Leveraging Deep Learning's Potential: A CNN and LSTM Network-Based Severity Classification of Mustard Downy Mildew

Mustard downy mildew poses a huge threat to global farming of the mustards for an effective approach towards disease management, timely and accurate detection is important. This study presents a novel method based on Convolutional Neural Network (CNNs) and Long Short-Term Memory (LSTM) networks that allows the classification of mustard downy mildew severity into five discrete levels. A high-resolution imaging dataset containing 12,000 images at different stages of disease progression was carefully created and used for model training and validation. The suggested model scored a highly satisfactory overall accuracy of 96.48%. Detailed analysis of precision, recall, and F1 score per class as presented in Table I suggested the model's predictive nature. It is important to note that the model had a high degree of accuracy at all levels of severity, indicating its ability to reduce false positives. The analysis of the confusion matrix. Gave a deep insight into the model's capacity to differentiate among severity levels, pointing out its practical applicability in agriculture environments. This is a breakthrough in the detection of mustard downy mildew since it develops an efficient automated tool for accurate severity classifications. Indeed, the success of this model creates opportunities for evidence-based crop management that is likely to lead towards sustainable agriculture and pave the way for more improvement in deep learning plant pathology.