Deep Learning Based Effective Rice Leaf Disease Classification using MobileNet- Attention

Abstract

Agriculture is one of the most promising fields for contributing to the growth of the economy of the country. Rice is considered as one of the important crops that contributes to the economy and food demand. The production of rice is mainly affected by disease in plants which severely impacts the production. The diseases include bacterial blight, blast, browspot, tungro. The dataset was obtained from the internet that contains a total of 5932 images. In this research, we proposed a method based on Mobilenet with an attention block to classify four different diseases of rice. The mobilenet architecture is effective for mobile devices. The proposed approach is a lightweight model with the combination of an attention block that uses squeeze excite block. The mobilenet was used for the feature extraction process. The Squeeze-and-Excitation Block allows a network to execute dynamic channel-wise feature recalibration, hence increasing its representational power. The proposed approach recognizes the diseases effectively and increases the model accuracy and is computationally effective. The model achieved an accuracy of 100% on rice dataset.