SSA-ECNet: Semantic Segmentation Architecture with Enhanced Cross-Attention Mechanism
Minghui Li, Zengmin Xu, Ruxing Meng, Lingli Wei, Weisen Luo
2024 7th International Conference on Advanced Algorithms and Control Engineering (ICAACE) (2024)
EI
Abstract
This paper presents a novel Convolutional Neural Network (CNN) semantic segmentation architecture for detecting water leakage defects in house images. Recent semantic segmentation architectures have predominantly focused on RGB images, where water leakage traces are often vague and surface features insufficiently distinct. Traditional semantic segmentation architectures exhibit insufficient edge clarity. These challenges have spurred the proposal of an enhanced model for multispectral image segmentation. To benchmark our approach, we established an RGB thermal dataset and devised a new fused image attention module to better extract features. Our findings indicate a significant improvement in segmentation accuracy by incorporating thermal infrared information.