| 引用本文: | 聂壮壮,汪 军,黄翔翔.基于改进 YOLOv5 的护目镜佩戴检测算法(J/M/D/N,J:杂志，M：书，D：论文，N：报纸).期刊名称,2025，42（1）：48-56 |
| CHEN X. Adap tive slidingmode contr ol for discrete2ti me multi2inputmulti2 out put systems[ J ]. Aut omatica, 2006, 42(6): 4272-435 |
|
| 摘要: |
| 目的 解决目前危化实验室、工厂等危险环境下护目镜佩戴情况检测存在的人工检查效率低下、无法有效保
障人员眼部安全等问题。 方法 首先构建出护目镜佩戴检测数据集,其中包含 4 个真实场景图片与部分网络爬取数
据集,并通过数据增强等手段将原始的 3 383 张扩充至 5 462 张图片,构成最终数据集,使各个样本数量达到均衡,
有效预防了因样本不均衡导致的模型精度低的问题;接着提出改进型 YOLOv5 目标检测算法来实现对护目镜佩戴
情况的自动检测,算法在 YOLOv5 中添加 SPD 小目标检测模块,该模块完全消除了传统卷积模块中导致信息丢失
的步长卷积和池化操作,使网络保留更多信息,引入坐标注意力机制解决了因添加 SPD 带来的相邻位置关系无法
有效提取的问题;同时,将原本的损失函数替换为 SIoU 损失函数,有效解决了真实框与目标框相互包含情况下的
IoU 计算问题,减少了计算自由度,降低了模型计算量,提升了模型准确率。 结果 在护目镜配戴检测数据集上的实
验结果表明:改进型的 YOLOv5 模型在护目镜佩戴检测数据集上的平均精度为 72. 7%,相较于原始 YOLOv5 模型
平均精度提高了 5. 6%。 结论 该模型实现了对复杂环境下护目镜佩戴情况的基本检测。 |
| 关键词: 护目镜佩戴检测 数据增强 目标检测 SPD 坐标注意力机制 |
| DOI： |
| 分类号: |
| 基金项目: |
|
| Goggle Wearing Detection Algorithm Based on Improved YOLOv5 |
|
NIE Zhuangzhuang WANG Jun HUANG Xiangxiang
|
|
School of Computer and Information Anhui Polytechnic University Anhui Wuhu 241000 China
|
| Abstract: |
| Objective In response to the problems of low efficiency in manual inspection and the inability to effectively
ensure the eye safety of personnel in hazardous environments such as chemical laboratories and factories this study aims to
address these issues in the detection of goggles wearing. Methods Firstly a dataset for goggles wearing detection was
constructed including four real-scene images and a portion of data obtained by web crawling. By means of data
augmentation the original dataset of 3383 images was expanded to 5462 images to form the final dataset ensuring the
balance of sample quantities and effectively preventing the problem of low model accuracy caused by sample imbalance.
Then an improved YOLOv5 object detection algorithm was proposed to automatically detect the wearing status of goggles.
In the YOLOv5 algorithm an SPD small target detection module was added to completely eliminate the stride convolution
and pooling operations that lead to information loss in traditional convolution modules allowing the network to retain more
information. Subsequently a coordinate attention mechanism was introduced to address the problem of ineffective
extraction of neighboring position relationships caused by the addition of SPD. Moreover the original loss function was
replaced with the SIoU loss function to effectively solve the IoU calculation problem when the real box and the target box
contain each other reducing the degrees of freedom in calculations decreasing the model?? s computational complexity, and improving model accuracy. Results Experimental results on the goggles wearing detection dataset show that the
improved YOLOv5 model has an average precision of 72. 7% on the goggle wearing detection dataset which is 5. 6%
higher than the average precision of the original YOLOv5 model on the same dataset. Conclusion This model realizes the
basic detection of goggles wearing status in complex environments. |
| Key words: goggles wearing detection data augmentation object detection SPD coordinate attention mechanism |
