Click on “Download PDF” for the PDF version or on the title for the HTML version.

If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.

Discrimination of Tomato Yellow Leaf Curl Disease using Hyperspectral Imaging

Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan

Citation:  Paper number  131597646,  2013 Kansas City, Missouri, July 21 - July 24, 2013. (doi: @2013
Authors:   Jinzhu LU, Di CUI, Huanyu JIANG
Keywords:   Hyperspectral imaging spectral analysis imaging processing plant disease reflectance.

Abstract. Tomato yellow leaf curl disease spreads very fast and often causes huge yield losses. It usually represents obvious symptoms on leaves at least 15 days after being infected. This paper investigated the possibility of discriminating tomato yellow leaf curl disease by hyperspectral imaging technique. A hyperspecral imaging system was used to collect hyperspectral images of healthy and diseased tomato leaves. The edge and midrib of leaves were chosen as regions of interest (ROI). Reflectance spectra in the range of 450-1000 nm of ROIs of healthy and infected leaves were measured. The raw, first-derivative, and absolute difference spectra were analyzed to select the feature wavelength for discriminating diseased and healthy leaves. Mean grey values and eight texture features of diseased and healthy leaves images at feature wavelength extracted by grey level co-occurrence matrix (GLCM) were analyzed by receiver operator characteristic curve (ROC). Nine features in total were analyzed. The results showed leaf edge was more prone to disease than leaf midrib area. Red edge shifting in the range of 695-750 nm and blue shifting in the range of 515-560 nm presented in the first-derivative reflectance. The highest difference value of diseased and healthy leaves was in the rang of 710-730 nm. The image at 853 nm was used for segmenting leaf area out and the image at 720 nm was used for acquiring grey information of leaves. ENT_DEV, ENT_MEAN, INE_DEV features were the top three for their areas under the curves (AUC) were more than 0.9.

(Download PDF)    (Export to EndNotes)