Detection of plant leaf disease using advanced deep learning architectures

Food is the basic necessity for human survival and plant diseases act as a critical hindrance to the quality of harvested crops. Timely identification and administration of plant diseases are remarkably essential owing to the ever-increasing population and global climate change. Though many algorithms have been designed for early diagnosis of plant leaf diseases in existing literature, the bulk of those lack a large enough dataset for accurate detection and diagnosis. This work aimed to determine the significance and perform fine-tuning of state-of-the-art models to detect and classify diseases in plant leaves. These models are required to perform early detection of diseases to prevent the loss of crops. In this paper, a dataset containing 39 classes of diseased and healthy leaves of 14 plants is used. We perform fine-tuning of various deep learning architectures including VGG16, AlexNet, ResNet18, and MobileNetV2. The results were evaluated based on four different metrics vis-à-vis accuracy, F1-score, precision, and recall. The best results were received using MobileNetV2 with an accuracy of 94.4%.