Optimization of Convolutional Neural Network in Paddy Disease Detection

Abstract

In Sabah, agriculture is an important economic sector. The situation has recently worsened due to paddy cultivation and rice production management issues. A traditional form, such as detecting disease with the naked eye, is susceptible to high error rates and incorrect classification. To improve the long-term sustainability of the paddy industry, disease detection is critical. The collaboration between Universiti Malaysia Sabah and other government agency has opened new research opportunities in agricultural programs in Sabah and this has sparked the initiative to use advanced technology in Smart Farming. It allowed researchers to utilize digital technologies to jumpstart sustainable and competitive agriculture development. This paper proposes a Paddy Leaf disease detection and classification algorithm that applies deep learning approach. The proposed solution can later assist farmers to diagnose the frequently occurring disease automatically. CNN architectures have successfully been developed to solve various prediction and classification tasks. Due to its excellent performance, the aim of this paper is to formulate a deep learning approach to recognize and detect paddy disease based on the paddy leaf images. In this work, an optimized deep convolutional neural network model will be utilized and assessed to diagnose the health of the paddy based on its leave condition. Based on the results obtained, the number of epochs and the dropout rate have a great influence on the performance of the CNN models. For instance, having a high number of epoch’s value and smaller percentage of dropout rate, the proposed CNN model is able to classify the type of paddy disease with performance accuracy of 64.80%.