Transfer Learning-based Deep Neural Networks for Rice Leaf Disease Classification: A Comparative Study of Squeeze Net and Google Net

Rice is considerably impacted by various diseases, demands an early and precise diagnosis to ensure sustainable agriculture. The adaptability and scalability of traditional machine learning approaches are restricted by the use of handcrafted features. This paper analyzes deep neural networks based on transfer learning for the classification of rice leaf diseases, with particular focus on the Squeeze Net and Google Net architectures. Model's performance was assessed using a dataset of 4,834 images from field and repository data. The three main rice diseases—bacterial leaf blight, brown spot, and leaf smut—were successfully classified using both networks. Experimental results demonstrate that Google Net achieved 98.29% accuracy, outperforming Squeeze Net at 97.86%, due to its deeper architecture and multi-scale feature extraction. However, Squeeze Net, with its lightweight design, provides comparable accuracy while being computationally efficient, making it more suitable for realtime deployment. This paper establishes transfer learning as a robust approach for intelligent paddy disease diagnosis in precision agriculture.