Analysis of Using Regularization Technique in The Convolutional Neural Network Architecture to Detect Paddy Disease for Small Dataset

Abstract

In some instances, convolutional neural network (CNN) methods such as large numbers of datasets can show very high accuracy. However, in cases such as the small number of datasets, the accuracy performance of CNN often decreases. This also intersects with the constraints of applying CNN to recognize types of diseases in rice with a small number of datasets. This research applied the combination of regularization techniques and CNN methods to recognize rice disease types with a total dataset reaching 120 leaf images. It is found that using CNN-regularization techniques shows better performance than standard CNN architecture. With an accuracy value reaching an average of 85, 878%, it is expected to open further research opportunities.