FLVAEGWO-CNN: Grey Wolf Optimisation-Based CNN for Classification of Olive Leaf Disease via Focal Loss Variational Autoencoder

Abstract

Agriculture is crucial for the economic stability of developing countries, with olive trees in the Mediterranean region providing significant benefits through olive oil and table olive production. However, olive trees are susceptible to diseases that threaten their productivity. Traditional methods of disease detection are time-consuming and impractical on a large scale. Advances in artificial intelligence (AI) and machine learning now offer efficient solutions for rapid and accurate disease identification, which can enhance disease management and increase yields. This study introduces the FLVAEGWO-CNN, an innovative deep learning model designed to improve classification accuracy of olive diseases, particularly in dealing with imbalanced datasets. The model integrates focal loss, variational autoencoders (VAE), grey wolf optimisation (GWO), and convolutional neural networks (CNNs) into a unified framework. The focal loss component addresses class imbalance by assigning more weight to hard-to-classify examples, while the VAE component improves data representation. GWO optimises the CNN's hyperparameters for robust performance. The FLVAEGWO-CNN model was evaluated on a dataset with significant class imbalance, achieving an exceptional accuracy of 99.2% in binary classification and excelling in multiclass classification, particularly in recognising minority classes like ‘Aculus olearius’. The results suggest that this model outperforms existing models and provides a viable solution for imbalanced datasets in classification tasks. To ensure the model's validity, further investigation into potential challenges like overfitting and generalisability is necessary. Future work will focus on validating the model across diverse datasets and refining its architecture. The FLVAEGWO-CNN model sets a new standard for accuracy and reliability in deep learning-based disease classification, with implications for various applications, including medical diagnosis and fraud detection.