Enhancing the Performance of ResNet50V2 for Rice Leaf Disease Detection Using Squeeze-and-Excitation (SE) Blocks

Rice is eaten by millions of people worldwide, especially in the Asian region and its production is prone to plant diseases that affect production, food security, and economic returns. Due to this, several major diseases such backlash the production of rice; Bacterial Blight, Blast, Brown Spot, Scald, Narrow Brown Leaf Spot, Neck Blast, Rice Hispa, and Sheath Blight encourage the need for early and efficient disease identification. In this study, we integrated the Squeeze-and-Excitation (SE) block to improve the accuracy of the ResNet50V2 deep-learning model for rice disease identification. Using the Kaggle rice leaf diseases dataset, we trained the model on a dataset of 2880 images and then tested it on 720 images, utilizing the SE block to enhance the model's focus on the features that matter, by channel-wise attention recalibration. Among the outputs, we see that the modified model testing accuracy is 93.33, original ResNet50V2 accuracy is 83.61. The new model had a few more parameters (25,140,368 versus 24,620,289) than the prior version, but that minor increase was greatly outshined by the new model's diagnostic accuracy. The results indicate that the inclusion of the SE block enhances the performance of the models with almost no additional computation, therefore serving as a promising approach for agricultural disease diagnosis, especially in regions with limited resources.