EAF-Net: An Image Enhancement Method and Adaptive Fusion Network for Crack Detection

Read the full article See related articles

Listed in

This article is not in any list yet, why not save it to one of your lists.
Log in to save this article

Abstract

Pavement crack detection is a critical and challenging task in infrastructure maintenance. To address these challenges, we introduce EAF-Net (an image enhancement method and adaptive fusion network). EAF-Net incorporates three advanced modules: the hybrid enhanced convolution module (HECM), the multi-branch attention module (MBAM), and the dynamic fusion module (DFM). The HECM leverages hybrid convolutional operations and adaptive feature weighting to capture detailed crack information. The MBAM enhances the model's capability to process intricate crack patterns. The DFM dynamically integrates features from five network stages. We rigorously evaluate EAF-Net on three crack detection datasets: DeepCrack, Crack500, and CFD. Experimental results show that EAF-Net consistently outperforms existing methods across all datasets. Specifically, on the DeepCrack dataset, it achieves an F1 of 0.881 and an MIoU of 0.888; on the CFD dataset, it scores an F1 of 0.625 and an MIoU of 0.721; and on the Crack500 dataset, it gets an F1 of 0.754 and an MIoU of 0.787.

Article activity feed