Sudhir Kumar, Man Mohan Deo, Kuldeep Kumar, Meenal Rathore, Mohd. Akram and Aditya Pratap
Abstract
The yield of the black gram crop is negatively impacted by Yellow Mosaic Disease (YMD). Both quantity and quality of the black gramsuffer significantly from this disease. Accurate diagnosis, flawless identification, and early detection guide the grower for proper and timely management of the disease. Deep learning-based pre-trained models have revolutionized the classification and identification of plant leaf disease in recent times. In the present study yellow mosaic disease of black gram has been classified using four deep learning models namely; DarkNet-19, SqueezeNet, AlexNet, and GoogLeNet. A total of 1100 images were collected from field experiments for each of three classes namely healthy, moderate, and susceptible plants. During the field investigation, datasets with images of three classes; healthy, moderate, and susceptible were collected, pre-processed, and augmented to create a set of 1100 images of each class. Seventy percent of the images were used for the training of the models and thirty percent of the images were used for validation. The results obtained for different deep learning architectures DarkNet-19, SqueezeNet, AlexNet, and GoogLeNet showed validation accuracy and loss scores of 96.09, 60.74, 94.41, and 93.85%, and 0.2319,0.6923, 0.2429 and 0.1399, respectively. For the YMD classification in black gram, DarkNet-19 showed the highest accuracy and Squeeze Net showed the lowest accuracy.