Application of Pre-Trained Deep Convolutional Neural Networks for Rice Plant Disease Classification

Abstract

Rice is a primary food and encounters an essential role in providing food security worldwide. However, several diseases affect this crop that significantly reduces its production and quality. Therefore, early detection of diseases is much needed task to prevent spreading of diseases. Hence, it is desirable to develop an automatic system which will help agronomist, pathologist and even the farmers to diagnose the rice diseases more efficiently and take preventive measures in time. In the present era of advanced artificial intelligence, various learning techniques have been explored for rice plant disease classification. Among various machine learning techniques, deep learning has been widely applied in various domains of computer vision and image analysis recently. It has successfully delivered promising results with large potential. However, training the deep learning model from the scratch requires huge labeled data and collection of huge labeled data is expensive, laborious and time taking process. Transfer learning of pre-trained deep learning model is a technique to overcome such problems. This paper has explored the performance of various pre-trained deep CNN models such as: (i) AlexNet; (ii) Vgg16; (iii) ResNet152V2; (iv) InceptionV3; (V) InceptionResNetV2; (vi) Xception; (vii) MobileNet; (viii) DenseNet169; (ix) NasNetMobile; and (x) NasNetLarge for image based rice plant disease classification. The dataset used in this paper consist of 1216 rice plant diseased images and these have been collected from the real agricultural field having seven classes: (i) rice blast; (ii) bacterial leaf blight; (iii) brown spot; (iv) sheath blight; (v) sheath rot; (vi) false smut; (vii) healthy leaves. The Vgg16 model resulted highest classification accuracy of 93.11%. The outcome of the model can be used as an advisory and as an early detection tool in the real agriculture domain.