Detecting and grading severity of bacterial spot caused by Xanthomonas spp. in tomato (Solanum lycopersicon) fields using visible spectrum images

Abstract

We introduce a novel method to detect and classify the severity of bacterial spot (Xanthomonas spp.) in tomato (Solanum lycopersicon) fields. Visual spectrum images were used as inputs and they were taken at 85 days of the plantation. Two hybrids, Hypeel 108 and U2006, were planted and then inoculated separately with X. perforans and X. gardneri, respectively at 37 and 57 days. Ten (10) different plantation areas were then evaluated taking 18 image samples of each in sub-areas, which were analyzed by 7 experts to grade them and be used as comparison. Productivity was also measured in the areas in order to correlate those to the different severities of the disease in the experiment. Visual spectrum images were preprocessed to area size adjustment and brightness correction and then transformed to a CIELab color space for more stable chroma analysis. A clustering process was applied in the a channel in order to group regions related to healthy leaves, unhealthy ones, bare soil and other artifacts. Post-filtering was applied to channels L and b to evaluate regions with over and underexposition of light and reddish fruits being detected. All of the processed regions were then measured and compared using a novel Severity Index SI, which automatically grades, from 1.0 to 5.0, the presence and the severity of the disease. Sixteen classes of severity SC are also proposed, as equal intervals of SI index. Images were taken in a variety of conditions and results showed besides strong correlation with experts analysis, better explanation and smaller error when analyzing the productivity affected by the disease measurements. Results indicate potential for using this methodology for detecting and grading the severity of bacterial spot in tomato fields, with advantages such as capability of repeatable results with low variance, speed and direct field-based applicability.