目的 在桥梁缆索和隧道环境中进行表面病害检测面临多重挑战：人工病害定位在复杂结构中的高投入成本；隧道内存在光照不足，病害区域的表面特征复杂，范围广阔和易遭遮挡等问题，这些因素使得传统图像处理技术在病害区域检测上表现出较低的抗干扰能力、识别精度不足和分割效果欠佳，本文因此提出一种基于融合注意力及增强感受野的深度网络模型。方法 该模型使用融合Transformer注意力机制的骨干网络提取目标特征信息，获得更为密切联系的全局特征表示，以解决光照不足导致局部特征缺少的问题；引入空间序列缩减法降低骨干网络的参数量；改进使用具有串并联关系的空洞卷积池化金字塔（Series-parallel Atrous Convolutional Pyramid, SACP）模块，使得卷积感受野进一步扩大并且彼此融合，更好地感知完整的病害范围；解码阶段融合卷积注意力模块（Concentration-Based Attention Module, CBAM），提高浅层特征的有效边界特征权重。改进损失函数加快模型收敛速度。结果 本文实地采集隧道内衬以及桥梁缆索病害部位图像构建数据集展开实验，结果表明：本文提出的网络模型在隧道内衬病害提取上准确率（Accuracy, Acc）达到94.4 %，平均交并比（mean Intersection over Union, mIoU）达到78.14 %，F1分数（F1-score）达到76.45%。在桥梁缆索病害提取上Acc达到97.15%，mIoU达到80.41%，F1-score达到77.92%。结论 相较于主流的分割网络本文模型在桥隧表面病害提取的精度上均有提升，具有更有效的提取效果和抗干扰能力，能更好满足复杂环境下病害检测工程需求。

Objectives Using computer image technology to identify defects in bridge cables and tunnel linings is an efficient and convenient method. Surface defect detection on bridge cables and tunnel linings faces multiple challenges: high costs of manual defect localization in complex structures; insufficient tunnel lighting, complex defect features, and imbalanced target-background ratios often reveal significant shortcomings in traditional image technology for defect area detection, such as poor anti-interference ability, low recognition rates, and subpar segmentation effects. Therefore, this paper proposes a deep network model based on the fusion attention and enhanced receptive field. Method The model employs a backbone network integrating a fused Transformer to extract target feature information, establishing more closely related global texture feature representation, which solve the problem of lack of local features caused by insufficient lighting. Additionally, Spatial Reduction Attention (SRA) is introduced to reduce the parameter count of the backbone. The Series-parallel Atrous Convolutional Pyramid (SACP) module has been introduced to further expand the convolutional receptive field and integrate it, thereby enhancing the perception of the complete defect scope with multi-scale characteristics. The decoder fused Concentration-Based Attention Module (CBAM) to strengthens the effective boundary feature weight of shallow features and suppress some occluded features, which sharpen segmentation results. The model employs a composite loss function combining cross-entropy loss and Dice loss to balance the contributions of positive and negative samples. Datasets for experimentation were constructed through on-site collection of tunnel lining defect images and bridge cable climbing robot-acquired cable defect images. Result The results demonstrate that: (1). The proposed Multi-Transformer backbone reduces computational complexity while maintaining feature extraction capabilities. (2). The optimized Series-parallel Atrous Convolutional Pyramid (SACP) module improves segmentation accuracy by 2%, and the use of Depthwise Separable Convolution reduces complexity effectively. (3).Due to the use of improved composite loss function, the convergence speed of the network has been accelerated. (4).The proposed network model achieves an accuracy (Acc) of 94.4%, the mean intersection over union (mIoU) of 78.14% and F1-score of 76.45% in tunnel lining dataset. Our model"s generalization ability was also demonstrated in the bridge cable dataset. Conclusion Compared to mainstream segmentation networks, the model we proposed improves the accuracy of bridge cable and tunnel lining defect detection, demonstrating enhanced extraction efficacy and anti-interference capability. Therefore, this model better meets the engineering requirements for defect detection in complex environments.