EB-UNet++: An enhanced crack segmentation network combining EfficientNet-B2 and UNet++ with boundary extraction module

Pavement crack detection is a crucial task in intelligent transportation systems and infrastructure maintenance. However, accurate segmentation of cracks remains challenging due to their irregular shapes, low contrast against the background, and varying lighting or surface conditions. In this study, we propose EB-UNet++, a novel deep learning architecture designed to enhance crack segmentation performance. EB-UNet++ integrates the powerful feature encoding capabilities of EfficientNet-B2 into the UNet++ encoder structure, enabling more efficient and robust multi-scale feature extraction. To further refine the crack boundaries and suppress false detections, we incorporate a Boundary Extraction Module into the network. Experimental results on benchmark pavement crack datasets demonstrate that EB-UNet++ outperforms several state-of-the-art models in both segmentation accuracy and boundary delineation, achieving higher IoU and F1-scores. The proposed architecture shows strong potential for practical deployment and scalability in automated road inspection and infrastructure monitoring systems.