Deep Learning-Based Automated Feature Engineering for Rice Leaf Disease Prediction

Abstract

Though the financial impact of agriculture is gradually decreasing with the extensive economic growth of the country, still agriculture is one of the most wide economic sectors that play an important role in the global socio-economic structure of India. There are various diseases that affect food crops and cause huge damages to crops, frightening the livings of helpless farmers and the food and nutrition security of masses. So, identification of diseases that affect the plants and quick recovery from the diseases are the major concerns of the researchers. Here, we have developed a deep learning-based automated feature engineering for rice leaf disease prediction. The objective of this work is to enhance the crop production by early prediction of leaf diseases and taking precautions to remove or at least not to spread the diseases over the surroundings of the infected areas. Initially, the diseased portion of the leaves of the rice plant images is identified and separated from the leaves. These diseased images are feed into the convolution neural network (CNN) model, a modern family of deep learning model. The architecture of the CNN model consists of four convolution layers, two fully connected layers, and at the last softmax output layer. The main motivation for using CNN is that it has the power of automated feature engineering to create infinite number of features without any human bias. Also, it has the capability of capturing all possible complex nonlinear interactions among the features. Next, a dimension reduction method is applied to remove the redundant features and finally, the diseases are classified using various classifiers. The method is experimented using 10,500 infected leaves. The experimental results and comparative study based on the performance evaluation show the effectiveness of the proposed method and help to select the appropriate classifier for rice leaf disease prediction.