Black Gram Leaf Disease Detection Model Using Combination of Hybrid-cnn Network and Transformer-based Classification Model

The king of pulses, black gram is also referred to as urad in India, where it has been grown since ancient times. The diseases that affect the leaves of the black gram plant are the reason behind the annual decrease in production of this crop, which is grown mostly in India. Therefore, detection of diseases in of utmost importance. Earlier detection and classification of plant leaf diseases utilized CNN based methods. These methods often constrained by economic losses making them less reliable. To overcome this issue, this paper presents a transformer-based model for identification of black gram leaves diseases. The proposed work utilized a hybrid CNN model which combines VGGNet and Inception-V3 for image feature extraction. The extracted features are further utilized by transformer-based classification network for efficient classification of plant leaf diseases for black gram. To validate the effectiveness of the proposed model extensive experiments on BPLD dataset are carried out. Further, to avoid data related and overfitting issues, the dataset was improved and increased to 15,000 images using different augmentation techniques. Also, the proposed hybrid CNN transformer model provided superior results on the BPLD dataset with an accuracy of 99.56%.