Fine-Tuning Model Transfer Learning VGG16 Untuk Klasifikasi Citra Penyakit Tanaman Padi

Abstract

The high level of demand for this commodity encourages improvements in agricultural yields by overcoming diseases in rice plants. Detection of disease in rice plants from the beginning of planting will reduce the impact of plant growth significantly. With proper treatment from the results of the identification of disease cases since this will increase the productivity of agricultural products. This study aims to analyze the performance of the rice plant disease classification convolution neural net (CNN) with the VGG16 architecture using fine-tuning. To process the dataset and classify the data into four classes (BrownSpoty, Healthy, Hispa, and LeafBlast), this study used several methodological steps. The stages include data preparation, feature extraction, training, comparing and evaluating models. As a result, VGG16 without fine tuning gets an accuracy of 50.88% while VGG 16 with fine tuning gets an accuracy of 63.50% in the training process. In the validation process, VGG16 without fine tuning gets an accuracy of 52.50% while VGG 16 with fine tuning gets an accuracy of 62.08%. In the testing process, VGG16 without fine tuning got 54.19% accuracy while VGG16 with fine tuning got 62.21% accuracy.