Multi-Head Attention-Based Transfer Learning Approach for Potato Disease Detection

Potatoes are widely consumed all over the world. Being one of the most cultivated crops around the world, they also attract various diseases. Hence, the early identification of such diseases using machine learning-based automated methods, is necessary. This paper proposes the solution for the early detection of two most commonly occurring diseases in potato leaves, i.e. Early blight and Late blight. In this work, a VGG16 model has been fine-tuned with a multi-head attention layer for identifying useful patterns for the classification of potato plant leaf diseases. The multi-head attention mechanism is useful since it can capture the relationship that exists in different parts of an input potato disease leaf image. The proposed model has attained an accuracy of 91 % with an F1-score of 0.9103. The better performance of the proposed model is a testimony to its effectiveness in the early identification of potato leaf disease.