Abstract: The off-peak night period is the prime time for bridge defect inspections. Traditional computer vision technology suffers a significant drop in crack recognition accuracy in dark nighttime environments due to insufficient illumination. To address this issue, this study proposes a bridge deck crack recognition method based on pressure-type bionic tactile perception for extremely dark environments. Drawing on the biological tactile perception mechanism, a pressure-type bionic tactile sensor is designed. The deformation generated by the contact between the flexible silicone material and the bridge deck, combined with an RGB three-light source system to capture reflected light changes, and the normal integral algorithm reconstructs a 3D surface height map of cracks, achieving tactile perception completely decoupled from ambient light. On this basis, a deep learning model TSBC-Net (Tactile Segmentation of Bridge Deck Cracks-Net) is constructed. Adopting a U-shaped architecture, the model effectively improves the ability to extract crack texture features from tactile images by integrating residual blocks, Dilated Gated Attention (DGA), Dual Vision Mamba (DVM), Efficient Atrous Spatial Pyramid Pooling (EASPP), and Dual Level Decoder (DLD) modules. Experimental results show that the designed tactile sensor can accurately restore bridge deck textures. Compared with traditional semantic segmentation models, the segmentation accuracy has been significantly improved, taking into account both model lightweighting and inference efficiency. This study provides a new solution for bridge deck crack recognition in extremely dark environments.