Hyperparameter Tuning of Convolution Neural Network for Paddy Leaves Disease Detection

Abstract

The traditional approach to identifying paddy leaf diseases in Malaysia's agriculture sector involves visual inspection by farmers or agricultural experts. However, this method heavily relies on human expertise to recognize visual symptoms like discoloration, spots, or lesions on the leaves, leading to subjective assessments and delayed diagnosis. While various deep learning models have been utilized to detect paddy leaf diseases, there is a need for a more comprehensive discussion on the appropriate hyperparameters setting. This study focused on utilizing Convolutional Neural Network (CNN) models, specifically ResNet and VGG architectures, to identify paddy leaf diseases in Sekinchan, Selangor. The models were trained on a dataset consisting of both online and real-world images, which underwent several augmentation processes to enhance the model's robustness. The study has highlighted the optimal hyperparameters of the CNN model, resulting in an impressive accuracy rate of 98.71% and a minimal loss rate of 0.001. The integration of the final model into a specific Graphical User Interface has further enhanced its utility.