DYNAMIC CALIBRATION AND IMAGE SEGMENTATION METHODS FOR MULTISPECTRAL IMAGING CROP NITROGEN DEFICIENCY SENSORS

Abstract

Site-specific variable-rate nitrogen application is one of the core operations in precision crop management. Thedetermination of an appropriate nitrogen application rate relies greatly on the capability of assessing crop nitrogen stress.A machinery-mounted multispectral imaging sensor has been developed for real-time crop nitrogen deficiency detection onthe sprayer during fertilization operations. While field tests indicated that this image-based sensor was capable of detectingcrop nitrogen deficiency on the go, the test results also showed that this sensor was very sensitive to ambient light changesand needed a considerably long image processing time to extract crop nitrogen deficiency data. To solve these problems, theresearch has developed a dynamic calibration method to compensate for ambient illumination variation on crop canopy reflectance,an image segmentation algorithm to eliminate the soil background noise, and a correlation model to estimate theSPAD values from the calibrated multispectral crop canopy reflectance. Field validation tests demonstrated that the developedsensor calibration and image processing algorithms improved the performance of the multispectral sensor on detectingcorn nitrogen stress. Using the modified sensor resulted in a reasonable correlation between the estimated and measuredSPAD values (R2 > 0.72). This research confirmed that it is technically feasible to design a machinery-mounted multispectralimaging sensor to detect crop nitrogen stress reliably and accurately.