Amila Fadhila Rahmaniati, Fitri Utaminingrum, Budi Darma Setiawan
Smart wheelchair navigation with camera based head movement navigation has challenges in multi-lighting conditions and high resource requirements. The purpose of this study is to improve the image-based head movement navigation features under varying lighting (bright, dim, and dark), by optimizing its performance and model complexity. This study compares the addition of spatial and channel attention to EfficientNetv2 version S with frozen layers to optimize the performance and complexity of the classification model in multi-lighting conditions. The results showed that EfficientNetv2 with spatial attention has better performance with an accuracy of 0.93 and a more efficient model with a total of 6,408 parameters. This result is compared to the EfficientNetv2 model with channel attention, which only has an accuracy of 0.79 and a number of parameters still at 2,465,605. This shows that spatial attention is better at significantly improving the performance of the EfficientNetv2 model, without increasing the complexity of the model. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.
Faculty of Computer Science, Universitas Brawijaya, Malang, Indonesia
