Wheat Disease Severity Estimation: A Deep Learning Approach

Abstract

In the agriculture domain, automatic and accurate estimation of disease severity in plants is a very challenging research field and most crucial for disease management, crop yield loss prediction and world food security. Deep learning, the latest breakthrough in artificial intelligence era, is promising for fine-grained plant disease severity classification, as it avoids manual feature extraction and labor-intensive segmentation. In this work, the authors have developed a deep learning model for evaluating the image-based stem rust disease severity in wheat crop. Real-life experimental field conditions were considered by the authors for the image dataset collection. The stem rust severity is further classified into four different severity stages named as healthy stage, early stage, middle stage, and end-stage. A deep learning model based on convolutional neural network architecture is developed to estimate the severity of the disease from the images. The training and testing accuracy of the model reached 98.41% and 96.42% respectively. This proposed model may have a great potential in stem rust severity estimation with higher accuracy and much less computational cost. The experimental results demonstrate the utility and efficiency of the network.