Implementing Deep NN for Plant Disease Detection and Diagnosis

India, known for its diverse agricultural climatic zones and rich agricultural heritage, holds a crucial place in global food production. However, the agricultural sector confronts numerous challenges, with a significant menace being the proliferation of plant diseases. These diseases not only imperil food security but also jeopardize the livelihoods of millions of farmers nationwide. The extensive geographical and climatic variations in India foster a conducive environment for a plethora of plant diseases. Fungal, bacterial, viral, and nematode infections present substantial threats to various crops, including staple cereals such as potato, tomato, cotton, and sugarcane. Prominent among these concerns are diseases like rust, blight, wilt, and leaf spot, which, if not effectively managed, can result in substantial yield losses. This research evaluates the capability of deep learning (DL) models to generalize in identifying plant diseases across diverse datasets and environmental conditions. The study concentrates on foliar disease images in potato and tomato, utilizing datasets such as PlantVillage, PlantDoc, Northern Leaf Blight (NLB) dataset, and a customized CD&S dataset. Multiple DL-based image classification models underwent training and assessment using various combinations of these datasets. Transfer learning was employed, utilizing pre-trained deep neural network (DNN) architectures—including InceptionV3, ResNet50, VGG16, DenseNet169, and Xception—in four distinct experiments. Following model training, the test data comprised images of different potato and tomato diseases from various datasets to gauge the generalization ability of each DL model. The outcomes revealed that the DenseNet169 model exhibited superior performance.