Elevating Crop Protection: Vision Transformer Models for Brassica Black Rot Detection

The agricultural sector is currently faced with the urgent problem of reducing the incidence of crop diseases to maintain food security and environmentally responsible farming practices. Among these, the Brassica black rot disease poses a considerable risk to crops in the Brassicaceae family, which includes several different cabbage and broccoli varieties. This study aims to revolutionize the diagnosis and classification of Brassica black rot disease at varied severity levels by tapping into the capabilities of cutting-edge deep learning (DL) techniques, more specifically ViT models. This study is based on a rigorously curated dataset that contains 5,000 high-resolution photos of Brassica leaves. These images have been categorized into six different severity levels, ranging from "Class 0" to "Class 5", and the researchers have annotated each of these images. The accuracy and reliability of the annotations were ensured under the oversight of seasoned plant pathologists. To aid in the evaluation and development of the model, a training set, a validation set, and a test set were meticulously collected from the dataset. As our research developed, the ViT-based model, a sophisticated DL architecture, became the key focus of our attention. Excellent accuracy was achieved in classifying the varying degrees of disease severity by meticulous training and tuning the model using the dataset containing cases of Brassica black rot. By incorporating Cohen's Upsilon (), a new performance indicator, we gained a more nuanced understanding of the model's propensity to correctly classify diseases. That was possible because of the additional context provided regarding the model's propensity to correctly classify diseases. Our research showed that the ViT-based model achieved a fantastic level of classification accuracy, with an average rate of 96.17%. In addition, the significant Cohen's Upsilon value of 0.94 demonstrated that the model was able to distinguish between different degrees of severity of Brassica black rot with a high level of concordance.