TransMF: Transformer-Based Multi-Scale Fusion Model for Crack Detection

Cracks are widespread in infrastructure that closely related to human activity. It is very popular use artificial intelligence detect cracks intelligently, which known as crack detection. The noise the background of images, discontinuity and other problems make detection task a huge challenge. Although many approaches have been proposed, there still two challenges: (1) long complex shape, making it difficult capture long-range continuity; (2) most images dataset noise, only ignore noise. In this paper, we propose novel method called Transformer-based Multi-scale Fusion Model (TransMF) for detection, including an Encoder Module (EM), Decoder (DM) (FM). uses hybrid convolution blocks Swin Transformer block model dependencies different parts image from local global perspective. designed with symmetrical structure Module. Module, output each layer unique scales fused form convolution, can release effect strengthen correlations between relevant context order enhance Finally, concatenated achieve purpose Extensive experiments on three benchmark datasets (CrackLS315, CRKWH100 DeepCrack) demonstrate proposed TransMF paper exceeds best performance present baselines.