Analytical Study of Models for Rice Leaf Disease Detection Using AI and Deep Learning

Abstract

Rice is one of the most significant crops across the world, especially in Asian countries, where it majorly contributes to the national economy. With the constant change of human activities, the quality of the atmosphere, land and water bodies has also changed this led to an increase in plant diseases. The production of rice is seriously affected by various leaf diseases such as bacterial blight, brown spot, and leaf smut. As these diseases are numerous, they are difficult to identify by traditional manual techniques, which are more time-consuming and led to human errors. It is crucial to detect these diseases early and accurately in order to minimize losses in yield and make cultivation sustainable.

Deep learning models are used in this research to determine the best paradigm to use in automatic rice leaf disease detection. Convolutional neural network (CNN) models such as DenseNet121 (Densely Connected) are evaluated on a publicly available leaf images dataset. CNN has proven to be most significant and functional in image processing due to its potential to extract the components of an image. In this work, we inspect different models of CNN, especially targeting on Desnet121 architecture. Dense Net initiate dense connectivity between different layers. Layer 1 sends output to the layer 2,3,4. Layer 2 sends output to the 3,4,5…so on. Each layer receives information from the previous layer.  The models are taken away on the basis of accuracy, precision, recall, F1-score, and computational efficiency. Experimental results demonstrate classification results much better than using a standard Convolutional Neural Network (CNN) trained from scratch. As well, it emphasizes on the potential of CNNs with optimization and synthetic data generation techniques to further improve and enhance model accuracy. In general, this study suggests that CNN models like Dense Net can provide a speedy and dependable way to detect rice leaf diseases and help farmers and experts to detect these diseases initially and reduce batch loss.  The discovery of this research emphasizes on the deep-learning models for real-time and rice disease detection, offering profound observations for future agriculture systems in developing countries.