Road crack detection presents critical challenges, including diverse defect patterns and complex anomaly characteristics. The current object detection algorithms demonstrate deficiencies in considering feature redundancy across channel-spatial dimensions, employ indiscriminate fusion strategies for multi-stage feature information, and particularly neglect the high-frequency characteristics inherent in crack features, leading to inefficient network performance and a loss of crucial information. Building upon the identified limitations, this paper proposes a dual-domain perception gate-controlled adaptive fusion network (DP-DETR) that achieves dynamic perception of salient features across channel and spatial domains within latent space. To enhance focus on critical features, a dual-domain dynamic perception information distillation mechanism is constructed, which distills redundant features separately across channel and spatial domains, effectively reducing architectural processing redundancy while achieving discriminative characteristic representation efficiency. In order to address the challenge of coarse-grained fusion in multi-stage feature integration, a feature information gating-adaptive fusion module (FGAF-Fusion) is proposed, which facilitates interactive channel-spatial information fusion through mixed local channel attention while employing gated adaptive fusion operations to selectively retain critical semantic information of small-scale targets. In response to the persistent high-frequency signature identified within crack feature distributions, a dual-domain structural feature enhancement loss function is designed, which elevates the weighting of high-frequency information by leveraging a spectral weighting matrix, while complementarily enhancing crack edge texture features in the spatial domain through gradient map integration. The experimental results obtained on the public RDD2022 dataset demonstrate that the proposed DP-DETR (Dual-Domain Perception Gate-Controlled Adaptive Fusion Network) approach mAP50 and mAP50:95 values of 54.2% and 25.8%, respectively, representing improvements of 6.7 and 4.2 percentage points over RT-DETR. In road crack object detection tasks, the proposed DP-DETR method can effectively detect various types of road defects, demonstrating highly competitive detection results and good robustness. The code will be released at https://github.com/jiangsu415/DP-DETR .