Grape Disease Detection Network based on Multi-task Learning and Attention Features

The disease-free growth of a plant is highly influential for both environment and human life. However, there are numerous plant diseases such as viruses, fungus, and micro-organisms that affect the growth and agricultural production of a plant. Grape esca, black-rot, and isariopsis are multi-symptomatic soil-borne diseases. Often, these diseases may cause leaves drop or sometimes even vanishes the plant/plant vicinity. Hence, early detection and prevention becomes necessary and must be treated on time for better grape growth and productivity. The state-of-the-art either involve classical computer vision techniques such as edge detection/segmentation or regression-based object detection applied over UAV images. In addition, the treatment is not viable until detected leaves are classified for actual disease/symptoms. This results in increased time and cost consumption. Therefore, in this paper, a grape leaf disease detection network (GLDDN) is proposed that utilizes dual attention mechanisms for feature evaluation, detection, and classification. At evaluation stage, the experimentation performed over benchmark dataset confirms that disease detection network could be fairly befitting than the existing methods since it recognizes as well as detects the infected/diseased regions. With the proposed disease detection mechanism, we achieved an overall accuracy of 99.93% accuracy for esca, black-rot and isariopsis detection.