Convolutional Neural Network Untuk Perbandingan Optimizer Pada Citra Batang Pohon

Abstract

In the surrounding environment, there are various types of trees with different characteristics. One characteristic of a tree that is difficult to distinguish is its trunk. After researchers made observations, the trunks of Pine and Tabebuya trees had the same characteristics, namely cracking. The problem of incorrectly identifying the characteristics of a tree's trunk can be overcome by classification. Deep Learning with the Convolutional Neural Network (CNN) algorithm is a method commonly used in image classification. The stages in this research include image data retrieval, data preprocessing, CNN architecture formation, model training, and model validation. Image retrieval was carried out directly by researchers, then the 1000 best images were selected. The image dataset is then divided into 75% training data and 25% validation data. Testing was carried out by comparing the Stochastic Gradient Descent (SGD) optimizer and the Adaptive Learning Rate Optimization Method (RMSprop) using epochs 10, 15, 20, 30, 50, and 80. The results showed that the SGD optimizer produced the highest accuracy compared to the RMSProp optimizer. The most optimal result when applying the SGD optimizer is 0.9360 with epochs 80, while for the RMSProp optimizer it is 0.9160 with epochs 20.