Improving Human Character Recognition Performance Based on Facial Images with the Addition of Channel Attention Module in ResNet50 Model

Abstract

This study investigates the performance of several convolutional neural network (CNN) architectures in classifying human characters based on facial images, considering the addition of the Channel Attention Module (CBAM) to the ResNet50 model. This research aims to evaluate and compare the capabilities of the ResNet50 model with and without the addition of CBAM, and to compare it with the EfficientNetB3 and GoogleNet architectures in recognizing human characters based on facial images. This research uses an experimental approach by utilizing a tagged facial image dataset. Accuracy, precision, recall, and F1-score metrics are used to quantitatively evaluate the performance of the models. The addition of CBAM to the ResNet50 model successfully improves its performance in classifying human characters, especially in identifying the Savory and Unsavory classes. ResNet50 with CBAM demonstrates higher accuracy compared to ResNet50 without CBAM, and outperforms EfficientNetB3 and GoogleNet. This research indicates that the addition of CBAM to the ResNet50 model can enhance accuracy in recognizing human characters based on facial images. These results provide valuable insights into the importance of integrating enrichment techniques into CNN architectures. However, this research has limitations in dataset variation and further research is needed with more varied datasets and additional experiments to understand the factors that affect model performance more deeply.