International Journal of Science, Engineering and TechnologyAuthors: Battu Swarupa, Assistant Professor M.Prasanna Kumar
Abstract: Many foliar diseases may reduce the quality of yields and overall production, and citrus crops are especially susceptible to these diseases. Conventional manual diagnosis may be tedious, time-consuming, and subjective, especially when symptoms seem visually similar. By developing a deep learning system for the automatic detection and categorisation of diseases affecting citrus leaves using EfficientNet-B7, this study addresses a need in the current market. The model enhances generalisation across multiple disease categories, including canker, black spots, greening, melanose, and healthy leaves, by applying transfer learning, large data augmentation, and fine-tuning. To make it seem like there are real-world differences in lighting, orientation, and background conditions, the dataset is preprocessed using normalisation and augmentation. The model is improved by using AdamW and cosine annealing after it has been trained using weighted categorical cross-entropy. When looking at F1-scores, recall, precision, and accuracy within classes, the experimental findings show that it works very well. On average, 97.2% accuracy is attained across all disease categories. The effectiveness and scalability of the proposed technique are confirmed by comparing it to existing designs. Based on the findings, EfficientNet-B7 may be used to smart precision farming and agricultural diagnostics in order to make better use of available resources.
