基于YOLOv8s的多尺度瓦楞纸板缺陷检测算法

Multi-scale Corrugated Cardboards Defect Detection Algorithm Based on YOLOv8s

关键词:  YOLOv8s  瓦楞纸板  缺陷检测  CCFM  DySample上采样算子

Key Words:YOLOv8s  corrugated cardboards  defect detection  CCFM  DySample up-sampling operator

摘要:针对瓦楞纸板表面缺陷检测中特征不明显、跨尺度缺陷（如水渍、划痕、压裂、机械损伤等）易漏检等问题，提出了一种基于YOLOv8s的跨尺度缺陷检测算法。通过跨尺度特征融合模块（CCFM）改进网络颈部结构，结合DySample上采样算子增强对划痕的跨尺度缺陷和低分辨率水渍的敏感度，有效整合了细节特征和上下文信息，减少了计算量；改进WIOUv3损失函数优化压裂和机械破坏的锚框质量；引入SPDConv模块保留特征不明显缺陷的细粒度特征，进一步提升特征学习效率。结果表明，改进后的模型在精确率、召回率和平均精度上分别提高了1.3、1.6和1.5个百分点，参数量减少了42.3%，帧率提高了25.8%，显著减少了假阳性和假阴性情况，提高了运算速度。该算法能够实现高效、准确的瓦楞纸板表面缺陷检测，为瓦楞纸板生产质量监控提供了可靠的技术支持，具有广泛的工业应用价值。

Abstract:A multi-scale defect detection algorithm based on YOLOv8s was proposed to address the problems of inconspicuous features and easy missed detection of cross-scale defects such as water stain， scratch， crush， mechanical sabotage， and so on， in the detection of surface defects of corrugated cardboard. The neck structure of network was improved by cross-scale feature fusion module （CCFM）， combined with DySample up-sampling operator to enhance the sensitivity to cross-scale defects of scratches and low-resolution water stains， which effectively integrated the detailed features and contextual information and reduced the computational amount. The WIOUv3 loss function was improved to optimize the quality of anchor frames for fracking and mechanical breaking. The SPDConv module was introduced to retain the fine-grained features of the defects with inconspicuous features， which further improved the feature learning efficiency. The results showed that the improved model improved 1.3， 1.6 and 1.5 percentage points in precision， recall， and mean average precision， respectively， reduced the amount of parameters by 42.3%， improved the frame rate by 25.8%， significantly reduced the false-positive and false-negative cases， and improved the computing speed. The algorithm was able to realize efficient and accurate corrugated cardboard surface defect detection， which provided reliable technical support for corrugated cardboard production quality monitoring and had wide industrial application value.