CLASSIFICATION OF ROSE DISEASES THROUGH LEAF IMAGE USING CONVOLUTIONAL NEURAL NETWORK METHOD

• Main Author: Safitri, Ira
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/68972
• Institution: Institut Teknologi Bandung

Classification of rose diseases is one of the keys to improving the rose population and provides useful knowledge about growth rates and development, and cultivation of rose plants. Methods and classification of diseases Rose planting is done manually and traditionally which takes a long time. Therefore, an automated approach is needed for cost and time so that the process is fast. Convolutional Neural Networks (CNN) is one of the artificial intelligence methods that can analyze images. In Agricultural Science, Convolutional Neural Networks have been used for the classification of flower plant species using pictures. This research used the Convolutional Neural Networks method with a transfer learning model. Models used the VGG16 and Resnet50 architectural models which were tested using the Adam optimizer and RMSProp with 50 and 100 epoch variations which aim to get better results efficiency and high accuracy. The dataset used is 4342 rose leaf image data in png format which consists of 3 types, namely black spot, downy mildew, and fresh leaf. Study It uses 80% training comparison, 10% validation, and 10% testing done offering imagery by GPU from Google Colab Pro. The model performance results in The best in testing using the transfer learning model is ResNet50 with epoch 100 using adam optimizer, learning rate 0,0001 and batch size 32, and dropout 0,5. The resulting ResNet50 model has an accuracy value of 100%, a loss of 1.486x10-5, and an F1 the score reached 1,00 from the classification of 3 types of rose plant disease classes. While VGG16 research model got the best model with epoch 100 using optimizer RMSProp with a learning rate of 0,0001 and a batch size of 32 and a dropout of 0,5. The model has an accuracy value of 99,77%, a loss of 0,0549 and an F1 the score of 0,99. In this study, the model ResNet50 architecture is superior to the VGG16 architectural model because ResNet50 produces a faster accuracy value than VGG16. The addition of epoch affects the level of accuracy produced.