An Ensemble Convolutional Neural Network Approach for Image Classification of Indonesian Endemic Fruits

Indonesia has a diverse range of endemic fruits that grow in its various regions. These fruits have their own distinctive characteristics, which can sometimes lead to confusion in the sorting process. Classification can be used as a solution to this problem. Several similar studies have classified fruits; however, there has been no research specifically using deep learning methods for Indonesia’s endemic fruits. The designed system is expected to classify fruits accurately based on their unique characteristics. The classification models used consist of three CNN architecture models: AlexNet, ResNet-50, and InceptionV3, which are then combined with an ensemble method. Each model is compared by evaluating the use of transfer learning and without it. The three models with the most optimal results are implemented in an ensemble application. The best results were obtained from the AlexNet model, with an accuracy of 99.67%, the InceptionV3 model, with an accuracy of 99.81%, and the ResNet-50 model, with an accuracy of 100%. All three models are implemented in an ensemble using the majority voting method. The results of the ensemble implementation yield an accuracy of 100% on the test dataset.