Cross-Dataset Generalization in -Based Plant Disease Recognition

Plant diseases recognition large crop losses and have negative economic effects, which makes them a serious danger to the world's food security. Early and accurate disease diagnosis is essential for efficient disease control and mitigation. Convolutional neural networks (CNNs) are a viable method for diagnosing plant diseases based on visual symptoms because of their impressive performance in image recognition tasks in recent years. In this work, we present a novel CNN algorithm created especially for the diagnosis of plant diseases. The accuracy and resilience of disease classification are increased by the suggested CNN architecture's tuning, which is designed to extract the intricate patterns and variations of plant illnesses from photos. To enhance the model's capacity for generalisation, extensive datasets including various plant species and disease symptoms are gathered and subjected to pre-processing techniques that standardise image quality. Using a stochastic gradient descent technique on the pre-processed CNN weights, the training procedure The model's hyper parameters are fine-tuned to prevent overfitting and improve performance on a separate validation set. An independent test data set is used to thoroughly assess the trained CNN in order to calculate a confusion matrix and quantify accuracy, recall, F1 score, and precision.