Real-time Disease Detection in Rice Fields in the Vietnamese Mekong Delta

Abstract

In rice cultivation, the majority of pesticides are used focusing on three main problems: herbicides for weed management, fungicides for disease management, and insecticides for pest management. The Food and Agricultural Organization reports ratios of pesticide application in rice fields in Vietnam on 2002 of 25.3% herbicides, 32.6% fungicides, and 40.3% insecticides (FAO, 2005). Fungal leaf blast (LB) disease and bacterial leaf blight (BB) disease are common and well-known diseases found in the rice fields of Vietnam. The rice blast fungus Pyricularia oryzae causes a fungal disease common in rice fields (Francisco and Zahirul, 2003). Depending on the site of the symptoms, the rice blast disease is referred to as leaf blast, collar blast, node blast or neck blast. In the early stages of LB, the lesions on the leaf blade are elliptical or spindle-shaped, with brown borders and gray centers, as shown in Fig. 1a. The bacterium Xanthomonas oryzae causes a bacterial disease (Francisco and Zahirul, 2003) characterized by a water-soaked lesion that usually starts at the leaf margins, a few centimeters away from the tip, and spreads towards the leaf base. The affected areas increase in length and width, and become yellowish to light brown due to dryness, with a yellowish border between dead and green areas of the leaf, as shown in Fig. 1b. Several studies have reported that LB and BB are the most harmful rice diseases and have caused yield losses. For example, rice disease resulted in a yield reduction of 1– 10% from 456 farmer’s fields surveyed across tropical Asia on during 1987–1997 (Savary et al., 2000), and rice yield losses ranging from 50% to 85% have been reported in the Philippines by the International Rice Research Institute (IRRI, 2020). In Vietnam, grain yield losses of 38.21% to 64.57% due to neck blast disease have been reported (Hai et al., 2007). Thus, plant protection focusing on managing diseases and controlling the amount of fungicides to be applied has been an important part of research over the last few years. The literature includes many reports on the detection of rice diseases. Ks and Sahayadhas (2018) report on the prediction of early symptoms of BB and brown spots on rice plants by separating leaf color, signs and illumination from different color channels. This algorithm makes it easy to perform final feature analyses, however, it cannot predict diseases with symptoms in similar colors. Bakar et al. (2018) describe an integrated method for the detection of LB using three categories: infection stage, spreading stage, and worst stage. This is possible by analyzing the Hue, Saturation and Value color spaces with multi-level thresholding, and identifying classified regions of interest during image segmentation. This technique successfully detects the disease based on images taken in uncontrolled environments, however, it is not suitable for the detection of other diseases with similar features. In another study, Islam et al. (2018) present the Gaussian Naïve Bayes method to classify the disease based on the percentage of RGB values of the affected portion using image processing. This method has successfully detected three rice diseases, brown spot, rice bacterial blight, and rice blast, and has a fast processing time and high accuracy, however, it cannot detect a disease with similar color features but a different shape.