A Hybrid Deep Learning Framework for Robust Detection of Rice Leaf Diseases Using CNN and Transformer-Based Attention

Pathogenic microbiological organisms that invade the leaves of cereal crops such as rice could be a disease. In particular, in the areas where rice is the main food and economic crop, the rice leaf disease is a problem. Visual identification of crop diseases is becoming more and more difficult due to the following three reasons: This method is instrumentally demanding, making it not only expensive but also very slow, requires domain competence, and is subject to the human bias of experts in the form of wrong or inconsistent identification. Modern developments in the field of deep learning have significantly advanced the automated classification of plant diseases, particularly the use of Convolutional Neural Networks (CNNs). Therefore, the difficulty of these methods lies in the fact that most of the time they are subjected to the so-called phenomenon of overfitting in controlled situations, lack interpretability, and aren't robust enough in real-world situations with lots of plants but differing backgrounds, and variable lighting. But the main focus of this paper lies in the designof a hybrid deep learning method. In this work, we use CNN for spatialfeatures and transformer-based attention mechanism for modeling long-range dependencies among the leaf image data. In the proposal, by Dual-GAN-based image augmentation, the model is enhanced to make sure that the training dataset is still realistic andmimics actual field conditions. The experimental results in the rice leaf disease datasets that are publicly available and even verified through real field images have shown the proposed technique outperforming standard CNN architectures like ResNet, VGG16, and DenseNet, in terms of both classification accuracy, at 97.8%, and the ability to generalize to multiple disease categories. The new method offers both a verifiable and a scalable response to rice disease detection that has great potential to be used in precision agriculture systems, such as the surveillance of UAV connected to mobile farming applications.