Ensemble of Visual Transformer and Deep Neural Networks for Recognizing Sunflower Diseases from Photographs

Abstract

The world's population is growing steadily, but already more than 1.5 billion people are experiencing acute food shortages. Oilseeds, such as sunflower, are important sources of vegetable fat and contain large amounts of calories, amino acids, vitamin D, minerals and antioxidants, making them essential in the diet. Sunflower is grown in most climate zones and is a crop rotation crop that restores soil fertility. Plants are susceptible to various diseases that directly affect the quality and quantity of the crop. To prevent the appearance of diseases and pests, it is necessary to recognize the problem in time and take measures. Timely diagnosis of plant diseases is very labor-intensive even in a small area. The work describes an ensemble system that includes Visual Transformer and deep convolutional networks, which can recognize sunflower diseases such as graymold, leaf scars and downy mildew with a 97.02% accuracy. The results showed the advantage of the ensemble system in comparison with known methods. The proposed system for visual diagnosis of sunflower diseases will help farmers quickly apply measures to eliminate pests and diseases to preserve the crop.