Abstract

Agriculture remains the cornerstone of global food security, and olives are at the core not only as a food staple but also in cosmetics, pharmaceuticals, and other industrial applications. However, the presence of olive trees risks hindering agricultural productivity and economic stability severely, and this necessitates the creation of better and more reliable diagnosis solutions. As a response to such challenges, deep learning-based methods have been successful plant disease identification tools with great success in recent years.

We have designed a hybrid deep learning model for computer-aided diagnosis of olive leaf disease. The first two stages of the model employ Inception Next blocks, which employ filters of varying sizes to extract automatically multi-scale features caused by disease symptoms in olive leaves. This permits a wide feature extraction, such as fine structural information and large regional variations. Transformer blocks are employed to implement a self-attention mechanism to establish dependencies and important regions in the input data. The proposed hybrid Convolutional Neural Network (CNN) and Vision Transformer (ViT) deep learning model for olive leaf disease auto-diagnosis is to employ separable self-attention to improve both computation efficiency and classification accuracy.

The approach was experimented with a publicly available downloadable dataset, which contains olive leaf images collected in spring. To perform a robust evaluation, the dataset was split into 70% training, 15% validation, and 15% testing. This balanced division of data ensures that the model is exposed to the right training and that there is sufficient validation and test samples to provide an objective performance evaluation.

The hybrid model achieved an impressive 98.23% classification accuracy outperforming state-of-the-art 15 trained models such as ConvNext-Base, DenseNet169 and Inceptionv4. Besides the increased level of accuracy, the model was also more successful in terms of precision, recall, and F1-score, which indicates that the model can be considered highly effective in the detection of the olive leaf disease. These results suggest a potential of hybrid deep learning models in crop diagnosis to provide quality, scalable, and automated detection of plant diseases aiming at increasing the model performance, expanding its usage to additional diseases in plants, and implementing the real-time detection capability.