A lightweight deep learning based approach for rice leaf disease classification

Abstract

Efficient and timely identification of diseases and pests in rice plants is crucial for farmers to apply timely interventions and mitigate economic losses. The recent ad vancements in deep learning, particularly in convolutional neural networks (CNN), have significantly enhanced the accuracy of image classification. Drawing inspiration from the success of CNNs in image classification tasks, this paper introduces deep learning-based approaches for the detection of diseases and pests in rice plants. The primary contributions of this paper can be delineated into two aspects. Firstly, state-of-the-art large-scale architectures like VGG16 and InceptionV3 have been adopted and fine-tuned to effectively detect and recognize rice diseases and pests. Experimen tal results underscore the efficacy of these models when applied to real datasets, show casing their potential for accurate identification in agricultural settings. Secondly, recognizing the constraints posed by the deployment of large-scale archi tectures on mobile devices, the paper proposes a two-stage small CNN architecture. This novel architecture is then compared with established memory-efficient CNN ar chitectures such as MobileNet, NasNet Mobile, and SqueezeNet. The experimental findings reveal that the proposed architecture not only achieves the desired accuracy of 93.3% but also significantly reduces the model size, boasting a remarkable 99% re duction compared to VGG16. This reduction in model size enhances the feasibility of deploying the proposed solution on resource-constrained mobile devices, ensuring accessibility and applicability for farmers in real-world scenarios.