期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2021
卷号：99
期号：2
页码：279
出版社：Journal of Theoretical and Applied
摘要：Early diagnosis of plant diseases carried out by experts in laboratory tests is often not applicable for fast and inexpensive implementation. Using deep learning, leaf images are used as data input. Training deep learning models require large, hard-to-come datasets to perform the task to achieve optimal results. In this study, the PlantVillage dataset was used totaling 2700 training data and 300 validation data. Data were trained using 100 epoch iterations using the transfer learning method with the VGG16 and InceptionV3 models. At the top layer of both models, the same MLP is applied to several parameters, namely the size of FC and the dropout rate to compare the model's performance. Based on testing using 150 IVEGRI data, the VGG16 model can generalize data better than InceptionV3. VGG16 by tuning block-3 using parameters 4096x2 and Dropout 0.4 shows the best performance with an average score of 1 precision, an average recall of 1, an average f1-score of 1, and 100% accuracy. Then, with the same parameters, the Inception-v3 model with tuning in the mixed6 inception module shows the best performance with an average score of 0.93 precision, an average recall of 0.92, an average f1-score of 0.92, and an average accuracy of 92%.
关键词：Deep Learning; Transfer Learning; VGG16; InceptionV3; Potato Leaf Diseases Classification
