【摘要】 针对字符重叠、错位及偏移等圆棒类产品表面字符缺陷检测难度大、检测精度低的问题，提出一种基于时域频域特征融合的轻量化圆棒类产品字符缺陷检测网络。将Scharr算子、时域特征和频域特征融合，获得清晰度高的字符边缘和细节信息。引入共享卷积设计轻量化检测头，在保证精度的前提下提高检测速度。实验结果表明，改进算法的平均精度均值达96.37%、召回率达91.13%、精确率达95.47%，较原始网络分别提高6.97%,4.62%,6.3%；检测速度达87.9 fps，较原始网络提高6.6 fps。更多还原

【Abstract】 To overcome the challenges of low detection accuracy in identifying surface character defects on cylindrical products such as cigarettes, a lightweight defect-detection network based on time-and frequency-domain feature fusion is proposed. High-definition character edges and detailed information are obtained by integrating the Scharr operator with time-and frequency-domain features, thereby effectively capturing both local image variations and structural periodicity characteristics. A shared convolution mechanism is incorporated to develop a lightweight detection head, thus improving inference speed while maintaining detection accuracy. The optimized network achieves a mean average precision of 96. 37%, a recall rate of 91. 13%, and a precision of 95. 47%, corresponding to enhancements of 6. 97%, 4. 62%, and 6. 3% over those of the baseline model, respectively. Furthermore, the detection speed increases to 87. 9 fps, which is 6. 6 fps higher than that of the original implementation. This method provides an effective solution for industrial defect detection in cylindrical products and demonstrates strong applicability to cigarette quality-control systems.更多还原