PMC-YOLO:一种烧结复杂环境下的篦条缺陷检测方法  

PMC-YOLO Method for Detecting Grate Bar Defects in Complex Sintering Environment

作　　者：申旭桓 马宗方[1] 袁立新 陈陆义 崔衡 董武刚 殷浩 SHEN Xuhuan;MA Zongfang;YUAN Lixin;CHEN Luyi;CUI Heng;DONG Wugang;YIN Hao(College of Information and Control engineering,Xi′an University of Architecture and Technology,Xi′an 710055,China;MCC Changtian(Changsha)Intelligent Technology Co.,Ltd.,Changsha 410205,China)

机构地区：[1]西安建筑科技大学信息与控制工程学院,西安710055 [2]中冶长天(长沙)智能科技有限公司,长沙410205

出　　处：《机械科学与技术》2025年第1期84-91,共8页Mechanical Science and Technology for Aerospace Engineering

摘　　要：为应对传统篦条缺陷检测方法效率低、易干扰等问题,该文提出一种PMC-YOLO篦条缺陷检测算法,以翅冀的篦条故障作为实验检测对象。首先,通过渐进特征融合网络AFPN(Asymptotic feature pyramid network)结合路径聚合网络PANet(Path aggregation network)的思想,提出一种P-AFPN特征融合方式,使模型适应不同层次的特征信息;其次,设计MC-SimSPPF模块,在SimSPPF(Simplified spatial pyramid pooling-fast)模块引入混合局部通道注意力机制MLCA(Mixed local channel attention),增强网络对有用特征的捕捉;之后,利用PP-LCNet(Pyramid lightweight convolutional neural network)实现主干网络的轻量级设计,确保高速度的同时维持高精度;最后,使用Focal-EIoU损失函数,准确地描述边界框之间的差异,提升目标定位精度。结果显示,改进的PMC-YOLO模型大小仅9.9 MB,平均检测精度达到93.8%,提升5.3%,检测速度达87 frame/s,适合在嵌入式设备上部署,且满足烧结环境下实时篦条缺陷检测需求。In order to solve the problems of low efficiency and easy interference of traditional grate bar defect detection methods,this paper proposes a PMC-YOLO grate bar defect detection algorithm,which takes the grate fault as the experimental detection object.Firstly,through the idea of asymptotic feature pyramid network(AFPN)combined with path aggregation network(PANet),a P-AFPN feature fusion method is proposed to adapt the model to different levels of feature information.Secondly,the MC-SimSPPF module is designed,and the mixed local channel attention(MLCA)mechanism is introduced into the simplified spatial pyramid pooling-fast(SimSPPF)module to enhance the capture of useful features by the network.After that,the pyramid lightweight convolutional neural network(PP-LCNet)is used to realize the lightweight design of the backbone network,ensuring high speed and maintaining high accuracy.Finally,the Focal-EIoU loss function is used to accurately describe the differences between bounding boxes and improve the accuracy of target positioning.The results show that the size of the improved PMC-YOLO model is only 9.9 MB,the average detection accuracy reaches 93.8%,the detection speed is increased by 5.3%,and the detection speed reaches 87 frame/s.The PMC-YOLO method is suitable for deployment on embedded devices and meets the requirements of real-time grate bar defect detection in sintering environment.

关 键 词：缺陷检测 篦条 PMC-YOLO 烧结复杂环境 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]