A Neural Network Model of Maize Crop Nitrogen Stress Assessment for a Multi-spectral Imaging Sensor

Abstract

Site-specific variable nitrogen application is one of the major operations in precision crop production management. Estimating crop nitrogen stress accurately during side-dressing operations is essential for managing site-specific nitrogen applications effectively. Various off-line assessing methods of nitrogen stress, such as aerial images of the field, crop tissue analysis, soil sampling analysis and soil plant analysis development (SPAD) meter readings of crop leaf, have been applied in current site-specific crop production management. To support a machinery-mounted multi-spectral imaging sensor to detect nitrogen stress during side-dressing operations, a neural-network model was created to estimate maize leaf SPAD values based on the sensed reflectance of maize canopy in three channels of green (G), red (R) and near-infrared (NIR) of a multi-spectral charge-coupled device (CCD) camera. Based on results obtained from both off-line post-processing and real-time field tests, it was verified that the developed neural-network model was capable of extracting stress information with reasonable accuracy in real-time in terms of sensed crop canopy reflectance in G, R and NIR channels. The coefficient of determination r2 between estimated and measured SPAD values was 0·89 and a root mean square (RMS) error of 2·52 was observed from validation tests.