A precise model for accurate rice disease diagnosis: a transfer learning approach

Abstract

Rice being a principal cereal, is consumed by a large portion of the world's population as a staple food. Like other plants, it is susceptible to a variety of diseases associated with themselves which can affect greatly the quantity and quality of yield. It is essential to detect the disease in its early stage and take appropriate treatment measures to ensure proper growth of the plant as well as maximum quality yield. Detection of the diseased plants through the naked eye seems to be an impossible task taking into consideration the large range of leaf diseases and the vast areas of farmlands. Technological advancement in the areas of machine learning, computer vision, and IoT has led to the development of automated systems for the detection of diagnosis of plant diseases from the image of diseased leaves. Though machine learning (ML) based solutions are available for this task, still there is potential for further improvement in the automated systems build on transfer learning approaches based on different CNN architectures. This paper proposes a transfer learning-based AlexNet model for automatic detection and diagnosis task. Three of the most common rice plant disease namely bacterial leaf blight, brown spot, and leaf smut which are very difficult to distinguish by the naked eye have been considered for investigation. The dataset is obtained from the 'Kaggle' website. The proposed AlexNet model performs significantly well for the detection and classification task on 'Kaggle' dataset in terms of accuracy, precision, recall, F1 score, and kappa coefficient in comparison with other ML and transfer learning-based approaches reported in the literature for the same task on same or different datasets.