VISION TRANSFORMER WITH PATCH ATTENTION FOR FINE-GRAINED BLACK GRAM LEAF DISEASE IDENTIFICATION AND SEVERITY ESTIMATION

Abstract

Black gram, commonly known as urad bean, is a crop that is popularly cultivated and has substantial economic value in India, particularly in the southern and central regions.But black gram is vulnerable to several leaf diseases, and farmers face significant crop losses and financial difficulties. One of the mostprevalent leaf diseases in black gram is yellow mosaic disease (YMD),whichhinders healthy productionand causes significant economic losses to local farmers.  The quality and quantity of blackgram are drastically decreased by this disease.Therefore, early and correct diagnosis is needed to control the disease appropriately and promptly.Plant leaf disease classification and identification have recently undergone a revolution thanks to deep learning-based pre-trained models.To resolve these problems, the study developed a Vision Transformer with Patch Attention for Fine-Grained Black Gram Leaf Disease Identification and Severity Estimation (VTPA-BGLDI) Model. To increase the number of image samples, data augmentation approaches were usedat the initial phase for effective training. In the second phase, we designed a VT-driven classification model specifically tailored for Blackgram disease identificationin analyzing yellow mosaic disease imagery.Before processing, the Vision Transformer (VT) splits the input image into smaller patches, which are then fed sequentially to the model in a manner similar to word embeddings.Lastly, the input imagesare classified into the appropriate class using a Multi-Layer Perceptron (MLP).The BPLD dataset comprises five distinct classes of plant disease images. Based on the experimental findings obtained from the BPLD dataset, we determined that the proposed VTPA-BGLDItechnique outperforms existing approaches in classifying Yellow Mosaic Disease in Blackgram.