基于改进YOLOv5的织物瑕疵检测方法

Research on fabric defect detection method based on improved YOLOv5

针对织物瑕疵中部分瑕疵目标小、长宽比极端等问题,提出一种基于改进YOLOv5的织物瑕疵检测方法。该方法在YOLOv5模型基础上引入自注意力机制CoTNet网络,并将颈部网络中的PAFPN网络优化为BiFPN网络,同时将目标损失函数改进为CIoU损失函数,加强模型对邻近键以及上下文之间特征信息的收集,在增强模型对小目标和尺寸变化大类型瑕疵检测能力的同时可获得更准确的边界框回归,加快收敛速度。实验证明,本文改进的模型在织物瑕疵检测数据集上的检测效果和YOLOv5模型相比平均精度均值提升了6.8%,准确率提升了6.7%,模型验证有效。

A fabric defect detection method based on an improved YOLOv5 was proposed to address issues such as small target size and extreme aspect ratio in fabric defects.In this method,the self-attention mechanism CotNet network was introduced on the basis of the original network model.The PAFPN network in the neck network was optimized to a BiFPN network.Additionally,the target loss function was improved to a CIoU loss function to enhance the model′s ability to collect contextual information between adjacent keys and to detect small targets and large defects with significant size changes more accurately.The proposed model achieves more accurate boundary box regression while enhancing the detection ability of small targets and large defects with significant size changes.It is experimentally demonstrated that the improved model in this paper improves by 6.8%and accuracy by 6.7%on the fabric defect detection dataset compared with YOLOv5 model,which validates the model.