Corn Plant Disease Detection Using CNN Model with Resnet50 Architecture

Abstract

Plant diseases are a significant problem in agricultural production. Plants affected by the disease can experience growth disorders, declining yields, and declining quality. This study aims to detect diseases in corn plants using a Convolutional Neural Network (CNN) model with ResNet50 architecture. Several scenarios with hyperparameter variations are tested to determine their effect on model accuracy. The first scenario using Adam's optimization algorithm, GlobalAveragePooling2D operation, dropout 0.5, and batch size 64 resulted in an accuracy of 85.97%. The second scenario uses the Flatten operation and results in 85.45% accuracy with a 0.5 dropout and 87.54% with a 0.2 dropout. The use of the SGD optimization algorithm in the third and fourth scenarios resulted in an accuracy of 61.30% and 60.09%, respectively. However, in the fifth scenario, with a dropout of 0.2, the accuracy increases to 73.25%. The results show that hyperparameter variations have a significant influence on model performance.