A hybrid deep learning approach integrating capsule networks and BiLSTM for plant leaf disease classification

Plant leaf disease classification presents significant challenges due to the extensive variation in disease symptoms and the diverse morphological characteristics of plant leaves. These variations complicate model training and hinder classification accuracy. This study proposes a hybrid deep learning (DL) model that combines SE-Residual blocks for feature enhancement, capsule networks (CN) for preserving spatial relationships, BiLSTM for processing sequential data, and attention mechanisms (AM) for feature prioritization, aiming to improve classification performance. SE-Residual blocks enhance feature extraction while minimizing information loss, and CN capture spatial relationships with reduced dependency on large datasets. BiLSTM processes sequential data, supported by AM to focus on critical features. The proposed model was trained and evaluated using the corn leaf disease dataset (CLDD) and the rice leaf disease dataset (RLDD). Its performance was compared with existing state-of-the-art models. The experimental results demonstrate that the proposed model achieved the highest training accuracy of 99.88% for CLDD and classification accuracies of 99.29% for blight, 100% for common rust, 100% for leaf spot, and 100% for healthy samples. Additionally, it achieved the highest training accuracy of 100% for RLDD and classification accuracies of 78.95% for bacterial leaf blight, 81.58% for brown spot, 89.77% for healthy, 77.27% for leaf blight, 100% for leaf scald, and 97.73% for narrow brown spot. These results highlight the effectiveness of the proposed model in achieving high accuracy for plant leaf disease classification.