Abstract

A method to estimate the leaf nitrogen concentration based on image processing techniques was developed. The images came from maize leaves with known levels of nitrogen, measured with a SPAD meter. In order to establish a correlation between the N concentration and the SPAD measurement, an average value in SPAD units and a representative color value were assigned to every leaf. In fact, the representative color values were computed by creating a weighting average of three representative leaf color groups. Accordingly, a prototype of every group and the pixel's area of each group were used to compute the weighting average. Thereby, these groups were created according to their degree of similarity using the k-means clustering technique, which provides a prototype of every group. In specific, the clustering was performed over images in the L∗a∗b∗ color space. In addition, correlation models to predict SPAD values, using a single color channel and two color channels, from the RGB, HSV and L∗a∗b∗ color spaces were tested. The analysis shows that it is possible to correlate SPAD values and color data with the concentration of N, furthermore, shows that when single color channel correlations were performed, the channels G, b∗ and V provide the better modeling accuracy. Moreover, it was likewise shown that better correlations were obtained when a combination of two color channels, from the same or different color spaces, were applied in the correlation. In particular, the most accurate prediction models were obtained held the pairs of color channels S-V and G-b∗.