基于改进YOLOv5s的建筑护栏目标检测

Building guardrail object detection based on improved YOLOv5s algorithm

目前,建筑场所上仍存在因建筑护栏缺失或建筑护栏安全性降低而导致的建筑工人高空坠亡事件。针对该问题,提出一种基于改进YOLOv5s的建筑护栏检测算法。首先,针对建筑护栏普遍存在的安全隐患,收集影响护栏安全性较大的情况的图像,例如:建筑护栏栏板的存在图像、建筑护栏栏板的缺失图像、护栏网图像、护栏栏板衔接错位图像和护栏栏板衔接正确图像等,并且制作成训练数据集。为提升YOLOv5s在复杂环境下多目标检测任务和区分任务结果的准确率,将新型的Biformer注意力机制与SE注意力机制相结合,嵌入到原模型的特征提取网络中,并利用CBAMC3取代原特征提取网络的C3模块。最后,使用CLAHE算法较大程度地解决部分图像亮度偏暗,影响检测精度的问题。实验结果表明,所提检测算法的mAP50值和召回率分别达到了79.6%和83%,相比于原YOLOv5s算法分别提高了3.7%和6.8%。

There are many incidents of construction workers falling from heights on construction sites due to the reduced safety of building guardrails.On this basis,a building guardrail detection algorithm based on improved YOLOv5s is proposed.In allusion to the safety hazards that are prevalent in building guardrails,images that have an impact on guardrail safety,such as images of the presence of building guardrails,images of missing building guardrails,images of guardrail nets,images of misaligned guardrail panel connections,and images of correct guardrail panel connections are collected,and these images are created for a training dataset.In order to improve the accuracy of YOLOv5s detection results when performing multi-target detection and discrimination tasks in complex environments,a novel Biformer attention mechanism combined with SE attention mechanism is embedded into the feature extraction network of the original model,and CBAMC3 is used to replace the C3 module of the original feature extraction network.The use of CLAHE algorithm can largely solve the problem of dim brightness in some images,which affects detection accuracy.The experimental results show that the mAP50 value and recall of the proposed detection algorithm can reach 79.6%and 83%,respectively,which are 3.7%and 6.8%higher than those of the original YOLOv5s algorithm.