Leveraging Squeeze Aggregation Excitation and Positional Encoding in ResNet‐50 for Rice Leaf Disease Classification

ABSTRACT Approximately fifty percent of the world's population depend on rice as a primary food source. This situation illustrates the importance of consistent and sustainable rice cultivation for food security. A significant issue in rice cultivation is the prevalence of leaf diseases, which can substantially impact plant growth and yield. Rapid and precise identification of these diseases is crucial to mitigate crop losses and reduce excessive pesticide application. Historically, disease identification through manual inspection was based on expert visual evaluation, which is challenging, subjective, and difficult to implement on a broad scale in agricultural environments. This work proposes an advanced deep learning model named ResNet‐50+SaE‐PE, a modified edition of the ResNet‐50 architecture, to address this issue. This model uses positional encoding (PE) to improve its ability to give more attention to spatial information and adds squeeze‐aggregatio‐excitation (SaE) modules to better focus on important features. These enhancements facilitate the identification and classification of disease‐affected regions in rice leaves. The proposed model has been tested on a publicly available standard Rice Leaf Disease dataset consisting of 4624 images. We partitioned the dataset into 80:20 ratios for training and testing, respectively. The performance of the model has been evaluated using accuracy, recall, specificity, precision, and F1‐score. The modified version of ResNet‐50+SaE‐PE shows that it is better at being accurate, precise, and reliable compared to older models. After experimenting with several optimizers, Stochastic Gradient Descent (SGD) with a learning rate of 0.001 emerged as the most reliable option. The average accuracy of 99.89% indicates that the model is effective for use in intelligent and scalable systems to monitor agricultural diseases.