A vision transformer-based approach for recognizing seven prevalent mango leaf diseases

Plant diseases, particularly affecting fruit crops, pose a significant challenge to the worldwide supply of fresh food due to their direct impact on the quality of fruits, resulting in an overall decline in agricultural production. The traditional approach of detecting leaf diseases in fruit plants requires farmers to undertake manual inspection which exhibits a lack of reliability and consistency. Moreover, the manual inspection procedure is prone to errors due to its reliance on the farmer’s knowledge and skill. Mango referred to as the "king of all fruits", is renowned for its rich composition of various vitamins and vital nutrients. Mangoes are susceptible to many diseases that adversely damage their visual appeal, and flavor, and have significant implications on the overall economy. The identification of diseases affecting mango plant leaves using automated recognition remains a challenge due to the diverse range of symptoms and limited availability of data. There have been several deep learning-based research studies focused on identifying diseases in mango leaves; however, the majority of these studies have employed a convolutional neural network (CNN) trained on a small number of data. This study presents a Vision Transformer (ViT) based approach to detect diseases in mango leaves using publicly available data namely MangoLeafBD. The ViT model has been selected as the detection model due to its parameter efficiency compared to deep CNN models. The ViT has produced remarkable overall classification accuracy of 100%, precision of 100%, recall of 100%, and f1-score of 100% for disease detection on mango leaves which is better than the existing CNN approaches on the MangoLeafBD dataset. This demonstrates that our approach has the potential to assist farmers in the field by providing automated, simple, and more reliable mango leaf disease diagnosis.