Quantifying the Severity of Loose Smut in Wheat Using MRCNN

Abstract

Wheat is one of the most common staple foods in the world. According to national agricultural research, an entire 40% of wheat grain quality has been decreased due to wheat diseases. The wheat disease mainly affects the wheat plant. Sometimes, the wheat disease damages the whole wheat plant. When the whole wheat plant is damaged, the quality of wheat grain is decreased. There are different varieties of wheat diseases such as fungal, bacterial, and virus-based diseases. One of the most common loose smut fungal diseases that decrease the grain quality in each wheat spikelet. Therefore, the identification of wheat diseases is important. In this paper, an automatic system for the identification of loose smut wheat disease along with its severity is proposed. The identification system uses 2000 RGB images which have been collected from secondary sources. Among all datasets, a total of 800, 700 images have been randomly selected for training and testing purposes in the Mask RCNN model. Through Labelme software, the images are labelled with ground and truth labels. During identification, our proposed system achieves a 97.8% F1 score for loose smut identification with bounding boxes. The severity of loose smut has been calculated through the disease severity index. With the help of DSI, a total of 63% severity for loose smut has been estimated in different wheat spikelets.