Detecting Plant Diseases Utilizing a Convolutional Neural Network and Transfer-Learning Methodology

Timely detection of diseases in plants significantly impacts the production of high-quality and abundant grains. These precision techniques have shown impressive results in plant leaf-based image classification. This paper uses the Convolutional Neural Network and Transfer-Learning application EfficientNetB7 to identify fresh and diseased leaves. Various parameters such as accuracy, block size, dropouts, loss, and time required for epochs were incorporated to appraise the model performance. This paper consists of several phases Implementation, i.e., creation of datasets, feature extraction, and training the classifier. Image dataset from Kaggle village plant datasets used for training and testing which contains several fresh and diseased leaves images. The deep CNN model is promising accurate results in identifying diseases has a notable impact on effective disease identification and could potentially enable disease detection in agricultural systems in real time. EfficientNetB7 compared to other Transfer-Learning applications such as ResNet, Inceptions, MobileNet, and DenseNet has a smaller number of parameters and achieved 99.89% performance accuracy is also higher than others.