【摘要】 针对矿山提升系统钢丝绳表面断丝、磨损等缺陷检测中存在的识别困难、模型复杂度高等技术问题，提出一种基于实时检测Transformer(RT-DETR)改进的钢丝绳缺陷检测模型。首先，设计了部分多尺度特征聚合（PMSA）结构，实现跨层级特征融合与空间细节保持，以增强多尺度特征提取能力，有效提升微小损伤的识别能力；然后，设计层次化选择路径特征聚合（SPFA），优化特征选择与多尺度融合策略，从而提升复杂缺陷的检测能力；最后，采用多尺度渐进距离交并比（MPDIoU）损失函数提高检测稳定性。实验结果表明，改进后的算法平均精度均值（mAP）更大、稳定性更高。更多还原

【Abstract】 Aiming at the technical problems of difficult identification, high model complexity and other in the detection of steel wire rope surface defects such as broken strands and wear in mining hoisting system, a defect detection model of steel wire rope based on an improved Real-time Detection Transformer(RT-DETR) was proposed. First, a Partial Multi-scale Feature Aggregation(PMSA)structure was designed to achieve cross-layer feature fusion and spatial detail preservation, enhancing the ability to extract multi-scale features and effectively improving the recognition of small defects.Then, a hierarchical Selective Path Feature Aggregatio(SPFA) was designed to optimize feature selection and multi-scale fusion strategies, thereby improving the detection ability of complex defects. Finally, a Multi-scale Progressive Distance Intersection over Union(MPDIoU) loss function was used to improve detection stability. Experimental results show that the improved algorithm achieves higher Mean Average Precision(mAP) and better stability.更多还原