融合轻量化神经网络的矿用输送带钢芯损伤检测方法  

Steel Core Damage Detection Method for Mining Conveyor Belts with Lightweight Neural Network

作　　者：盛彬 吴利刚 张楠 SHENG Bin;WU Ligang;ZHANG Nan(School of Mechanical and Electrical Engineering,Shanxi Datong University,Datong 037003,China;Shanxi General Aviation Polytechnic,Datong 037003,China)

机构地区：[1]山西大同大学机电工程学院,山西大同037003 [2]山西通用航空职业技术学院,山西大同037003

出　　处：《控制工程》2024年第7期1254-1262,共9页Control Engineering of China

基　　金：山西省高等学校科技创新计划项目(2023L276);山西大同大学教学改革创新项目(XJG2022216);山西大同大学科研基金资助项目(2020Q13,2022K1,2022K06)。

摘　　要：为了提高矿用输送带钢芯损伤的检测准确度和实时性,对传统YOLOv5算法进行了改进。首先,引入轻量化神经网络,大幅度降低模型复杂度和运算量;其次,引入高效通道注意力(efficient channel attention,ECA)机制,提升检测准确度,并加快损失函数的收敛速度;再次,采用加权双向特征金字塔网络(bi-directional feature pyramid network,BiFPN),融合高分辨率和低分辨率的图像特征,提升模型的综合性能。实验结果表明,与YOLOv5模型相比,改进模型的参数量和浮点运算量分别减少了约64.52%和69.07%,网络层数由468层降低至295层,检测精确度和召回率分别提升了约15.83%和3.93%,检测速度达到了109.89帧/s。In order to improve the accuracy and real-time performance of steel core damage detection of mining conveyor belt,the conventional YOLOv5 algorithm is improved.Firstly,the lightweight neural network is introduced to greatly reduce the complexity and computation of the model.Secondly,the efficient channel attention(ECA)mechanism is introduced to improve the detection accuracy and accelerate the convergence of the loss function.Thirdly,the bi-directional feature pyramid network(BiFPN)is used to integrate high-resolution and low-resolution image features to improve the comprehensive performance of the model.The experimental results show that compared with the YOLOv5 model,the number of parameters and floating-point operations of the improved model are reduced by about 64.52% and 69.07% respectively,the number of network layers is reduced from 468 to 295,the detection precision and recall are improved by about 15.83% and 3.93% respectively,and the detection speed reaches 109.89 frames/s.

关 键 词：轻量化神经网络 注意力机制 跨通道特征融合 实时检测 深度学习 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]