Classification of Banana Plant Leaves Based on Nutrient Deficiency Using Vision Transformer
Sumit Kumar Sharma, Dinesh Kumar Vishwakarma
Abstract
Banana is one of major fruit crop consumed in India and its production is merely impacted over the year by many factors where malnutrition is one of the key causes which impact banana plants growth and production of banana. Identification of particular type of malnutrition in early stage avoids the diseases and lower the impact of that particular malnutrition on the plant. Traditional monitoring of plants requires intensive time and relies on humans to judge the type of malnutrition suffered a plant, whereas computer vision techniques currently using in similar tasks such as diseases detection, computer vision also performs well on the discussed task which discloses in this study here we proposes a vision transformers-based solution to classify the leaves based on the particular type of malnutrition it has. Vision transformer models have low inductive biases which makes them not to focus on local regions but also consider global patterns in image which makes to produce better results over images taken from real world crop fields. Due to which classification task able to classify with accuracy of 90 percentage on the benchmark dataset which is taken from Maland university of India in Karnataka, which includes 7000 plus images of banana plant leaves when leaves are under deficiency of 8 unique nutrition.