Rice Leaf Disease Classification Using Deep Transfer Learning Convolutional Neural Network: MobileNet + Bidirectional GRU
by Sarah Yim
Abstract – Agriculture, the supporting backbone to the economic growth of many countries, has played a significant global role in the economy. With rapid technological advancements in various fields today, it is important to invest time and effort into developing advanced methods to preserve agricultural practices. However, the lack of technology and expertise in crop disease identification is a notorious problem troubling the agricultural industry, especially in developing countries. This often leads to severe crop loss and waste, affecting not only farmers’ yield but also consumers’ food intake. Rice leaf disease identification, in particular, rises as an important issue as rice is a staple food for a large proportion of the global population. Specifically, timely and accurate diagnosis of rice leaf diseases is crucial. To address this issue, this paper implements an image-based deep learning approach to identify and classify rice leaf diseases presented in a dataset derived from a rice field in Sherta located in Gujarat, India. This dataset contains 120 images belonging to three distinct classes: Brown Spot, Leaf Smut, and Bacterial. Evaluation performances followed by statistical analysis are conducted using eight different convolutional neural network (CNN) models: Inception V3, Vgg16, Vgg19, MobileNet, DenseNet121, ResNet101, NASNetMobile, and MobileNet+Bi-GRU. The best performing model was the MobileNet+Bi-GRU model with an accuracy score of 87.24%. The experimental results from the performance evaluations revealed great potential in incorporating deep learning techniques for rice leaf disease identification and classification.
Not yet available.