Plant Leaf Disease Segmentation and Classification Using Channel and Spatial Attention Block With UNet Model

ABSTRACT Agriculture plays a crucial role in global food security, yet the health of crops is continuously threatened by plant diseases, particularly leaf diseases that reduce crop yields and quality. Accurate early detection of these diseases is vital for minimizing losses and promoting sustainable agricultural practices. While existing deep learning‐based approaches have shown promise in disease identification, they often struggle with precision and computational efficiency. This paper proposes a robust plant leaf disease segmentation system using a Channel and Spatial Attention Block (CSAB) integrated with a UNet model. The system leverages pre‐processing with a Median Filter (MF) for noise reduction and feature enhancement. CSAB introduces dual attention mechanisms—channel‐wise and spatial‐wise—ensuring that the model focuses on critical disease‐affected regions, thereby improving segmentation performance. Furthermore, the Black Kite Optimizer (BKO) is applied for hyperparameter tuning to enhance the model's convergence speed and accuracy. Extensive experiments conducted on the Leaf Disease Segmentation dataset show that the proposed system outperforms traditional segmentation models, achieving a high accuracy of 99.5% and an F‐score of 99.2%. Despite these promising results, the model's evaluation is limited to a single dataset and a narrow range of disease categories. Future work will address these limitations by expanding the dataset and optimizing the model for real‐time applications in various agricultural environments.