Precision Agriculture: Rice Leaf Disease Detection By Deep Learning

Many plant diseases constitute a severe threat to the world’s food security and reduce agricultural output because it directly or indirectly impacts the quality and quantity of grains and other plant products. To determine whether a leaf is infected, farmers must watch and rely on their senses. This approach is unstable, unpredictable, and error-prone. In a variety of papers, DL techniques for diagnosing leaf diseases have been suggested. The majority of them used convolutional neural networks (CNNs) to build their models. Some models are the most effective in terms of efficiency, but they are either slow or computationally difficult, or unable to function on devices with constrained resources, such as smartphones. Using image classification, rice leaf illnesses such as Bacterial Leaf Blight (BLB), Brown Spot Rice Disease (BSR), and hispa and healthy are categorised. This paper aims to do a comprehensive study of research papers that are published in this field and evaluate their work. In our proposed work, we have taken two types of datasets: a high-resolution image dataset and a low-resolution dataset, each having the four classes mentioned above and containing approx. 3500 and 2000 images, respectively, and then trained the model first on the high-resolution dataset and then on the low-resolution dataset. It increases the accuracy of Resnet 50 from 83.49 percent to 91.64 percent and from 81.90 percent to 91.04 percent for Inception_V3.