【摘要】 【目标】隧道裂缝漏水一直是工程维护中的关键难题。复杂的隧道环境，如低光照、粉尘干扰及表面反光等因素，导致裂缝漏水区域特征难以准确识别，传统检测方法效率低下且受人为因素影响较大，难以实现快速、精准的检测。针对这些问题，提出了一种基于YOLO11s优化的目标检测模型，旨在提高隧道裂缝漏水检测的效率和准确性。【方法】首先，将一种高效的自注意力机制ESSA集成至C3k2模块的Bottleneck中，设计而成一种全新的特征提取模块C3k2-ESSA,增强渗漏区域的特征表达能力。然后，使用风车形卷积替换颈部网络的普通下采样层，通过其独特卷积核结构强化渗漏区域的高斯分布特征提取，同时抑制背景噪声。最后，引入变化边界感知模块增强裂缝漏水区域的边界信息，通过多尺度特征融合增强裂缝漏水区域的轮廓表征，减少误检与漏检。【结果】在公开数据集上的系统评估表明，改进后的模型相比于原YOLO11s模型，AP0.5,AP0.75和AP0.5∶0.95分别提高了3%,6.5%和4.1%,满足实时检测的需求。【结论】与其他先进模型的对比试验进一步证明了改进模型的有效性，为隧道维护和安全管理提供了有力的技术支持。更多还原

【Abstract】 [Objective]Tunnel crack leak has been a critical issue in engineering maintenance. The complex tunnel environment(e.g., low illumination, dust interference, and surface reflection) makes it difficult to accurately identify the characteristics of crack leak area. Traditional detection methods suffer from low efficiency, significantly affected by human factors. To address these problems, an improved target detection model based on YOLO11s is proposed to improve the efficiency and accuracy of tunnel crack leak detection. [Method]First, an efficient self-attention mechanism, ESSA, was integrated into Bottleneck of C3k2 module, resulting in a novel feature extraction module named C3k2-ESSA. It strengthened the feature representation capability of leak area. Second, the ordinary downsampling layers in neck network were replaced with windmill-shaped convolutions. Their unique kernel structure improved the extraction of Gaussian distribution features in leak area. Simultaneously, it suppressed background noise. Finally, a variable boundary awareness module was introduced. It strengthened the boundary information of crack leak area. Multi-scale feature fusion was employed to enhance contour representation, reducing false and missed detections. [Result]The systematic evaluation on a public dataset indicates that the improved model shows significant gains compared with the original YOLO11s model. AP0.5 increases by 3%, AP0.75 increases by 6.5%, and AP0.5∶0.95 increases by 4.1%, meeting the requirements for real-time detection. [Conclusion]Comparative tests with other state-of-the-art models further validate the effectiveness of the proposed model. It provides technical support for tunnel maintenance and safety management.更多还原