改进YOLOv11的轻量型建筑表面裂缝检测算法

Improved Lightweight Algorithm for Building Surface Crack Detection Based on YOLOv11

摘要:

建筑表面裂缝检测是保障建筑工程质量的关键环节；为应对现有算法在小尺度目标检测准确性不足以及内存占用过高限制了其在边缘设备上应用的问题，提出了一种基于YOLOv11的轻量型裂缝智能检测算法YOLOv11-LCS；通过广泛收集建筑表面裂缝图像数据构建BSCRACK数据集，在算法设计上结合轻量化策略与多尺度特征融合，引入LDConv模块和添加额外卷积操作的CCFM模块，优化了主干特征提取网络和颈部特征融合网络，显著提升了检测效率和精度；此外，采用无参注意力机制SimAM构建C3k2-SimAM模块，在不增加额外计算成本的前提下，进一步增强对小尺度裂缝的检测能力；实验结果表明，YOLOv11-LCS的参数量和计算量分别减少了42.2%和30.2%，参数量仅为1.49 M，同时 mAP@0.5提升至97.9%，有效满足了建筑表面裂缝检测的精度需求，展现了其在边缘设备上的应用潜力。

Abstract:

Surface crack detection in buildings is a critical aspect of ensuring the quality of construction projects. To address the issues of insufficient accuracy in detecting small-scale targets and high-memory usage, which limit the application of existing algorithms on edge devices, a lightweight intelligent crack detection algorithm based on YOLOv11, called YOLOv11-LCS, was proposed. By extensively collecting images of building surface cracks, the BSCRACK dataset was created, the lightweight strategy and multi-scale feature fusion was combined in the algorithm design. It introduces the LDConv module and an additional convolutional operation in the CCFM module, which optimizes both the backbone feature extraction network and the neck feature fusion network, and significantly enhances detection efficiency and accuracy. Furthermore, it employs the parameter-free attention mechanism SimAM to construct the C3k2-SimAM module, which enhances the detection capability for small-scale cracks without increasing additional computational costs. Experimental results show that, YOLOv11-LCS reduces the number of parameters and computational load by 42.2% and 30.2%, respectively, with the number of parameters being only 1.49 M, while the mAP@0.5 is improved to 97.9%. This effectively meets the precision requirements for building surface crack detection, and demonstrates its potential for application on edge devices.