SPAD meter had an accuracy of 60%.  Our results showed a significant linear correlation between the areal density of N deficient leaves detected and the amount of N fertilizer applied, lending support to the feasibility of using high resolution computer vision for VRT N management strategies.