<p>Asphalt pavement cracking is a common problem that significantly impacts road durability and safety. Accurate and efficient crack detection is of paramount importance to road maintenance and management. This paper proposes an improved network architecture called ShuttleCrackNet, which is derived from the original ShuttleNet, to enhance global modeling and detail retrieval capabilities. Three major modifications are introduced in ShuttleCrackNet. First, efficient attention mechanisms are adopted to enhance focus on cracks, reduce attention on background, and improve the representation capability of crack features. Second, a feature pyramid-like structure is added to enhance the network's ability to learn complex features, resulting in a more precise crack identification. Finally, the self-attention module that serves as a critical component of Transformer networks is deployed in the proposed ShuttleCrackNet to capture essential global contextual relationships and enhance generalization capability. Experimental results on 1500 test images demonstrate that ShuttleCrackNet yields an F-measure of 91.97% and an intersection-over-union of 85.56%. Performance evaluation on public and private datasets indicates that the proposed ShuttleCrackNet outperforms other efficient networks, including SegNet, PSPNet, U-Net, DeepLabV3 + and ShuttleNet, in terms of crack detection, particularly in addressing crack continuity and detecting fine cracks.</p>