Tutorial and applications of convolutional neural network models in image classification

Abstract

Image classification is a supervised learning problem in the machine learning area. We apply deep learning models to classify image data. In particular, we discuss the advantages of the various types of convolutional neural networks competed in the ImageNet large-scale visual recognition challenge (ILSVRC). First, we provide a review of the CNN models to be applied and explain the details of models to be employed. In general, we keep the core structure of the models in the same form proposed in ILSVRC. We investigate the models via four popular image data sets of various sizes. To compare the performance of the models, we adopt top-1 accuracy, top-5 accuracy, and f1-score as the measures of accuracy. We employ AdamW for an optimizer that is a fast algorithm and often yields precise learning. As a result, we show that the Inception-ResNet-v2 model has excellent performance, and the ResNet is robust to imbalanced data.