An Efficient Convolutional Neural Network Model for Multiclass Rice Leaf Disease Detection

In India as well as in other countries around the world, rice is the main staple food. Along with the increase in global population, the demand for rice production has also rapidly increased. It is currently very challenging to ensure that the complete population has access to enough food. The production of rice is also reducing due to various kinds of rice leaf diseases. These diseases can cause countries and farmers to suffer large financial losses which directly impacts the global economy. The objective of the proposed work is to reduce and early predict rice leaf diseases. This will lead to an increase in rice production resulting in securing the income source of the farmers and encouraging sustainable economic growth. The CNN model has been employed in this article to categorize the rice leaf disease. A dataset containing 9 different classes of rice leaf diseases and a healthy leaf has been collected from Kaggle. The leaf disease classes include blast, hispa, dead-heart, tungro, brown spot, bacterial_leaf_blight, downy_mildew, and bacterial_leaf-streak. To estimate the performance accuracy of the suggested model, this model has been adjusted and implemented on a range of epoch values. The results have identified that the model exhibits the highest accuracy and least loss during the training and validation stages when the epoch value has been set to 15. The accuracy and loss in the training phase using epoch 15 are 97.11% and 0.0886, respectively, whereas, on the validation side the accuracy of 92.98% and loss of 0.2321 have been identified.