基于改进YOLOv5s的并条棉网杂质检测

Detection of impurity in drawing cotton web based on improved YOLOv5s

针对目前并条环节人工检测棉网杂质速度缓慢、检测精度低和主观随机性大等问题,提出一种基于改进YOLOv5s算法的并条棉网杂质检测方法。首先设计了一套离线图像采集系统连续采集并条棉网图像,并建立包含杂质的并条棉网图像数据集;然后在C3模块中引入RFE感受野增强模块,通过更广阔的感受野来增强微小杂质的特征。同时添加SE注意力机制,学习多尺度的通道依赖关系,强化对微小杂质的特征提取能力,提升算法检测效果。结果表明:与YOLOv5s算法相比,改进算法的精确率、召回率和mAP@0.5分别提升了3.9个百分点、2.3个百分点和3.8个百分点。该研究为自动识别并条棉网杂质提供了有效方案。

Aimed at problems of sluggish speed,low detection accuracy,and subjective unpredictability in the current drawing process with manual detection on cotton web impurity,a method to detect impurities in the drawing web based on the modified YOLOv5s algorithm was put forward.Firstly,an offline image acquisition system was designed to continuously collect images of drawing cotton web and establish an image dataset of drawing cotton web containing impurities.Then,the RFE receptive field enhancement module was introduced into the C3 module,which enhanced the features of tiny impurities via a broader receptive field.Simultaneously,the SE attention mechanism was included to learn the multi-scale channel dependence,strengthen the feature extraction capacity of microscopic contaminants,and improve the algorithm′s detection effect.Results showed that,compared with YOLOv5s algorithm,the improved algorithm was improved by 3.9 percentage points in accuracy,2.3 percentage points in recall rate,and 3.8 percentage points in mAP@0.5 respectively.An effective scheme was provided in the study for automatically identifying contaminants in fishing nets.