Publication

Abstract : More than half of the people in the world rely on rice as their primary energy source. Two main challenges in rice cultivation are plant diseases and nutrient deficiency. Brown spots on leaves caused by the fungus and Cochliobolus miyabeanus, are one of the predominant plant diseases. Nitrogen is the most significant nutrient required in rice crops. Farmers monitor the plant’s growth by matching the leaf colour with the colour as in the leaf colour chart (LCC) (an indicator of the amount of nitrogen concentration in leaves). LCC is mostly used manually in order to estimate the nitrogen content and in turn the fertilizer amount, resulting in overestimating or underestimating this required portion. But, the other crop health factors like diseases are mostly ignored. Hence, a novel model based on deep learning is proposed to simultaneously detect brown spots and interpret leaf colour, thus estimating nitrogen amount as well as the occurrence of disease. The dataset consists of 1607 training and 402 test images, divided into three classes by observing the LCC chart as well as the presence of brown spots. Image pre-processing (gaussian blur, colour conversion, thresholding, and augmentation), K-means segmentation and feature extraction using InceptionV3 followed by classification using VGG-19 were done, which gave an accuracy of approximately 75%. This novel model is an exception to the existing models which focuses on one particular feature, as the proposed model detects two of the most relevant issues seen in rice cultivation, i.e. brown spots and nitrogen deficiency.