Wavelet based multi-spectral image analysis of maize leaf chlorophyll content

Abstract

With recent advancement in precision farming, the need for variable rate technology has become apparent. Variable rate technology can improve the efficiency of farm operations and lessen farm's environmental impact. To implement effective variable rate applications, it is essential to gather and process information on crop nitrogen level reliably. This research is intended to develop an image-processing method to assess crop nitrogen level based on multi-spectral images of maize plants. This method first removed unnecessary information from the image and then converted the image into a one-dimensional (1D) signal representing the reflectance of the maize plant across leaves. The obtained data was further processed using the wavelet packet transform to find specific patterns that correspond to crop nitrogen stress. To implement wavelet analysis, the 1D signal was deconstructed into packets of narrow frequency bands to find the lowest level approximations at different levels. The maximum wavelet coefficients were identified for interested signal bands and then compared to SPAD meter readings, which were used as the ground-truth corn nitrogen level. Analysis results indicated that the db4 wavelet at a level 8 deconstruction had the highest linear regression coefficient ( R 2 = 0.78) with a high correlation coefficient ( r = 0.88) for corn nitrogen levels.