A Hybrid Model for Enhanced Detection of Microbial Diseases in Rice Plants Using ResNet50 and Vision LSTM

In this paper, a deep learning approach is used to design a rice plant disease diagnosis model. In this approach hybridization of ResNet50 is done with vision LSTM (ViL). The high level features are extracted out of ResNet50 and fed into ViL for further classification of rice plant disease classification caused due to microbes. Cascading ResNet50 with a ViL combines the strengths of both architectures to enhance image classification. ResNet50 extracts the spatial features and patterns and then Vision LSTM shows the sequential and spatial relationships between image patches through positional embeddings and LSTM layers. This hybrid approach is designed to preserves spatial information with reduced computational complexity and higher accuracy for such computer vision applications. The result analysis shows that the proposed ResNet50+ViL shows an performance accuracy of 91% and also outperforms better over state-of-the-art methods. This shows that the proposed model is robust and efficient model for rice disease detection.