Rapid in-field diagnosis of huanglongbing disease using computer vision

Abstract

Citrus greening or Huanglongbing (HLB) is an extremely destructive disease, which has had an undesirable impact on the quality and quantity of the citrus production in Florida during the past few years. No effective treatment has been reported for this disease yet; therefore, rapid diagnosis and removal of the affected trees can protect the entire grove from further infection. One of the early symptoms of HLB is the accumulation of starch in affected citrus leaves which appears as an uneven yellow blotch mottled pattern on the leaf surface. This symptom looks highly analogous to some nutrient deficiency symptoms; however, starch accumulation in the HLB symptomatic area has a unique capability of rotating the polarization planar of light at a certain waveband. A computer vision approach was employed in this study to highlight the starch accumulation and differentiate it from nutrient deficiency symptoms. This system was examined under a real in-field condition to detect HLB positive, HLB negative, and zinc deficient samples. Two simple image descriptors (mean and standard deviation) were extracted from the sample images and utilized in a step-by step classification model. The overall accuracy of 97% was achieved for identification of citrus leaves in three classes using this method.