Automatic blight disease detection in potato (Solanum tuberosum L.) and tomato (Solanum lycopersicum, L. 1753) plants using deep learning

Early and late blight are two diseases which pose a huge risk to both potato (Solanum tuberosum L.) tomato lycopersicum, L. 1753) crops make farmers run at loss. The early automatic detection of these would save time as well enable act quickly on have been affected. Machine learning deep technology provide many solutions for the in affected crops, common literature. However, explanation methods such not common, but necessary, considering machine models seen black boxes. This study proposes ResNet-9 model detects disease state leaf images, can leverage. With data obtained from popular “Plant Village Dataset”, there were 3990 initial training samples, increased 12009 after augmentation. A rigorous hyperparameter optimization procedure was followed, trained with values, examined test set, contained 1331 images. accuracy 99.25%, 99.67% overall precision, 99.33% recall f1-score values achieved. To fully understand model, explanations proposed provided through saliency maps, showed reasoning behind predictions model. It observed that considered shape leaf, diseased areas present general green its this makes us better see behaves expected. Our results could contribute testing deployment convolutional neural network (CNN) classification proximal sensing images plant leaves. Further studies benefit modeling framework chance several other variables determine infections an earlier stage crop protection.