Using a Convolutional Neural Network to Diagnose Common Diseases in Oryza Sativa
Diseases affecting rice plants have become a major inhibitor of crop cultivation, resulting in the loss of 20-40% of rice crops annually. This has caused an increased reliance on other crops and an increase in food insecurity in developing countries. Furthermore, many farmers lack the knowledge and resources to curb the exacerbating effect that pathogens have on rice plants. One key step in reducing this effect is providing an effective and reliable diagnosis. Over the past decade, convolutional neural networks (CNN) have increased in popularity due to their success in diagnostic technology. The purpose of this project was to create an effective CNN model that uses image classification to diagnose images of diseased rice plants. Four classes were identified within the experiment, including the rice blast, sheath blight, and brown spot diseases. A dataset of healthy rice plants was also used within the CNN as a control variable. Over 3000 images of rice plants were trained into the neural network, with 499 images, or approximately 15%, used in the testing dataset. An accuracy rate of 97.39% was achieved with the best CNN model on the testing dataset. This project is applicable in poor rural areas, where access to diagnostic technologies is limited, and it shows how the use of machine learning is promising in the field of plant pathology.