Hyperparameter Optimalization on Transfer Learning Method to Classify Mango Varieties

Mango is a garden plant that comes from Asia and has many varieties. Mango fruit varieties can be identified through the characteristics of fruit shape, fruit color, and leaf texture. However, these characteristics have the same tendency so it is difficult to distinguish for some people. This research aims to classify mango fruit variety that implement convolutional neural networks method with transfer learning using DenseNet201, InceptionResNetV2 and Xception architecture. The datasets consist of 1600 images divided by 8 classes. To get the highest accuracy, Experiments were carried out on different transfer learning architectures, input shape, optimizer algorithms, and learning rate values. The result of this study is a machine learning model with 100% testing accuracy obtained through the use of DenseNet201 architecture, Adam optimizer and 0.0001 learning rate value.